Merge branch 'for-linus' of git://neil.brown.name/md
[sfrench/cifs-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                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
102                 .procname       = "speed_limit_min",
103                 .data           = &sysctl_speed_limit_min,
104                 .maxlen         = sizeof(int),
105                 .mode           = S_IRUGO|S_IWUSR,
106                 .proc_handler   = &proc_dointvec,
107         },
108         {
109                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
110                 .procname       = "speed_limit_max",
111                 .data           = &sysctl_speed_limit_max,
112                 .maxlen         = sizeof(int),
113                 .mode           = S_IRUGO|S_IWUSR,
114                 .proc_handler   = &proc_dointvec,
115         },
116         { .ctl_name = 0 }
117 };
118
119 static ctl_table raid_dir_table[] = {
120         {
121                 .ctl_name       = DEV_RAID,
122                 .procname       = "raid",
123                 .maxlen         = 0,
124                 .mode           = S_IRUGO|S_IXUGO,
125                 .child          = raid_table,
126         },
127         { .ctl_name = 0 }
128 };
129
130 static ctl_table raid_root_table[] = {
131         {
132                 .ctl_name       = CTL_DEV,
133                 .procname       = "dev",
134                 .maxlen         = 0,
135                 .mode           = 0555,
136                 .child          = raid_dir_table,
137         },
138         { .ctl_name = 0 }
139 };
140
141 static struct block_device_operations md_fops;
142
143 static int start_readonly;
144
145 /*
146  * We have a system wide 'event count' that is incremented
147  * on any 'interesting' event, and readers of /proc/mdstat
148  * can use 'poll' or 'select' to find out when the event
149  * count increases.
150  *
151  * Events are:
152  *  start array, stop array, error, add device, remove device,
153  *  start build, activate spare
154  */
155 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
156 static atomic_t md_event_count;
157 void md_new_event(mddev_t *mddev)
158 {
159         atomic_inc(&md_event_count);
160         wake_up(&md_event_waiters);
161 }
162 EXPORT_SYMBOL_GPL(md_new_event);
163
164 /* Alternate version that can be called from interrupts
165  * when calling sysfs_notify isn't needed.
166  */
167 static void md_new_event_inintr(mddev_t *mddev)
168 {
169         atomic_inc(&md_event_count);
170         wake_up(&md_event_waiters);
171 }
172
173 /*
174  * Enables to iterate over all existing md arrays
175  * all_mddevs_lock protects this list.
176  */
177 static LIST_HEAD(all_mddevs);
178 static DEFINE_SPINLOCK(all_mddevs_lock);
179
180
181 /*
182  * iterates through all used mddevs in the system.
183  * We take care to grab the all_mddevs_lock whenever navigating
184  * the list, and to always hold a refcount when unlocked.
185  * Any code which breaks out of this loop while own
186  * a reference to the current mddev and must mddev_put it.
187  */
188 #define for_each_mddev(mddev,tmp)                                       \
189                                                                         \
190         for (({ spin_lock(&all_mddevs_lock);                            \
191                 tmp = all_mddevs.next;                                  \
192                 mddev = NULL;});                                        \
193              ({ if (tmp != &all_mddevs)                                 \
194                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
195                 spin_unlock(&all_mddevs_lock);                          \
196                 if (mddev) mddev_put(mddev);                            \
197                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
198                 tmp != &all_mddevs;});                                  \
199              ({ spin_lock(&all_mddevs_lock);                            \
200                 tmp = tmp->next;})                                      \
201                 )
202
203
204 /* Rather than calling directly into the personality make_request function,
205  * IO requests come here first so that we can check if the device is
206  * being suspended pending a reconfiguration.
207  * We hold a refcount over the call to ->make_request.  By the time that
208  * call has finished, the bio has been linked into some internal structure
209  * and so is visible to ->quiesce(), so we don't need the refcount any more.
210  */
211 static int md_make_request(struct request_queue *q, struct bio *bio)
212 {
213         mddev_t *mddev = q->queuedata;
214         int rv;
215         if (mddev == NULL || mddev->pers == NULL) {
216                 bio_io_error(bio);
217                 return 0;
218         }
219         rcu_read_lock();
220         if (mddev->suspended) {
221                 DEFINE_WAIT(__wait);
222                 for (;;) {
223                         prepare_to_wait(&mddev->sb_wait, &__wait,
224                                         TASK_UNINTERRUPTIBLE);
225                         if (!mddev->suspended)
226                                 break;
227                         rcu_read_unlock();
228                         schedule();
229                         rcu_read_lock();
230                 }
231                 finish_wait(&mddev->sb_wait, &__wait);
232         }
233         atomic_inc(&mddev->active_io);
234         rcu_read_unlock();
235         rv = mddev->pers->make_request(q, bio);
236         if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
237                 wake_up(&mddev->sb_wait);
238
239         return rv;
240 }
241
242 static void mddev_suspend(mddev_t *mddev)
243 {
244         BUG_ON(mddev->suspended);
245         mddev->suspended = 1;
246         synchronize_rcu();
247         wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
248         mddev->pers->quiesce(mddev, 1);
249         md_unregister_thread(mddev->thread);
250         mddev->thread = NULL;
251         /* we now know that no code is executing in the personality module,
252          * except possibly the tail end of a ->bi_end_io function, but that
253          * is certain to complete before the module has a chance to get
254          * unloaded
255          */
256 }
257
258 static void mddev_resume(mddev_t *mddev)
259 {
260         mddev->suspended = 0;
261         wake_up(&mddev->sb_wait);
262         mddev->pers->quiesce(mddev, 0);
263 }
264
265
266 static inline mddev_t *mddev_get(mddev_t *mddev)
267 {
268         atomic_inc(&mddev->active);
269         return mddev;
270 }
271
272 static void mddev_delayed_delete(struct work_struct *ws);
273
274 static void mddev_put(mddev_t *mddev)
275 {
276         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
277                 return;
278         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
279             !mddev->hold_active) {
280                 list_del(&mddev->all_mddevs);
281                 if (mddev->gendisk) {
282                         /* we did a probe so need to clean up.
283                          * Call schedule_work inside the spinlock
284                          * so that flush_scheduled_work() after
285                          * mddev_find will succeed in waiting for the
286                          * work to be done.
287                          */
288                         INIT_WORK(&mddev->del_work, mddev_delayed_delete);
289                         schedule_work(&mddev->del_work);
290                 } else
291                         kfree(mddev);
292         }
293         spin_unlock(&all_mddevs_lock);
294 }
295
296 static mddev_t * mddev_find(dev_t unit)
297 {
298         mddev_t *mddev, *new = NULL;
299
300  retry:
301         spin_lock(&all_mddevs_lock);
302
303         if (unit) {
304                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
305                         if (mddev->unit == unit) {
306                                 mddev_get(mddev);
307                                 spin_unlock(&all_mddevs_lock);
308                                 kfree(new);
309                                 return mddev;
310                         }
311
312                 if (new) {
313                         list_add(&new->all_mddevs, &all_mddevs);
314                         spin_unlock(&all_mddevs_lock);
315                         new->hold_active = UNTIL_IOCTL;
316                         return new;
317                 }
318         } else if (new) {
319                 /* find an unused unit number */
320                 static int next_minor = 512;
321                 int start = next_minor;
322                 int is_free = 0;
323                 int dev = 0;
324                 while (!is_free) {
325                         dev = MKDEV(MD_MAJOR, next_minor);
326                         next_minor++;
327                         if (next_minor > MINORMASK)
328                                 next_minor = 0;
329                         if (next_minor == start) {
330                                 /* Oh dear, all in use. */
331                                 spin_unlock(&all_mddevs_lock);
332                                 kfree(new);
333                                 return NULL;
334                         }
335                                 
336                         is_free = 1;
337                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
338                                 if (mddev->unit == dev) {
339                                         is_free = 0;
340                                         break;
341                                 }
342                 }
343                 new->unit = dev;
344                 new->md_minor = MINOR(dev);
345                 new->hold_active = UNTIL_STOP;
346                 list_add(&new->all_mddevs, &all_mddevs);
347                 spin_unlock(&all_mddevs_lock);
348                 return new;
349         }
350         spin_unlock(&all_mddevs_lock);
351
352         new = kzalloc(sizeof(*new), GFP_KERNEL);
353         if (!new)
354                 return NULL;
355
356         new->unit = unit;
357         if (MAJOR(unit) == MD_MAJOR)
358                 new->md_minor = MINOR(unit);
359         else
360                 new->md_minor = MINOR(unit) >> MdpMinorShift;
361
362         mutex_init(&new->reconfig_mutex);
363         INIT_LIST_HEAD(&new->disks);
364         INIT_LIST_HEAD(&new->all_mddevs);
365         init_timer(&new->safemode_timer);
366         atomic_set(&new->active, 1);
367         atomic_set(&new->openers, 0);
368         atomic_set(&new->active_io, 0);
369         spin_lock_init(&new->write_lock);
370         init_waitqueue_head(&new->sb_wait);
371         init_waitqueue_head(&new->recovery_wait);
372         new->reshape_position = MaxSector;
373         new->resync_min = 0;
374         new->resync_max = MaxSector;
375         new->level = LEVEL_NONE;
376
377         goto retry;
378 }
379
380 static inline int mddev_lock(mddev_t * mddev)
381 {
382         return mutex_lock_interruptible(&mddev->reconfig_mutex);
383 }
384
385 static inline int mddev_is_locked(mddev_t *mddev)
386 {
387         return mutex_is_locked(&mddev->reconfig_mutex);
388 }
389
390 static inline int mddev_trylock(mddev_t * mddev)
391 {
392         return mutex_trylock(&mddev->reconfig_mutex);
393 }
394
395 static inline void mddev_unlock(mddev_t * mddev)
396 {
397         mutex_unlock(&mddev->reconfig_mutex);
398
399         md_wakeup_thread(mddev->thread);
400 }
401
402 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
403 {
404         mdk_rdev_t *rdev;
405
406         list_for_each_entry(rdev, &mddev->disks, same_set)
407                 if (rdev->desc_nr == nr)
408                         return rdev;
409
410         return NULL;
411 }
412
413 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
414 {
415         mdk_rdev_t *rdev;
416
417         list_for_each_entry(rdev, &mddev->disks, same_set)
418                 if (rdev->bdev->bd_dev == dev)
419                         return rdev;
420
421         return NULL;
422 }
423
424 static struct mdk_personality *find_pers(int level, char *clevel)
425 {
426         struct mdk_personality *pers;
427         list_for_each_entry(pers, &pers_list, list) {
428                 if (level != LEVEL_NONE && pers->level == level)
429                         return pers;
430                 if (strcmp(pers->name, clevel)==0)
431                         return pers;
432         }
433         return NULL;
434 }
435
436 /* return the offset of the super block in 512byte sectors */
437 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
438 {
439         sector_t num_sectors = bdev->bd_inode->i_size / 512;
440         return MD_NEW_SIZE_SECTORS(num_sectors);
441 }
442
443 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
444 {
445         sector_t num_sectors = rdev->sb_start;
446
447         if (chunk_size)
448                 num_sectors &= ~((sector_t)chunk_size/512 - 1);
449         return num_sectors;
450 }
451
452 static int alloc_disk_sb(mdk_rdev_t * rdev)
453 {
454         if (rdev->sb_page)
455                 MD_BUG();
456
457         rdev->sb_page = alloc_page(GFP_KERNEL);
458         if (!rdev->sb_page) {
459                 printk(KERN_ALERT "md: out of memory.\n");
460                 return -ENOMEM;
461         }
462
463         return 0;
464 }
465
466 static void free_disk_sb(mdk_rdev_t * rdev)
467 {
468         if (rdev->sb_page) {
469                 put_page(rdev->sb_page);
470                 rdev->sb_loaded = 0;
471                 rdev->sb_page = NULL;
472                 rdev->sb_start = 0;
473                 rdev->sectors = 0;
474         }
475 }
476
477
478 static void super_written(struct bio *bio, int error)
479 {
480         mdk_rdev_t *rdev = bio->bi_private;
481         mddev_t *mddev = rdev->mddev;
482
483         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
484                 printk("md: super_written gets error=%d, uptodate=%d\n",
485                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
486                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
487                 md_error(mddev, rdev);
488         }
489
490         if (atomic_dec_and_test(&mddev->pending_writes))
491                 wake_up(&mddev->sb_wait);
492         bio_put(bio);
493 }
494
495 static void super_written_barrier(struct bio *bio, int error)
496 {
497         struct bio *bio2 = bio->bi_private;
498         mdk_rdev_t *rdev = bio2->bi_private;
499         mddev_t *mddev = rdev->mddev;
500
501         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
502             error == -EOPNOTSUPP) {
503                 unsigned long flags;
504                 /* barriers don't appear to be supported :-( */
505                 set_bit(BarriersNotsupp, &rdev->flags);
506                 mddev->barriers_work = 0;
507                 spin_lock_irqsave(&mddev->write_lock, flags);
508                 bio2->bi_next = mddev->biolist;
509                 mddev->biolist = bio2;
510                 spin_unlock_irqrestore(&mddev->write_lock, flags);
511                 wake_up(&mddev->sb_wait);
512                 bio_put(bio);
513         } else {
514                 bio_put(bio2);
515                 bio->bi_private = rdev;
516                 super_written(bio, error);
517         }
518 }
519
520 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
521                    sector_t sector, int size, struct page *page)
522 {
523         /* write first size bytes of page to sector of rdev
524          * Increment mddev->pending_writes before returning
525          * and decrement it on completion, waking up sb_wait
526          * if zero is reached.
527          * If an error occurred, call md_error
528          *
529          * As we might need to resubmit the request if BIO_RW_BARRIER
530          * causes ENOTSUPP, we allocate a spare bio...
531          */
532         struct bio *bio = bio_alloc(GFP_NOIO, 1);
533         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
534
535         bio->bi_bdev = rdev->bdev;
536         bio->bi_sector = sector;
537         bio_add_page(bio, page, size, 0);
538         bio->bi_private = rdev;
539         bio->bi_end_io = super_written;
540         bio->bi_rw = rw;
541
542         atomic_inc(&mddev->pending_writes);
543         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
544                 struct bio *rbio;
545                 rw |= (1<<BIO_RW_BARRIER);
546                 rbio = bio_clone(bio, GFP_NOIO);
547                 rbio->bi_private = bio;
548                 rbio->bi_end_io = super_written_barrier;
549                 submit_bio(rw, rbio);
550         } else
551                 submit_bio(rw, bio);
552 }
553
554 void md_super_wait(mddev_t *mddev)
555 {
556         /* wait for all superblock writes that were scheduled to complete.
557          * if any had to be retried (due to BARRIER problems), retry them
558          */
559         DEFINE_WAIT(wq);
560         for(;;) {
561                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
562                 if (atomic_read(&mddev->pending_writes)==0)
563                         break;
564                 while (mddev->biolist) {
565                         struct bio *bio;
566                         spin_lock_irq(&mddev->write_lock);
567                         bio = mddev->biolist;
568                         mddev->biolist = bio->bi_next ;
569                         bio->bi_next = NULL;
570                         spin_unlock_irq(&mddev->write_lock);
571                         submit_bio(bio->bi_rw, bio);
572                 }
573                 schedule();
574         }
575         finish_wait(&mddev->sb_wait, &wq);
576 }
577
578 static void bi_complete(struct bio *bio, int error)
579 {
580         complete((struct completion*)bio->bi_private);
581 }
582
583 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
584                    struct page *page, int rw)
585 {
586         struct bio *bio = bio_alloc(GFP_NOIO, 1);
587         struct completion event;
588         int ret;
589
590         rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
591
592         bio->bi_bdev = bdev;
593         bio->bi_sector = sector;
594         bio_add_page(bio, page, size, 0);
595         init_completion(&event);
596         bio->bi_private = &event;
597         bio->bi_end_io = bi_complete;
598         submit_bio(rw, bio);
599         wait_for_completion(&event);
600
601         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
602         bio_put(bio);
603         return ret;
604 }
605 EXPORT_SYMBOL_GPL(sync_page_io);
606
607 static int read_disk_sb(mdk_rdev_t * rdev, int size)
608 {
609         char b[BDEVNAME_SIZE];
610         if (!rdev->sb_page) {
611                 MD_BUG();
612                 return -EINVAL;
613         }
614         if (rdev->sb_loaded)
615                 return 0;
616
617
618         if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
619                 goto fail;
620         rdev->sb_loaded = 1;
621         return 0;
622
623 fail:
624         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
625                 bdevname(rdev->bdev,b));
626         return -EINVAL;
627 }
628
629 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
630 {
631         return  sb1->set_uuid0 == sb2->set_uuid0 &&
632                 sb1->set_uuid1 == sb2->set_uuid1 &&
633                 sb1->set_uuid2 == sb2->set_uuid2 &&
634                 sb1->set_uuid3 == sb2->set_uuid3;
635 }
636
637 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
638 {
639         int ret;
640         mdp_super_t *tmp1, *tmp2;
641
642         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
643         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
644
645         if (!tmp1 || !tmp2) {
646                 ret = 0;
647                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
648                 goto abort;
649         }
650
651         *tmp1 = *sb1;
652         *tmp2 = *sb2;
653
654         /*
655          * nr_disks is not constant
656          */
657         tmp1->nr_disks = 0;
658         tmp2->nr_disks = 0;
659
660         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
661 abort:
662         kfree(tmp1);
663         kfree(tmp2);
664         return ret;
665 }
666
667
668 static u32 md_csum_fold(u32 csum)
669 {
670         csum = (csum & 0xffff) + (csum >> 16);
671         return (csum & 0xffff) + (csum >> 16);
672 }
673
674 static unsigned int calc_sb_csum(mdp_super_t * sb)
675 {
676         u64 newcsum = 0;
677         u32 *sb32 = (u32*)sb;
678         int i;
679         unsigned int disk_csum, csum;
680
681         disk_csum = sb->sb_csum;
682         sb->sb_csum = 0;
683
684         for (i = 0; i < MD_SB_BYTES/4 ; i++)
685                 newcsum += sb32[i];
686         csum = (newcsum & 0xffffffff) + (newcsum>>32);
687
688
689 #ifdef CONFIG_ALPHA
690         /* This used to use csum_partial, which was wrong for several
691          * reasons including that different results are returned on
692          * different architectures.  It isn't critical that we get exactly
693          * the same return value as before (we always csum_fold before
694          * testing, and that removes any differences).  However as we
695          * know that csum_partial always returned a 16bit value on
696          * alphas, do a fold to maximise conformity to previous behaviour.
697          */
698         sb->sb_csum = md_csum_fold(disk_csum);
699 #else
700         sb->sb_csum = disk_csum;
701 #endif
702         return csum;
703 }
704
705
706 /*
707  * Handle superblock details.
708  * We want to be able to handle multiple superblock formats
709  * so we have a common interface to them all, and an array of
710  * different handlers.
711  * We rely on user-space to write the initial superblock, and support
712  * reading and updating of superblocks.
713  * Interface methods are:
714  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
715  *      loads and validates a superblock on dev.
716  *      if refdev != NULL, compare superblocks on both devices
717  *    Return:
718  *      0 - dev has a superblock that is compatible with refdev
719  *      1 - dev has a superblock that is compatible and newer than refdev
720  *          so dev should be used as the refdev in future
721  *     -EINVAL superblock incompatible or invalid
722  *     -othererror e.g. -EIO
723  *
724  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
725  *      Verify that dev is acceptable into mddev.
726  *       The first time, mddev->raid_disks will be 0, and data from
727  *       dev should be merged in.  Subsequent calls check that dev
728  *       is new enough.  Return 0 or -EINVAL
729  *
730  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
731  *     Update the superblock for rdev with data in mddev
732  *     This does not write to disc.
733  *
734  */
735
736 struct super_type  {
737         char                *name;
738         struct module       *owner;
739         int                 (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
740                                           int minor_version);
741         int                 (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
742         void                (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
743         unsigned long long  (*rdev_size_change)(mdk_rdev_t *rdev,
744                                                 sector_t num_sectors);
745 };
746
747 /*
748  * load_super for 0.90.0 
749  */
750 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
751 {
752         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
753         mdp_super_t *sb;
754         int ret;
755
756         /*
757          * Calculate the position of the superblock (512byte sectors),
758          * it's at the end of the disk.
759          *
760          * It also happens to be a multiple of 4Kb.
761          */
762         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
763
764         ret = read_disk_sb(rdev, MD_SB_BYTES);
765         if (ret) return ret;
766
767         ret = -EINVAL;
768
769         bdevname(rdev->bdev, b);
770         sb = (mdp_super_t*)page_address(rdev->sb_page);
771
772         if (sb->md_magic != MD_SB_MAGIC) {
773                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
774                        b);
775                 goto abort;
776         }
777
778         if (sb->major_version != 0 ||
779             sb->minor_version < 90 ||
780             sb->minor_version > 91) {
781                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
782                         sb->major_version, sb->minor_version,
783                         b);
784                 goto abort;
785         }
786
787         if (sb->raid_disks <= 0)
788                 goto abort;
789
790         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
791                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
792                         b);
793                 goto abort;
794         }
795
796         rdev->preferred_minor = sb->md_minor;
797         rdev->data_offset = 0;
798         rdev->sb_size = MD_SB_BYTES;
799
800         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
801                 if (sb->level != 1 && sb->level != 4
802                     && sb->level != 5 && sb->level != 6
803                     && sb->level != 10) {
804                         /* FIXME use a better test */
805                         printk(KERN_WARNING
806                                "md: bitmaps not supported for this level.\n");
807                         goto abort;
808                 }
809         }
810
811         if (sb->level == LEVEL_MULTIPATH)
812                 rdev->desc_nr = -1;
813         else
814                 rdev->desc_nr = sb->this_disk.number;
815
816         if (!refdev) {
817                 ret = 1;
818         } else {
819                 __u64 ev1, ev2;
820                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
821                 if (!uuid_equal(refsb, sb)) {
822                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
823                                 b, bdevname(refdev->bdev,b2));
824                         goto abort;
825                 }
826                 if (!sb_equal(refsb, sb)) {
827                         printk(KERN_WARNING "md: %s has same UUID"
828                                " but different superblock to %s\n",
829                                b, bdevname(refdev->bdev, b2));
830                         goto abort;
831                 }
832                 ev1 = md_event(sb);
833                 ev2 = md_event(refsb);
834                 if (ev1 > ev2)
835                         ret = 1;
836                 else 
837                         ret = 0;
838         }
839         rdev->sectors = calc_num_sectors(rdev, sb->chunk_size);
840
841         if (rdev->sectors < sb->size * 2 && sb->level > 1)
842                 /* "this cannot possibly happen" ... */
843                 ret = -EINVAL;
844
845  abort:
846         return ret;
847 }
848
849 /*
850  * validate_super for 0.90.0
851  */
852 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
853 {
854         mdp_disk_t *desc;
855         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
856         __u64 ev1 = md_event(sb);
857
858         rdev->raid_disk = -1;
859         clear_bit(Faulty, &rdev->flags);
860         clear_bit(In_sync, &rdev->flags);
861         clear_bit(WriteMostly, &rdev->flags);
862         clear_bit(BarriersNotsupp, &rdev->flags);
863
864         if (mddev->raid_disks == 0) {
865                 mddev->major_version = 0;
866                 mddev->minor_version = sb->minor_version;
867                 mddev->patch_version = sb->patch_version;
868                 mddev->external = 0;
869                 mddev->chunk_size = sb->chunk_size;
870                 mddev->ctime = sb->ctime;
871                 mddev->utime = sb->utime;
872                 mddev->level = sb->level;
873                 mddev->clevel[0] = 0;
874                 mddev->layout = sb->layout;
875                 mddev->raid_disks = sb->raid_disks;
876                 mddev->dev_sectors = sb->size * 2;
877                 mddev->events = ev1;
878                 mddev->bitmap_offset = 0;
879                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
880
881                 if (mddev->minor_version >= 91) {
882                         mddev->reshape_position = sb->reshape_position;
883                         mddev->delta_disks = sb->delta_disks;
884                         mddev->new_level = sb->new_level;
885                         mddev->new_layout = sb->new_layout;
886                         mddev->new_chunk = sb->new_chunk;
887                 } else {
888                         mddev->reshape_position = MaxSector;
889                         mddev->delta_disks = 0;
890                         mddev->new_level = mddev->level;
891                         mddev->new_layout = mddev->layout;
892                         mddev->new_chunk = mddev->chunk_size;
893                 }
894
895                 if (sb->state & (1<<MD_SB_CLEAN))
896                         mddev->recovery_cp = MaxSector;
897                 else {
898                         if (sb->events_hi == sb->cp_events_hi && 
899                                 sb->events_lo == sb->cp_events_lo) {
900                                 mddev->recovery_cp = sb->recovery_cp;
901                         } else
902                                 mddev->recovery_cp = 0;
903                 }
904
905                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
906                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
907                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
908                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
909
910                 mddev->max_disks = MD_SB_DISKS;
911
912                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
913                     mddev->bitmap_file == NULL)
914                         mddev->bitmap_offset = mddev->default_bitmap_offset;
915
916         } else if (mddev->pers == NULL) {
917                 /* Insist on good event counter while assembling */
918                 ++ev1;
919                 if (ev1 < mddev->events) 
920                         return -EINVAL;
921         } else if (mddev->bitmap) {
922                 /* if adding to array with a bitmap, then we can accept an
923                  * older device ... but not too old.
924                  */
925                 if (ev1 < mddev->bitmap->events_cleared)
926                         return 0;
927         } else {
928                 if (ev1 < mddev->events)
929                         /* just a hot-add of a new device, leave raid_disk at -1 */
930                         return 0;
931         }
932
933         if (mddev->level != LEVEL_MULTIPATH) {
934                 desc = sb->disks + rdev->desc_nr;
935
936                 if (desc->state & (1<<MD_DISK_FAULTY))
937                         set_bit(Faulty, &rdev->flags);
938                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
939                             desc->raid_disk < mddev->raid_disks */) {
940                         set_bit(In_sync, &rdev->flags);
941                         rdev->raid_disk = desc->raid_disk;
942                 }
943                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
944                         set_bit(WriteMostly, &rdev->flags);
945         } else /* MULTIPATH are always insync */
946                 set_bit(In_sync, &rdev->flags);
947         return 0;
948 }
949
950 /*
951  * sync_super for 0.90.0
952  */
953 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
954 {
955         mdp_super_t *sb;
956         mdk_rdev_t *rdev2;
957         int next_spare = mddev->raid_disks;
958
959
960         /* make rdev->sb match mddev data..
961          *
962          * 1/ zero out disks
963          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
964          * 3/ any empty disks < next_spare become removed
965          *
966          * disks[0] gets initialised to REMOVED because
967          * we cannot be sure from other fields if it has
968          * been initialised or not.
969          */
970         int i;
971         int active=0, working=0,failed=0,spare=0,nr_disks=0;
972
973         rdev->sb_size = MD_SB_BYTES;
974
975         sb = (mdp_super_t*)page_address(rdev->sb_page);
976
977         memset(sb, 0, sizeof(*sb));
978
979         sb->md_magic = MD_SB_MAGIC;
980         sb->major_version = mddev->major_version;
981         sb->patch_version = mddev->patch_version;
982         sb->gvalid_words  = 0; /* ignored */
983         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
984         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
985         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
986         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
987
988         sb->ctime = mddev->ctime;
989         sb->level = mddev->level;
990         sb->size = mddev->dev_sectors / 2;
991         sb->raid_disks = mddev->raid_disks;
992         sb->md_minor = mddev->md_minor;
993         sb->not_persistent = 0;
994         sb->utime = mddev->utime;
995         sb->state = 0;
996         sb->events_hi = (mddev->events>>32);
997         sb->events_lo = (u32)mddev->events;
998
999         if (mddev->reshape_position == MaxSector)
1000                 sb->minor_version = 90;
1001         else {
1002                 sb->minor_version = 91;
1003                 sb->reshape_position = mddev->reshape_position;
1004                 sb->new_level = mddev->new_level;
1005                 sb->delta_disks = mddev->delta_disks;
1006                 sb->new_layout = mddev->new_layout;
1007                 sb->new_chunk = mddev->new_chunk;
1008         }
1009         mddev->minor_version = sb->minor_version;
1010         if (mddev->in_sync)
1011         {
1012                 sb->recovery_cp = mddev->recovery_cp;
1013                 sb->cp_events_hi = (mddev->events>>32);
1014                 sb->cp_events_lo = (u32)mddev->events;
1015                 if (mddev->recovery_cp == MaxSector)
1016                         sb->state = (1<< MD_SB_CLEAN);
1017         } else
1018                 sb->recovery_cp = 0;
1019
1020         sb->layout = mddev->layout;
1021         sb->chunk_size = mddev->chunk_size;
1022
1023         if (mddev->bitmap && mddev->bitmap_file == NULL)
1024                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1025
1026         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1027         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1028                 mdp_disk_t *d;
1029                 int desc_nr;
1030                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1031                     && !test_bit(Faulty, &rdev2->flags))
1032                         desc_nr = rdev2->raid_disk;
1033                 else
1034                         desc_nr = next_spare++;
1035                 rdev2->desc_nr = desc_nr;
1036                 d = &sb->disks[rdev2->desc_nr];
1037                 nr_disks++;
1038                 d->number = rdev2->desc_nr;
1039                 d->major = MAJOR(rdev2->bdev->bd_dev);
1040                 d->minor = MINOR(rdev2->bdev->bd_dev);
1041                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1042                     && !test_bit(Faulty, &rdev2->flags))
1043                         d->raid_disk = rdev2->raid_disk;
1044                 else
1045                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1046                 if (test_bit(Faulty, &rdev2->flags))
1047                         d->state = (1<<MD_DISK_FAULTY);
1048                 else if (test_bit(In_sync, &rdev2->flags)) {
1049                         d->state = (1<<MD_DISK_ACTIVE);
1050                         d->state |= (1<<MD_DISK_SYNC);
1051                         active++;
1052                         working++;
1053                 } else {
1054                         d->state = 0;
1055                         spare++;
1056                         working++;
1057                 }
1058                 if (test_bit(WriteMostly, &rdev2->flags))
1059                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1060         }
1061         /* now set the "removed" and "faulty" bits on any missing devices */
1062         for (i=0 ; i < mddev->raid_disks ; i++) {
1063                 mdp_disk_t *d = &sb->disks[i];
1064                 if (d->state == 0 && d->number == 0) {
1065                         d->number = i;
1066                         d->raid_disk = i;
1067                         d->state = (1<<MD_DISK_REMOVED);
1068                         d->state |= (1<<MD_DISK_FAULTY);
1069                         failed++;
1070                 }
1071         }
1072         sb->nr_disks = nr_disks;
1073         sb->active_disks = active;
1074         sb->working_disks = working;
1075         sb->failed_disks = failed;
1076         sb->spare_disks = spare;
1077
1078         sb->this_disk = sb->disks[rdev->desc_nr];
1079         sb->sb_csum = calc_sb_csum(sb);
1080 }
1081
1082 /*
1083  * rdev_size_change for 0.90.0
1084  */
1085 static unsigned long long
1086 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1087 {
1088         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1089                 return 0; /* component must fit device */
1090         if (rdev->mddev->bitmap_offset)
1091                 return 0; /* can't move bitmap */
1092         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1093         if (!num_sectors || num_sectors > rdev->sb_start)
1094                 num_sectors = rdev->sb_start;
1095         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1096                        rdev->sb_page);
1097         md_super_wait(rdev->mddev);
1098         return num_sectors / 2; /* kB for sysfs */
1099 }
1100
1101
1102 /*
1103  * version 1 superblock
1104  */
1105
1106 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1107 {
1108         __le32 disk_csum;
1109         u32 csum;
1110         unsigned long long newcsum;
1111         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1112         __le32 *isuper = (__le32*)sb;
1113         int i;
1114
1115         disk_csum = sb->sb_csum;
1116         sb->sb_csum = 0;
1117         newcsum = 0;
1118         for (i=0; size>=4; size -= 4 )
1119                 newcsum += le32_to_cpu(*isuper++);
1120
1121         if (size == 2)
1122                 newcsum += le16_to_cpu(*(__le16*) isuper);
1123
1124         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1125         sb->sb_csum = disk_csum;
1126         return cpu_to_le32(csum);
1127 }
1128
1129 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1130 {
1131         struct mdp_superblock_1 *sb;
1132         int ret;
1133         sector_t sb_start;
1134         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1135         int bmask;
1136
1137         /*
1138          * Calculate the position of the superblock in 512byte sectors.
1139          * It is always aligned to a 4K boundary and
1140          * depeding on minor_version, it can be:
1141          * 0: At least 8K, but less than 12K, from end of device
1142          * 1: At start of device
1143          * 2: 4K from start of device.
1144          */
1145         switch(minor_version) {
1146         case 0:
1147                 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1148                 sb_start -= 8*2;
1149                 sb_start &= ~(sector_t)(4*2-1);
1150                 break;
1151         case 1:
1152                 sb_start = 0;
1153                 break;
1154         case 2:
1155                 sb_start = 8;
1156                 break;
1157         default:
1158                 return -EINVAL;
1159         }
1160         rdev->sb_start = sb_start;
1161
1162         /* superblock is rarely larger than 1K, but it can be larger,
1163          * and it is safe to read 4k, so we do that
1164          */
1165         ret = read_disk_sb(rdev, 4096);
1166         if (ret) return ret;
1167
1168
1169         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1170
1171         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1172             sb->major_version != cpu_to_le32(1) ||
1173             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1174             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1175             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1176                 return -EINVAL;
1177
1178         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1179                 printk("md: invalid superblock checksum on %s\n",
1180                         bdevname(rdev->bdev,b));
1181                 return -EINVAL;
1182         }
1183         if (le64_to_cpu(sb->data_size) < 10) {
1184                 printk("md: data_size too small on %s\n",
1185                        bdevname(rdev->bdev,b));
1186                 return -EINVAL;
1187         }
1188         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1189                 if (sb->level != cpu_to_le32(1) &&
1190                     sb->level != cpu_to_le32(4) &&
1191                     sb->level != cpu_to_le32(5) &&
1192                     sb->level != cpu_to_le32(6) &&
1193                     sb->level != cpu_to_le32(10)) {
1194                         printk(KERN_WARNING
1195                                "md: bitmaps not supported for this level.\n");
1196                         return -EINVAL;
1197                 }
1198         }
1199
1200         rdev->preferred_minor = 0xffff;
1201         rdev->data_offset = le64_to_cpu(sb->data_offset);
1202         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1203
1204         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1205         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1206         if (rdev->sb_size & bmask)
1207                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1208
1209         if (minor_version
1210             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1211                 return -EINVAL;
1212
1213         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1214                 rdev->desc_nr = -1;
1215         else
1216                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1217
1218         if (!refdev) {
1219                 ret = 1;
1220         } else {
1221                 __u64 ev1, ev2;
1222                 struct mdp_superblock_1 *refsb = 
1223                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1224
1225                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1226                     sb->level != refsb->level ||
1227                     sb->layout != refsb->layout ||
1228                     sb->chunksize != refsb->chunksize) {
1229                         printk(KERN_WARNING "md: %s has strangely different"
1230                                 " superblock to %s\n",
1231                                 bdevname(rdev->bdev,b),
1232                                 bdevname(refdev->bdev,b2));
1233                         return -EINVAL;
1234                 }
1235                 ev1 = le64_to_cpu(sb->events);
1236                 ev2 = le64_to_cpu(refsb->events);
1237
1238                 if (ev1 > ev2)
1239                         ret = 1;
1240                 else
1241                         ret = 0;
1242         }
1243         if (minor_version)
1244                 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1245                         le64_to_cpu(sb->data_offset);
1246         else
1247                 rdev->sectors = rdev->sb_start;
1248         if (rdev->sectors < le64_to_cpu(sb->data_size))
1249                 return -EINVAL;
1250         rdev->sectors = le64_to_cpu(sb->data_size);
1251         if (le32_to_cpu(sb->chunksize))
1252                 rdev->sectors &= ~((sector_t)le32_to_cpu(sb->chunksize) - 1);
1253
1254         if (le64_to_cpu(sb->size) > rdev->sectors)
1255                 return -EINVAL;
1256         return ret;
1257 }
1258
1259 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1260 {
1261         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1262         __u64 ev1 = le64_to_cpu(sb->events);
1263
1264         rdev->raid_disk = -1;
1265         clear_bit(Faulty, &rdev->flags);
1266         clear_bit(In_sync, &rdev->flags);
1267         clear_bit(WriteMostly, &rdev->flags);
1268         clear_bit(BarriersNotsupp, &rdev->flags);
1269
1270         if (mddev->raid_disks == 0) {
1271                 mddev->major_version = 1;
1272                 mddev->patch_version = 0;
1273                 mddev->external = 0;
1274                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1275                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1276                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1277                 mddev->level = le32_to_cpu(sb->level);
1278                 mddev->clevel[0] = 0;
1279                 mddev->layout = le32_to_cpu(sb->layout);
1280                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1281                 mddev->dev_sectors = le64_to_cpu(sb->size);
1282                 mddev->events = ev1;
1283                 mddev->bitmap_offset = 0;
1284                 mddev->default_bitmap_offset = 1024 >> 9;
1285                 
1286                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1287                 memcpy(mddev->uuid, sb->set_uuid, 16);
1288
1289                 mddev->max_disks =  (4096-256)/2;
1290
1291                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1292                     mddev->bitmap_file == NULL )
1293                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1294
1295                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1296                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1297                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1298                         mddev->new_level = le32_to_cpu(sb->new_level);
1299                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1300                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1301                 } else {
1302                         mddev->reshape_position = MaxSector;
1303                         mddev->delta_disks = 0;
1304                         mddev->new_level = mddev->level;
1305                         mddev->new_layout = mddev->layout;
1306                         mddev->new_chunk = mddev->chunk_size;
1307                 }
1308
1309         } else if (mddev->pers == NULL) {
1310                 /* Insist of good event counter while assembling */
1311                 ++ev1;
1312                 if (ev1 < mddev->events)
1313                         return -EINVAL;
1314         } else if (mddev->bitmap) {
1315                 /* If adding to array with a bitmap, then we can accept an
1316                  * older device, but not too old.
1317                  */
1318                 if (ev1 < mddev->bitmap->events_cleared)
1319                         return 0;
1320         } else {
1321                 if (ev1 < mddev->events)
1322                         /* just a hot-add of a new device, leave raid_disk at -1 */
1323                         return 0;
1324         }
1325         if (mddev->level != LEVEL_MULTIPATH) {
1326                 int role;
1327                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1328                 switch(role) {
1329                 case 0xffff: /* spare */
1330                         break;
1331                 case 0xfffe: /* faulty */
1332                         set_bit(Faulty, &rdev->flags);
1333                         break;
1334                 default:
1335                         if ((le32_to_cpu(sb->feature_map) &
1336                              MD_FEATURE_RECOVERY_OFFSET))
1337                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1338                         else
1339                                 set_bit(In_sync, &rdev->flags);
1340                         rdev->raid_disk = role;
1341                         break;
1342                 }
1343                 if (sb->devflags & WriteMostly1)
1344                         set_bit(WriteMostly, &rdev->flags);
1345         } else /* MULTIPATH are always insync */
1346                 set_bit(In_sync, &rdev->flags);
1347
1348         return 0;
1349 }
1350
1351 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1352 {
1353         struct mdp_superblock_1 *sb;
1354         mdk_rdev_t *rdev2;
1355         int max_dev, i;
1356         /* make rdev->sb match mddev and rdev data. */
1357
1358         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1359
1360         sb->feature_map = 0;
1361         sb->pad0 = 0;
1362         sb->recovery_offset = cpu_to_le64(0);
1363         memset(sb->pad1, 0, sizeof(sb->pad1));
1364         memset(sb->pad2, 0, sizeof(sb->pad2));
1365         memset(sb->pad3, 0, sizeof(sb->pad3));
1366
1367         sb->utime = cpu_to_le64((__u64)mddev->utime);
1368         sb->events = cpu_to_le64(mddev->events);
1369         if (mddev->in_sync)
1370                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1371         else
1372                 sb->resync_offset = cpu_to_le64(0);
1373
1374         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1375
1376         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1377         sb->size = cpu_to_le64(mddev->dev_sectors);
1378
1379         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1380                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1381                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1382         }
1383
1384         if (rdev->raid_disk >= 0 &&
1385             !test_bit(In_sync, &rdev->flags)) {
1386                 if (mddev->curr_resync_completed > rdev->recovery_offset)
1387                         rdev->recovery_offset = mddev->curr_resync_completed;
1388                 if (rdev->recovery_offset > 0) {
1389                         sb->feature_map |=
1390                                 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1391                         sb->recovery_offset =
1392                                 cpu_to_le64(rdev->recovery_offset);
1393                 }
1394         }
1395
1396         if (mddev->reshape_position != MaxSector) {
1397                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1398                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1399                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1400                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1401                 sb->new_level = cpu_to_le32(mddev->new_level);
1402                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1403         }
1404
1405         max_dev = 0;
1406         list_for_each_entry(rdev2, &mddev->disks, same_set)
1407                 if (rdev2->desc_nr+1 > max_dev)
1408                         max_dev = rdev2->desc_nr+1;
1409
1410         if (max_dev > le32_to_cpu(sb->max_dev))
1411                 sb->max_dev = cpu_to_le32(max_dev);
1412         for (i=0; i<max_dev;i++)
1413                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1414         
1415         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1416                 i = rdev2->desc_nr;
1417                 if (test_bit(Faulty, &rdev2->flags))
1418                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1419                 else if (test_bit(In_sync, &rdev2->flags))
1420                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1421                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1422                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1423                 else
1424                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1425         }
1426
1427         sb->sb_csum = calc_sb_1_csum(sb);
1428 }
1429
1430 static unsigned long long
1431 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1432 {
1433         struct mdp_superblock_1 *sb;
1434         sector_t max_sectors;
1435         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1436                 return 0; /* component must fit device */
1437         if (rdev->sb_start < rdev->data_offset) {
1438                 /* minor versions 1 and 2; superblock before data */
1439                 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1440                 max_sectors -= rdev->data_offset;
1441                 if (!num_sectors || num_sectors > max_sectors)
1442                         num_sectors = max_sectors;
1443         } else if (rdev->mddev->bitmap_offset) {
1444                 /* minor version 0 with bitmap we can't move */
1445                 return 0;
1446         } else {
1447                 /* minor version 0; superblock after data */
1448                 sector_t sb_start;
1449                 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1450                 sb_start &= ~(sector_t)(4*2 - 1);
1451                 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1452                 if (!num_sectors || num_sectors > max_sectors)
1453                         num_sectors = max_sectors;
1454                 rdev->sb_start = sb_start;
1455         }
1456         sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1457         sb->data_size = cpu_to_le64(num_sectors);
1458         sb->super_offset = rdev->sb_start;
1459         sb->sb_csum = calc_sb_1_csum(sb);
1460         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1461                        rdev->sb_page);
1462         md_super_wait(rdev->mddev);
1463         return num_sectors / 2; /* kB for sysfs */
1464 }
1465
1466 static struct super_type super_types[] = {
1467         [0] = {
1468                 .name   = "0.90.0",
1469                 .owner  = THIS_MODULE,
1470                 .load_super         = super_90_load,
1471                 .validate_super     = super_90_validate,
1472                 .sync_super         = super_90_sync,
1473                 .rdev_size_change   = super_90_rdev_size_change,
1474         },
1475         [1] = {
1476                 .name   = "md-1",
1477                 .owner  = THIS_MODULE,
1478                 .load_super         = super_1_load,
1479                 .validate_super     = super_1_validate,
1480                 .sync_super         = super_1_sync,
1481                 .rdev_size_change   = super_1_rdev_size_change,
1482         },
1483 };
1484
1485 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1486 {
1487         mdk_rdev_t *rdev, *rdev2;
1488
1489         rcu_read_lock();
1490         rdev_for_each_rcu(rdev, mddev1)
1491                 rdev_for_each_rcu(rdev2, mddev2)
1492                         if (rdev->bdev->bd_contains ==
1493                             rdev2->bdev->bd_contains) {
1494                                 rcu_read_unlock();
1495                                 return 1;
1496                         }
1497         rcu_read_unlock();
1498         return 0;
1499 }
1500
1501 static LIST_HEAD(pending_raid_disks);
1502
1503 static void md_integrity_check(mdk_rdev_t *rdev, mddev_t *mddev)
1504 {
1505         struct mdk_personality *pers = mddev->pers;
1506         struct gendisk *disk = mddev->gendisk;
1507         struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1508         struct blk_integrity *bi_mddev = blk_get_integrity(disk);
1509
1510         /* Data integrity passthrough not supported on RAID 4, 5 and 6 */
1511         if (pers && pers->level >= 4 && pers->level <= 6)
1512                 return;
1513
1514         /* If rdev is integrity capable, register profile for mddev */
1515         if (!bi_mddev && bi_rdev) {
1516                 if (blk_integrity_register(disk, bi_rdev))
1517                         printk(KERN_ERR "%s: %s Could not register integrity!\n",
1518                                __func__, disk->disk_name);
1519                 else
1520                         printk(KERN_NOTICE "Enabling data integrity on %s\n",
1521                                disk->disk_name);
1522                 return;
1523         }
1524
1525         /* Check that mddev and rdev have matching profiles */
1526         if (blk_integrity_compare(disk, rdev->bdev->bd_disk) < 0) {
1527                 printk(KERN_ERR "%s: %s/%s integrity mismatch!\n", __func__,
1528                        disk->disk_name, rdev->bdev->bd_disk->disk_name);
1529                 printk(KERN_NOTICE "Disabling data integrity on %s\n",
1530                        disk->disk_name);
1531                 blk_integrity_unregister(disk);
1532         }
1533 }
1534
1535 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1536 {
1537         char b[BDEVNAME_SIZE];
1538         struct kobject *ko;
1539         char *s;
1540         int err;
1541
1542         if (rdev->mddev) {
1543                 MD_BUG();
1544                 return -EINVAL;
1545         }
1546
1547         /* prevent duplicates */
1548         if (find_rdev(mddev, rdev->bdev->bd_dev))
1549                 return -EEXIST;
1550
1551         /* make sure rdev->sectors exceeds mddev->dev_sectors */
1552         if (rdev->sectors && (mddev->dev_sectors == 0 ||
1553                         rdev->sectors < mddev->dev_sectors)) {
1554                 if (mddev->pers) {
1555                         /* Cannot change size, so fail
1556                          * If mddev->level <= 0, then we don't care
1557                          * about aligning sizes (e.g. linear)
1558                          */
1559                         if (mddev->level > 0)
1560                                 return -ENOSPC;
1561                 } else
1562                         mddev->dev_sectors = rdev->sectors;
1563         }
1564
1565         /* Verify rdev->desc_nr is unique.
1566          * If it is -1, assign a free number, else
1567          * check number is not in use
1568          */
1569         if (rdev->desc_nr < 0) {
1570                 int choice = 0;
1571                 if (mddev->pers) choice = mddev->raid_disks;
1572                 while (find_rdev_nr(mddev, choice))
1573                         choice++;
1574                 rdev->desc_nr = choice;
1575         } else {
1576                 if (find_rdev_nr(mddev, rdev->desc_nr))
1577                         return -EBUSY;
1578         }
1579         if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1580                 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1581                        mdname(mddev), mddev->max_disks);
1582                 return -EBUSY;
1583         }
1584         bdevname(rdev->bdev,b);
1585         while ( (s=strchr(b, '/')) != NULL)
1586                 *s = '!';
1587
1588         rdev->mddev = mddev;
1589         printk(KERN_INFO "md: bind<%s>\n", b);
1590
1591         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1592                 goto fail;
1593
1594         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1595         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1596                 kobject_del(&rdev->kobj);
1597                 goto fail;
1598         }
1599         rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1600
1601         list_add_rcu(&rdev->same_set, &mddev->disks);
1602         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1603
1604         /* May as well allow recovery to be retried once */
1605         mddev->recovery_disabled = 0;
1606
1607         md_integrity_check(rdev, mddev);
1608         return 0;
1609
1610  fail:
1611         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1612                b, mdname(mddev));
1613         return err;
1614 }
1615
1616 static void md_delayed_delete(struct work_struct *ws)
1617 {
1618         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1619         kobject_del(&rdev->kobj);
1620         kobject_put(&rdev->kobj);
1621 }
1622
1623 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1624 {
1625         char b[BDEVNAME_SIZE];
1626         if (!rdev->mddev) {
1627                 MD_BUG();
1628                 return;
1629         }
1630         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1631         list_del_rcu(&rdev->same_set);
1632         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1633         rdev->mddev = NULL;
1634         sysfs_remove_link(&rdev->kobj, "block");
1635         sysfs_put(rdev->sysfs_state);
1636         rdev->sysfs_state = NULL;
1637         /* We need to delay this, otherwise we can deadlock when
1638          * writing to 'remove' to "dev/state".  We also need
1639          * to delay it due to rcu usage.
1640          */
1641         synchronize_rcu();
1642         INIT_WORK(&rdev->del_work, md_delayed_delete);
1643         kobject_get(&rdev->kobj);
1644         schedule_work(&rdev->del_work);
1645 }
1646
1647 /*
1648  * prevent the device from being mounted, repartitioned or
1649  * otherwise reused by a RAID array (or any other kernel
1650  * subsystem), by bd_claiming the device.
1651  */
1652 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1653 {
1654         int err = 0;
1655         struct block_device *bdev;
1656         char b[BDEVNAME_SIZE];
1657
1658         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1659         if (IS_ERR(bdev)) {
1660                 printk(KERN_ERR "md: could not open %s.\n",
1661                         __bdevname(dev, b));
1662                 return PTR_ERR(bdev);
1663         }
1664         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1665         if (err) {
1666                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1667                         bdevname(bdev, b));
1668                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1669                 return err;
1670         }
1671         if (!shared)
1672                 set_bit(AllReserved, &rdev->flags);
1673         rdev->bdev = bdev;
1674         return err;
1675 }
1676
1677 static void unlock_rdev(mdk_rdev_t *rdev)
1678 {
1679         struct block_device *bdev = rdev->bdev;
1680         rdev->bdev = NULL;
1681         if (!bdev)
1682                 MD_BUG();
1683         bd_release(bdev);
1684         blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1685 }
1686
1687 void md_autodetect_dev(dev_t dev);
1688
1689 static void export_rdev(mdk_rdev_t * rdev)
1690 {
1691         char b[BDEVNAME_SIZE];
1692         printk(KERN_INFO "md: export_rdev(%s)\n",
1693                 bdevname(rdev->bdev,b));
1694         if (rdev->mddev)
1695                 MD_BUG();
1696         free_disk_sb(rdev);
1697 #ifndef MODULE
1698         if (test_bit(AutoDetected, &rdev->flags))
1699                 md_autodetect_dev(rdev->bdev->bd_dev);
1700 #endif
1701         unlock_rdev(rdev);
1702         kobject_put(&rdev->kobj);
1703 }
1704
1705 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1706 {
1707         unbind_rdev_from_array(rdev);
1708         export_rdev(rdev);
1709 }
1710
1711 static void export_array(mddev_t *mddev)
1712 {
1713         mdk_rdev_t *rdev, *tmp;
1714
1715         rdev_for_each(rdev, tmp, mddev) {
1716                 if (!rdev->mddev) {
1717                         MD_BUG();
1718                         continue;
1719                 }
1720                 kick_rdev_from_array(rdev);
1721         }
1722         if (!list_empty(&mddev->disks))
1723                 MD_BUG();
1724         mddev->raid_disks = 0;
1725         mddev->major_version = 0;
1726 }
1727
1728 static void print_desc(mdp_disk_t *desc)
1729 {
1730         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1731                 desc->major,desc->minor,desc->raid_disk,desc->state);
1732 }
1733
1734 static void print_sb_90(mdp_super_t *sb)
1735 {
1736         int i;
1737
1738         printk(KERN_INFO 
1739                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1740                 sb->major_version, sb->minor_version, sb->patch_version,
1741                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1742                 sb->ctime);
1743         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1744                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1745                 sb->md_minor, sb->layout, sb->chunk_size);
1746         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1747                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1748                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1749                 sb->failed_disks, sb->spare_disks,
1750                 sb->sb_csum, (unsigned long)sb->events_lo);
1751
1752         printk(KERN_INFO);
1753         for (i = 0; i < MD_SB_DISKS; i++) {
1754                 mdp_disk_t *desc;
1755
1756                 desc = sb->disks + i;
1757                 if (desc->number || desc->major || desc->minor ||
1758                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1759                         printk("     D %2d: ", i);
1760                         print_desc(desc);
1761                 }
1762         }
1763         printk(KERN_INFO "md:     THIS: ");
1764         print_desc(&sb->this_disk);
1765 }
1766
1767 static void print_sb_1(struct mdp_superblock_1 *sb)
1768 {
1769         __u8 *uuid;
1770
1771         uuid = sb->set_uuid;
1772         printk(KERN_INFO "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1773                         ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1774                KERN_INFO "md:    Name: \"%s\" CT:%llu\n",
1775                 le32_to_cpu(sb->major_version),
1776                 le32_to_cpu(sb->feature_map),
1777                 uuid[0], uuid[1], uuid[2], uuid[3],
1778                 uuid[4], uuid[5], uuid[6], uuid[7],
1779                 uuid[8], uuid[9], uuid[10], uuid[11],
1780                 uuid[12], uuid[13], uuid[14], uuid[15],
1781                 sb->set_name,
1782                 (unsigned long long)le64_to_cpu(sb->ctime)
1783                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1784
1785         uuid = sb->device_uuid;
1786         printk(KERN_INFO "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1787                         " RO:%llu\n"
1788                KERN_INFO "md:     Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1789                         ":%02x%02x%02x%02x%02x%02x\n"
1790                KERN_INFO "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1791                KERN_INFO "md:         (MaxDev:%u) \n",
1792                 le32_to_cpu(sb->level),
1793                 (unsigned long long)le64_to_cpu(sb->size),
1794                 le32_to_cpu(sb->raid_disks),
1795                 le32_to_cpu(sb->layout),
1796                 le32_to_cpu(sb->chunksize),
1797                 (unsigned long long)le64_to_cpu(sb->data_offset),
1798                 (unsigned long long)le64_to_cpu(sb->data_size),
1799                 (unsigned long long)le64_to_cpu(sb->super_offset),
1800                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1801                 le32_to_cpu(sb->dev_number),
1802                 uuid[0], uuid[1], uuid[2], uuid[3],
1803                 uuid[4], uuid[5], uuid[6], uuid[7],
1804                 uuid[8], uuid[9], uuid[10], uuid[11],
1805                 uuid[12], uuid[13], uuid[14], uuid[15],
1806                 sb->devflags,
1807                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1808                 (unsigned long long)le64_to_cpu(sb->events),
1809                 (unsigned long long)le64_to_cpu(sb->resync_offset),
1810                 le32_to_cpu(sb->sb_csum),
1811                 le32_to_cpu(sb->max_dev)
1812                 );
1813 }
1814
1815 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1816 {
1817         char b[BDEVNAME_SIZE];
1818         printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1819                 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1820                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1821                 rdev->desc_nr);
1822         if (rdev->sb_loaded) {
1823                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1824                 switch (major_version) {
1825                 case 0:
1826                         print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1827                         break;
1828                 case 1:
1829                         print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1830                         break;
1831                 }
1832         } else
1833                 printk(KERN_INFO "md: no rdev superblock!\n");
1834 }
1835
1836 static void md_print_devices(void)
1837 {
1838         struct list_head *tmp;
1839         mdk_rdev_t *rdev;
1840         mddev_t *mddev;
1841         char b[BDEVNAME_SIZE];
1842
1843         printk("\n");
1844         printk("md:     **********************************\n");
1845         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1846         printk("md:     **********************************\n");
1847         for_each_mddev(mddev, tmp) {
1848
1849                 if (mddev->bitmap)
1850                         bitmap_print_sb(mddev->bitmap);
1851                 else
1852                         printk("%s: ", mdname(mddev));
1853                 list_for_each_entry(rdev, &mddev->disks, same_set)
1854                         printk("<%s>", bdevname(rdev->bdev,b));
1855                 printk("\n");
1856
1857                 list_for_each_entry(rdev, &mddev->disks, same_set)
1858                         print_rdev(rdev, mddev->major_version);
1859         }
1860         printk("md:     **********************************\n");
1861         printk("\n");
1862 }
1863
1864
1865 static void sync_sbs(mddev_t * mddev, int nospares)
1866 {
1867         /* Update each superblock (in-memory image), but
1868          * if we are allowed to, skip spares which already
1869          * have the right event counter, or have one earlier
1870          * (which would mean they aren't being marked as dirty
1871          * with the rest of the array)
1872          */
1873         mdk_rdev_t *rdev;
1874
1875         list_for_each_entry(rdev, &mddev->disks, same_set) {
1876                 if (rdev->sb_events == mddev->events ||
1877                     (nospares &&
1878                      rdev->raid_disk < 0 &&
1879                      (rdev->sb_events&1)==0 &&
1880                      rdev->sb_events+1 == mddev->events)) {
1881                         /* Don't update this superblock */
1882                         rdev->sb_loaded = 2;
1883                 } else {
1884                         super_types[mddev->major_version].
1885                                 sync_super(mddev, rdev);
1886                         rdev->sb_loaded = 1;
1887                 }
1888         }
1889 }
1890
1891 static void md_update_sb(mddev_t * mddev, int force_change)
1892 {
1893         mdk_rdev_t *rdev;
1894         int sync_req;
1895         int nospares = 0;
1896
1897         if (mddev->external)
1898                 return;
1899 repeat:
1900         spin_lock_irq(&mddev->write_lock);
1901
1902         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1903         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1904                 force_change = 1;
1905         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1906                 /* just a clean<-> dirty transition, possibly leave spares alone,
1907                  * though if events isn't the right even/odd, we will have to do
1908                  * spares after all
1909                  */
1910                 nospares = 1;
1911         if (force_change)
1912                 nospares = 0;
1913         if (mddev->degraded)
1914                 /* If the array is degraded, then skipping spares is both
1915                  * dangerous and fairly pointless.
1916                  * Dangerous because a device that was removed from the array
1917                  * might have a event_count that still looks up-to-date,
1918                  * so it can be re-added without a resync.
1919                  * Pointless because if there are any spares to skip,
1920                  * then a recovery will happen and soon that array won't
1921                  * be degraded any more and the spare can go back to sleep then.
1922                  */
1923                 nospares = 0;
1924
1925         sync_req = mddev->in_sync;
1926         mddev->utime = get_seconds();
1927
1928         /* If this is just a dirty<->clean transition, and the array is clean
1929          * and 'events' is odd, we can roll back to the previous clean state */
1930         if (nospares
1931             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1932             && (mddev->events & 1)
1933             && mddev->events != 1)
1934                 mddev->events--;
1935         else {
1936                 /* otherwise we have to go forward and ... */
1937                 mddev->events ++;
1938                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1939                         /* .. if the array isn't clean, insist on an odd 'events' */
1940                         if ((mddev->events&1)==0) {
1941                                 mddev->events++;
1942                                 nospares = 0;
1943                         }
1944                 } else {
1945                         /* otherwise insist on an even 'events' (for clean states) */
1946                         if ((mddev->events&1)) {
1947                                 mddev->events++;
1948                                 nospares = 0;
1949                         }
1950                 }
1951         }
1952
1953         if (!mddev->events) {
1954                 /*
1955                  * oops, this 64-bit counter should never wrap.
1956                  * Either we are in around ~1 trillion A.C., assuming
1957                  * 1 reboot per second, or we have a bug:
1958                  */
1959                 MD_BUG();
1960                 mddev->events --;
1961         }
1962
1963         /*
1964          * do not write anything to disk if using
1965          * nonpersistent superblocks
1966          */
1967         if (!mddev->persistent) {
1968                 if (!mddev->external)
1969                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1970
1971                 spin_unlock_irq(&mddev->write_lock);
1972                 wake_up(&mddev->sb_wait);
1973                 return;
1974         }
1975         sync_sbs(mddev, nospares);
1976         spin_unlock_irq(&mddev->write_lock);
1977
1978         dprintk(KERN_INFO 
1979                 "md: updating %s RAID superblock on device (in sync %d)\n",
1980                 mdname(mddev),mddev->in_sync);
1981
1982         bitmap_update_sb(mddev->bitmap);
1983         list_for_each_entry(rdev, &mddev->disks, same_set) {
1984                 char b[BDEVNAME_SIZE];
1985                 dprintk(KERN_INFO "md: ");
1986                 if (rdev->sb_loaded != 1)
1987                         continue; /* no noise on spare devices */
1988                 if (test_bit(Faulty, &rdev->flags))
1989                         dprintk("(skipping faulty ");
1990
1991                 dprintk("%s ", bdevname(rdev->bdev,b));
1992                 if (!test_bit(Faulty, &rdev->flags)) {
1993                         md_super_write(mddev,rdev,
1994                                        rdev->sb_start, rdev->sb_size,
1995                                        rdev->sb_page);
1996                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1997                                 bdevname(rdev->bdev,b),
1998                                 (unsigned long long)rdev->sb_start);
1999                         rdev->sb_events = mddev->events;
2000
2001                 } else
2002                         dprintk(")\n");
2003                 if (mddev->level == LEVEL_MULTIPATH)
2004                         /* only need to write one superblock... */
2005                         break;
2006         }
2007         md_super_wait(mddev);
2008         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2009
2010         spin_lock_irq(&mddev->write_lock);
2011         if (mddev->in_sync != sync_req ||
2012             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2013                 /* have to write it out again */
2014                 spin_unlock_irq(&mddev->write_lock);
2015                 goto repeat;
2016         }
2017         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2018         spin_unlock_irq(&mddev->write_lock);
2019         wake_up(&mddev->sb_wait);
2020
2021 }
2022
2023 /* words written to sysfs files may, or may not, be \n terminated.
2024  * We want to accept with case. For this we use cmd_match.
2025  */
2026 static int cmd_match(const char *cmd, const char *str)
2027 {
2028         /* See if cmd, written into a sysfs file, matches
2029          * str.  They must either be the same, or cmd can
2030          * have a trailing newline
2031          */
2032         while (*cmd && *str && *cmd == *str) {
2033                 cmd++;
2034                 str++;
2035         }
2036         if (*cmd == '\n')
2037                 cmd++;
2038         if (*str || *cmd)
2039                 return 0;
2040         return 1;
2041 }
2042
2043 struct rdev_sysfs_entry {
2044         struct attribute attr;
2045         ssize_t (*show)(mdk_rdev_t *, char *);
2046         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2047 };
2048
2049 static ssize_t
2050 state_show(mdk_rdev_t *rdev, char *page)
2051 {
2052         char *sep = "";
2053         size_t len = 0;
2054
2055         if (test_bit(Faulty, &rdev->flags)) {
2056                 len+= sprintf(page+len, "%sfaulty",sep);
2057                 sep = ",";
2058         }
2059         if (test_bit(In_sync, &rdev->flags)) {
2060                 len += sprintf(page+len, "%sin_sync",sep);
2061                 sep = ",";
2062         }
2063         if (test_bit(WriteMostly, &rdev->flags)) {
2064                 len += sprintf(page+len, "%swrite_mostly",sep);
2065                 sep = ",";
2066         }
2067         if (test_bit(Blocked, &rdev->flags)) {
2068                 len += sprintf(page+len, "%sblocked", sep);
2069                 sep = ",";
2070         }
2071         if (!test_bit(Faulty, &rdev->flags) &&
2072             !test_bit(In_sync, &rdev->flags)) {
2073                 len += sprintf(page+len, "%sspare", sep);
2074                 sep = ",";
2075         }
2076         return len+sprintf(page+len, "\n");
2077 }
2078
2079 static ssize_t
2080 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2081 {
2082         /* can write
2083          *  faulty  - simulates and error
2084          *  remove  - disconnects the device
2085          *  writemostly - sets write_mostly
2086          *  -writemostly - clears write_mostly
2087          *  blocked - sets the Blocked flag
2088          *  -blocked - clears the Blocked flag
2089          */
2090         int err = -EINVAL;
2091         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2092                 md_error(rdev->mddev, rdev);
2093                 err = 0;
2094         } else if (cmd_match(buf, "remove")) {
2095                 if (rdev->raid_disk >= 0)
2096                         err = -EBUSY;
2097                 else {
2098                         mddev_t *mddev = rdev->mddev;
2099                         kick_rdev_from_array(rdev);
2100                         if (mddev->pers)
2101                                 md_update_sb(mddev, 1);
2102                         md_new_event(mddev);
2103                         err = 0;
2104                 }
2105         } else if (cmd_match(buf, "writemostly")) {
2106                 set_bit(WriteMostly, &rdev->flags);
2107                 err = 0;
2108         } else if (cmd_match(buf, "-writemostly")) {
2109                 clear_bit(WriteMostly, &rdev->flags);
2110                 err = 0;
2111         } else if (cmd_match(buf, "blocked")) {
2112                 set_bit(Blocked, &rdev->flags);
2113                 err = 0;
2114         } else if (cmd_match(buf, "-blocked")) {
2115                 clear_bit(Blocked, &rdev->flags);
2116                 wake_up(&rdev->blocked_wait);
2117                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2118                 md_wakeup_thread(rdev->mddev->thread);
2119
2120                 err = 0;
2121         }
2122         if (!err && rdev->sysfs_state)
2123                 sysfs_notify_dirent(rdev->sysfs_state);
2124         return err ? err : len;
2125 }
2126 static struct rdev_sysfs_entry rdev_state =
2127 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2128
2129 static ssize_t
2130 errors_show(mdk_rdev_t *rdev, char *page)
2131 {
2132         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2133 }
2134
2135 static ssize_t
2136 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2137 {
2138         char *e;
2139         unsigned long n = simple_strtoul(buf, &e, 10);
2140         if (*buf && (*e == 0 || *e == '\n')) {
2141                 atomic_set(&rdev->corrected_errors, n);
2142                 return len;
2143         }
2144         return -EINVAL;
2145 }
2146 static struct rdev_sysfs_entry rdev_errors =
2147 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2148
2149 static ssize_t
2150 slot_show(mdk_rdev_t *rdev, char *page)
2151 {
2152         if (rdev->raid_disk < 0)
2153                 return sprintf(page, "none\n");
2154         else
2155                 return sprintf(page, "%d\n", rdev->raid_disk);
2156 }
2157
2158 static ssize_t
2159 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2160 {
2161         char *e;
2162         int err;
2163         char nm[20];
2164         int slot = simple_strtoul(buf, &e, 10);
2165         if (strncmp(buf, "none", 4)==0)
2166                 slot = -1;
2167         else if (e==buf || (*e && *e!= '\n'))
2168                 return -EINVAL;
2169         if (rdev->mddev->pers && slot == -1) {
2170                 /* Setting 'slot' on an active array requires also
2171                  * updating the 'rd%d' link, and communicating
2172                  * with the personality with ->hot_*_disk.
2173                  * For now we only support removing
2174                  * failed/spare devices.  This normally happens automatically,
2175                  * but not when the metadata is externally managed.
2176                  */
2177                 if (rdev->raid_disk == -1)
2178                         return -EEXIST;
2179                 /* personality does all needed checks */
2180                 if (rdev->mddev->pers->hot_add_disk == NULL)
2181                         return -EINVAL;
2182                 err = rdev->mddev->pers->
2183                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2184                 if (err)
2185                         return err;
2186                 sprintf(nm, "rd%d", rdev->raid_disk);
2187                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2188                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2189                 md_wakeup_thread(rdev->mddev->thread);
2190         } else if (rdev->mddev->pers) {
2191                 mdk_rdev_t *rdev2;
2192                 /* Activating a spare .. or possibly reactivating
2193                  * if we every get bitmaps working here.
2194                  */
2195
2196                 if (rdev->raid_disk != -1)
2197                         return -EBUSY;
2198
2199                 if (rdev->mddev->pers->hot_add_disk == NULL)
2200                         return -EINVAL;
2201
2202                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2203                         if (rdev2->raid_disk == slot)
2204                                 return -EEXIST;
2205
2206                 rdev->raid_disk = slot;
2207                 if (test_bit(In_sync, &rdev->flags))
2208                         rdev->saved_raid_disk = slot;
2209                 else
2210                         rdev->saved_raid_disk = -1;
2211                 err = rdev->mddev->pers->
2212                         hot_add_disk(rdev->mddev, rdev);
2213                 if (err) {
2214                         rdev->raid_disk = -1;
2215                         return err;
2216                 } else
2217                         sysfs_notify_dirent(rdev->sysfs_state);
2218                 sprintf(nm, "rd%d", rdev->raid_disk);
2219                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2220                         printk(KERN_WARNING
2221                                "md: cannot register "
2222                                "%s for %s\n",
2223                                nm, mdname(rdev->mddev));
2224
2225                 /* don't wakeup anyone, leave that to userspace. */
2226         } else {
2227                 if (slot >= rdev->mddev->raid_disks)
2228                         return -ENOSPC;
2229                 rdev->raid_disk = slot;
2230                 /* assume it is working */
2231                 clear_bit(Faulty, &rdev->flags);
2232                 clear_bit(WriteMostly, &rdev->flags);
2233                 set_bit(In_sync, &rdev->flags);
2234                 sysfs_notify_dirent(rdev->sysfs_state);
2235         }
2236         return len;
2237 }
2238
2239
2240 static struct rdev_sysfs_entry rdev_slot =
2241 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2242
2243 static ssize_t
2244 offset_show(mdk_rdev_t *rdev, char *page)
2245 {
2246         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2247 }
2248
2249 static ssize_t
2250 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2251 {
2252         char *e;
2253         unsigned long long offset = simple_strtoull(buf, &e, 10);
2254         if (e==buf || (*e && *e != '\n'))
2255                 return -EINVAL;
2256         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2257                 return -EBUSY;
2258         if (rdev->sectors && rdev->mddev->external)
2259                 /* Must set offset before size, so overlap checks
2260                  * can be sane */
2261                 return -EBUSY;
2262         rdev->data_offset = offset;
2263         return len;
2264 }
2265
2266 static struct rdev_sysfs_entry rdev_offset =
2267 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2268
2269 static ssize_t
2270 rdev_size_show(mdk_rdev_t *rdev, char *page)
2271 {
2272         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2273 }
2274
2275 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2276 {
2277         /* check if two start/length pairs overlap */
2278         if (s1+l1 <= s2)
2279                 return 0;
2280         if (s2+l2 <= s1)
2281                 return 0;
2282         return 1;
2283 }
2284
2285 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2286 {
2287         unsigned long long blocks;
2288         sector_t new;
2289
2290         if (strict_strtoull(buf, 10, &blocks) < 0)
2291                 return -EINVAL;
2292
2293         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2294                 return -EINVAL; /* sector conversion overflow */
2295
2296         new = blocks * 2;
2297         if (new != blocks * 2)
2298                 return -EINVAL; /* unsigned long long to sector_t overflow */
2299
2300         *sectors = new;
2301         return 0;
2302 }
2303
2304 static ssize_t
2305 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2306 {
2307         mddev_t *my_mddev = rdev->mddev;
2308         sector_t oldsectors = rdev->sectors;
2309         sector_t sectors;
2310
2311         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2312                 return -EINVAL;
2313         if (my_mddev->pers && rdev->raid_disk >= 0) {
2314                 if (my_mddev->persistent) {
2315                         sectors = super_types[my_mddev->major_version].
2316                                 rdev_size_change(rdev, sectors);
2317                         if (!sectors)
2318                                 return -EBUSY;
2319                 } else if (!sectors)
2320                         sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2321                                 rdev->data_offset;
2322         }
2323         if (sectors < my_mddev->dev_sectors)
2324                 return -EINVAL; /* component must fit device */
2325
2326         rdev->sectors = sectors;
2327         if (sectors > oldsectors && my_mddev->external) {
2328                 /* need to check that all other rdevs with the same ->bdev
2329                  * do not overlap.  We need to unlock the mddev to avoid
2330                  * a deadlock.  We have already changed rdev->sectors, and if
2331                  * we have to change it back, we will have the lock again.
2332                  */
2333                 mddev_t *mddev;
2334                 int overlap = 0;
2335                 struct list_head *tmp;
2336
2337                 mddev_unlock(my_mddev);
2338                 for_each_mddev(mddev, tmp) {
2339                         mdk_rdev_t *rdev2;
2340
2341                         mddev_lock(mddev);
2342                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2343                                 if (test_bit(AllReserved, &rdev2->flags) ||
2344                                     (rdev->bdev == rdev2->bdev &&
2345                                      rdev != rdev2 &&
2346                                      overlaps(rdev->data_offset, rdev->sectors,
2347                                               rdev2->data_offset,
2348                                               rdev2->sectors))) {
2349                                         overlap = 1;
2350                                         break;
2351                                 }
2352                         mddev_unlock(mddev);
2353                         if (overlap) {
2354                                 mddev_put(mddev);
2355                                 break;
2356                         }
2357                 }
2358                 mddev_lock(my_mddev);
2359                 if (overlap) {
2360                         /* Someone else could have slipped in a size
2361                          * change here, but doing so is just silly.
2362                          * We put oldsectors back because we *know* it is
2363                          * safe, and trust userspace not to race with
2364                          * itself
2365                          */
2366                         rdev->sectors = oldsectors;
2367                         return -EBUSY;
2368                 }
2369         }
2370         return len;
2371 }
2372
2373 static struct rdev_sysfs_entry rdev_size =
2374 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2375
2376 static struct attribute *rdev_default_attrs[] = {
2377         &rdev_state.attr,
2378         &rdev_errors.attr,
2379         &rdev_slot.attr,
2380         &rdev_offset.attr,
2381         &rdev_size.attr,
2382         NULL,
2383 };
2384 static ssize_t
2385 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2386 {
2387         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2388         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2389         mddev_t *mddev = rdev->mddev;
2390         ssize_t rv;
2391
2392         if (!entry->show)
2393                 return -EIO;
2394
2395         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2396         if (!rv) {
2397                 if (rdev->mddev == NULL)
2398                         rv = -EBUSY;
2399                 else
2400                         rv = entry->show(rdev, page);
2401                 mddev_unlock(mddev);
2402         }
2403         return rv;
2404 }
2405
2406 static ssize_t
2407 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2408               const char *page, size_t length)
2409 {
2410         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2411         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2412         ssize_t rv;
2413         mddev_t *mddev = rdev->mddev;
2414
2415         if (!entry->store)
2416                 return -EIO;
2417         if (!capable(CAP_SYS_ADMIN))
2418                 return -EACCES;
2419         rv = mddev ? mddev_lock(mddev): -EBUSY;
2420         if (!rv) {
2421                 if (rdev->mddev == NULL)
2422                         rv = -EBUSY;
2423                 else
2424                         rv = entry->store(rdev, page, length);
2425                 mddev_unlock(mddev);
2426         }
2427         return rv;
2428 }
2429
2430 static void rdev_free(struct kobject *ko)
2431 {
2432         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2433         kfree(rdev);
2434 }
2435 static struct sysfs_ops rdev_sysfs_ops = {
2436         .show           = rdev_attr_show,
2437         .store          = rdev_attr_store,
2438 };
2439 static struct kobj_type rdev_ktype = {
2440         .release        = rdev_free,
2441         .sysfs_ops      = &rdev_sysfs_ops,
2442         .default_attrs  = rdev_default_attrs,
2443 };
2444
2445 /*
2446  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2447  *
2448  * mark the device faulty if:
2449  *
2450  *   - the device is nonexistent (zero size)
2451  *   - the device has no valid superblock
2452  *
2453  * a faulty rdev _never_ has rdev->sb set.
2454  */
2455 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2456 {
2457         char b[BDEVNAME_SIZE];
2458         int err;
2459         mdk_rdev_t *rdev;
2460         sector_t size;
2461
2462         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2463         if (!rdev) {
2464                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2465                 return ERR_PTR(-ENOMEM);
2466         }
2467
2468         if ((err = alloc_disk_sb(rdev)))
2469                 goto abort_free;
2470
2471         err = lock_rdev(rdev, newdev, super_format == -2);
2472         if (err)
2473                 goto abort_free;
2474
2475         kobject_init(&rdev->kobj, &rdev_ktype);
2476
2477         rdev->desc_nr = -1;
2478         rdev->saved_raid_disk = -1;
2479         rdev->raid_disk = -1;
2480         rdev->flags = 0;
2481         rdev->data_offset = 0;
2482         rdev->sb_events = 0;
2483         atomic_set(&rdev->nr_pending, 0);
2484         atomic_set(&rdev->read_errors, 0);
2485         atomic_set(&rdev->corrected_errors, 0);
2486
2487         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2488         if (!size) {
2489                 printk(KERN_WARNING 
2490                         "md: %s has zero or unknown size, marking faulty!\n",
2491                         bdevname(rdev->bdev,b));
2492                 err = -EINVAL;
2493                 goto abort_free;
2494         }
2495
2496         if (super_format >= 0) {
2497                 err = super_types[super_format].
2498                         load_super(rdev, NULL, super_minor);
2499                 if (err == -EINVAL) {
2500                         printk(KERN_WARNING
2501                                 "md: %s does not have a valid v%d.%d "
2502                                "superblock, not importing!\n",
2503                                 bdevname(rdev->bdev,b),
2504                                super_format, super_minor);
2505                         goto abort_free;
2506                 }
2507                 if (err < 0) {
2508                         printk(KERN_WARNING 
2509                                 "md: could not read %s's sb, not importing!\n",
2510                                 bdevname(rdev->bdev,b));
2511                         goto abort_free;
2512                 }
2513         }
2514
2515         INIT_LIST_HEAD(&rdev->same_set);
2516         init_waitqueue_head(&rdev->blocked_wait);
2517
2518         return rdev;
2519
2520 abort_free:
2521         if (rdev->sb_page) {
2522                 if (rdev->bdev)
2523                         unlock_rdev(rdev);
2524                 free_disk_sb(rdev);
2525         }
2526         kfree(rdev);
2527         return ERR_PTR(err);
2528 }
2529
2530 /*
2531  * Check a full RAID array for plausibility
2532  */
2533
2534
2535 static void analyze_sbs(mddev_t * mddev)
2536 {
2537         int i;
2538         mdk_rdev_t *rdev, *freshest, *tmp;
2539         char b[BDEVNAME_SIZE];
2540
2541         freshest = NULL;
2542         rdev_for_each(rdev, tmp, mddev)
2543                 switch (super_types[mddev->major_version].
2544                         load_super(rdev, freshest, mddev->minor_version)) {
2545                 case 1:
2546                         freshest = rdev;
2547                         break;
2548                 case 0:
2549                         break;
2550                 default:
2551                         printk( KERN_ERR \
2552                                 "md: fatal superblock inconsistency in %s"
2553                                 " -- removing from array\n", 
2554                                 bdevname(rdev->bdev,b));
2555                         kick_rdev_from_array(rdev);
2556                 }
2557
2558
2559         super_types[mddev->major_version].
2560                 validate_super(mddev, freshest);
2561
2562         i = 0;
2563         rdev_for_each(rdev, tmp, mddev) {
2564                 if (rdev->desc_nr >= mddev->max_disks ||
2565                     i > mddev->max_disks) {
2566                         printk(KERN_WARNING
2567                                "md: %s: %s: only %d devices permitted\n",
2568                                mdname(mddev), bdevname(rdev->bdev, b),
2569                                mddev->max_disks);
2570                         kick_rdev_from_array(rdev);
2571                         continue;
2572                 }
2573                 if (rdev != freshest)
2574                         if (super_types[mddev->major_version].
2575                             validate_super(mddev, rdev)) {
2576                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2577                                         " from array!\n",
2578                                         bdevname(rdev->bdev,b));
2579                                 kick_rdev_from_array(rdev);
2580                                 continue;
2581                         }
2582                 if (mddev->level == LEVEL_MULTIPATH) {
2583                         rdev->desc_nr = i++;
2584                         rdev->raid_disk = rdev->desc_nr;
2585                         set_bit(In_sync, &rdev->flags);
2586                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2587                         rdev->raid_disk = -1;
2588                         clear_bit(In_sync, &rdev->flags);
2589                 }
2590         }
2591
2592
2593
2594         if (mddev->recovery_cp != MaxSector &&
2595             mddev->level >= 1)
2596                 printk(KERN_ERR "md: %s: raid array is not clean"
2597                        " -- starting background reconstruction\n",
2598                        mdname(mddev));
2599
2600 }
2601
2602 static void md_safemode_timeout(unsigned long data);
2603
2604 static ssize_t
2605 safe_delay_show(mddev_t *mddev, char *page)
2606 {
2607         int msec = (mddev->safemode_delay*1000)/HZ;
2608         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2609 }
2610 static ssize_t
2611 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2612 {
2613         int scale=1;
2614         int dot=0;
2615         int i;
2616         unsigned long msec;
2617         char buf[30];
2618
2619         /* remove a period, and count digits after it */
2620         if (len >= sizeof(buf))
2621                 return -EINVAL;
2622         strlcpy(buf, cbuf, sizeof(buf));
2623         for (i=0; i<len; i++) {
2624                 if (dot) {
2625                         if (isdigit(buf[i])) {
2626                                 buf[i-1] = buf[i];
2627                                 scale *= 10;
2628                         }
2629                         buf[i] = 0;
2630                 } else if (buf[i] == '.') {
2631                         dot=1;
2632                         buf[i] = 0;
2633                 }
2634         }
2635         if (strict_strtoul(buf, 10, &msec) < 0)
2636                 return -EINVAL;
2637         msec = (msec * 1000) / scale;
2638         if (msec == 0)
2639                 mddev->safemode_delay = 0;
2640         else {
2641                 unsigned long old_delay = mddev->safemode_delay;
2642                 mddev->safemode_delay = (msec*HZ)/1000;
2643                 if (mddev->safemode_delay == 0)
2644                         mddev->safemode_delay = 1;
2645                 if (mddev->safemode_delay < old_delay)
2646                         md_safemode_timeout((unsigned long)mddev);
2647         }
2648         return len;
2649 }
2650 static struct md_sysfs_entry md_safe_delay =
2651 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2652
2653 static ssize_t
2654 level_show(mddev_t *mddev, char *page)
2655 {
2656         struct mdk_personality *p = mddev->pers;
2657         if (p)
2658                 return sprintf(page, "%s\n", p->name);
2659         else if (mddev->clevel[0])
2660                 return sprintf(page, "%s\n", mddev->clevel);
2661         else if (mddev->level != LEVEL_NONE)
2662                 return sprintf(page, "%d\n", mddev->level);
2663         else
2664                 return 0;
2665 }
2666
2667 static ssize_t
2668 level_store(mddev_t *mddev, const char *buf, size_t len)
2669 {
2670         char level[16];
2671         ssize_t rv = len;
2672         struct mdk_personality *pers;
2673         void *priv;
2674
2675         if (mddev->pers == NULL) {
2676                 if (len == 0)
2677                         return 0;
2678                 if (len >= sizeof(mddev->clevel))
2679                         return -ENOSPC;
2680                 strncpy(mddev->clevel, buf, len);
2681                 if (mddev->clevel[len-1] == '\n')
2682                         len--;
2683                 mddev->clevel[len] = 0;
2684                 mddev->level = LEVEL_NONE;
2685                 return rv;
2686         }
2687
2688         /* request to change the personality.  Need to ensure:
2689          *  - array is not engaged in resync/recovery/reshape
2690          *  - old personality can be suspended
2691          *  - new personality will access other array.
2692          */
2693
2694         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2695                 return -EBUSY;
2696
2697         if (!mddev->pers->quiesce) {
2698                 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2699                        mdname(mddev), mddev->pers->name);
2700                 return -EINVAL;
2701         }
2702
2703         /* Now find the new personality */
2704         if (len == 0 || len >= sizeof(level))
2705                 return -EINVAL;
2706         strncpy(level, buf, len);
2707         if (level[len-1] == '\n')
2708                 len--;
2709         level[len] = 0;
2710
2711         request_module("md-%s", level);
2712         spin_lock(&pers_lock);
2713         pers = find_pers(LEVEL_NONE, level);
2714         if (!pers || !try_module_get(pers->owner)) {
2715                 spin_unlock(&pers_lock);
2716                 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2717                 return -EINVAL;
2718         }
2719         spin_unlock(&pers_lock);
2720
2721         if (pers == mddev->pers) {
2722                 /* Nothing to do! */
2723                 module_put(pers->owner);
2724                 return rv;
2725         }
2726         if (!pers->takeover) {
2727                 module_put(pers->owner);
2728                 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2729                        mdname(mddev), level);
2730                 return -EINVAL;
2731         }
2732
2733         /* ->takeover must set new_* and/or delta_disks
2734          * if it succeeds, and may set them when it fails.
2735          */
2736         priv = pers->takeover(mddev);
2737         if (IS_ERR(priv)) {
2738                 mddev->new_level = mddev->level;
2739                 mddev->new_layout = mddev->layout;
2740                 mddev->new_chunk = mddev->chunk_size;
2741                 mddev->raid_disks -= mddev->delta_disks;
2742                 mddev->delta_disks = 0;
2743                 module_put(pers->owner);
2744                 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2745                        mdname(mddev), level);
2746                 return PTR_ERR(priv);
2747         }
2748
2749         /* Looks like we have a winner */
2750         mddev_suspend(mddev);
2751         mddev->pers->stop(mddev);
2752         module_put(mddev->pers->owner);
2753         mddev->pers = pers;
2754         mddev->private = priv;
2755         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2756         mddev->level = mddev->new_level;
2757         mddev->layout = mddev->new_layout;
2758         mddev->chunk_size = mddev->new_chunk;
2759         mddev->delta_disks = 0;
2760         pers->run(mddev);
2761         mddev_resume(mddev);
2762         set_bit(MD_CHANGE_DEVS, &mddev->flags);
2763         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2764         md_wakeup_thread(mddev->thread);
2765         return rv;
2766 }
2767
2768 static struct md_sysfs_entry md_level =
2769 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2770
2771
2772 static ssize_t
2773 layout_show(mddev_t *mddev, char *page)
2774 {
2775         /* just a number, not meaningful for all levels */
2776         if (mddev->reshape_position != MaxSector &&
2777             mddev->layout != mddev->new_layout)
2778                 return sprintf(page, "%d (%d)\n",
2779                                mddev->new_layout, mddev->layout);
2780         return sprintf(page, "%d\n", mddev->layout);
2781 }
2782
2783 static ssize_t
2784 layout_store(mddev_t *mddev, const char *buf, size_t len)
2785 {
2786         char *e;
2787         unsigned long n = simple_strtoul(buf, &e, 10);
2788
2789         if (!*buf || (*e && *e != '\n'))
2790                 return -EINVAL;
2791
2792         if (mddev->pers) {
2793                 int err;
2794                 if (mddev->pers->reconfig == NULL)
2795                         return -EBUSY;
2796                 err = mddev->pers->reconfig(mddev, n, -1);
2797                 if (err)
2798                         return err;
2799         } else {
2800                 mddev->new_layout = n;
2801                 if (mddev->reshape_position == MaxSector)
2802                         mddev->layout = n;
2803         }
2804         return len;
2805 }
2806 static struct md_sysfs_entry md_layout =
2807 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2808
2809
2810 static ssize_t
2811 raid_disks_show(mddev_t *mddev, char *page)
2812 {
2813         if (mddev->raid_disks == 0)
2814                 return 0;
2815         if (mddev->reshape_position != MaxSector &&
2816             mddev->delta_disks != 0)
2817                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2818                                mddev->raid_disks - mddev->delta_disks);
2819         return sprintf(page, "%d\n", mddev->raid_disks);
2820 }
2821
2822 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2823
2824 static ssize_t
2825 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2826 {
2827         char *e;
2828         int rv = 0;
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                 rv = update_raid_disks(mddev, n);
2836         else if (mddev->reshape_position != MaxSector) {
2837                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2838                 mddev->delta_disks = n - olddisks;
2839                 mddev->raid_disks = n;
2840         } else
2841                 mddev->raid_disks = n;
2842         return rv ? rv : len;
2843 }
2844 static struct md_sysfs_entry md_raid_disks =
2845 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2846
2847 static ssize_t
2848 chunk_size_show(mddev_t *mddev, char *page)
2849 {
2850         if (mddev->reshape_position != MaxSector &&
2851             mddev->chunk_size != mddev->new_chunk)
2852                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2853                                mddev->chunk_size);
2854         return sprintf(page, "%d\n", mddev->chunk_size);
2855 }
2856
2857 static ssize_t
2858 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2859 {
2860         char *e;
2861         unsigned long n = simple_strtoul(buf, &e, 10);
2862
2863         if (!*buf || (*e && *e != '\n'))
2864                 return -EINVAL;
2865
2866         if (mddev->pers) {
2867                 int err;
2868                 if (mddev->pers->reconfig == NULL)
2869                         return -EBUSY;
2870                 err = mddev->pers->reconfig(mddev, -1, n);
2871                 if (err)
2872                         return err;
2873         } else {
2874                 mddev->new_chunk = n;
2875                 if (mddev->reshape_position == MaxSector)
2876                         mddev->chunk_size = n;
2877         }
2878         return len;
2879 }
2880 static struct md_sysfs_entry md_chunk_size =
2881 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2882
2883 static ssize_t
2884 resync_start_show(mddev_t *mddev, char *page)
2885 {
2886         if (mddev->recovery_cp == MaxSector)
2887                 return sprintf(page, "none\n");
2888         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2889 }
2890
2891 static ssize_t
2892 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2893 {
2894         char *e;
2895         unsigned long long n = simple_strtoull(buf, &e, 10);
2896
2897         if (mddev->pers)
2898                 return -EBUSY;
2899         if (!*buf || (*e && *e != '\n'))
2900                 return -EINVAL;
2901
2902         mddev->recovery_cp = n;
2903         return len;
2904 }
2905 static struct md_sysfs_entry md_resync_start =
2906 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2907
2908 /*
2909  * The array state can be:
2910  *
2911  * clear
2912  *     No devices, no size, no level
2913  *     Equivalent to STOP_ARRAY ioctl
2914  * inactive
2915  *     May have some settings, but array is not active
2916  *        all IO results in error
2917  *     When written, doesn't tear down array, but just stops it
2918  * suspended (not supported yet)
2919  *     All IO requests will block. The array can be reconfigured.
2920  *     Writing this, if accepted, will block until array is quiescent
2921  * readonly
2922  *     no resync can happen.  no superblocks get written.
2923  *     write requests fail
2924  * read-auto
2925  *     like readonly, but behaves like 'clean' on a write request.
2926  *
2927  * clean - no pending writes, but otherwise active.
2928  *     When written to inactive array, starts without resync
2929  *     If a write request arrives then
2930  *       if metadata is known, mark 'dirty' and switch to 'active'.
2931  *       if not known, block and switch to write-pending
2932  *     If written to an active array that has pending writes, then fails.
2933  * active
2934  *     fully active: IO and resync can be happening.
2935  *     When written to inactive array, starts with resync
2936  *
2937  * write-pending
2938  *     clean, but writes are blocked waiting for 'active' to be written.
2939  *
2940  * active-idle
2941  *     like active, but no writes have been seen for a while (100msec).
2942  *
2943  */
2944 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2945                    write_pending, active_idle, bad_word};
2946 static char *array_states[] = {
2947         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2948         "write-pending", "active-idle", NULL };
2949
2950 static int match_word(const char *word, char **list)
2951 {
2952         int n;
2953         for (n=0; list[n]; n++)
2954                 if (cmd_match(word, list[n]))
2955                         break;
2956         return n;
2957 }
2958
2959 static ssize_t
2960 array_state_show(mddev_t *mddev, char *page)
2961 {
2962         enum array_state st = inactive;
2963
2964         if (mddev->pers)
2965                 switch(mddev->ro) {
2966                 case 1:
2967                         st = readonly;
2968                         break;
2969                 case 2:
2970                         st = read_auto;
2971                         break;
2972                 case 0:
2973                         if (mddev->in_sync)
2974                                 st = clean;
2975                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2976                                 st = write_pending;
2977                         else if (mddev->safemode)
2978                                 st = active_idle;
2979                         else
2980                                 st = active;
2981                 }
2982         else {
2983                 if (list_empty(&mddev->disks) &&
2984                     mddev->raid_disks == 0 &&
2985                     mddev->dev_sectors == 0)
2986                         st = clear;
2987                 else
2988                         st = inactive;
2989         }
2990         return sprintf(page, "%s\n", array_states[st]);
2991 }
2992
2993 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2994 static int do_md_run(mddev_t * mddev);
2995 static int restart_array(mddev_t *mddev);
2996
2997 static ssize_t
2998 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2999 {
3000         int err = -EINVAL;
3001         enum array_state st = match_word(buf, array_states);
3002         switch(st) {
3003         case bad_word:
3004                 break;
3005         case clear:
3006                 /* stopping an active array */
3007                 if (atomic_read(&mddev->openers) > 0)
3008                         return -EBUSY;
3009                 err = do_md_stop(mddev, 0, 0);
3010                 break;
3011         case inactive:
3012                 /* stopping an active array */
3013                 if (mddev->pers) {
3014                         if (atomic_read(&mddev->openers) > 0)
3015                                 return -EBUSY;
3016                         err = do_md_stop(mddev, 2, 0);
3017                 } else
3018                         err = 0; /* already inactive */
3019                 break;
3020         case suspended:
3021                 break; /* not supported yet */
3022         case readonly:
3023                 if (mddev->pers)
3024                         err = do_md_stop(mddev, 1, 0);
3025                 else {
3026                         mddev->ro = 1;
3027                         set_disk_ro(mddev->gendisk, 1);
3028                         err = do_md_run(mddev);
3029                 }
3030                 break;
3031         case read_auto:
3032                 if (mddev->pers) {
3033                         if (mddev->ro == 0)
3034                                 err = do_md_stop(mddev, 1, 0);
3035                         else if (mddev->ro == 1)
3036                                 err = restart_array(mddev);
3037                         if (err == 0) {
3038                                 mddev->ro = 2;
3039                                 set_disk_ro(mddev->gendisk, 0);
3040                         }
3041                 } else {
3042                         mddev->ro = 2;
3043                         err = do_md_run(mddev);
3044                 }
3045                 break;
3046         case clean:
3047                 if (mddev->pers) {
3048                         restart_array(mddev);
3049                         spin_lock_irq(&mddev->write_lock);
3050                         if (atomic_read(&mddev->writes_pending) == 0) {
3051                                 if (mddev->in_sync == 0) {
3052                                         mddev->in_sync = 1;
3053                                         if (mddev->safemode == 1)
3054                                                 mddev->safemode = 0;
3055                                         if (mddev->persistent)
3056                                                 set_bit(MD_CHANGE_CLEAN,
3057                                                         &mddev->flags);
3058                                 }
3059                                 err = 0;
3060                         } else
3061                                 err = -EBUSY;
3062                         spin_unlock_irq(&mddev->write_lock);
3063                 } else {
3064                         mddev->ro = 0;
3065                         mddev->recovery_cp = MaxSector;
3066                         err = do_md_run(mddev);
3067                 }
3068                 break;
3069         case active:
3070                 if (mddev->pers) {
3071                         restart_array(mddev);
3072                         if (mddev->external)
3073                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3074                         wake_up(&mddev->sb_wait);
3075                         err = 0;
3076                 } else {
3077                         mddev->ro = 0;
3078                         set_disk_ro(mddev->gendisk, 0);
3079                         err = do_md_run(mddev);
3080                 }
3081                 break;
3082         case write_pending:
3083         case active_idle:
3084                 /* these cannot be set */
3085                 break;
3086         }
3087         if (err)
3088                 return err;
3089         else {
3090                 sysfs_notify_dirent(mddev->sysfs_state);
3091                 return len;
3092         }
3093 }
3094 static struct md_sysfs_entry md_array_state =
3095 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3096
3097 static ssize_t
3098 null_show(mddev_t *mddev, char *page)
3099 {
3100         return -EINVAL;
3101 }
3102
3103 static ssize_t
3104 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3105 {
3106         /* buf must be %d:%d\n? giving major and minor numbers */
3107         /* The new device is added to the array.
3108          * If the array has a persistent superblock, we read the
3109          * superblock to initialise info and check validity.
3110          * Otherwise, only checking done is that in bind_rdev_to_array,
3111          * which mainly checks size.
3112          */
3113         char *e;
3114         int major = simple_strtoul(buf, &e, 10);
3115         int minor;
3116         dev_t dev;
3117         mdk_rdev_t *rdev;
3118         int err;
3119
3120         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3121                 return -EINVAL;
3122         minor = simple_strtoul(e+1, &e, 10);
3123         if (*e && *e != '\n')
3124                 return -EINVAL;
3125         dev = MKDEV(major, minor);
3126         if (major != MAJOR(dev) ||
3127             minor != MINOR(dev))
3128                 return -EOVERFLOW;
3129
3130
3131         if (mddev->persistent) {
3132                 rdev = md_import_device(dev, mddev->major_version,
3133                                         mddev->minor_version);
3134                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3135                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3136                                                        mdk_rdev_t, same_set);
3137                         err = super_types[mddev->major_version]
3138                                 .load_super(rdev, rdev0, mddev->minor_version);
3139                         if (err < 0)
3140                                 goto out;
3141                 }
3142         } else if (mddev->external)
3143                 rdev = md_import_device(dev, -2, -1);
3144         else
3145                 rdev = md_import_device(dev, -1, -1);
3146
3147         if (IS_ERR(rdev))
3148                 return PTR_ERR(rdev);
3149         err = bind_rdev_to_array(rdev, mddev);
3150  out:
3151         if (err)
3152                 export_rdev(rdev);
3153         return err ? err : len;
3154 }
3155
3156 static struct md_sysfs_entry md_new_device =
3157 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3158
3159 static ssize_t
3160 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3161 {
3162         char *end;
3163         unsigned long chunk, end_chunk;
3164
3165         if (!mddev->bitmap)
3166                 goto out;
3167         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3168         while (*buf) {
3169                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3170                 if (buf == end) break;
3171                 if (*end == '-') { /* range */
3172                         buf = end + 1;
3173                         end_chunk = simple_strtoul(buf, &end, 0);
3174                         if (buf == end) break;
3175                 }
3176                 if (*end && !isspace(*end)) break;
3177                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3178                 buf = end;
3179                 while (isspace(*buf)) buf++;
3180         }
3181         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3182 out:
3183         return len;
3184 }
3185
3186 static struct md_sysfs_entry md_bitmap =
3187 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3188
3189 static ssize_t
3190 size_show(mddev_t *mddev, char *page)
3191 {
3192         return sprintf(page, "%llu\n",
3193                 (unsigned long long)mddev->dev_sectors / 2);
3194 }
3195
3196 static int update_size(mddev_t *mddev, sector_t num_sectors);
3197
3198 static ssize_t
3199 size_store(mddev_t *mddev, const char *buf, size_t len)
3200 {
3201         /* If array is inactive, we can reduce the component size, but
3202          * not increase it (except from 0).
3203          * If array is active, we can try an on-line resize
3204          */
3205         sector_t sectors;
3206         int err = strict_blocks_to_sectors(buf, &sectors);
3207
3208         if (err < 0)
3209                 return err;
3210         if (mddev->pers) {
3211                 err = update_size(mddev, sectors);
3212                 md_update_sb(mddev, 1);
3213         } else {
3214                 if (mddev->dev_sectors == 0 ||
3215                     mddev->dev_sectors > sectors)
3216                         mddev->dev_sectors = sectors;
3217                 else
3218                         err = -ENOSPC;
3219         }
3220         return err ? err : len;
3221 }
3222
3223 static struct md_sysfs_entry md_size =
3224 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3225
3226
3227 /* Metdata version.
3228  * This is one of
3229  *   'none' for arrays with no metadata (good luck...)
3230  *   'external' for arrays with externally managed metadata,
3231  * or N.M for internally known formats
3232  */
3233 static ssize_t
3234 metadata_show(mddev_t *mddev, char *page)
3235 {
3236         if (mddev->persistent)
3237                 return sprintf(page, "%d.%d\n",
3238                                mddev->major_version, mddev->minor_version);
3239         else if (mddev->external)
3240                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3241         else
3242                 return sprintf(page, "none\n");
3243 }
3244
3245 static ssize_t
3246 metadata_store(mddev_t *mddev, const char *buf, size_t len)