Merge git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable
[linux-2.6] / drivers / md / md.c
1 /*
2    md.c : Multiple Devices driver for Linux
3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5      completely rewritten, based on the MD driver code from Marc Zyngier
6
7    Changes:
8
9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13    - kmod support by: Cyrus Durgin
14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19
20      Neil Brown <neilb@cse.unsw.edu.au>.
21
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25    This program is free software; you can redistribute it and/or modify
26    it under the terms of the GNU General Public License as published by
27    the Free Software Foundation; either version 2, or (at your option)
28    any later version.
29
30    You should have received a copy of the GNU General Public License
31    (for example /usr/src/linux/COPYING); if not, write to the Free
32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/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         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2021                 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2022
2023 }
2024
2025 /* words written to sysfs files may, or may not, be \n terminated.
2026  * We want to accept with case. For this we use cmd_match.
2027  */
2028 static int cmd_match(const char *cmd, const char *str)
2029 {
2030         /* See if cmd, written into a sysfs file, matches
2031          * str.  They must either be the same, or cmd can
2032          * have a trailing newline
2033          */
2034         while (*cmd && *str && *cmd == *str) {
2035                 cmd++;
2036                 str++;
2037         }
2038         if (*cmd == '\n')
2039                 cmd++;
2040         if (*str || *cmd)
2041                 return 0;
2042         return 1;
2043 }
2044
2045 struct rdev_sysfs_entry {
2046         struct attribute attr;
2047         ssize_t (*show)(mdk_rdev_t *, char *);
2048         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2049 };
2050
2051 static ssize_t
2052 state_show(mdk_rdev_t *rdev, char *page)
2053 {
2054         char *sep = "";
2055         size_t len = 0;
2056
2057         if (test_bit(Faulty, &rdev->flags)) {
2058                 len+= sprintf(page+len, "%sfaulty",sep);
2059                 sep = ",";
2060         }
2061         if (test_bit(In_sync, &rdev->flags)) {
2062                 len += sprintf(page+len, "%sin_sync",sep);
2063                 sep = ",";
2064         }
2065         if (test_bit(WriteMostly, &rdev->flags)) {
2066                 len += sprintf(page+len, "%swrite_mostly",sep);
2067                 sep = ",";
2068         }
2069         if (test_bit(Blocked, &rdev->flags)) {
2070                 len += sprintf(page+len, "%sblocked", sep);
2071                 sep = ",";
2072         }
2073         if (!test_bit(Faulty, &rdev->flags) &&
2074             !test_bit(In_sync, &rdev->flags)) {
2075                 len += sprintf(page+len, "%sspare", sep);
2076                 sep = ",";
2077         }
2078         return len+sprintf(page+len, "\n");
2079 }
2080
2081 static ssize_t
2082 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2083 {
2084         /* can write
2085          *  faulty  - simulates and error
2086          *  remove  - disconnects the device
2087          *  writemostly - sets write_mostly
2088          *  -writemostly - clears write_mostly
2089          *  blocked - sets the Blocked flag
2090          *  -blocked - clears the Blocked flag
2091          *  insync - sets Insync providing device isn't active
2092          */
2093         int err = -EINVAL;
2094         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2095                 md_error(rdev->mddev, rdev);
2096                 err = 0;
2097         } else if (cmd_match(buf, "remove")) {
2098                 if (rdev->raid_disk >= 0)
2099                         err = -EBUSY;
2100                 else {
2101                         mddev_t *mddev = rdev->mddev;
2102                         kick_rdev_from_array(rdev);
2103                         if (mddev->pers)
2104                                 md_update_sb(mddev, 1);
2105                         md_new_event(mddev);
2106                         err = 0;
2107                 }
2108         } else if (cmd_match(buf, "writemostly")) {
2109                 set_bit(WriteMostly, &rdev->flags);
2110                 err = 0;
2111         } else if (cmd_match(buf, "-writemostly")) {
2112                 clear_bit(WriteMostly, &rdev->flags);
2113                 err = 0;
2114         } else if (cmd_match(buf, "blocked")) {
2115                 set_bit(Blocked, &rdev->flags);
2116                 err = 0;
2117         } else if (cmd_match(buf, "-blocked")) {
2118                 clear_bit(Blocked, &rdev->flags);
2119                 wake_up(&rdev->blocked_wait);
2120                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2121                 md_wakeup_thread(rdev->mddev->thread);
2122
2123                 err = 0;
2124         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2125                 set_bit(In_sync, &rdev->flags);
2126                 err = 0;
2127         }
2128         if (!err && rdev->sysfs_state)
2129                 sysfs_notify_dirent(rdev->sysfs_state);
2130         return err ? err : len;
2131 }
2132 static struct rdev_sysfs_entry rdev_state =
2133 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2134
2135 static ssize_t
2136 errors_show(mdk_rdev_t *rdev, char *page)
2137 {
2138         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2139 }
2140
2141 static ssize_t
2142 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2143 {
2144         char *e;
2145         unsigned long n = simple_strtoul(buf, &e, 10);
2146         if (*buf && (*e == 0 || *e == '\n')) {
2147                 atomic_set(&rdev->corrected_errors, n);
2148                 return len;
2149         }
2150         return -EINVAL;
2151 }
2152 static struct rdev_sysfs_entry rdev_errors =
2153 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2154
2155 static ssize_t
2156 slot_show(mdk_rdev_t *rdev, char *page)
2157 {
2158         if (rdev->raid_disk < 0)
2159                 return sprintf(page, "none\n");
2160         else
2161                 return sprintf(page, "%d\n", rdev->raid_disk);
2162 }
2163
2164 static ssize_t
2165 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2166 {
2167         char *e;
2168         int err;
2169         char nm[20];
2170         int slot = simple_strtoul(buf, &e, 10);
2171         if (strncmp(buf, "none", 4)==0)
2172                 slot = -1;
2173         else if (e==buf || (*e && *e!= '\n'))
2174                 return -EINVAL;
2175         if (rdev->mddev->pers && slot == -1) {
2176                 /* Setting 'slot' on an active array requires also
2177                  * updating the 'rd%d' link, and communicating
2178                  * with the personality with ->hot_*_disk.
2179                  * For now we only support removing
2180                  * failed/spare devices.  This normally happens automatically,
2181                  * but not when the metadata is externally managed.
2182                  */
2183                 if (rdev->raid_disk == -1)
2184                         return -EEXIST;
2185                 /* personality does all needed checks */
2186                 if (rdev->mddev->pers->hot_add_disk == NULL)
2187                         return -EINVAL;
2188                 err = rdev->mddev->pers->
2189                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2190                 if (err)
2191                         return err;
2192                 sprintf(nm, "rd%d", rdev->raid_disk);
2193                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2194                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2195                 md_wakeup_thread(rdev->mddev->thread);
2196         } else if (rdev->mddev->pers) {
2197                 mdk_rdev_t *rdev2;
2198                 /* Activating a spare .. or possibly reactivating
2199                  * if we ever get bitmaps working here.
2200                  */
2201
2202                 if (rdev->raid_disk != -1)
2203                         return -EBUSY;
2204
2205                 if (rdev->mddev->pers->hot_add_disk == NULL)
2206                         return -EINVAL;
2207
2208                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2209                         if (rdev2->raid_disk == slot)
2210                                 return -EEXIST;
2211
2212                 rdev->raid_disk = slot;
2213                 if (test_bit(In_sync, &rdev->flags))
2214                         rdev->saved_raid_disk = slot;
2215                 else
2216                         rdev->saved_raid_disk = -1;
2217                 err = rdev->mddev->pers->
2218                         hot_add_disk(rdev->mddev, rdev);
2219                 if (err) {
2220                         rdev->raid_disk = -1;
2221                         return err;
2222                 } else
2223                         sysfs_notify_dirent(rdev->sysfs_state);
2224                 sprintf(nm, "rd%d", rdev->raid_disk);
2225                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2226                         printk(KERN_WARNING
2227                                "md: cannot register "
2228                                "%s for %s\n",
2229                                nm, mdname(rdev->mddev));
2230
2231                 /* don't wakeup anyone, leave that to userspace. */
2232         } else {
2233                 if (slot >= rdev->mddev->raid_disks)
2234                         return -ENOSPC;
2235                 rdev->raid_disk = slot;
2236                 /* assume it is working */
2237                 clear_bit(Faulty, &rdev->flags);
2238                 clear_bit(WriteMostly, &rdev->flags);
2239                 set_bit(In_sync, &rdev->flags);
2240                 sysfs_notify_dirent(rdev->sysfs_state);
2241         }
2242         return len;
2243 }
2244
2245
2246 static struct rdev_sysfs_entry rdev_slot =
2247 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2248
2249 static ssize_t
2250 offset_show(mdk_rdev_t *rdev, char *page)
2251 {
2252         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2253 }
2254
2255 static ssize_t
2256 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2257 {
2258         char *e;
2259         unsigned long long offset = simple_strtoull(buf, &e, 10);
2260         if (e==buf || (*e && *e != '\n'))
2261                 return -EINVAL;
2262         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2263                 return -EBUSY;
2264         if (rdev->sectors && rdev->mddev->external)
2265                 /* Must set offset before size, so overlap checks
2266                  * can be sane */
2267                 return -EBUSY;
2268         rdev->data_offset = offset;
2269         return len;
2270 }
2271
2272 static struct rdev_sysfs_entry rdev_offset =
2273 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2274
2275 static ssize_t
2276 rdev_size_show(mdk_rdev_t *rdev, char *page)
2277 {
2278         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2279 }
2280
2281 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2282 {
2283         /* check if two start/length pairs overlap */
2284         if (s1+l1 <= s2)
2285                 return 0;
2286         if (s2+l2 <= s1)
2287                 return 0;
2288         return 1;
2289 }
2290
2291 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2292 {
2293         unsigned long long blocks;
2294         sector_t new;
2295
2296         if (strict_strtoull(buf, 10, &blocks) < 0)
2297                 return -EINVAL;
2298
2299         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2300                 return -EINVAL; /* sector conversion overflow */
2301
2302         new = blocks * 2;
2303         if (new != blocks * 2)
2304                 return -EINVAL; /* unsigned long long to sector_t overflow */
2305
2306         *sectors = new;
2307         return 0;
2308 }
2309
2310 static ssize_t
2311 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2312 {
2313         mddev_t *my_mddev = rdev->mddev;
2314         sector_t oldsectors = rdev->sectors;
2315         sector_t sectors;
2316
2317         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2318                 return -EINVAL;
2319         if (my_mddev->pers && rdev->raid_disk >= 0) {
2320                 if (my_mddev->persistent) {
2321                         sectors = super_types[my_mddev->major_version].
2322                                 rdev_size_change(rdev, sectors);
2323                         if (!sectors)
2324                                 return -EBUSY;
2325                 } else if (!sectors)
2326                         sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2327                                 rdev->data_offset;
2328         }
2329         if (sectors < my_mddev->dev_sectors)
2330                 return -EINVAL; /* component must fit device */
2331
2332         rdev->sectors = sectors;
2333         if (sectors > oldsectors && my_mddev->external) {
2334                 /* need to check that all other rdevs with the same ->bdev
2335                  * do not overlap.  We need to unlock the mddev to avoid
2336                  * a deadlock.  We have already changed rdev->sectors, and if
2337                  * we have to change it back, we will have the lock again.
2338                  */
2339                 mddev_t *mddev;
2340                 int overlap = 0;
2341                 struct list_head *tmp;
2342
2343                 mddev_unlock(my_mddev);
2344                 for_each_mddev(mddev, tmp) {
2345                         mdk_rdev_t *rdev2;
2346
2347                         mddev_lock(mddev);
2348                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2349                                 if (test_bit(AllReserved, &rdev2->flags) ||
2350                                     (rdev->bdev == rdev2->bdev &&
2351                                      rdev != rdev2 &&
2352                                      overlaps(rdev->data_offset, rdev->sectors,
2353                                               rdev2->data_offset,
2354                                               rdev2->sectors))) {
2355                                         overlap = 1;
2356                                         break;
2357                                 }
2358                         mddev_unlock(mddev);
2359                         if (overlap) {
2360                                 mddev_put(mddev);
2361                                 break;
2362                         }
2363                 }
2364                 mddev_lock(my_mddev);
2365                 if (overlap) {
2366                         /* Someone else could have slipped in a size
2367                          * change here, but doing so is just silly.
2368                          * We put oldsectors back because we *know* it is
2369                          * safe, and trust userspace not to race with
2370                          * itself
2371                          */
2372                         rdev->sectors = oldsectors;
2373                         return -EBUSY;
2374                 }
2375         }
2376         return len;
2377 }
2378
2379 static struct rdev_sysfs_entry rdev_size =
2380 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2381
2382 static struct attribute *rdev_default_attrs[] = {
2383         &rdev_state.attr,
2384         &rdev_errors.attr,
2385         &rdev_slot.attr,
2386         &rdev_offset.attr,
2387         &rdev_size.attr,
2388         NULL,
2389 };
2390 static ssize_t
2391 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2392 {
2393         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2394         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2395         mddev_t *mddev = rdev->mddev;
2396         ssize_t rv;
2397
2398         if (!entry->show)
2399                 return -EIO;
2400
2401         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2402         if (!rv) {
2403                 if (rdev->mddev == NULL)
2404                         rv = -EBUSY;
2405                 else
2406                         rv = entry->show(rdev, page);
2407                 mddev_unlock(mddev);
2408         }
2409         return rv;
2410 }
2411
2412 static ssize_t
2413 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2414               const char *page, size_t length)
2415 {
2416         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2417         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2418         ssize_t rv;
2419         mddev_t *mddev = rdev->mddev;
2420
2421         if (!entry->store)
2422                 return -EIO;
2423         if (!capable(CAP_SYS_ADMIN))
2424                 return -EACCES;
2425         rv = mddev ? mddev_lock(mddev): -EBUSY;
2426         if (!rv) {
2427                 if (rdev->mddev == NULL)
2428                         rv = -EBUSY;
2429                 else
2430                         rv = entry->store(rdev, page, length);
2431                 mddev_unlock(mddev);
2432         }
2433         return rv;
2434 }
2435
2436 static void rdev_free(struct kobject *ko)
2437 {
2438         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2439         kfree(rdev);
2440 }
2441 static struct sysfs_ops rdev_sysfs_ops = {
2442         .show           = rdev_attr_show,
2443         .store          = rdev_attr_store,
2444 };
2445 static struct kobj_type rdev_ktype = {
2446         .release        = rdev_free,
2447         .sysfs_ops      = &rdev_sysfs_ops,
2448         .default_attrs  = rdev_default_attrs,
2449 };
2450
2451 /*
2452  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2453  *
2454  * mark the device faulty if:
2455  *
2456  *   - the device is nonexistent (zero size)
2457  *   - the device has no valid superblock
2458  *
2459  * a faulty rdev _never_ has rdev->sb set.
2460  */
2461 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2462 {
2463         char b[BDEVNAME_SIZE];
2464         int err;
2465         mdk_rdev_t *rdev;
2466         sector_t size;
2467
2468         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2469         if (!rdev) {
2470                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2471                 return ERR_PTR(-ENOMEM);
2472         }
2473
2474         if ((err = alloc_disk_sb(rdev)))
2475                 goto abort_free;
2476
2477         err = lock_rdev(rdev, newdev, super_format == -2);
2478         if (err)
2479                 goto abort_free;
2480
2481         kobject_init(&rdev->kobj, &rdev_ktype);
2482
2483         rdev->desc_nr = -1;
2484         rdev->saved_raid_disk = -1;
2485         rdev->raid_disk = -1;
2486         rdev->flags = 0;
2487         rdev->data_offset = 0;
2488         rdev->sb_events = 0;
2489         atomic_set(&rdev->nr_pending, 0);
2490         atomic_set(&rdev->read_errors, 0);
2491         atomic_set(&rdev->corrected_errors, 0);
2492
2493         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2494         if (!size) {
2495                 printk(KERN_WARNING 
2496                         "md: %s has zero or unknown size, marking faulty!\n",
2497                         bdevname(rdev->bdev,b));
2498                 err = -EINVAL;
2499                 goto abort_free;
2500         }
2501
2502         if (super_format >= 0) {
2503                 err = super_types[super_format].
2504                         load_super(rdev, NULL, super_minor);
2505                 if (err == -EINVAL) {
2506                         printk(KERN_WARNING
2507                                 "md: %s does not have a valid v%d.%d "
2508                                "superblock, not importing!\n",
2509                                 bdevname(rdev->bdev,b),
2510                                super_format, super_minor);
2511                         goto abort_free;
2512                 }
2513                 if (err < 0) {
2514                         printk(KERN_WARNING 
2515                                 "md: could not read %s's sb, not importing!\n",
2516                                 bdevname(rdev->bdev,b));
2517                         goto abort_free;
2518                 }
2519         }
2520
2521         INIT_LIST_HEAD(&rdev->same_set);
2522         init_waitqueue_head(&rdev->blocked_wait);
2523
2524         return rdev;
2525
2526 abort_free:
2527         if (rdev->sb_page) {
2528                 if (rdev->bdev)
2529                         unlock_rdev(rdev);
2530                 free_disk_sb(rdev);
2531         }
2532         kfree(rdev);
2533         return ERR_PTR(err);
2534 }
2535
2536 /*
2537  * Check a full RAID array for plausibility
2538  */
2539
2540
2541 static void analyze_sbs(mddev_t * mddev)
2542 {
2543         int i;
2544         mdk_rdev_t *rdev, *freshest, *tmp;
2545         char b[BDEVNAME_SIZE];
2546
2547         freshest = NULL;
2548         rdev_for_each(rdev, tmp, mddev)
2549                 switch (super_types[mddev->major_version].
2550                         load_super(rdev, freshest, mddev->minor_version)) {
2551                 case 1:
2552                         freshest = rdev;
2553                         break;
2554                 case 0:
2555                         break;
2556                 default:
2557                         printk( KERN_ERR \
2558                                 "md: fatal superblock inconsistency in %s"
2559                                 " -- removing from array\n", 
2560                                 bdevname(rdev->bdev,b));
2561                         kick_rdev_from_array(rdev);
2562                 }
2563
2564
2565         super_types[mddev->major_version].
2566                 validate_super(mddev, freshest);
2567
2568         i = 0;
2569         rdev_for_each(rdev, tmp, mddev) {
2570                 if (rdev->desc_nr >= mddev->max_disks ||
2571                     i > mddev->max_disks) {
2572                         printk(KERN_WARNING
2573                                "md: %s: %s: only %d devices permitted\n",
2574                                mdname(mddev), bdevname(rdev->bdev, b),
2575                                mddev->max_disks);
2576                         kick_rdev_from_array(rdev);
2577                         continue;
2578                 }
2579                 if (rdev != freshest)
2580                         if (super_types[mddev->major_version].
2581                             validate_super(mddev, rdev)) {
2582                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2583                                         " from array!\n",
2584                                         bdevname(rdev->bdev,b));
2585                                 kick_rdev_from_array(rdev);
2586                                 continue;
2587                         }
2588                 if (mddev->level == LEVEL_MULTIPATH) {
2589                         rdev->desc_nr = i++;
2590                         rdev->raid_disk = rdev->desc_nr;
2591                         set_bit(In_sync, &rdev->flags);
2592                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2593                         rdev->raid_disk = -1;
2594                         clear_bit(In_sync, &rdev->flags);
2595                 }
2596         }
2597
2598
2599
2600         if (mddev->recovery_cp != MaxSector &&
2601             mddev->level >= 1)
2602                 printk(KERN_ERR "md: %s: raid array is not clean"
2603                        " -- starting background reconstruction\n",
2604                        mdname(mddev));
2605
2606 }
2607
2608 static void md_safemode_timeout(unsigned long data);
2609
2610 static ssize_t
2611 safe_delay_show(mddev_t *mddev, char *page)
2612 {
2613         int msec = (mddev->safemode_delay*1000)/HZ;
2614         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2615 }
2616 static ssize_t
2617 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2618 {
2619         int scale=1;
2620         int dot=0;
2621         int i;
2622         unsigned long msec;
2623         char buf[30];
2624
2625         /* remove a period, and count digits after it */
2626         if (len >= sizeof(buf))
2627                 return -EINVAL;
2628         strlcpy(buf, cbuf, sizeof(buf));
2629         for (i=0; i<len; i++) {
2630                 if (dot) {
2631                         if (isdigit(buf[i])) {
2632                                 buf[i-1] = buf[i];
2633                                 scale *= 10;
2634                         }
2635                         buf[i] = 0;
2636                 } else if (buf[i] == '.') {
2637                         dot=1;
2638                         buf[i] = 0;
2639                 }
2640         }
2641         if (strict_strtoul(buf, 10, &msec) < 0)
2642                 return -EINVAL;
2643         msec = (msec * 1000) / scale;
2644         if (msec == 0)
2645                 mddev->safemode_delay = 0;
2646         else {
2647                 unsigned long old_delay = mddev->safemode_delay;
2648                 mddev->safemode_delay = (msec*HZ)/1000;
2649                 if (mddev->safemode_delay == 0)
2650                         mddev->safemode_delay = 1;
2651                 if (mddev->safemode_delay < old_delay)
2652                         md_safemode_timeout((unsigned long)mddev);
2653         }
2654         return len;
2655 }
2656 static struct md_sysfs_entry md_safe_delay =
2657 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2658
2659 static ssize_t
2660 level_show(mddev_t *mddev, char *page)
2661 {
2662         struct mdk_personality *p = mddev->pers;
2663         if (p)
2664                 return sprintf(page, "%s\n", p->name);
2665         else if (mddev->clevel[0])
2666                 return sprintf(page, "%s\n", mddev->clevel);
2667         else if (mddev->level != LEVEL_NONE)
2668                 return sprintf(page, "%d\n", mddev->level);
2669         else
2670                 return 0;
2671 }
2672
2673 static ssize_t
2674 level_store(mddev_t *mddev, const char *buf, size_t len)
2675 {
2676         char level[16];
2677         ssize_t rv = len;
2678         struct mdk_personality *pers;
2679         void *priv;
2680
2681         if (mddev->pers == NULL) {
2682                 if (len == 0)
2683                         return 0;
2684                 if (len >= sizeof(mddev->clevel))
2685                         return -ENOSPC;
2686                 strncpy(mddev->clevel, buf, len);
2687                 if (mddev->clevel[len-1] == '\n')
2688                         len--;
2689                 mddev->clevel[len] = 0;
2690                 mddev->level = LEVEL_NONE;
2691                 return rv;
2692         }
2693
2694         /* request to change the personality.  Need to ensure:
2695          *  - array is not engaged in resync/recovery/reshape
2696          *  - old personality can be suspended
2697          *  - new personality will access other array.
2698          */
2699
2700         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2701                 return -EBUSY;
2702
2703         if (!mddev->pers->quiesce) {
2704                 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2705                        mdname(mddev), mddev->pers->name);
2706                 return -EINVAL;
2707         }
2708
2709         /* Now find the new personality */
2710         if (len == 0 || len >= sizeof(level))
2711                 return -EINVAL;
2712         strncpy(level, buf, len);
2713         if (level[len-1] == '\n')
2714                 len--;
2715         level[len] = 0;
2716
2717         request_module("md-%s", level);
2718         spin_lock(&pers_lock);
2719         pers = find_pers(LEVEL_NONE, level);
2720         if (!pers || !try_module_get(pers->owner)) {
2721                 spin_unlock(&pers_lock);
2722                 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2723                 return -EINVAL;
2724         }
2725         spin_unlock(&pers_lock);
2726
2727         if (pers == mddev->pers) {
2728                 /* Nothing to do! */
2729                 module_put(pers->owner);
2730                 return rv;
2731         }
2732         if (!pers->takeover) {
2733                 module_put(pers->owner);
2734                 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2735                        mdname(mddev), level);
2736                 return -EINVAL;
2737         }
2738
2739         /* ->takeover must set new_* and/or delta_disks
2740          * if it succeeds, and may set them when it fails.
2741          */
2742         priv = pers->takeover(mddev);
2743         if (IS_ERR(priv)) {
2744                 mddev->new_level = mddev->level;
2745                 mddev->new_layout = mddev->layout;
2746                 mddev->new_chunk = mddev->chunk_size;
2747                 mddev->raid_disks -= mddev->delta_disks;
2748                 mddev->delta_disks = 0;
2749                 module_put(pers->owner);
2750                 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2751                        mdname(mddev), level);
2752                 return PTR_ERR(priv);
2753         }
2754
2755         /* Looks like we have a winner */
2756         mddev_suspend(mddev);
2757         mddev->pers->stop(mddev);
2758         module_put(mddev->pers->owner);
2759         mddev->pers = pers;
2760         mddev->private = priv;
2761         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2762         mddev->level = mddev->new_level;
2763         mddev->layout = mddev->new_layout;
2764         mddev->chunk_size = mddev->new_chunk;
2765         mddev->delta_disks = 0;
2766         pers->run(mddev);
2767         mddev_resume(mddev);
2768         set_bit(MD_CHANGE_DEVS, &mddev->flags);
2769         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2770         md_wakeup_thread(mddev->thread);
2771         return rv;
2772 }
2773
2774 static struct md_sysfs_entry md_level =
2775 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2776
2777
2778 static ssize_t
2779 layout_show(mddev_t *mddev, char *page)
2780 {
2781         /* just a number, not meaningful for all levels */
2782         if (mddev->reshape_position != MaxSector &&
2783             mddev->layout != mddev->new_layout)
2784                 return sprintf(page, "%d (%d)\n",
2785                                mddev->new_layout, mddev->layout);
2786         return sprintf(page, "%d\n", mddev->layout);
2787 }
2788
2789 static ssize_t
2790 layout_store(mddev_t *mddev, const char *buf, size_t len)
2791 {
2792         char *e;
2793         unsigned long n = simple_strtoul(buf, &e, 10);
2794
2795         if (!*buf || (*e && *e != '\n'))
2796                 return -EINVAL;
2797
2798         if (mddev->pers) {
2799                 int err;
2800                 if (mddev->pers->reconfig == NULL)
2801                         return -EBUSY;
2802                 err = mddev->pers->reconfig(mddev, n, -1);
2803                 if (err)
2804                         return err;
2805         } else {
2806                 mddev->new_layout = n;
2807                 if (mddev->reshape_position == MaxSector)
2808                         mddev->layout = n;
2809         }
2810         return len;
2811 }
2812 static struct md_sysfs_entry md_layout =
2813 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2814
2815
2816 static ssize_t
2817 raid_disks_show(mddev_t *mddev, char *page)
2818 {
2819         if (mddev->raid_disks == 0)
2820                 return 0;
2821         if (mddev->reshape_position != MaxSector &&
2822             mddev->delta_disks != 0)
2823                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2824                                mddev->raid_disks - mddev->delta_disks);
2825         return sprintf(page, "%d\n", mddev->raid_disks);
2826 }
2827
2828 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2829
2830 static ssize_t
2831 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2832 {
2833         char *e;
2834         int rv = 0;
2835         unsigned long n = simple_strtoul(buf, &e, 10);
2836
2837         if (!*buf || (*e && *e != '\n'))
2838                 return -EINVAL;
2839
2840         if (mddev->pers)
2841                 rv = update_raid_disks(mddev, n);
2842         else if (mddev->reshape_position != MaxSector) {
2843                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2844                 mddev->delta_disks = n - olddisks;
2845                 mddev->raid_disks = n;
2846         } else
2847                 mddev->raid_disks = n;
2848         return rv ? rv : len;
2849 }
2850 static struct md_sysfs_entry md_raid_disks =
2851 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2852
2853 static ssize_t
2854 chunk_size_show(mddev_t *mddev, char *page)
2855 {
2856         if (mddev->reshape_position != MaxSector &&
2857             mddev->chunk_size != mddev->new_chunk)
2858                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2859                                mddev->chunk_size);
2860         return sprintf(page, "%d\n", mddev->chunk_size);
2861 }
2862
2863 static ssize_t
2864 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2865 {
2866         char *e;
2867         unsigned long n = simple_strtoul(buf, &e, 10);
2868
2869         if (!*buf || (*e && *e != '\n'))
2870                 return -EINVAL;
2871
2872         if (mddev->pers) {
2873                 int err;
2874                 if (mddev->pers->reconfig == NULL)
2875                         return -EBUSY;
2876                 err = mddev->pers->reconfig(mddev, -1, n);
2877                 if (err)
2878                         return err;
2879         } else {
2880                 mddev->new_chunk = n;
2881                 if (mddev->reshape_position == MaxSector)
2882                         mddev->chunk_size = n;
2883         }
2884         return len;
2885 }
2886 static struct md_sysfs_entry md_chunk_size =
2887 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2888
2889 static ssize_t
2890 resync_start_show(mddev_t *mddev, char *page)
2891 {
2892         if (mddev->recovery_cp == MaxSector)
2893                 return sprintf(page, "none\n");
2894         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2895 }
2896
2897 static ssize_t
2898 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2899 {
2900         char *e;
2901         unsigned long long n = simple_strtoull(buf, &e, 10);
2902
2903         if (mddev->pers)
2904                 return -EBUSY;
2905         if (!*buf || (*e && *e != '\n'))
2906                 return -EINVAL;
2907
2908         mddev->recovery_cp = n;
2909         return len;
2910 }
2911 static struct md_sysfs_entry md_resync_start =
2912 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2913
2914 /*
2915  * The array state can be:
2916  *
2917  * clear
2918  *     No devices, no size, no level
2919  *     Equivalent to STOP_ARRAY ioctl
2920  * inactive
2921  *     May have some settings, but array is not active
2922  *        all IO results in error
2923  *     When written, doesn't tear down array, but just stops it
2924  * suspended (not supported yet)
2925  *     All IO requests will block. The array can be reconfigured.
2926  *     Writing this, if accepted, will block until array is quiescent
2927  * readonly
2928  *     no resync can happen.  no superblocks get written.
2929  *     write requests fail
2930  * read-auto
2931  *     like readonly, but behaves like 'clean' on a write request.
2932  *
2933  * clean - no pending writes, but otherwise active.
2934  *     When written to inactive array, starts without resync
2935  *     If a write request arrives then
2936  *       if metadata is known, mark 'dirty' and switch to 'active'.
2937  *       if not known, block and switch to write-pending
2938  *     If written to an active array that has pending writes, then fails.
2939  * active
2940  *     fully active: IO and resync can be happening.
2941  *     When written to inactive array, starts with resync
2942  *
2943  * write-pending
2944  *     clean, but writes are blocked waiting for 'active' to be written.
2945  *
2946  * active-idle
2947  *     like active, but no writes have been seen for a while (100msec).
2948  *
2949  */
2950 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2951                    write_pending, active_idle, bad_word};
2952 static char *array_states[] = {
2953         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2954         "write-pending", "active-idle", NULL };
2955
2956 static int match_word(const char *word, char **list)
2957 {
2958         int n;
2959         for (n=0; list[n]; n++)
2960                 if (cmd_match(word, list[n]))
2961                         break;
2962         return n;
2963 }
2964
2965 static ssize_t
2966 array_state_show(mddev_t *mddev, char *page)
2967 {
2968         enum array_state st = inactive;
2969
2970         if (mddev->pers)
2971                 switch(mddev->ro) {
2972                 case 1:
2973                         st = readonly;
2974                         break;
2975                 case 2:
2976                         st = read_auto;
2977                         break;
2978                 case 0:
2979                         if (mddev->in_sync)
2980                                 st = clean;
2981                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2982                                 st = write_pending;
2983                         else if (mddev->safemode)
2984                                 st = active_idle;
2985                         else
2986                                 st = active;
2987                 }
2988         else {
2989                 if (list_empty(&mddev->disks) &&
2990                     mddev->raid_disks == 0 &&
2991                     mddev->dev_sectors == 0)
2992                         st = clear;
2993                 else
2994                         st = inactive;
2995         }
2996         return sprintf(page, "%s\n", array_states[st]);
2997 }
2998
2999 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3000 static int do_md_run(mddev_t * mddev);
3001 static int restart_array(mddev_t *mddev);
3002
3003 static ssize_t
3004 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3005 {
3006         int err = -EINVAL;
3007         enum array_state st = match_word(buf, array_states);
3008         switch(st) {
3009         case bad_word:
3010                 break;
3011         case clear:
3012                 /* stopping an active array */
3013                 if (atomic_read(&mddev->openers) > 0)
3014                         return -EBUSY;
3015                 err = do_md_stop(mddev, 0, 0);
3016                 break;
3017         case inactive:
3018                 /* stopping an active array */
3019                 if (mddev->pers) {
3020                         if (atomic_read(&mddev->openers) > 0)
3021                                 return -EBUSY;
3022                         err = do_md_stop(mddev, 2, 0);
3023                 } else
3024                         err = 0; /* already inactive */
3025                 break;
3026         case suspended:
3027                 break; /* not supported yet */
3028         case readonly:
3029                 if (mddev->pers)
3030                         err = do_md_stop(mddev, 1, 0);
3031                 else {
3032                         mddev->ro = 1;
3033                         set_disk_ro(mddev->gendisk, 1);
3034                         err = do_md_run(mddev);
3035                 }
3036                 break;
3037         case read_auto:
3038                 if (mddev->pers) {
3039                         if (mddev->ro == 0)
3040                                 err = do_md_stop(mddev, 1, 0);
3041                         else if (mddev->ro == 1)
3042                                 err = restart_array(mddev);
3043                         if (err == 0) {
3044                                 mddev->ro = 2;
3045                                 set_disk_ro(mddev->gendisk, 0);
3046                         }
3047                 } else {
3048                         mddev->ro = 2;
3049                         err = do_md_run(mddev);
3050                 }
3051                 break;
3052         case clean:
3053                 if (mddev->pers) {
3054                         restart_array(mddev);
3055                         spin_lock_irq(&mddev->write_lock);
3056                         if (atomic_read(&mddev->writes_pending) == 0) {
3057                                 if (mddev->in_sync == 0) {
3058                                         mddev->in_sync = 1;
3059                                         if (mddev->safemode == 1)
3060                                                 mddev->safemode = 0;
3061                                         if (mddev->persistent)
3062                                                 set_bit(MD_CHANGE_CLEAN,
3063                                                         &mddev->flags);
3064                                 }
3065                                 err = 0;
3066                         } else
3067                                 err = -EBUSY;
3068                         spin_unlock_irq(&mddev->write_lock);
3069                 } else {
3070                         mddev->ro = 0;
3071                         mddev->recovery_cp = MaxSector;
3072                         err = do_md_run(mddev);
3073                 }
3074                 break;
3075         case active:
3076                 if (mddev->pers) {
3077                         restart_array(mddev);
3078                         if (mddev->external)
3079                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3080                         wake_up(&mddev->sb_wait);
3081                         err = 0;
3082                 } else {
3083                         mddev->ro = 0;
3084                         set_disk_ro(mddev->gendisk, 0);
3085                         err = do_md_run(mddev);
3086                 }
3087                 break;
3088         case write_pending:
3089         case active_idle:
3090                 /* these cannot be set */
3091                 break;
3092         }
3093         if (err)
3094                 return err;
3095         else {
3096                 sysfs_notify_dirent(mddev->sysfs_state);
3097                 return len;
3098         }
3099 }
3100 static struct md_sysfs_entry md_array_state =
3101 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3102
3103 static ssize_t
3104 null_show(mddev_t *mddev, char *page)
3105 {
3106         return -EINVAL;
3107 }
3108
3109 static ssize_t
3110 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3111 {
3112         /* buf must be %d:%d\n? giving major and minor numbers */
3113         /* The new device is added to the array.
3114          * If the array has a persistent superblock, we read the
3115          * superblock to initialise info and check validity.
3116          * Otherwise, only checking done is that in bind_rdev_to_array,
3117          * which mainly checks size.
3118          */
3119         char *e;
3120         int major = simple_strtoul(buf, &e, 10);
3121         int minor;
3122         dev_t dev;
3123         mdk_rdev_t *rdev;
3124         int err;
3125
3126         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3127                 return -EINVAL;
3128         minor = simple_strtoul(e+1, &e, 10);
3129         if (*e && *e != '\n')
3130                 return -EINVAL;
3131         dev = MKDEV(major, minor);
3132         if (major != MAJOR(dev) ||
3133             minor != MINOR(dev))
3134                 return -EOVERFLOW;
3135
3136
3137         if (mddev->persistent) {
3138                 rdev = md_import_device(dev, mddev->major_version,
3139                                         mddev->minor_version);
3140                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3141                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3142                                                        mdk_rdev_t, same_set);
3143                         err = super_types[mddev->major_version]
3144                                 .load_super(rdev, rdev0, mddev->minor_version);
3145                         if (err < 0)
3146                                 goto out;
3147                 }
3148         } else if (mddev->external)
3149                 rdev = md_import_device(dev, -2, -1);
3150         else
3151                 rdev = md_import_device(dev, -1, -1);
3152
3153         if (IS_ERR(rdev))
3154                 return PTR_ERR(rdev);
3155         err = bind_rdev_to_array(rdev, mddev);
3156  out:
3157         if (err)
3158                 export_rdev(rdev);
3159         return err ? err : len;
3160 }
3161
3162 static struct md_sysfs_entry md_new_device =
3163 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3164
3165 static ssize_t
3166 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3167 {
3168         char *end;
3169         unsigned long chunk, end_chunk;
3170
3171         if (!mddev->bitmap)
3172                 goto out;
3173         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3174         while (*buf) {
3175                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3176                 if (buf == end) break;
3177                 if (*end == '-') { /* range */
3178                         buf = end + 1;
3179                         end_chunk = simple_strtoul(buf, &end, 0);
3180                         if (buf == end) break;
3181                 }
3182                 if (*end && !isspace(*end)) break;
3183                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3184                 buf = end;
3185                 while (isspace(*buf)) buf++;
3186         }
3187         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3188 out:
3189         return len;
3190 }
3191
3192 static struct md_sysfs_entry md_bitmap =
3193 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3194
3195 static ssize_t
3196 size_show(mddev_t *mddev, char *page)
3197 {
3198         return sprintf(page, "%llu\n",
3199                 (unsigned long long)mddev->dev_sectors / 2);
3200 }
3201
3202 static int update_size(mddev_t *mddev, sector_t num_sectors);
3203
3204 static ssize_t
3205 size_store(mddev_t *mddev, const char *buf, size_t len)
3206 {
3207         /* If array is inactive, we can reduce the component size, but
3208          * not increase it (except from 0).
3209          * If array is active, we can try an on-line resize
3210          */
3211         sector_t sectors;
3212         int err = strict_blocks_to_sectors(buf, &sectors);
3213
3214         if (err < 0)
3215                 return err;
3216         if (mddev->pers) {
3217                 err = update_size(mddev, sectors);
3218                 md_update_sb(mddev, 1);
3219         } else {
3220                 if (mddev->dev_sectors == 0 ||
3221                     mddev->dev_sectors > sectors)
3222                         mddev->dev_sectors = sectors;
3223                 else
3224                         err = -ENOSPC;
3225         }
3226         return err ? err : len;
3227 }
3228
3229 static struct md_sysfs_entry md_size =
3230 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3231
3232
3233 /* Metdata version.
3234  * This is one of
3235  *   'none' for arrays with no metadata (good luck...)
3236  *   'external' for arrays with externally managed metadata,
3237  * or N.M for internally known formats
3238  */
3239 static ssize_t
3240 metadata_show(mddev_t *mddev, char *page)
3241 {
3242         if (mddev->persistent)
3243                 return sprintf(page, "%d.%d\n",
3244                                mddev->major_version, mddev->minor_version);
3245         else if (mddev->external)
3246                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3247         else
3248                 return sprintf(page, "none\n");
3249 }
3250
3251 static ssize_t
3252 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3253 {
3254         int major, minor;
3255         char *e;
3256         /* Changing the details of 'external' metadata is
3257          * always permitted.  Otherwise there must be
3258          * no devices attached to the array.
3259          */
3260         if (mddev->external && strncmp(buf, "external:", 9) == 0)
3261                 ;
3262         else if (!list_empty(&mddev->disks))
3263                 return -EBUSY;
3264
3265         if (cmd_match(buf, "none")) {
3266                 mddev->persistent = 0;
3267                 mddev->external = 0;
3268                 mddev->major_version = 0;
3269                 mddev->minor_version = 90;
3270                 return len;
3271         }
3272         if (strncmp(buf, "external:", 9) == 0) {
3273                 size_t namelen = len-9;
3274                 if (namelen >= sizeof(mddev->metadata_type))
3275                         namelen = sizeof(mddev->metadata_type)-1;
3276                 strncpy(mddev->metadata_type, buf+9, namelen);
3277                 mddev->metadata_type[namelen] = 0;
3278                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3279                         mddev->metadata_type[--namelen] = 0;
3280                 mddev->persistent = 0;
3281                 mddev->external = 1;
3282                 mddev->major_version = 0;
3283                 mddev->minor_version = 90;
3284                 return len;
3285         }
3286         major = simple_strtoul(buf, &e, 10);
3287         if (e==buf || *e != '.')
3288                 return -EINVAL;
3289         buf = e+1;
3290         minor = simple_strtoul(buf, &e, 10);
3291         if (e==buf || (*e && *e != '\n') )
3292                 return -EINVAL;
3293         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3294                 return -ENOENT;
3295         mddev->major_version = major;
3296         mddev->minor_version = minor;
3297         mddev->persistent = 1;
3298         mddev->external = 0;
3299         return len;
3300 }
3301
3302 static struct md_sysfs_entry md_metadata =
3303 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3304
3305 static ssize_t
3306 action_show(mddev_t *mddev, char *page)
3307 {
3308         char *type = "idle";
3309         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3310             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3311                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3312                         type = "reshape";
3313                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3314                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3315                                 type = "resync";
3316                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3317                                 type = "check";
3318                         else
3319                                 type = "repair";
3320                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3321                         type = "recover";
3322         }
3323         return sprintf(page, "%s\n", type);
3324 }
3325
3326 static ssize_t
3327 action_store(mddev_t *mddev, const char *page, size_t len)
3328 {
3329         if (!mddev->pers || !mddev->pers->sync_request)
3330                 return -EINVAL;
3331
3332         if (cmd_match(page, "idle")) {
3333                 if (mddev->sync_thread) {
3334                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3335                         md_unregister_thread(mddev->sync_thread);
3336                         mddev->sync_thread = NULL;
3337                         mddev->recovery = 0;
3338                 }
3339         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3340                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3341                 return -EBUSY;
3342         else if (cmd_match(page, "resync"))
3343                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3344         else if (cmd_match(page, "recover")) {
3345                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3346                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3347         } else if (cmd_match(page, "reshape")) {
3348                 int err;
3349                 if (mddev->pers->start_reshape == NULL)
3350                         return -EINVAL;
3351                 err = mddev->pers->start_reshape(mddev);
3352                 if (err)
3353                         return err;
3354                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3355         } else {
3356                 if (cmd_match(page, "check"))
3357                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3358                 else if (!cmd_match(page, "repair"))
3359                         return -EINVAL;
3360                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3361                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3362         }
3363         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3364         md_wakeup_thread(mddev->thread);
3365         sysfs_notify_dirent(mddev->sysfs_action);
3366         return len;
3367 }
3368
3369 static ssize_t
3370 mismatch_cnt_show(mddev_t *mddev, char *page)
3371 {
3372         return sprintf(page, "%llu\n",
3373                        (unsigned long long) mddev->resync_mismatches);
3374 }
3375
3376 static struct md_sysfs_entry md_scan_mode =
3377 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3378
3379
3380 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3381
3382 static ssize_t
3383 sync_min_show(mddev_t *mddev, char *page)
3384 {
3385         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3386                        mddev->sync_speed_min ? "local": "system");
3387 }
3388
3389 static ssize_t
3390 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3391 {
3392         int min;
3393         char *e;
3394         if (strncmp(buf, "system", 6)==0) {
3395                 mddev->sync_speed_min = 0;
3396                 return len;
3397         }
3398         min = simple_strtoul(buf, &e, 10);
3399         if (buf == e || (*e && *e != '\n') || min <= 0)
3400                 return -EINVAL;
3401         mddev->sync_speed_min = min;
3402         return len;
3403 }
3404
3405 static struct md_sysfs_entry md_sync_min =
3406 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3407
3408 static ssize_t
3409 sync_max_show(mddev_t *mddev, char *page)
3410 {
3411         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3412                        mddev->sync_speed_max ? "local": "system");
3413 }
3414
3415 static ssize_t
3416 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3417 {
3418         int max;
3419         char *e;
3420         if (strncmp(buf, "system", 6)==0) {
3421                 mddev->sync_speed_max = 0;
3422                 return len;
3423         }
3424         max = simple_strtoul(buf, &e, 10);
3425         if (buf == e || (*e && *e != '\n') || max <= 0)
3426                 return -EINVAL;
3427         mddev->sync_speed_max = max;
3428         return len;
3429 }
3430
3431 static struct md_sysfs_entry md_sync_max =
3432 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3433
3434 static ssize_t
3435 degraded_show(mddev_t *mddev, char *page)
3436 {
3437         return sprintf(page, "%d\n", mddev->degraded);
3438 }
3439 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3440
3441 static ssize_t
3442 sync_force_parallel_show(mddev_t *mddev, char *page)
3443 {
3444         return sprintf(page, "%d\n", mddev->parallel_resync);
3445 }
3446
3447 static ssize_t
3448 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3449 {
3450         long n;
3451
3452         if (strict_strtol(buf, 10, &n))
3453                 return -EINVAL;
3454
3455         if (n != 0 && n != 1)
3456                 return -EINVAL;
3457
3458         mddev->parallel_resync = n;
3459
3460         if (mddev->sync_thread)
3461                 wake_up(&resync_wait);
3462
3463         return len;
3464 }
3465
3466 /* force parallel resync, even with shared block devices */
3467 static struct md_sysfs_entry md_sync_force_parallel =
3468 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3469        sync_force_parallel_show, sync_force_parallel_store);
3470
3471 static ssize_t
3472 sync_speed_show(mddev_t *mddev, char *page)
3473 {
3474         unsigned long resync, dt, db;
3475         if (mddev->curr_resync == 0)
3476                 return sprintf(page, "none\n");
3477         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3478         dt = (jiffies - mddev->resync_mark) / HZ;
3479         if (!dt) dt++;
3480         db = resync - mddev->resync_mark_cnt;
3481         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3482 }
3483
3484 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3485
3486 static ssize_t
3487 sync_completed_show(mddev_t *mddev, char *page)
3488 {
3489         unsigned long max_sectors, resync;
3490
3491         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3492                 return sprintf(page, "none\n");
3493
3494         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3495                 max_sectors = mddev->resync_max_sectors;
3496         else
3497                 max_sectors = mddev->dev_sectors;
3498
3499         resync = mddev->curr_resync_completed;
3500         return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3501 }
3502
3503 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3504
3505 static ssize_t
3506 min_sync_show(mddev_t *mddev, char *page)
3507 {
3508         return sprintf(page, "%llu\n",
3509                        (unsigned long long)mddev->resync_min);
3510 }
3511 static ssize_t
3512 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3513 {
3514         unsigned long long min;
3515         if (strict_strtoull(buf, 10, &min))
3516                 return -EINVAL;
3517         if (min > mddev->resync_max)
3518                 return -EINVAL;
3519         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3520                 return -EBUSY;
3521
3522         /* Must be a multiple of chunk_size */
3523         if (mddev->chunk_size) {
3524                 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3525                         return -EINVAL;
3526         }
3527         mddev->resync_min = min;
3528
3529         return len;
3530 }
3531
3532 static struct md_sysfs_entry md_min_sync =
3533 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3534
3535 static ssize_t
3536 max_sync_show(mddev_t *mddev, char *page)
3537 {
3538         if (mddev->resync_max == MaxSector)
3539                 return sprintf(page, "max\n");
3540         else
3541                 return sprintf(page, "%llu\n",
3542                                (unsigned long long)mddev->resync_max);
3543 }
3544 static ssize_t
3545 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3546 {
3547         if (strncmp(buf, "max", 3) == 0)
3548                 mddev->resync_max = MaxSector;
3549         else {
3550                 unsigned long long max;
3551                 if (strict_strtoull(buf, 10, &max))
3552                         return -EINVAL;
3553                 if (max < mddev->resync_min)
3554                         return -EINVAL;
3555                 if (max < mddev->resync_max &&
3556                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3557                         return -EBUSY;
3558
3559                 /* Must be a multiple of chunk_size */
3560                 if (mddev->chunk_size) {
3561                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3562                                 return -EINVAL;
3563                 }
3564                 mddev->resync_max = max;
3565         }
3566         wake_up(&mddev->recovery_wait);
3567         return len;
3568 }
3569
3570 static struct md_sysfs_entry md_max_sync =
3571 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3572
3573 static ssize_t
3574 suspend_lo_show(mddev_t *mddev, char *page)
3575 {
3576         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3577 }
3578
3579 static ssize_t
3580 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3581 {
3582         char *e;
3583         unsigned long long new = simple_strtoull(buf, &e, 10);
3584
3585         if (mddev->pers->quiesce == NULL)
3586                 return -EINVAL;
3587         if (buf == e || (*e && *e != '\n'))
3588                 return -EINVAL;
3589         if (new >= mddev->suspend_hi ||
3590             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3591                 mddev->suspend_lo = new;
3592                 mddev->pers->quiesce(mddev, 2);
3593                 return len;
3594         } else
3595                 return -EINVAL;
3596 }
3597 static struct md_sysfs_entry md_suspend_lo =
3598 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3599
3600
3601 static ssize_t
3602 suspend_hi_show(mddev_t *mddev, char *page)
3603 {
3604         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3605 }
3606
3607 static ssize_t
3608 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3609 {
3610         char *e;
3611         unsigned long long new = simple_strtoull(buf, &e, 10);
3612
3613         if (mddev->pers->quiesce == NULL)
3614                 return -EINVAL;
3615         if (buf == e || (*e && *e != '\n'))
3616                 return -EINVAL;
3617         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3618             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3619                 mddev->suspend_hi = new;
3620                 mddev->pers->quiesce(mddev, 1);
3621                 mddev->pers->quiesce(mddev, 0);
3622                 return len;
3623         } else
3624                 return -EINVAL;
3625 }
3626 static struct md_sysfs_entry md_suspend_hi =
3627 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3628
3629 static ssize_t
3630 reshape_position_show(mddev_t *mddev, char *page)
3631 {
3632         if (mddev->reshape_position != MaxSector)
3633                 return sprintf(page, "%llu\n",
3634                                (unsigned long long)mddev->reshape_position);
3635         strcpy(page, "none\n");
3636         return 5;
3637 }
3638
3639 static ssize_t
3640 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3641 {
3642         char *e;
3643         unsigned long long new = simple_strtoull(buf, &e, 10);
3644         if (mddev->pers)
3645                 return -EBUSY;
3646         if (buf == e || (*e && *e != '\n'))
3647                 return -EINVAL;
3648         mddev->reshape_position = new;
3649         mddev->delta_disks = 0;
3650         mddev->new_level = mddev->level;
3651         mddev->new_layout = mddev->layout;
3652         mddev->new_chunk = mddev->chunk_size;
3653         return len;
3654 }
3655
3656 static struct md_sysfs_entry md_reshape_position =
3657 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3658        reshape_position_store);
3659
3660 static ssize_t
3661 array_size_show(mddev_t *mddev, char *page)
3662 {
3663         if (mddev->external_size)
3664                 return sprintf(page, "%llu\n",
3665                                (unsigned long long)mddev->array_sectors/2);
3666         else
3667                 return sprintf(page, "default\n");
3668 }
3669
3670 static ssize_t
3671 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3672 {
3673         sector_t sectors;
3674
3675         if (strncmp(buf, "default", 7) == 0) {
3676                 if (mddev->pers)
3677                         sectors = mddev->pers->size(mddev, 0, 0);
3678                 else
3679                         sectors = mddev->array_sectors;
3680
3681                 mddev->external_size = 0;
3682         } else {
3683                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3684                         return -EINVAL;
3685                 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3686                         return -EINVAL;
3687
3688                 mddev->external_size = 1;
3689         }
3690
3691         mddev->array_sectors = sectors;
3692         set_capacity(mddev->gendisk, mddev->array_sectors);
3693         if (mddev->pers) {
3694                 struct block_device *bdev = bdget_disk(mddev->gendisk, 0);
3695
3696                 if (bdev) {
3697                         mutex_lock(&bdev->bd_inode->i_mutex);
3698                         i_size_write(bdev->bd_inode,
3699                                      (loff_t)mddev->array_sectors << 9);
3700                         mutex_unlock(&bdev->bd_inode->i_mutex);
3701                         bdput(bdev);
3702                 }
3703         }
3704
3705         return len;
3706 }
3707
3708 static struct md_sysfs_entry md_array_size =
3709 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3710        array_size_store);
3711
3712 static struct attribute *md_default_attrs[] = {
3713         &md_level.attr,
3714         &md_layout.attr,
3715         &md_raid_disks.attr,
3716         &md_chunk_size.attr,
3717         &md_size.attr,
3718         &md_resync_start.attr,
3719         &md_metadata.attr,
3720         &md_new_device.attr,
3721         &md_safe_delay.attr,
3722         &md_array_state.attr,
3723         &md_reshape_position.attr,
3724         &md_array_size.attr,
3725         NULL,
3726 };
3727
3728 static struct attribute *md_redundancy_attrs[] = {
3729         &md_scan_mode.attr,
3730         &md_mismatches.attr,
3731         &md_sync_min.attr,
3732         &md_sync_max.attr,
3733         &md_sync_speed.attr,
3734         &md_sync_force_parallel.attr,
3735         &md_sync_completed.attr,
3736         &md_min_sync.attr,
3737         &md_max_sync.attr,
3738         &md_suspend_lo.attr,
3739         &md_suspend_hi.attr,
3740         &md_bitmap.attr,
3741         &md_degraded.attr,
3742         NULL,
3743 };
3744 static struct attribute_group md_redundancy_group = {
3745         .name = NULL,
3746         .attrs = md_redundancy_attrs,
3747 };
3748
3749
3750 static ssize_t
3751 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3752 {
3753         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3754         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3755         ssize_t rv;
3756
3757         if (!entry->show)
3758                 return -EIO;
3759         rv = mddev_lock(mddev);
3760         if (!rv) {
3761                 rv = entry->show(mddev, page);
3762                 mddev_unlock(mddev);
3763         }
3764         return rv;
3765 }
3766
3767 static ssize_t
3768 md_attr_store(struct kobject *kobj, struct attribute *attr,
3769               const char *page, size_t length)
3770 {
3771         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3772         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3773         ssize_t rv;
3774
3775         if (!entry->store)
3776                 return -EIO;
3777         if (!capable(CAP_SYS_ADMIN))
3778                 return -EACCES;
3779         rv = mddev_lock(mddev);
3780         if (mddev->hold_active == UNTIL_IOCTL)
3781                 mddev->hold_active = 0;
3782         if (!rv) {
3783                 rv = entry->store(mddev, page, length);
3784                 mddev_unlock(mddev);
3785         }
3786         return rv;
3787 }
3788
3789 static void md_free(struct kobject *ko)
3790 {
3791         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3792
3793         if (mddev->sysfs_state)
3794                 sysfs_put(mddev->sysfs_state);
3795
3796         if (mddev->gendisk) {
3797                 del_gendisk(mddev->gendisk);
3798                 put_disk(mddev->gendisk);
3799         }
3800         if (mddev->queue)
3801                 blk_cleanup_queue(mddev->queue);
3802
3803         kfree(mddev);
3804 }
3805
3806 static struct sysfs_ops md_sysfs_ops = {
3807         .show   = md_attr_show,
3808         .store  = md_attr_store,
3809 };
3810 static struct kobj_type md_ktype = {
3811         .release        = md_free,
3812         .sysfs_ops      = &md_sysfs_ops,
3813         .default_attrs  = md_default_attrs,
3814 };
3815
3816 int mdp_major = 0;
3817
3818 static void mddev_delayed_delete(struct work_struct *ws)
3819 {
3820         mddev_t *mddev = container_of(ws, mddev_t, del_work);
3821
3822         if (mddev->private == &md_redundancy_group) {
3823                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3824                 if (mddev->sysfs_action)
3825                         sysfs_put(mddev->sysfs_action);
3826                 mddev->sysfs_action = NULL;
3827                 mddev->private = NULL;
3828         }
3829         kobject_del(&mddev->kobj);
3830         kobject_put(&mddev->kobj);
3831 }
3832
3833 static int md_alloc(dev_t dev, char *name)
3834 {
3835         static DEFINE_MUTEX(disks_mutex);
3836         mddev_t *mddev = mddev_find(dev);
3837         struct gendisk *disk;
3838         int partitioned;
3839         int shift;
3840         int unit;
3841         int error;
3842
3843         if (!mddev)
3844                 return -ENODEV;
3845
3846         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3847         shift = partitioned ? MdpMinorShift : 0;
3848         unit = MINOR(mddev->unit) >> shift;
3849
3850         /* wait for any previous instance if this device
3851          * to be completed removed (mddev_delayed_delete).
3852          */
3853         flush_scheduled_work();
3854
3855         mutex_lock(&disks_mutex);
3856         if (mddev->gendisk) {
3857                 mutex_unlock(&disks_mutex);
3858                 mddev_put(mddev);
3859                 return -EEXIST;
3860         }
3861
3862         if (name) {
3863                 /* Need to ensure that 'name' is not a duplicate.
3864                  */
3865                 mddev_t *mddev2;
3866                 spin_lock(&all_mddevs_lock);
3867
3868                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3869                         if (mddev2->gendisk &&
3870                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
3871                                 spin_unlock(&all_mddevs_lock);
3872                                 return -EEXIST;
3873                         }
3874                 spin_unlock(&all_mddevs_lock);
3875         }
3876
3877         mddev->queue = blk_alloc_queue(GFP_KERNEL);
3878         if (!mddev->queue) {
3879                 mutex_unlock(&disks_mutex);
3880                 mddev_put(mddev);
3881                 return -ENOMEM;
3882         }
3883         mddev->queue->queuedata = mddev;
3884
3885         /* Can be unlocked because the queue is new: no concurrency */
3886         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3887
3888         blk_queue_make_request(mddev->queue, md_make_request);
3889
3890         disk = alloc_disk(1 << shift);
3891         if (!disk) {
3892                 mutex_unlock(&disks_mutex);
3893                 blk_cleanup_queue(mddev->queue);
3894                 mddev->queue = NULL;
3895                 mddev_put(mddev);
3896                 return -ENOMEM;
3897         }
3898         disk->major = MAJOR(mddev->unit);
3899         disk->first_minor = unit << shift;
3900         if (name)
3901                 strcpy(disk->disk_name, name);
3902         else if (partitioned)
3903                 sprintf(disk->disk_name, "md_d%d", unit);
3904         else
3905                 sprintf(disk->disk_name, "md%d", unit);
3906         disk->fops = &md_fops;
3907         disk->private_data = mddev;
3908         disk->queue = mddev->queue;
3909         /* Allow extended partitions.  This makes the
3910          * 'mdp' device redundant, but we can't really
3911          * remove it now.
3912          */
3913         disk->flags |= GENHD_FL_EXT_DEVT;
3914         add_disk(disk);
3915         mddev->gendisk = disk;
3916         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3917                                      &disk_to_dev(disk)->kobj, "%s", "md");
3918         mutex_unlock(&disks_mutex);
3919         if (error)
3920                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3921                        disk->disk_name);
3922         else {
3923                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3924                 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3925         }
3926         mddev_put(mddev);
3927         return 0;
3928 }
3929
3930 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3931 {
3932         md_alloc(dev, NULL);
3933         return NULL;
3934 }
3935
3936 static int add_named_array(const char *val, struct kernel_param *kp)
3937 {
3938         /* val must be "md_*" where * is not all digits.
3939          * We allocate an array with a large free minor number, and
3940          * set the name to val.  val must not already be an active name.
3941          */
3942         int len = strlen(val);
3943         char buf[DISK_NAME_LEN];
3944
3945         while (len && val[len-1] == '\n')
3946                 len--;
3947         if (len >= DISK_NAME_LEN)
3948                 return -E2BIG;
3949         strlcpy(buf, val, len+1);
3950         if (strncmp(buf, "md_", 3) != 0)
3951                 return -EINVAL;
3952         return md_alloc(0, buf);
3953 }
3954
3955 static void md_safemode_timeout(unsigned long data)
3956 {
3957         mddev_t *mddev = (mddev_t *) data;
3958
3959         if (!atomic_read(&mddev->writes_pending)) {
3960                 mddev->safemode = 1;
3961                 if (mddev->external)
3962                         sysfs_notify_dirent(mddev->sysfs_state);
3963         }
3964         md_wakeup_thread(mddev->thread);
3965 }
3966
3967 static int start_dirty_degraded;
3968
3969 static int do_md_run(mddev_t * mddev)
3970 {
3971         int err;
3972         int chunk_size;
3973         mdk_rdev_t *rdev;
3974         struct gendisk *disk;
3975         struct mdk_personality *pers;
3976         char b[BDEVNAME_SIZE];
3977
3978         if (list_empty(&mddev->disks))
3979                 /* cannot run an array with no devices.. */
3980                 return -EINVAL;
3981
3982         if (mddev->pers)
3983                 return -EBUSY;
3984
3985         /*
3986          * Analyze all RAID superblock(s)
3987          */
3988         if (!mddev->raid_disks) {
3989                 if (!mddev->persistent)
3990                         return -EINVAL;
3991                 analyze_sbs(mddev);
3992         }
3993
3994         chunk_size = mddev->chunk_size;
3995
3996         if (chunk_size) {
3997                 if (chunk_size > MAX_CHUNK_SIZE) {
3998                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3999                                 chunk_size, MAX_CHUNK_SIZE);
4000                         return -EINVAL;
4001                 }
4002                 /*
4003                  * chunk-size has to be a power of 2
4004                  */
4005                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
4006                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
4007                         return -EINVAL;
4008                 }
4009
4010                 /* devices must have minimum size of one chunk */
4011                 list_for_each_entry(rdev, &mddev->disks, same_set) {
4012                         if (test_bit(Faulty, &rdev->flags))
4013                                 continue;
4014                         if (rdev->sectors < chunk_size / 512) {
4015                                 printk(KERN_WARNING
4016                                         "md: Dev %s smaller than chunk_size:"
4017                                         " %llu < %d\n",
4018                                         bdevname(rdev->bdev,b),
4019                                         (unsigned long long)rdev->sectors,
4020                                         chunk_size / 512);
4021                                 return -EINVAL;
4022                         }
4023                 }
4024         }
4025
4026         if (mddev->level != LEVEL_NONE)
4027                 request_module("md-level-%d", mddev->level);
4028         else if (mddev->clevel[0])
4029                 request_module("md-%s", mddev->clevel);
4030
4031         /*
4032          * Drop all container device buffers, from now on
4033          * the only valid external interface is through the md
4034          * device.
4035          */
4036         list_for_each_entry(rdev, &mddev->disks, same_set) {
4037                 if (test_bit(Faulty, &rdev->flags))
4038                         continue;
4039                 sync_blockdev(rdev->bdev);
4040                 invalidate_bdev(rdev->bdev);
4041
4042                 /* perform some consistency tests on the device.
4043                  * We don't want the data to overlap the metadata,
4044                  * Internal Bitmap issues have been handled elsewhere.
4045                  */
4046                 if (rdev->data_offset < rdev->sb_start) {
4047                         if (mddev->dev_sectors &&
4048                             rdev->data_offset + mddev->dev_sectors
4049                             > rdev->sb_start) {
4050                                 printk("md: %s: data overlaps metadata\n",
4051                                        mdname(mddev));
4052                                 return -EINVAL;
4053                         }
4054                 } else {
4055                         if (rdev->sb_start + rdev->sb_size/512
4056                             > rdev->data_offset) {
4057                                 printk("md: %s: metadata overlaps data\n",
4058                                        mdname(mddev));
4059                                 return -EINVAL;
4060                         }
4061                 }
4062                 sysfs_notify_dirent(rdev->sysfs_state);
4063         }
4064
4065         md_probe(mddev->unit, NULL, NULL);
4066         disk = mddev->gendisk;
4067         if (!disk)
4068                 return -ENOMEM;
4069
4070         spin_lock(&pers_lock);
4071         pers = find_pers(mddev->level, mddev->clevel);
4072         if (!pers || !try_module_get(pers->owner)) {
4073                 spin_unlock(&pers_lock);
4074                 if (mddev->level != LEVEL_NONE)
4075                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4076                                mddev->level);
4077                 else
4078                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4079                                mddev->clevel);
4080                 return -EINVAL;
4081         }
4082         mddev->pers = pers;
4083         spin_unlock(&pers_lock);
4084         if (mddev->level != pers->level) {
4085                 mddev->level = pers->level;
4086                 mddev->new_level = pers->level;
4087         }
4088         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4089
4090         if (pers->level >= 4 && pers->level <= 6)
4091                 /* Cannot support integrity (yet) */
4092                 blk_integrity_unregister(mddev->gendisk);
4093
4094         if (mddev->reshape_position != MaxSector &&
4095             pers->start_reshape == NULL) {
4096                 /* This personality cannot handle reshaping... */
4097                 mddev->pers = NULL;
4098                 module_put(pers->owner);
4099                 return -EINVAL;
4100         }
4101
4102         if (pers->sync_request) {
4103                 /* Warn if this is a potentially silly
4104                  * configuration.
4105                  */
4106                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4107                 mdk_rdev_t *rdev2;
4108                 int warned = 0;
4109
4110                 list_for_each_entry(rdev, &mddev->disks, same_set)
4111                         list_for_each_entry(rdev2, &mddev->disks, same_set) {
4112                                 if (rdev < rdev2 &&
4113                                     rdev->bdev->bd_contains ==
4114                                     rdev2->bdev->bd_contains) {
4115                                         printk(KERN_WARNING
4116                                                "%s: WARNING: %s appears to be"
4117                                                " on the same physical disk as"
4118                                                " %s.\n",
4119                                                mdname(mddev),
4120                                                bdevname(rdev->bdev,b),
4121                                                bdevname(rdev2->bdev,b2));
4122                                         warned = 1;
4123                                 }
4124                         }
4125
4126                 if (warned)
4127                         printk(KERN_WARNING
4128                                "True protection against single-disk"
4129                                " failure might be compromised.\n");
4130         }
4131
4132         mddev->recovery = 0;
4133         /* may be over-ridden by personality */
4134         mddev->resync_max_sectors = mddev->dev_sectors;
4135
4136         mddev->barriers_work = 1;
4137         mddev->ok_start_degraded = start_dirty_degraded;
4138
4139         if (start_readonly)
4140                 mddev->ro = 2; /* read-only, but switch on first write */
4141
4142         err = mddev->pers->run(mddev);
4143         if (err)
4144                 printk(KERN_ERR "md: pers->run() failed ...\n");
4145         else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4146                 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4147                           " but 'external_size' not in effect?\n", __func__);
4148                 printk(KERN_ERR
4149                        "md: invalid array_size %llu > default size %llu\n",
4150                        (unsigned long long)mddev->array_sectors / 2,
4151                        (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4152                 err = -EINVAL;
4153                 mddev->pers->stop(mddev);
4154         }
4155         if (err == 0 && mddev->pers->sync_request) {
4156                 err = bitmap_create(mddev);
4157                 if (err) {
4158                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4159                                mdname(mddev), err);
4160                         mddev->pers->stop(mddev);
4161                 }
4162         }
4163         if (err) {
4164                 module_put(mddev->pers->owner);
4165                 mddev->pers = NULL;
4166                 bitmap_destroy(mddev);
4167                 return err;
4168         }
4169         if (mddev->pers->sync_request) {
4170                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4171                         printk(KERN_WARNING
4172                                "md: cannot register extra attributes for %s\n",
4173                                mdname(mddev));
4174                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4175         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4176                 mddev->ro = 0;
4177
4178         atomic_set(&mddev->writes_pending,0);
4179         mddev->safemode = 0;
4180         mddev->safemode_timer.function = md_safemode_timeout;
4181         mddev->safemode_timer.data = (unsigned long) mddev;
4182         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4183         mddev->in_sync = 1;
4184
4185         list_for_each_entry(rdev, &mddev->disks, same_set)
4186                 if (rdev->raid_disk >= 0) {
4187                         char nm[20];
4188                         sprintf(nm, "rd%d", rdev->raid_disk);
4189                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4190                                 printk("md: cannot register %s for %s\n",
4191                                        nm, mdname(mddev));
4192                 }
4193         
4194         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4195         
4196         if (mddev->flags)
4197                 md_update_sb(mddev, 0);
4198
4199         set_capacity(disk, mddev->array_sectors);
4200
4201         /* If there is a partially-recovered drive we need to
4202          * start recovery here.  If we leave it to md_check_recovery,
4203          * it will remove the drives and not do the right thing
4204          */
4205         if (mddev->degraded && !mddev->sync_thread) {
4206                 int spares = 0;
4207                 list_for_each_entry(rdev, &mddev->disks, same_set)
4208                         if (rdev->raid_disk >= 0 &&
4209                             !test_bit(In_sync, &rdev->flags) &&
4210                             !test_bit(Faulty, &rdev->flags))
4211                                 /* complete an interrupted recovery */
4212                                 spares++;
4213                 if (spares && mddev->pers->sync_request) {
4214                         mddev->recovery = 0;
4215                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4216                         mddev->sync_thread = md_register_thread(md_do_sync,
4217                                                                 mddev,
4218                                                                 "%s_resync");
4219                         if (!mddev->sync_thread) {
4220                                 printk(KERN_ERR "%s: could not start resync"
4221                                        " thread...\n",
4222                                        mdname(mddev));
4223                                 /* leave the spares where they are, it shouldn't hurt */
4224                                 mddev->recovery = 0;
4225                         }
4226                 }
4227         }
4228         md_wakeup_thread(mddev->thread);
4229         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4230
4231         mddev->changed = 1;
4232         md_new_event(mddev);
4233         sysfs_notify_dirent(mddev->sysfs_state);
4234         if (mddev->sysfs_action)
4235                 sysfs_notify_dirent(mddev->sysfs_action);
4236         sysfs_notify(&mddev->kobj, NULL, "degraded");
4237         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4238         return 0;
4239 }
4240
4241 static int restart_array(mddev_t *mddev)
4242 {
4243         struct gendisk *disk = mddev->gendisk;
4244
4245         /* Complain if it has no devices */
4246         if (list_empty(&mddev->disks))
4247                 return -ENXIO;
4248         if (!mddev->pers)
4249                 return -EINVAL;
4250         if (!mddev->ro)
4251                 return -EBUSY;
4252         mddev->safemode = 0;
4253         mddev->ro = 0;
4254         set_disk_ro(disk, 0);
4255         printk(KERN_INFO "md: %s switched to read-write mode.\n",
4256                 mdname(mddev));
4257         /* Kick recovery or resync if necessary */
4258         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4259         md_wakeup_thread(mddev->thread);
4260         md_wakeup_thread(mddev->sync_thread);
4261         sysfs_notify_dirent(mddev->sysfs_state);
4262         return 0;
4263 }
4264
4265 /* similar to deny_write_access, but accounts for our holding a reference
4266  * to the file ourselves */
4267 static int deny_bitmap_write_access(struct file * file)
4268 {
4269         struct inode *inode = file->f_mapping->host;
4270
4271         spin_lock(&inode->i_lock);
4272         if (atomic_read(&inode->i_writecount) > 1) {
4273                 spin_unlock(&inode->i_lock);
4274                 return -ETXTBSY;
4275         }
4276         atomic_set(&inode->i_writecount, -1);
4277         spin_unlock(&inode->i_lock);
4278
4279         return 0;
4280 }
4281
4282 static void restore_bitmap_write_access(struct file *file)
4283 {
4284         struct inode *inode = file->f_mapping->host;
4285
4286         spin_lock(&inode->i_lock);
4287         atomic_set(&inode->i_writecount, 1);
4288         spin_unlock(&inode->i_lock);
4289 }
4290
4291 /* mode:
4292  *   0 - completely stop and dis-assemble array
4293  *   1 - switch to readonly
4294  *   2 - stop but do not disassemble array
4295  */
4296 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4297 {
4298         int err = 0;
4299         struct gendisk *disk = mddev->gendisk;
4300
4301         if (atomic_read(&mddev->openers) > is_open) {
4302                 printk("md: %s still in use.\n",mdname(mddev));
4303                 return -EBUSY;
4304         }
4305
4306         if (mddev->pers) {
4307
4308                 if (mddev->sync_thread) {
4309                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4310                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4311                         md_unregister_thread(mddev->sync_thread);
4312                         mddev->sync_thread = NULL;
4313                 }
4314
4315                 del_timer_sync(&mddev->safemode_timer);
4316
4317                 switch(mode) {
4318                 case 1: /* readonly */
4319                         err  = -ENXIO;
4320                         if (mddev->ro==1)
4321                                 goto out;
4322                         mddev->ro = 1;
4323                         break;
4324                 case 0: /* disassemble */
4325                 case 2: /* stop */
4326                         bitmap_flush(mddev);
4327                         md_super_wait(mddev);
4328                         if (mddev->ro)
4329                                 set_disk_ro(disk, 0);
4330
4331                         mddev->pers->stop(mddev);
4332                         mddev->queue->merge_bvec_fn = NULL;
4333                         mddev->queue->unplug_fn = NULL;
4334                         mddev->queue->backing_dev_info.congested_fn = NULL;
4335                         module_put(mddev->pers->owner);
4336                         if (mddev->pers->sync_request)
4337                                 mddev->private = &md_redundancy_group;
4338                         mddev->pers = NULL;
4339                         /* tell userspace to handle 'inactive' */
4340                         sysfs_notify_dirent(mddev->sysfs_state);
4341
4342                         set_capacity(disk, 0);
4343                         mddev->changed = 1;
4344
4345                         if (mddev->ro)
4346                                 mddev->ro = 0;
4347                 }
4348                 if (!mddev->in_sync || mddev->flags) {
4349                         /* mark array as shutdown cleanly */
4350                         mddev->in_sync = 1;
4351                         md_update_sb(mddev, 1);
4352                 }
4353                 if (mode == 1)
4354                         set_disk_ro(disk, 1);
4355                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4356         }
4357
4358         /*
4359          * Free resources if final stop
4360          */
4361         if (mode == 0) {
4362                 mdk_rdev_t *rdev;
4363
4364                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4365
4366                 bitmap_destroy(mddev);
4367                 if (mddev->bitmap_file) {
4368                         restore_bitmap_write_access(mddev->bitmap_file);
4369                         fput(mddev->bitmap_file);
4370                         mddev->bitmap_file = NULL;
4371                 }
4372                 mddev->bitmap_offset = 0;
4373
4374                 list_for_each_entry(rdev, &mddev->disks, same_set)
4375                         if (rdev->raid_disk >= 0) {
4376                                 char nm[20];
4377                                 sprintf(nm, "rd%d", rdev->raid_disk);
4378                                 sysfs_remove_link(&mddev->kobj, nm);
4379                         }
4380
4381                 /* make sure all md_delayed_delete calls have finished */
4382                 flush_scheduled_work();
4383
4384                 export_array(mddev);
4385
4386                 mddev->array_sectors = 0;
4387                 mddev->external_size = 0;
4388                 mddev->dev_sectors = 0;
4389                 mddev->raid_disks = 0;
4390                 mddev->recovery_cp = 0;
4391                 mddev->resync_min = 0;
4392                 mddev->resync_max = MaxSector;
4393                 mddev->reshape_position = MaxSector;
4394                 mddev->external = 0;
4395                 mddev->persistent = 0;
4396                 mddev->level = LEVEL_NONE;
4397                 mddev->clevel[0] = 0;
4398                 mddev->flags = 0;
4399                 mddev->ro = 0;
4400                 mddev->metadata_type[0] = 0;
4401                 mddev->chunk_size = 0;
4402                 mddev->ctime = mddev->utime = 0;
4403                 mddev->layout = 0;
4404                 mddev->max_disks = 0;
4405                 mddev->events = 0;
4406                 mddev->delta_disks = 0;
4407                 mddev->new_level = LEVEL_NONE;
4408                 mddev->new_layout = 0;
4409                 mddev->new_chunk = 0;
4410                 mddev->curr_resync = 0;
4411                 mddev->resync_mismatches = 0;
4412                 mddev->suspend_lo = mddev->suspend_hi = 0;
4413                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4414                 mddev->recovery = 0;
4415                 mddev->in_sync = 0;
4416                 mddev->changed = 0;
4417                 mddev->degraded = 0;
4418                 mddev->barriers_work = 0;
4419                 mddev->safemode = 0;
4420                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4421                 if (mddev->hold_active == UNTIL_STOP)
4422                         mddev->hold_active = 0;
4423
4424         } else if (mddev->pers)
4425                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4426                         mdname(mddev));
4427         err = 0;
4428         blk_integrity_unregister(disk);
4429         md_new_event(mddev);
4430         sysfs_notify_dirent(mddev->sysfs_state);
4431 out:
4432         return err;
4433 }
4434
4435 #ifndef MODULE
4436 static void autorun_array(mddev_t *mddev)
4437 {
4438         mdk_rdev_t *rdev;
4439         int err;
4440
4441         if (list_empty(&mddev->disks))
4442                 return;
4443
4444         printk(KERN_INFO "md: running: ");
4445
4446         list_for_each_entry(rdev, &mddev->disks, same_set) {
4447                 char b[BDEVNAME_SIZE];
4448                 printk("<%s>", bdevname(rdev->bdev,b));
4449         }
4450         printk("\n");
4451
4452         err = do_md_run(mddev);
4453         if (err) {
4454                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4455                 do_md_stop(mddev, 0, 0);
4456         }
4457 }
4458
4459 /*
4460  * lets try to run arrays based on all disks that have arrived
4461  * until now. (those are in pending_raid_disks)
4462  *
4463  * the method: pick the first pending disk, collect all disks with
4464  * the same UUID, remove all from the pending list and put them into
4465  * the 'same_array' list. Then order this list based on superblock
4466  * update time (freshest comes first), kick out 'old' disks and
4467  * compare superblocks. If everything's fine then run it.
4468  *
4469  * If "unit" is allocated, then bump its reference count
4470  */
4471 static void autorun_devices(int part)
4472 {
4473         mdk_rdev_t *rdev0, *rdev, *tmp;
4474         mddev_t *mddev;
4475         char b[BDEVNAME_SIZE];
4476
4477         printk(KERN_INFO "md: autorun ...\n");
4478         while (!list_empty(&pending_raid_disks)) {
4479                 int unit;
4480                 dev_t dev;
4481                 LIST_HEAD(candidates);
4482                 rdev0 = list_entry(pending_raid_disks.next,
4483                                          mdk_rdev_t, same_set);
4484
4485                 printk(KERN_INFO "md: considering %s ...\n",
4486                         bdevname(rdev0->bdev,b));
4487                 INIT_LIST_HEAD(&candidates);
4488                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4489                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4490                                 printk(KERN_INFO "md:  adding %s ...\n",
4491                                         bdevname(rdev->bdev,b));
4492                                 list_move(&rdev->same_set, &candidates);
4493                         }
4494                 /*
4495                  * now we have a set of devices, with all of them having
4496                  * mostly sane superblocks. It's time to allocate the
4497                  * mddev.
4498                  */
4499                 if (part) {
4500                         dev = MKDEV(mdp_major,
4501                                     rdev0->preferred_minor << MdpMinorShift);
4502                         unit = MINOR(dev) >> MdpMinorShift;
4503                 } else {
4504                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4505                         unit = MINOR(dev);
4506                 }
4507                 if (rdev0->preferred_minor != unit) {
4508                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4509                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4510                         break;
4511                 }
4512
4513                 md_probe(dev, NULL, NULL);
4514                 mddev = mddev_find(dev);
4515                 if (!mddev || !mddev->gendisk) {
4516                         if (mddev)
4517                                 mddev_put(mddev);
4518                         printk(KERN_ERR
4519                                 "md: cannot allocate memory for md drive.\n");
4520                         break;
4521                 }
4522                 if (mddev_lock(mddev)) 
4523                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4524                                mdname(mddev));
4525                 else if (mddev->raid_disks || mddev->major_version
4526                          || !list_empty(&mddev->disks)) {
4527                         printk(KERN_WARNING 
4528                                 "md: %s already running, cannot run %s\n",
4529                                 mdname(mddev), bdevname(rdev0->bdev,b));
4530                         mddev_unlock(mddev);
4531                 } else {
4532                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4533                         mddev->persistent = 1;
4534                         rdev_for_each_list(rdev, tmp, &candidates) {
4535                                 list_del_init(&rdev->same_set);
4536                                 if (bind_rdev_to_array(rdev, mddev))
4537                                         export_rdev(rdev);
4538                         }
4539                         autorun_array(mddev);
4540                         mddev_unlock(mddev);
4541                 }
4542                 /* on success, candidates will be empty, on error
4543                  * it won't...
4544                  */
4545                 rdev_for_each_list(rdev, tmp, &candidates) {
4546                         list_del_init(&rdev->same_set);
4547                         export_rdev(rdev);
4548                 }
4549                 mddev_put(mddev);
4550         }
4551         printk(KERN_INFO "md: ... autorun DONE.\n");
4552 }
4553 #endif /* !MODULE */
4554
4555 static int get_version(void __user * arg)
4556 {
4557         mdu_version_t ver;
4558
4559         ver.major = MD_MAJOR_VERSION;
4560         ver.minor = MD_MINOR_VERSION;
4561         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4562
4563         if (copy_to_user(arg, &ver, sizeof(ver)))
4564                 return -EFAULT;
4565
4566         return 0;
4567 }
4568
4569 static int get_array_info(mddev_t * mddev, void __user * arg)
4570 {
4571         mdu_array_info_t info;
4572         int nr,working,active,failed,spare;
4573         mdk_rdev_t *rdev;
4574
4575         nr=working=active=failed=spare=0;
4576         list_for_each_entry(rdev, &mddev->disks, same_set) {
4577                 nr++;
4578                 if (test_bit(Faulty, &rdev->flags))
4579                         failed++;
4580                 else {
4581                         working++;
4582                         if (test_bit(In_sync, &rdev->flags))
4583                                 active++;       
4584                         else
4585                                 spare++;
4586                 }
4587         }
4588
4589         info.major_version = mddev->major_version;
4590         info.minor_version = mddev->minor_version;
4591         info.patch_version = MD_PATCHLEVEL_VERSION;
4592         info.ctime         = mddev->ctime;
4593         info.level         = mddev->level;
4594         info.size          = mddev->dev_sectors / 2;
4595         if (info.size != mddev->dev_sectors / 2) /* overflow */
4596                 info.size = -1;
4597         info.nr_disks      = nr;
4598         info.raid_disks    = mddev->raid_disks;
4599         info.md_minor      = mddev->md_minor;
4600         info.not_persistent= !mddev->persistent;
4601
4602         info.utime         = mddev->utime;
4603         info.state         = 0;
4604         if (mddev->in_sync)
4605                 info.state = (1<<MD_SB_CLEAN);
4606         if (mddev->bitmap && mddev->bitmap_offset)
4607                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4608         info.active_disks  = active;
4609         info.working_disks = working;
4610         info.failed_disks  = failed;
4611         info.spare_disks   = spare;
4612
4613         info.layout        = mddev->layout;
4614         info.chunk_size    = mddev->chunk_size;
4615
4616         if (copy_to_user(arg, &info, sizeof(info)))
4617                 return -EFAULT;
4618
4619         return 0;
4620 }
4621
4622 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4623 {
4624         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4625         char *ptr, *buf = NULL;
4626         int err = -ENOMEM;
4627
4628         if (md_allow_write(mddev))
4629                 file = kmalloc(sizeof(*file), GFP_NOIO);
4630         else
4631                 file = kmalloc(sizeof(*file), GFP_KERNEL);
4632
4633         if (!file)
4634                 goto out;
4635
4636         /* bitmap disabled, zero the first byte and copy out */
4637         if (!mddev->bitmap || !mddev->bitmap->file) {
4638                 file->pathname[0] = '\0';
4639                 goto copy_out;
4640         }
4641
4642         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4643         if (!buf)
4644                 goto out;
4645
4646         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4647         if (IS_ERR(ptr))
4648                 goto out;
4649
4650         strcpy(file->pathname, ptr);
4651
4652 copy_out:
4653         err = 0;
4654         if (copy_to_user(arg, file, sizeof(*file)))
4655                 err = -EFAULT;
4656 out:
4657         kfree(buf);
4658         kfree(file);
4659         return err;
4660 }
4661
4662 static int get_disk_info(mddev_t * mddev, void __user * arg)
4663 {
4664         mdu_disk_info_t info;
4665         mdk_rdev_t *rdev;
4666
4667         if (copy_from_user(&info, arg, sizeof(info)))
4668                 return -EFAULT;
4669
4670         rdev = find_rdev_nr(mddev, info.number);
4671         if (rdev) {
4672                 info.major = MAJOR(rdev->bdev->bd_dev);
4673                 info.minor = MINOR(rdev->bdev->bd_dev);
4674                 info.raid_disk = rdev->raid_disk;
4675                 info.state = 0;
4676                 if (test_bit(Faulty, &rdev->flags))
4677                         info.state |= (1<<MD_DISK_FAULTY);
4678                 else if (test_bit(In_sync, &rdev->flags)) {
4679                         info.state |= (1<<MD_DISK_ACTIVE);
4680                         info.state |= (1<<MD_DISK_SYNC);
4681                 }
4682                 if (test_bit(WriteMostly, &rdev->flags))
4683                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4684         } else {
4685                 info.major = info.minor = 0;
4686                 info.raid_disk = -1;
4687                 info.state = (1<<MD_DISK_REMOVED);
4688         }
4689
4690         if (copy_to_user(arg, &info, sizeof(info)))
4691                 return -EFAULT;
4692
4693         return 0;
4694 }
4695
4696 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4697 {
4698         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4699         mdk_rdev_t *rdev;
4700         dev_t dev = MKDEV(info->major,info->minor);
4701
4702         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4703                 return -EOVERFLOW;
4704
4705         if (!mddev->raid_disks) {
4706                 int err;
4707                 /* expecting a device which has a superblock */
4708                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4709                 if (IS_ERR(rdev)) {
4710                         printk(KERN_WARNING 
4711                                 "md: md_import_device returned %ld\n",
4712                                 PTR_ERR(rdev));
4713                         return PTR_ERR(rdev);
4714                 }
4715                 if (!list_empty(&mddev->disks)) {
4716                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4717                                                         mdk_rdev_t, same_set);
4718                         int err = super_types[mddev->major_version]
4719                                 .load_super(rdev, rdev0, mddev->minor_version);
4720                         if (err < 0) {
4721                                 printk(KERN_WARNING 
4722                                         "md: %s has different UUID to %s\n",
4723                                         bdevname(rdev->bdev,b), 
4724                                         bdevname(rdev0->bdev,b2));
4725                                 export_rdev(rdev);
4726                                 return -EINVAL;
4727                         }
4728                 }
4729                 err = bind_rdev_to_array(rdev, mddev);
4730                 if (err)
4731                         export_rdev(rdev);
4732                 return err;
4733         }
4734
4735         /*
4736          * add_new_disk can be used once the array is assembled
4737          * to add "hot spares".  They must already have a superblock
4738          * written
4739          */
4740         if (mddev->pers) {
4741                 int err;
4742                 if (!mddev->pers->hot_add_disk) {
4743                         printk(KERN_WARNING 
4744                                 "%s: personality does not support diskops!\n",
4745                                mdname(mddev));
4746                         return -EINVAL;
4747                 }
4748                 if (mddev->persistent)
4749                         rdev = md_import_device(dev, mddev->major_version,
4750                                                 mddev->minor_version);
4751                 else
4752                         rdev = md_import_device(dev, -1, -1);
4753                 if (IS_ERR(rdev)) {
4754                         printk(KERN_WARNING 
4755                                 "md: md_import_device returned %ld\n",
4756                                 PTR_ERR(rdev));
4757                         return PTR_ERR(rdev);
4758                 }
4759                 /* set save_raid_disk if appropriate */
4760                 if (!mddev->persistent) {
4761                         if (info->state & (1<<MD_DISK_SYNC)  &&
4762                             info->raid_disk < mddev->raid_disks)
4763                                 rdev->raid_disk = info->raid_disk;
4764                         else
4765                                 rdev->raid_disk = -1;
4766                 } else
4767                         super_types[mddev->major_version].
4768                                 validate_super(mddev, rdev);
4769                 rdev->saved_raid_disk = rdev->raid_disk;
4770
4771                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4772                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4773                         set_bit(WriteMostly, &rdev->flags);
4774                 else
4775                         clear_bit(WriteMostly, &rdev->flags);
4776
4777                 rdev->raid_disk = -1;
4778                 err = bind_rdev_to_array(rdev, mddev);
4779                 if (!err && !mddev->pers->hot_remove_disk) {
4780                         /* If there is hot_add_disk but no hot_remove_disk
4781                          * then added disks for geometry changes,
4782                          * and should be added immediately.
4783                          */
4784                         super_types[mddev->major_version].
4785                                 validate_super(mddev, rdev);
4786                         err = mddev->pers->hot_add_disk(mddev, rdev);
4787                         if (err)
4788                                 unbind_rdev_from_array(rdev);
4789                 }
4790                 if (err)
4791                         export_rdev(rdev);
4792                 else
4793                         sysfs_notify_dirent(rdev->sysfs_state);
4794
4795                 md_update_sb(mddev, 1);
4796                 if (mddev->degraded)
4797                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4798                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4799                 md_wakeup_thread(mddev->thread);
4800                 return err;
4801         }
4802
4803         /* otherwise, add_new_disk is only allowed
4804          * for major_version==0 superblocks
4805          */
4806         if (mddev->major_version != 0) {
4807                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4808                        mdname(mddev));
4809                 return -EINVAL;
4810         }
4811
4812         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4813                 int err;
4814                 rdev = md_import_device(dev, -1, 0);
4815                 if (IS_ERR(rdev)) {
4816                         printk(KERN_WARNING 
4817                                 "md: error, md_import_device() returned %ld\n",
4818                                 PTR_ERR(rdev));
4819                         return PTR_ERR(rdev);
4820                 }
4821                 rdev->desc_nr = info->number;
4822                 if (info->raid_disk < mddev->raid_disks)
4823                         rdev->raid_disk = info->raid_disk;
4824                 else
4825                         rdev->raid_disk = -1;
4826
4827                 if (rdev->raid_disk < mddev->raid_disks)
4828                         if (info->state & (1<<MD_DISK_SYNC))
4829                                 set_bit(In_sync, &rdev->flags);
4830
4831                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4832                         set_bit(WriteMostly, &rdev->flags);
4833
4834                 if (!mddev->persistent) {
4835                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4836                         rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4837                 } else 
4838                         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4839                 rdev->sectors = calc_num_sectors(rdev, mddev->chunk_size);
4840
4841                 err = bind_rdev_to_array(rdev, mddev);
4842                 if (err) {
4843                         export_rdev(rdev);
4844                         return err;
4845                 }
4846         }
4847
4848         return 0;
4849 }
4850
4851 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4852 {
4853         char b[BDEVNAME_SIZE];
4854         mdk_rdev_t *rdev;
4855
4856         rdev = find_rdev(mddev, dev);
4857         if (!rdev)
4858                 return -ENXIO;
4859
4860         if (rdev->raid_disk >= 0)
4861                 goto busy;
4862
4863         kick_rdev_from_array(rdev);
4864         md_update_sb(mddev, 1);
4865         md_new_event(mddev);
4866
4867         return 0;
4868 busy:
4869         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4870                 bdevname(rdev->bdev,b), mdname(mddev));
4871         return -EBUSY;
4872 }
4873
4874 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4875 {
4876         char b[BDEVNAME_SIZE];
4877         int err;
4878         mdk_rdev_t *rdev;
4879
4880         if (!mddev->pers)
4881                 return -ENODEV;
4882
4883         if (mddev->major_version != 0) {
4884                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4885                         " version-0 superblocks.\n",
4886                         mdname(mddev));
4887                 return -EINVAL;
4888         }
4889         if (!mddev->pers->hot_add_disk) {
4890                 printk(KERN_WARNING 
4891                         "%s: personality does not support diskops!\n",
4892                         mdname(mddev));
4893                 return -EINVAL;
4894         }
4895
4896         rdev = md_import_device(dev, -1, 0);
4897         if (IS_ERR(rdev)) {
4898                 printk(KERN_WARNING 
4899                         "md: error, md_import_device() returned %ld\n",
4900                         PTR_ERR(rdev));
4901                 return -EINVAL;
4902         }
4903
4904         if (mddev->persistent)
4905                 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4906         else
4907                 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4908
4909         rdev->sectors = calc_num_sectors(rdev, mddev->chunk_size);
4910
4911         if (test_bit(Faulty, &rdev->flags)) {
4912                 printk(KERN_WARNING 
4913                         "md: can not hot-add faulty %s disk to %s!\n",
4914                         bdevname(rdev->bdev,b), mdname(mddev));
4915                 err = -EINVAL;
4916                 goto abort_export;
4917         }
4918         clear_bit(In_sync, &rdev->flags);
4919         rdev->desc_nr = -1;
4920         rdev->saved_raid_disk = -1;
4921         err = bind_rdev_to_array(rdev, mddev);
4922         if (err)
4923                 goto abort_export;
4924
4925         /*
4926          * The rest should better be atomic, we can have disk failures
4927          * noticed in interrupt contexts ...
4928          */
4929
4930         rdev->raid_disk = -1;
4931
4932         md_update_sb(mddev, 1);
4933
4934         /*
4935          * Kick recovery, maybe this spare has to be added to the
4936          * array immediately.
4937          */
4938         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4939         md_wakeup_thread(mddev->thread);
4940         md_new_event(mddev);
4941         return 0;
4942
4943 abort_export:
4944         export_rdev(rdev);
4945         return err;
4946 }
4947
4948 static int set_bitmap_file(mddev_t *mddev, int fd)
4949 {
4950         int err;
4951
4952         if (mddev->pers) {
4953                 if (!mddev->pers->quiesce)
4954                         return -EBUSY;
4955                 if (mddev->recovery || mddev->sync_thread)
4956                         return -EBUSY;
4957                 /* we should be able to change the bitmap.. */
4958         }
4959
4960
4961         if (fd >= 0) {
4962                 if (mddev->bitmap)
4963                         return -EEXIST; /* cannot add when bitmap is present */
4964                 mddev->bitmap_file = fget(fd);
4965
4966                 if (mddev->bitmap_file == NULL) {
4967                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4968                                mdname(mddev));
4969                         return -EBADF;
4970                 }
4971
4972                 err = deny_bitmap_write_access(mddev->bitmap_file);
4973                 if (err) {
4974                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4975                                mdname(mddev));
4976                         fput(mddev->bitmap_file);
4977                         mddev->bitmap_file = NULL;
4978                         return err;
4979                 }
4980                 mddev->bitmap_offset = 0; /* file overrides offset */
4981         } else if (mddev->bitmap == NULL)
4982                 return -ENOENT; /* cannot remove what isn't there */
4983         err = 0;
4984         if (mddev->pers) {
4985                 mddev->pers->quiesce(mddev, 1);
4986                 if (fd >= 0)
4987                         err = bitmap_create(mddev);
4988                 if (fd < 0 || err) {
4989                         bitmap_destroy(mddev);
4990                         fd = -1; /* make sure to put the file */
4991                 }
4992                 mddev->pers->quiesce(mddev, 0);
4993         }
4994         if (fd < 0) {
4995                 if (mddev->bitmap_file) {
4996                         restore_bitmap_write_access(mddev->bitmap_file);
4997                         fput(mddev->bitmap_file);
4998                 }
4999                 mddev->bitmap_file = NULL;
5000         }
5001
5002         return err;
5003 }
5004
5005 /*
5006  * set_array_info is used two different ways
5007  * The original usage is when creating a new array.
5008  * In this usage, raid_disks is > 0 and it together with
5009  *  level, size, not_persistent,layout,chunksize determine the
5010  *  shape of the array.
5011  *  This will always create an array with a type-0.90.0 superblock.
5012  * The newer usage is when assembling an array.
5013  *  In this case raid_disks will be 0, and the major_version field is
5014  *  use to determine which style super-blocks are to be found on the devices.
5015  *  The minor and patch _version numbers are also kept incase the
5016  *  super_block handler wishes to interpret them.
5017  */
5018 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5019 {
5020
5021         if (info->raid_disks == 0) {
5022                 /* just setting version number for superblock loading */
5023                 if (info->major_version < 0 ||
5024                     info->major_version >= ARRAY_SIZE(super_types) ||
5025                     super_types[info->major_version].name == NULL) {
5026                         /* maybe try to auto-load a module? */
5027                         printk(KERN_INFO 
5028                                 "md: superblock version %d not known\n",
5029                                 info->major_version);
5030                         return -EINVAL;
5031                 }
5032                 mddev->major_version = info->major_version;
5033                 mddev->minor_version = info->minor_version;
5034                 mddev->patch_version = info->patch_version;
5035                 mddev->persistent = !info->not_persistent;
5036                 return 0;
5037         }
5038         mddev->major_version = MD_MAJOR_VERSION;
5039         mddev->minor_version = MD_MINOR_VERSION;
5040         mddev->patch_version = MD_PATCHLEVEL_VERSION;
5041         mddev->ctime         = get_seconds();
5042
5043         mddev->level         = info->level;
5044         mddev->clevel[0]     = 0;
5045         mddev->dev_sectors   = 2 * (sector_t)info->size;
5046         mddev->raid_disks    = info->raid_disks;
5047         /* don't set md_minor, it is determined by which /dev/md* was
5048          * openned
5049          */
5050         if (info->state & (1<<MD_SB_CLEAN))
5051                 mddev->recovery_cp = MaxSector;
5052         else
5053                 mddev->recovery_cp = 0;
5054         mddev->persistent    = ! info->not_persistent;
5055         mddev->external      = 0;
5056
5057         mddev->layout        = info->layout;
5058         mddev->chunk_size    = info->chunk_size;
5059
5060         mddev->max_disks     = MD_SB_DISKS;
5061
5062         if (mddev->persistent)
5063                 mddev->flags         = 0;
5064         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5065
5066         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5067         mddev->bitmap_offset = 0;
5068
5069         mddev->reshape_position = MaxSector;
5070
5071         /*
5072          * Generate a 128 bit UUID
5073          */
5074         get_random_bytes(mddev->uuid, 16);
5075
5076         mddev->new_level = mddev->level;
5077         mddev->new_chunk = mddev->chunk_size;
5078         mddev->new_layout = mddev->layout;
5079         mddev->delta_disks = 0;
5080
5081         return 0;
5082 }
5083
5084 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5085 {
5086         WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5087
5088         if (mddev->external_size)
5089                 return;
5090
5091         mddev->array_sectors = array_sectors;
5092 }
5093 EXPORT_SYMBOL(md_set_array_sectors);
5094
5095 static int update_size(mddev_t *mddev, sector_t num_sectors)
5096 {
5097         mdk_rdev_t *rdev;
5098         int rv;
5099         int fit = (num_sectors == 0);
5100
5101         if (mddev->pers->resize == NULL)
5102                 return -EINVAL;
5103         /* The "num_sectors" is the number of sectors of each device that
5104          * is used.  This can only make sense for arrays with redundancy.
5105          * linear and raid0 always use whatever space is available. We can only
5106          * consider changing this number if no resync or reconstruction is
5107          * happening, and if the new size is acceptable. It must fit before the
5108          * sb_start or, if that is <data_offset, it must fit before the size
5109          * of each device.  If num_sectors is zero, we find the largest size
5110          * that fits.
5111
5112          */
5113         if (mddev->sync_thread)
5114                 return -EBUSY;
5115         if (mddev->bitmap)
5116                 /* Sorry, cannot grow a bitmap yet, just remove it,
5117                  * grow, and re-add.
5118                  */
5119                 return -EBUSY;
5120         list_for_each_entry(rdev, &mddev->disks, same_set) {
5121                 sector_t avail = rdev->sectors;
5122
5123                 if (fit && (num_sectors == 0 || num_sectors > avail))
5124                         num_sectors = avail;
5125                 if (avail < num_sectors)
5126                         return -ENOSPC;
5127         }
5128         rv = mddev->pers->resize(mddev, num_sectors);
5129         if (!rv) {
5130                 struct block_device *bdev;
5131
5132                 bdev = bdget_disk(mddev->gendisk, 0);
5133                 if (bdev) {
5134                         mutex_lock(&bdev->bd_inode->i_mutex);
5135                         i_size_write(bdev->bd_inode,
5136                                      (loff_t)mddev->array_sectors << 9);
5137                         mutex_unlock(&bdev->bd_inode->i_mutex);
5138                         bdput(bdev);
5139                 }
5140         }
5141         return rv;
5142 }
5143
5144 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5145 {
5146         int rv;
5147         /* change the number of raid disks */
5148         if (mddev->pers->check_reshape == NULL)
5149                 return -EINVAL;
5150         if (raid_disks <= 0 ||
5151             raid_disks >= mddev->max_disks)
5152                 return -EINVAL;
5153         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5154                 return -EBUSY;
5155         mddev->delta_disks = raid_disks - mddev->raid_disks;
5156
5157         rv = mddev->pers->check_reshape(mddev);
5158         return rv;
5159 }
5160
5161
5162 /*
5163  * update_array_info is used to change the configuration of an
5164  * on-line array.
5165  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5166  * fields in the info are checked against the array.
5167  * Any differences that cannot be handled will cause an error.
5168  * Normally, only one change can be managed at a time.
5169  */
5170 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5171 {
5172         int rv = 0;
5173         int cnt = 0;
5174         int state = 0;
5175
5176         /* calculate expected state,ignoring low bits */
5177         if (mddev->bitmap && mddev->bitmap_offset)
5178                 state |= (1 << MD_SB_BITMAP_PRESENT);
5179
5180         if (mddev->major_version != info->major_version ||
5181             mddev->minor_version != info->minor_version ||
5182 /*          mddev->patch_version != info->patch_version || */
5183             mddev->ctime         != info->ctime         ||
5184             mddev->level         != info->level         ||
5185 /*          mddev->layout        != info->layout        || */
5186             !mddev->persistent   != info->not_persistent||
5187             mddev->chunk_size    != info->chunk_size    ||
5188             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5189             ((state^info->state) & 0xfffffe00)
5190                 )
5191                 return -EINVAL;
5192         /* Check there is only one change */
5193         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5194                 cnt++;
5195         if (mddev->raid_disks != info->raid_disks)
5196                 cnt++;
5197         if (mddev->layout != info->layout)
5198                 cnt++;
5199         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5200                 cnt++;
5201         if (cnt == 0)
5202                 return 0;
5203         if (cnt > 1)
5204                 return -EINVAL;
5205
5206         if (mddev->layout != info->layout) {
5207                 /* Change layout
5208                  * we don't need to do anything at the md level, the
5209                  * personality will take care of it all.
5210                  */
5211                 if (mddev->pers->reconfig == NULL)
5212                         return -EINVAL;
5213                 else
5214                         return mddev->pers->reconfig(mddev, info->layout, -1);
5215         }
5216         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5217                 rv = update_size(mddev, (sector_t)info->size * 2);
5218
5219         if (mddev->raid_disks    != info->raid_disks)
5220                 rv = update_raid_disks(mddev, info->raid_disks);
5221
5222         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5223                 if (mddev->pers->quiesce == NULL)
5224                         return -EINVAL;
5225                 if (mddev->recovery || mddev->sync_thread)
5226                         return -EBUSY;
5227                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5228                         /* add the bitmap */
5229                         if (mddev->bitmap)
5230                                 return -EEXIST;
5231                         if (mddev->default_bitmap_offset == 0)
5232                                 return -EINVAL;
5233                         mddev->bitmap_offset = mddev->default_bitmap_offset;
5234                         mddev->pers->quiesce(mddev, 1);
5235                         rv = bitmap_create(mddev);
5236                         if (rv)
5237                                 bitmap_destroy(mddev);
5238                         mddev->pers->quiesce(mddev, 0);
5239                 } else {
5240                         /* remove the bitmap */
5241                         if (!mddev->bitmap)
5242                                 return -ENOENT;
5243                         if (mddev->bitmap->file)
5244                                 return -EINVAL;
5245                         mddev->pers->quiesce(mddev, 1);
5246                         bitmap_destroy(mddev);
5247                         mddev->pers->quiesce(mddev, 0);
5248                         mddev->bitmap_offset = 0;
5249                 }
5250         }
5251         md_update_sb(mddev, 1);
5252         return rv;
5253 }
5254
5255 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5256 {
5257         mdk_rdev_t *rdev;
5258
5259         if (mddev->pers == NULL)
5260                 return -ENODEV;
5261
5262         rdev = find_rdev(mddev, dev);
5263         if (!rdev)
5264                 return -ENODEV;
5265
5266         md_error(mddev, rdev);
5267         return 0;
5268 }
5269
5270 /*
5271  * We have a problem here : there is no easy way to give a CHS
5272  * virtual geometry. We currently pretend that we have a 2 heads
5273  * 4 sectors (with a BIG number of cylinders...). This drives
5274  * dosfs just mad... ;-)
5275  */
5276 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5277 {
5278         mddev_t *mddev = bdev->bd_disk->private_data;
5279
5280         geo->heads = 2;
5281         geo->sectors = 4;
5282         geo->cylinders = get_capacity(mddev->gendisk) / 8;
5283         return 0;
5284 }
5285
5286 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5287                         unsigned int cmd, unsigned long arg)
5288 {
5289         int err = 0;
5290         void __user *argp = (void __user *)arg;
5291         mddev_t *mddev = NULL;
5292
5293         if (!capable(CAP_SYS_ADMIN))
5294                 return -EACCES;
5295
5296         /*
5297          * Commands dealing with the RAID driver but not any
5298          * particular array:
5299          */
5300         switch (cmd)
5301         {
5302                 case RAID_VERSION:
5303                         err = get_version(argp);
5304                         goto done;
5305
5306                 case PRINT_RAID_DEBUG:
5307                         err = 0;
5308                         md_print_devices();
5309                         goto done;
5310
5311 #ifndef MODULE
5312                 case RAID_AUTORUN:
5313                         err = 0;
5314                         autostart_arrays(arg);
5315                         goto done;
5316 #endif
5317                 default:;
5318         }
5319
5320         /*
5321          * Commands creating/starting a new array:
5322          */
5323
5324         mddev = bdev->bd_disk->private_data;
5325
5326         if (!mddev) {
5327                 BUG();
5328                 goto abort;
5329         }
5330
5331         err = mddev_lock(mddev);
5332         if (err) {
5333                 printk(KERN_INFO 
5334                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
5335                         err, cmd);
5336                 goto abort;
5337         }
5338
5339         switch (cmd)
5340         {
5341                 case SET_ARRAY_INFO:
5342                         {
5343                                 mdu_array_info_t info;
5344                                 if (!arg)
5345                                         memset(&info, 0, sizeof(info));
5346                                 else if (copy_from_user(&info, argp, sizeof(info))) {
5347                                         err = -EFAULT;
5348                                         goto abort_unlock;
5349                                 }
5350                                 if (mddev->pers) {
5351                                         err = update_array_info(mddev, &info);
5352                                         if (err) {
5353                                                 printk(KERN_WARNING "md: couldn't update"
5354                                                        " array info. %d\n", err);
5355                                                 goto abort_unlock;
5356                                         }
5357                                         goto done_unlock;
5358                                 }
5359                                 if (!list_empty(&mddev->disks)) {
5360                                         printk(KERN_WARNING
5361                                                "md: array %s already has disks!\n",
5362                                                mdname(mddev));
5363                                         err = -EBUSY;
5364                                         goto abort_unlock;
5365                                 }
5366                                 if (mddev->raid_disks) {
5367                                         printk(KERN_WARNING
5368                                                "md: array %s already initialised!\n",
5369                                                mdname(mddev));
5370                                         err = -EBUSY;
5371                                         goto abort_unlock;
5372                                 }
5373                                 err = set_array_info(mddev, &info);
5374                                 if (err) {
5375                                         printk(KERN_WARNING "md: couldn't set"
5376                                                " array info. %d\n", err);
5377                                         goto abort_unlock;
5378                                 }
5379                         }
5380                         goto done_unlock;
5381
5382                 default:;
5383         }
5384
5385         /*
5386          * Commands querying/configuring an existing array:
5387          */
5388         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5389          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5390         if ((!mddev->raid_disks && !mddev->external)
5391             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5392             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5393             && cmd != GET_BITMAP_FILE) {
5394                 err = -ENODEV;
5395                 goto abort_unlock;
5396         }
5397
5398         /*
5399          * Commands even a read-only array can execute:
5400          */
5401         switch (cmd)
5402         {
5403                 case GET_ARRAY_INFO:
5404                         err = get_array_info(mddev, argp);
5405                         goto done_unlock;
5406
5407                 case GET_BITMAP_FILE:
5408                         err = get_bitmap_file(mddev, argp);
5409                         goto done_unlock;
5410
5411                 case GET_DISK_INFO:
5412                         err = get_disk_info(mddev, argp);
5413                         goto done_unlock;
5414
5415                 case RESTART_ARRAY_RW:
5416                         err = restart_array(mddev);
5417                         goto done_unlock;
5418
5419                 case STOP_ARRAY:
5420                         err = do_md_stop(mddev, 0, 1);
5421                         goto done_unlock;
5422
5423                 case STOP_ARRAY_RO:
5424                         err = do_md_stop(mddev, 1, 1);
5425                         goto done_unlock;
5426
5427         }
5428
5429         /*
5430          * The remaining ioctls are changing the state of the
5431          * superblock, so we do not allow them on read-only arrays.
5432          * However non-MD ioctls (e.g. get-size) will still come through
5433          * here and hit the 'default' below, so only disallow
5434          * 'md' ioctls, and switch to rw mode if started auto-readonly.
5435          */
5436         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5437                 if (mddev->ro == 2) {
5438                         mddev->ro = 0;
5439                         sysfs_notify_dirent(mddev->sysfs_state);
5440                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5441                         md_wakeup_thread(mddev->thread);
5442                 } else {
5443                         err = -EROFS;
5444                         goto abort_unlock;
5445                 }
5446         }
5447
5448         switch (cmd)
5449         {
5450                 case ADD_NEW_DISK:
5451                 {
5452                         mdu_disk_info_t info;
5453                         if (copy_from_user(&info, argp, sizeof(info)))
5454                                 err = -EFAULT;
5455                         else
5456                                 err = add_new_disk(mddev, &info);
5457                         goto done_unlock;
5458                 }
5459
5460                 case HOT_REMOVE_DISK:
5461                         err = hot_remove_disk(mddev, new_decode_dev(arg));
5462                         goto done_unlock;
5463
5464                 case HOT_ADD_DISK:
5465                         err = hot_add_disk(mddev, new_decode_dev(arg));
5466                         goto done_unlock;
5467
5468                 case SET_DISK_FAULTY:
5469                         err = set_disk_faulty(mddev, new_decode_dev(arg));
5470                         goto done_unlock;
5471
5472                 case RUN_ARRAY:
5473                         err = do_md_run(mddev);
5474                         goto done_unlock;
5475
5476                 case SET_BITMAP_FILE:
5477                         err = set_bitmap_file(mddev, (int)arg);
5478                         goto done_unlock;
5479
5480                 default:
5481                         err = -EINVAL;
5482                         goto abort_unlock;
5483         }
5484
5485 done_unlock:
5486 abort_unlock:
5487         if (mddev->hold_active == UNTIL_IOCTL &&
5488             err != -EINVAL)
5489                 mddev->hold_active = 0;
5490         mddev_unlock(mddev);
5491
5492         return err;
5493 done:
5494         if (err)
5495                 MD_BUG();
5496 abort:
5497         return err;
5498 }
5499
5500 static int md_open(struct block_device *bdev, fmode_t mode)
5501 {
5502         /*
5503          * Succeed if we can lock the mddev, which confirms that
5504          * it isn't being stopped right now.
5505          */
5506         mddev_t *mddev = mddev_find(bdev->bd_dev);
5507         int err;
5508
5509         if (mddev->gendisk != bdev->bd_disk) {
5510                 /* we are racing with mddev_put which is discarding this
5511                  * bd_disk.
5512                  */
5513                 mddev_put(mddev);
5514                 /* Wait until bdev->bd_disk is definitely gone */
5515                 flush_scheduled_work();
5516                 /* Then retry the open from the top */
5517                 return -ERESTARTSYS;
5518         }
5519         BUG_ON(mddev != bdev->bd_disk->private_data);
5520
5521         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5522                 goto out;
5523
5524         err = 0;
5525         atomic_inc(&mddev->openers);
5526         mddev_unlock(mddev);
5527
5528         check_disk_change(bdev);
5529  out:
5530         return err;
5531 }
5532
5533 static int md_release(struct gendisk *disk, fmode_t mode)
5534 {
5535         mddev_t *mddev = disk->private_data;
5536
5537         BUG_ON(!mddev);
5538         atomic_dec(&mddev->openers);
5539         mddev_put(mddev);
5540
5541         return 0;
5542 }
5543
5544 static int md_media_changed(struct gendisk *disk)
5545 {
5546         mddev_t *mddev = disk->private_data;
5547
5548         return mddev->changed;
5549 }
5550
5551 static int md_revalidate(struct gendisk *disk)
5552 {
5553         mddev_t *mddev = disk->private_data;
5554
5555         mddev->changed = 0;
5556         return 0;
5557 }
5558 static struct block_device_operations md_fops =
5559 {
5560         .owner          = THIS_MODULE,
5561         .open           = md_open,
5562         .release        = md_release,
5563         .locked_ioctl   = md_ioctl,
5564         .getgeo         = md_getgeo,
5565         .media_changed  = md_media_changed,
5566         .revalidate_disk= md_revalidate,
5567 };
5568
5569 static int md_thread(void * arg)
5570 {
5571         mdk_thread_t *thread = arg;
5572
5573         /*
5574          * md_thread is a 'system-thread', it's priority should be very
5575          * high. We avoid resource deadlocks individually in each
5576          * raid personality. (RAID5 does preallocation) We also use RR and
5577          * the very same RT priority as kswapd, thus we will never get
5578          * into a priority inversion deadlock.
5579          *
5580          * we definitely have to have equal or higher priority than
5581          * bdflush, otherwise bdflush will deadlock if there are too
5582          * many dirty RAID5 blocks.
5583          */
5584
5585         allow_signal(SIGKILL);
5586         while (!kthread_should_stop()) {
5587
5588                 /* We need to wait INTERRUPTIBLE so that
5589                  * we don't add to the load-average.
5590                  * That means we need to be sure no signals are
5591                  * pending
5592                  */
5593                 if (signal_pending(current))
5594                         flush_signals(current);
5595
5596                 wait_event_interruptible_timeout
5597                         (thread->wqueue,
5598                          test_bit(THREAD_WAKEUP, &thread->flags)
5599                          || kthread_should_stop(),
5600                          thread->timeout);
5601
5602                 clear_bit(THREAD_WAKEUP, &thread->flags);
5603
5604                 thread->run(thread->mddev);
5605         }
5606
5607         return 0;
5608 }
5609
5610 void md_wakeup_thread(mdk_thread_t *thread)
5611 {
5612         if (thread) {
5613                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5614                 set_bit(THREAD_WAKEUP, &thread->flags);
5615                 wake_up(&thread->wqueue);
5616         }
5617 }
5618
5619 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5620                                  const char *name)
5621 {
5622         mdk_thread_t *thread;
5623
5624         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5625         if (!thread)
5626                 return NULL;
5627
5628         init_waitqueue_head(&thread->wqueue);
5629
5630         thread->run = run;
5631         thread->mddev = mddev;
5632         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5633         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5634         if (IS_ERR(thread->tsk)) {
5635                 kfree(thread);
5636                 return NULL;
5637         }
5638         return thread;
5639 }
5640
5641 void md_unregister_thread(mdk_thread_t *thread)
5642 {
5643         if (!thread)
5644                 return;
5645         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5646
5647         kthread_stop(thread->tsk);
5648         kfree(thread);
5649 }
5650
5651 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5652 {
5653         if (!mddev) {
5654                 MD_BUG();
5655                 return;
5656         }
5657
5658         if (!rdev || test_bit(Faulty, &rdev->flags))
5659                 return;
5660
5661         if (mddev->external)
5662                 set_bit(Blocked, &rdev->flags);
5663 /*
5664         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5665                 mdname(mddev),
5666                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5667                 __builtin_return_address(0),__builtin_return_address(1),
5668                 __builtin_return_address(2),__builtin_return_address(3));
5669 */
5670         if (!mddev->pers)
5671                 return;
5672         if (!mddev->pers->error_handler)
5673                 return;
5674         mddev->pers->error_handler(mddev,rdev);
5675         if (mddev->degraded)
5676                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5677         set_bit(StateChanged, &rdev->flags);
5678         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5679         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5680         md_wakeup_thread(mddev->thread);
5681         md_new_event_inintr(mddev);
5682 }
5683
5684 /* seq_file implementation /proc/mdstat */
5685
5686 static void status_unused(struct seq_file *seq)
5687 {
5688         int i = 0;
5689         mdk_rdev_t *rdev;
5690
5691         seq_printf(seq, "unused devices: ");
5692
5693         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5694                 char b[BDEVNAME_SIZE];
5695                 i++;
5696                 seq_printf(seq, "%s ",
5697                               bdevname(rdev->bdev,b));
5698         }
5699         if (!i)
5700                 seq_printf(seq, "<none>");
5701
5702         seq_printf(seq, "\n");
5703 }
5704
5705
5706 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5707 {
5708         sector_t max_blocks, resync, res;
5709         unsigned long dt, db, rt;
5710         int scale;
5711         unsigned int per_milli;
5712
5713         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5714
5715         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5716                 max_blocks = mddev->resync_max_sectors >> 1;
5717         else
5718                 max_blocks = mddev->dev_sectors / 2;
5719
5720         /*
5721          * Should not happen.
5722          */
5723         if (!max_blocks) {
5724                 MD_BUG();
5725                 return;
5726         }
5727         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5728          * in a sector_t, and (max_blocks>>scale) will fit in a
5729          * u32, as those are the requirements for sector_div.
5730          * Thus 'scale' must be at least 10
5731          */
5732         scale = 10;
5733         if (sizeof(sector_t) > sizeof(unsigned long)) {
5734                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5735                         scale++;
5736         }
5737         res = (resync>>scale)*1000;
5738         sector_div(res, (u32)((max_blocks>>scale)+1));
5739
5740         per_milli = res;
5741         {
5742                 int i, x = per_milli/50, y = 20-x;
5743                 seq_printf(seq, "[");
5744                 for (i = 0; i < x; i++)
5745                         seq_printf(seq, "=");
5746                 seq_printf(seq, ">");
5747                 for (i = 0; i < y; i++)
5748                         seq_printf(seq, ".");
5749                 seq_printf(seq, "] ");
5750         }
5751         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5752                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5753                     "reshape" :
5754                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5755                      "check" :
5756                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5757                       "resync" : "recovery"))),
5758                    per_milli/10, per_milli % 10,
5759                    (unsigned long long) resync,
5760                    (unsigned long long) max_blocks);
5761
5762         /*
5763          * We do not want to overflow, so the order of operands and
5764          * the * 100 / 100 trick are important. We do a +1 to be
5765          * safe against division by zero. We only estimate anyway.
5766          *
5767          * dt: time from mark until now
5768          * db: blocks written from mark until now
5769          * rt: remaining time
5770          */
5771         dt = ((jiffies - mddev->resync_mark) / HZ);
5772         if (!dt) dt++;
5773         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5774                 - mddev->resync_mark_cnt;
5775         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5776
5777         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5778
5779         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5780 }
5781
5782 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5783 {
5784         struct list_head *tmp;
5785         loff_t l = *pos;
5786         mddev_t *mddev;
5787
5788         if (l >= 0x10000)
5789                 return NULL;
5790         if (!l--)
5791                 /* header */
5792                 return (void*)1;
5793
5794         spin_lock(&all_mddevs_lock);
5795         list_for_each(tmp,&all_mddevs)
5796                 if (!l--) {
5797                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5798                         mddev_get(mddev);
5799                         spin_unlock(&all_mddevs_lock);
5800                         return mddev;
5801                 }
5802         spin_unlock(&all_mddevs_lock);
5803         if (!l--)
5804                 return (void*)2;/* tail */
5805         return NULL;
5806 }
5807
5808 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5809 {
5810         struct list_head *tmp;
5811         mddev_t *next_mddev, *mddev = v;
5812         
5813         ++*pos;
5814         if (v == (void*)2)
5815                 return NULL;
5816
5817         spin_lock(&all_mddevs_lock);
5818         if (v == (void*)1)
5819                 tmp = all_mddevs.next;
5820         else
5821                 tmp = mddev->all_mddevs.next;
5822         if (tmp != &all_mddevs)
5823                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5824         else {
5825                 next_mddev = (void*)2;
5826                 *pos = 0x10000;
5827         }               
5828         spin_unlock(&all_mddevs_lock);
5829
5830         if (v != (void*)1)
5831                 mddev_put(mddev);
5832         return next_mddev;
5833
5834 }
5835
5836 static void md_seq_stop(struct seq_file *seq, void *v)
5837 {
5838         mddev_t *mddev = v;
5839
5840         if (mddev && v != (void*)1 && v != (void*)2)
5841                 mddev_put(mddev);
5842 }
5843
5844 struct mdstat_info {
5845         int event;
5846 };
5847
5848 static int md_seq_show(struct seq_file *seq, void *v)
5849 {
5850         mddev_t *mddev = v;
5851         sector_t sectors;
5852         mdk_rdev_t *rdev;
5853         struct mdstat_info *mi = seq->private;
5854         struct bitmap *bitmap;
5855
5856         if (v == (void*)1) {
5857                 struct mdk_personality *pers;
5858                 seq_printf(seq, "Personalities : ");
5859                 spin_lock(&pers_lock);
5860                 list_for_each_entry(pers, &pers_list, list)
5861                         seq_printf(seq, "[%s] ", pers->name);
5862
5863                 spin_unlock(&pers_lock);
5864                 seq_printf(seq, "\n");
5865                 mi->event = atomic_read(&md_event_count);
5866                 return 0;
5867         }
5868         if (v == (void*)2) {
5869                 status_unused(seq);
5870                 return 0;
5871         }
5872
5873         if (mddev_lock(mddev) < 0)
5874                 return -EINTR;
5875
5876         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5877                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5878                                                 mddev->pers ? "" : "in");
5879                 if (mddev->pers) {
5880                         if (mddev->ro==1)
5881                                 seq_printf(seq, " (read-only)");
5882                         if (mddev->ro==2)
5883                                 seq_printf(seq, " (auto-read-only)");
5884                         seq_printf(seq, " %s", mddev->pers->name);
5885                 }
5886
5887                 sectors = 0;
5888                 list_for_each_entry(rdev, &mddev->disks, same_set) {
5889                         char b[BDEVNAME_SIZE];
5890                         seq_printf(seq, " %s[%d]",
5891                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5892                         if (test_bit(WriteMostly, &rdev->flags))
5893                                 seq_printf(seq, "(W)");
5894                         if (test_bit(Faulty, &rdev->flags)) {
5895                                 seq_printf(seq, "(F)");
5896                                 continue;
5897                         } else if (rdev->raid_disk < 0)
5898                                 seq_printf(seq, "(S)"); /* spare */
5899                         sectors += rdev->sectors;
5900                 }
5901
5902                 if (!list_empty(&mddev->disks)) {
5903                         if (mddev->pers)
5904                                 seq_printf(seq, "\n      %llu blocks",
5905                                            (unsigned long long)
5906                                            mddev->array_sectors / 2);
5907                         else
5908                                 seq_printf(seq, "\n      %llu blocks",
5909                                            (unsigned long long)sectors / 2);
5910                 }
5911                 if (mddev->persistent) {
5912                         if (mddev->major_version != 0 ||
5913                             mddev->minor_version != 90) {
5914                                 seq_printf(seq," super %d.%d",
5915                                            mddev->major_version,
5916                                            mddev->minor_version);
5917                         }
5918                 } else if (mddev->external)
5919                         seq_printf(seq, " super external:%s",
5920                                    mddev->metadata_type);
5921                 else
5922                         seq_printf(seq, " super non-persistent");
5923
5924                 if (mddev->pers) {
5925                         mddev->pers->status(seq, mddev);
5926                         seq_printf(seq, "\n      ");
5927                         if (mddev->pers->sync_request) {
5928                                 if (mddev->curr_resync > 2) {
5929                                         status_resync(seq, mddev);
5930                                         seq_printf(seq, "\n      ");
5931                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5932                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5933                                 else if (mddev->recovery_cp < MaxSector)
5934                                         seq_printf(seq, "\tresync=PENDING\n      ");
5935                         }
5936                 } else
5937                         seq_printf(seq, "\n       ");
5938
5939                 if ((bitmap = mddev->bitmap)) {
5940                         unsigned long chunk_kb;
5941                         unsigned long flags;
5942                         spin_lock_irqsave(&bitmap->lock, flags);
5943                         chunk_kb = bitmap->chunksize >> 10;
5944                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5945                                 "%lu%s chunk",
5946                                 bitmap->pages - bitmap->missing_pages,
5947                                 bitmap->pages,
5948                                 (bitmap->pages - bitmap->missing_pages)
5949                                         << (PAGE_SHIFT - 10),
5950                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5951                                 chunk_kb ? "KB" : "B");
5952                         if (bitmap->file) {
5953                                 seq_printf(seq, ", file: ");
5954                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5955                         }
5956
5957                         seq_printf(seq, "\n");
5958                         spin_unlock_irqrestore(&bitmap->lock, flags);
5959                 }
5960
5961                 seq_printf(seq, "\n");
5962         }
5963         mddev_unlock(mddev);
5964         
5965         return 0;
5966 }
5967
5968 static struct seq_operations md_seq_ops = {
5969         .start  = md_seq_start,
5970         .next   = md_seq_next,
5971         .stop   = md_seq_stop,
5972         .show   = md_seq_show,
5973 };
5974
5975 static int md_seq_open(struct inode *inode, struct file *file)
5976 {
5977         int error;
5978         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5979         if (mi == NULL)
5980                 return -ENOMEM;
5981
5982         error = seq_open(file, &md_seq_ops);
5983         if (error)
5984                 kfree(mi);
5985         else {
5986                 struct seq_file *p = file->private_data;
5987                 p->private = mi;
5988                 mi->event = atomic_read(&md_event_count);
5989         }
5990         return error;
5991 }
5992
5993 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5994 {
5995         struct seq_file *m = filp->private_data;
5996         struct mdstat_info *mi = m->private;
5997         int mask;
5998
5999         poll_wait(filp, &md_event_waiters, wait);
6000
6001         /* always allow read */
6002         mask = POLLIN | POLLRDNORM;
6003
6004         if (mi->event != atomic_read(&md_event_count))
6005                 mask |= POLLERR | POLLPRI;
6006         return mask;
6007 }
6008
6009 static const struct file_operations md_seq_fops = {
6010         .owner          = THIS_MODULE,
6011         .open           = md_seq_open,
6012         .read           = seq_read,
6013         .llseek         = seq_lseek,
6014         .release        = seq_release_private,
6015         .poll           = mdstat_poll,
6016 };
6017
6018 int register_md_personality(struct mdk_personality *p)
6019 {
6020         spin_lock(&pers_lock);
6021         list_add_tail(&p->list, &pers_list);
6022         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6023         spin_unlock(&pers_lock);
6024         return 0;
6025 }
6026
6027 int unregister_md_personality(struct mdk_personality *p)
6028 {
6029         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6030         spin_lock(&pers_lock);
6031         list_del_init(&p->list);
6032         spin_unlock(&pers_lock);
6033         return 0;
6034 }
6035
6036 static int is_mddev_idle(mddev_t *mddev, int init)
6037 {
6038         mdk_rdev_t * rdev;
6039         int idle;
6040         int curr_events;
6041
6042         idle = 1;
6043         rcu_read_lock();
6044         rdev_for_each_rcu(rdev, mddev) {
6045                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6046                 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6047                               (int)part_stat_read(&disk->part0, sectors[1]) -
6048                               atomic_read(&disk->sync_io);
6049                 /* sync IO will cause sync_io to increase before the disk_stats
6050                  * as sync_io is counted when a request starts, and
6051                  * disk_stats is counted when it completes.
6052                  * So resync activity will cause curr_events to be smaller than
6053                  * when there was no such activity.
6054                  * non-sync IO will cause disk_stat to increase without
6055                  * increasing sync_io so curr_events will (eventually)
6056                  * be larger than it was before.  Once it becomes
6057                  * substantially larger, the test below will cause
6058                  * the array to appear non-idle, and resync will slow
6059                  * down.
6060                  * If there is a lot of outstanding resync activity when
6061                  * we set last_event to curr_events, then all that activity
6062                  * completing might cause the array to appear non-idle
6063                  * and resync will be slowed down even though there might
6064                  * not have been non-resync activity.  This will only
6065                  * happen once though.  'last_events' will soon reflect
6066                  * the state where there is little or no outstanding
6067                  * resync requests, and further resync activity will
6068                  * always make curr_events less than last_events.
6069                  *
6070                  */
6071                 if (init || curr_events - rdev->last_events > 64) {
6072                         rdev->last_events = curr_events;
6073                         idle = 0;
6074                 }
6075         }
6076         rcu_read_unlock();
6077         return idle;
6078 }
6079
6080 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6081 {
6082         /* another "blocks" (512byte) blocks have been synced */
6083         atomic_sub(blocks, &mddev->recovery_active);
6084         wake_up(&mddev->recovery_wait);
6085         if (!ok) {
6086                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6087                 md_wakeup_thread(mddev->thread);
6088                 // stop recovery, signal do_sync ....
6089         }
6090 }
6091
6092
6093 /* md_write_start(mddev, bi)
6094  * If we need to update some array metadata (e.g. 'active' flag
6095  * in superblock) before writing, schedule a superblock update
6096  * and wait for it to complete.
6097  */
6098 void md_write_start(mddev_t *mddev, struct bio *bi)
6099 {
6100         int did_change = 0;
6101         if (bio_data_dir(bi) != WRITE)
6102                 return;
6103
6104         BUG_ON(mddev->ro == 1);
6105         if (mddev->ro == 2) {
6106                 /* need to switch to read/write */
6107                 mddev->ro = 0;
6108                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6109                 md_wakeup_thread(mddev->thread);
6110                 md_wakeup_thread(mddev->sync_thread);
6111                 did_change = 1;
6112         }
6113         atomic_inc(&mddev->writes_pending);
6114         if (mddev->safemode == 1)
6115                 mddev->safemode = 0;
6116         if (mddev->in_sync) {
6117                 spin_lock_irq(&mddev->write_lock);
6118                 if (mddev->in_sync) {
6119                         mddev->in_sync = 0;
6120                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6121                         md_wakeup_thread(mddev->thread);
6122                         did_change = 1;
6123                 }
6124                 spin_unlock_irq(&mddev->write_lock);
6125         }
6126         if (did_change)
6127                 sysfs_notify_dirent(mddev->sysfs_state);
6128         wait_event(mddev->sb_wait,
6129                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6130                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6131 }
6132
6133 void md_write_end(mddev_t *mddev)
6134 {
6135         if (atomic_dec_and_test(&mddev->writes_pending)) {
6136                 if (mddev->safemode == 2)
6137                         md_wakeup_thread(mddev->thread);
6138                 else if (mddev->safemode_delay)
6139                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6140         }
6141 }
6142
6143 /* md_allow_write(mddev)
6144  * Calling this ensures that the array is marked 'active' so that writes
6145  * may proceed without blocking.  It is important to call this before
6146  * attempting a GFP_KERNEL allocation while holding the mddev lock.
6147  * Must be called with mddev_lock held.
6148  *
6149  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6150  * is dropped, so return -EAGAIN after notifying userspace.
6151  */
6152 int md_allow_write(mddev_t *mddev)
6153 {
6154         if (!mddev->pers)
6155                 return 0;
6156         if (mddev->ro)
6157                 return 0;
6158         if (!mddev->pers->sync_request)
6159                 return 0;
6160
6161         spin_lock_irq(&mddev->write_lock);
6162         if (mddev->in_sync) {
6163                 mddev->in_sync = 0;
6164                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6165                 if (mddev->safemode_delay &&
6166                     mddev->safemode == 0)
6167                         mddev->safemode = 1;
6168                 spin_unlock_irq(&mddev->write_lock);
6169                 md_update_sb(mddev, 0);
6170                 sysfs_notify_dirent(mddev->sysfs_state);
6171         } else
6172                 spin_unlock_irq(&mddev->write_lock);
6173
6174         if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6175                 return -EAGAIN;
6176         else
6177                 return 0;
6178 }
6179 EXPORT_SYMBOL_GPL(md_allow_write);
6180
6181 #define SYNC_MARKS      10
6182 #define SYNC_MARK_STEP  (3*HZ)
6183 void md_do_sync(mddev_t *mddev)
6184 {
6185         mddev_t *mddev2;
6186         unsigned int currspeed = 0,
6187                  window;
6188         sector_t max_sectors,j, io_sectors;
6189         unsigned long mark[SYNC_MARKS];
6190         sector_t mark_cnt[SYNC_MARKS];
6191         int last_mark,m;
6192         struct list_head *tmp;
6193         sector_t last_check;
6194         int skipped = 0;
6195         mdk_rdev_t *rdev;
6196         char *desc;
6197
6198         /* just incase thread restarts... */
6199         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6200                 return;
6201         if (mddev->ro) /* never try to sync a read-only array */
6202                 return;
6203
6204         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6205                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6206                         desc = "data-check";
6207                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6208                         desc = "requested-resync";
6209                 else
6210                         desc = "resync";
6211         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6212                 desc = "reshape";
6213         else
6214                 desc = "recovery";
6215
6216         /* we overload curr_resync somewhat here.
6217          * 0 == not engaged in resync at all
6218          * 2 == checking that there is no conflict with another sync
6219          * 1 == like 2, but have yielded to allow conflicting resync to
6220          *              commense
6221          * other == active in resync - this many blocks
6222          *
6223          * Before starting a resync we must have set curr_resync to
6224          * 2, and then checked that every "conflicting" array has curr_resync
6225          * less than ours.  When we find one that is the same or higher
6226          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
6227          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6228          * This will mean we have to start checking from the beginning again.
6229          *
6230          */
6231
6232         do {
6233                 mddev->curr_resync = 2;
6234
6235         try_again:
6236                 if (kthread_should_stop()) {
6237                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6238                         goto skip;
6239                 }
6240                 for_each_mddev(mddev2, tmp) {
6241                         if (mddev2 == mddev)
6242                                 continue;
6243                         if (!mddev->parallel_resync
6244                         &&  mddev2->curr_resync
6245                         &&  match_mddev_units(mddev, mddev2)) {
6246                                 DEFINE_WAIT(wq);
6247                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
6248                                         /* arbitrarily yield */
6249                                         mddev->curr_resync = 1;
6250                                         wake_up(&resync_wait);
6251                                 }
6252                                 if (mddev > mddev2 && mddev->curr_resync == 1)
6253                                         /* no need to wait here, we can wait the next
6254                                          * time 'round when curr_resync == 2
6255                                          */
6256                                         continue;
6257                                 /* We need to wait 'interruptible' so as not to
6258                                  * contribute to the load average, and not to
6259                                  * be caught by 'softlockup'
6260                                  */
6261                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6262                                 if (!kthread_should_stop() &&
6263                                     mddev2->curr_resync >= mddev->curr_resync) {
6264                                         printk(KERN_INFO "md: delaying %s of %s"
6265                                                " until %s has finished (they"
6266                                                " share one or more physical units)\n",
6267                                                desc, mdname(mddev), mdname(mddev2));
6268                                         mddev_put(mddev2);
6269                                         if (signal_pending(current))
6270                                                 flush_signals(current);
6271                                         schedule();
6272                                         finish_wait(&resync_wait, &wq);
6273                                         goto try_again;
6274                                 }
6275                                 finish_wait(&resync_wait, &wq);
6276                         }
6277                 }
6278         } while (mddev->curr_resync < 2);
6279
6280         j = 0;
6281         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6282                 /* resync follows the size requested by the personality,
6283                  * which defaults to physical size, but can be virtual size
6284                  */
6285                 max_sectors = mddev->resync_max_sectors;
6286                 mddev->resync_mismatches = 0;
6287                 /* we don't use the checkpoint if there's a bitmap */
6288                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6289                         j = mddev->resync_min;
6290                 else if (!mddev->bitmap)
6291                         j = mddev->recovery_cp;
6292
6293         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6294                 max_sectors = mddev->dev_sectors;
6295         else {
6296                 /* recovery follows the physical size of devices */
6297                 max_sectors = mddev->dev_sectors;
6298                 j = MaxSector;
6299                 list_for_each_entry(rdev, &mddev->disks, same_set)
6300                         if (rdev->raid_disk >= 0 &&
6301                             !test_bit(Faulty, &rdev->flags) &&
6302                             !test_bit(In_sync, &rdev->flags) &&
6303                             rdev->recovery_offset < j)
6304                                 j = rdev->recovery_offset;
6305         }
6306
6307         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6308         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
6309                 " %d KB/sec/disk.\n", speed_min(mddev));
6310         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6311                "(but not more than %d KB/sec) for %s.\n",
6312                speed_max(mddev), desc);
6313
6314         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6315
6316         io_sectors = 0;
6317         for (m = 0; m < SYNC_MARKS; m++) {
6318                 mark[m] = jiffies;
6319                 mark_cnt[m] = io_sectors;
6320         }
6321         last_mark = 0;
6322         mddev->resync_mark = mark[last_mark];
6323         mddev->resync_mark_cnt = mark_cnt[last_mark];
6324
6325         /*
6326          * Tune reconstruction:
6327          */
6328         window = 32*(PAGE_SIZE/512);
6329         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6330                 window/2,(unsigned long long) max_sectors/2);
6331
6332         atomic_set(&mddev->recovery_active, 0);
6333         last_check = 0;
6334
6335         if (j>2) {
6336                 printk(KERN_INFO 
6337                        "md: resuming %s of %s from checkpoint.\n",
6338                        desc, mdname(mddev));
6339                 mddev->curr_resync = j;
6340         }
6341
6342         while (j < max_sectors) {
6343                 sector_t sectors;
6344
6345                 skipped = 0;
6346
6347                 if ((mddev->curr_resync > mddev->curr_resync_completed &&
6348                      (mddev->curr_resync - mddev->curr_resync_completed)
6349                     > (max_sectors >> 4)) ||
6350                     (j - mddev->curr_resync_completed)*2
6351                     >= mddev->resync_max - mddev->curr_resync_completed
6352                         ) {
6353                         /* time to update curr_resync_completed */
6354                         blk_unplug(mddev->queue);
6355                         wait_event(mddev->recovery_wait,
6356                                    atomic_read(&mddev->recovery_active) == 0);
6357                         mddev->curr_resync_completed =
6358                                 mddev->curr_resync;
6359                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6360                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6361                 }
6362
6363                 if (j >= mddev->resync_max)
6364                         wait_event(mddev->recovery_wait,
6365                                    mddev->resync_max > j
6366                                    || kthread_should_stop());
6367
6368                 if (kthread_should_stop())
6369                         goto interrupted;
6370
6371                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6372                                                   currspeed < speed_min(mddev));
6373                 if (sectors == 0) {
6374                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6375                         goto out;
6376                 }
6377
6378                 if (!skipped) { /* actual IO requested */
6379                         io_sectors += sectors;
6380                         atomic_add(sectors, &mddev->recovery_active);
6381                 }
6382
6383                 j += sectors;
6384                 if (j>1) mddev->curr_resync = j;
6385                 mddev->curr_mark_cnt = io_sectors;
6386                 if (last_check == 0)
6387                         /* this is the earliers that rebuilt will be
6388                          * visible in /proc/mdstat
6389                          */
6390                         md_new_event(mddev);
6391
6392                 if (last_check + window > io_sectors || j == max_sectors)
6393                         continue;
6394
6395                 last_check = io_sectors;
6396
6397                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6398                         break;
6399
6400         repeat:
6401                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6402                         /* step marks */
6403                         int next = (last_mark+1) % SYNC_MARKS;
6404
6405                         mddev->resync_mark = mark[next];
6406                         mddev->resync_mark_cnt = mark_cnt[next];
6407                         mark[next] = jiffies;
6408                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6409                         last_mark = next;
6410                 }
6411
6412
6413                 if (kthread_should_stop())
6414                         goto interrupted;
6415
6416
6417                 /*
6418                  * this loop exits only if either when we are slower than
6419                  * the 'hard' speed limit, or the system was IO-idle for
6420                  * a jiffy.
6421                  * the system might be non-idle CPU-wise, but we only care
6422                  * about not overloading the IO subsystem. (things like an
6423                  * e2fsck being done on the RAID array should execute fast)
6424                  */
6425                 blk_unplug(mddev->queue);
6426                 cond_resched();
6427
6428                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6429                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
6430
6431                 if (currspeed > speed_min(mddev)) {
6432                         if ((currspeed > speed_max(mddev)) ||
6433                                         !is_mddev_idle(mddev, 0)) {
6434                                 msleep(500);
6435                                 goto repeat;
6436                         }
6437                 }
6438         }
6439         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6440         /*
6441          * this also signals 'finished resyncing' to md_stop
6442          */
6443  out:
6444         blk_unplug(mddev->queue);
6445
6446         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6447
6448         /* tell personality that we are finished */
6449         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6450
6451         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6452             mddev->curr_resync > 2) {
6453                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6454                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6455                                 if (mddev->curr_resync >= mddev->recovery_cp) {
6456                                         printk(KERN_INFO
6457                                                "md: checkpointing %s of %s.\n",
6458                                                desc, mdname(mddev));
6459                                         mddev->recovery_cp = mddev->curr_resync;
6460                                 }
6461                         } else
6462                                 mddev->recovery_cp = MaxSector;
6463                 } else {
6464                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6465                                 mddev->curr_resync = MaxSector;
6466                         list_for_each_entry(rdev, &mddev->disks, same_set)
6467                                 if (rdev->raid_disk >= 0 &&
6468                                     !test_bit(Faulty, &rdev->flags) &&
6469                                     !test_bit(In_sync, &rdev->flags) &&
6470                                     rdev->recovery_offset < mddev->curr_resync)
6471                                         rdev->recovery_offset = mddev->curr_resync;
6472                 }
6473         }
6474         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6475
6476  skip:
6477         mddev->curr_resync = 0;
6478         mddev->curr_resync_completed = 0;
6479         mddev->resync_min = 0;
6480         mddev->resync_max = MaxSector;
6481         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6482         wake_up(&resync_wait);
6483         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6484         md_wakeup_thread(mddev->thread);
6485         return;
6486
6487  interrupted:
6488         /*
6489          * got a signal, exit.
6490          */
6491         printk(KERN_INFO
6492                "md: md_do_sync() got signal ... exiting\n");
6493         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6494         goto out;
6495
6496 }
6497 EXPORT_SYMBOL_GPL(md_do_sync);
6498
6499
6500 static int remove_and_add_spares(mddev_t *mddev)
6501 {
6502         mdk_rdev_t *rdev;
6503         int spares = 0;
6504
6505         mddev->curr_resync_completed = 0;
6506
6507         list_for_each_entry(rdev, &mddev->disks, same_set)
6508                 if (rdev->raid_disk >= 0 &&
6509                     !test_bit(Blocked, &rdev->flags) &&
6510                     (test_bit(Faulty, &rdev->flags) ||
6511                      ! test_bit(In_sync, &rdev->flags)) &&
6512                     atomic_read(&rdev->nr_pending)==0) {
6513                         if (mddev->pers->hot_remove_disk(
6514                                     mddev, rdev->raid_disk)==0) {
6515                                 char nm[20];
6516                                 sprintf(nm,"rd%d", rdev->raid_disk);
6517                                 sysfs_remove_link(&mddev->kobj, nm);
6518                                 rdev->raid_disk = -1;
6519                         }
6520                 }
6521
6522         if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6523                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6524                         if (rdev->raid_disk >= 0 &&
6525                             !test_bit(In_sync, &rdev->flags) &&
6526                             !test_bit(Blocked, &rdev->flags))
6527                                 spares++;
6528                         if (rdev->raid_disk < 0
6529                             && !test_bit(Faulty, &rdev->flags)) {
6530                                 rdev->recovery_offset = 0;
6531                                 if (mddev->pers->
6532                                     hot_add_disk(mddev, rdev) == 0) {
6533                                         char nm[20];
6534                                         sprintf(nm, "rd%d", rdev->raid_disk);
6535                                         if (sysfs_create_link(&mddev->kobj,
6536                                                               &rdev->kobj, nm))
6537                                                 printk(KERN_WARNING
6538                                                        "md: cannot register "
6539                                                        "%s for %s\n",
6540                                                        nm, mdname(mddev));
6541                                         spares++;
6542                                         md_new_event(mddev);
6543                                 } else
6544                                         break;
6545                         }
6546                 }
6547         }
6548         return spares;
6549 }
6550 /*
6551  * This routine is regularly called by all per-raid-array threads to
6552  * deal with generic issues like resync and super-block update.
6553  * Raid personalities that don't have a thread (linear/raid0) do not
6554  * need this as they never do any recovery or update the superblock.
6555  *
6556  * It does not do any resync itself, but rather "forks" off other threads
6557  * to do that as needed.
6558  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6559  * "->recovery" and create a thread at ->sync_thread.
6560  * When the thread finishes it sets MD_RECOVERY_DONE
6561  * and wakeups up this thread which will reap the thread and finish up.
6562  * This thread also removes any faulty devices (with nr_pending == 0).
6563  *
6564  * The overall approach is:
6565  *  1/ if the superblock needs updating, update it.
6566  *  2/ If a recovery thread is running, don't do anything else.
6567  *  3/ If recovery has finished, clean up, possibly marking spares active.
6568  *  4/ If there are any faulty devices, remove them.
6569  *  5/ If array is degraded, try to add spares devices
6570  *  6/ If array has spares or is not in-sync, start a resync thread.
6571  */
6572 void md_check_recovery(mddev_t *mddev)
6573 {
6574         mdk_rdev_t *rdev;
6575
6576
6577         if (mddev->bitmap)
6578                 bitmap_daemon_work(mddev->bitmap);
6579
6580         if (mddev->ro)
6581                 return;
6582
6583         if (signal_pending(current)) {
6584                 if (mddev->pers->sync_request && !mddev->external) {
6585                         printk(KERN_INFO "md: %s in immediate safe mode\n",
6586                                mdname(mddev));
6587                         mddev->safemode = 2;
6588                 }
6589                 flush_signals(current);
6590         }
6591
6592         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6593                 return;
6594         if ( ! (
6595                 (mddev->flags && !mddev->external) ||
6596                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6597                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6598                 (mddev->external == 0 && mddev->safemode == 1) ||
6599                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6600                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6601                 ))
6602                 return;
6603
6604         if (mddev_trylock(mddev)) {
6605                 int spares = 0;
6606
6607                 if (mddev->ro) {
6608                         /* Only thing we do on a ro array is remove
6609                          * failed devices.
6610                          */
6611                         remove_and_add_spares(mddev);
6612                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6613                         goto unlock;
6614                 }
6615
6616                 if (!mddev->external) {
6617                         int did_change = 0;
6618                         spin_lock_irq(&mddev->write_lock);
6619                         if (mddev->safemode &&
6620                             !atomic_read(&mddev->writes_pending) &&
6621                             !mddev->in_sync &&
6622                             mddev->recovery_cp == MaxSector) {
6623                                 mddev->in_sync = 1;
6624                                 did_change = 1;
6625                                 if (mddev->persistent)
6626                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6627                         }
6628                         if (mddev->safemode == 1)
6629                                 mddev->safemode = 0;
6630                         spin_unlock_irq(&mddev->write_lock);
6631                         if (did_change)
6632                                 sysfs_notify_dirent(mddev->sysfs_state);
6633                 }
6634
6635                 if (mddev->flags)
6636                         md_update_sb(mddev, 0);
6637
6638                 list_for_each_entry(rdev, &mddev->disks, same_set)
6639                         if (test_and_clear_bit(StateChanged, &rdev->flags))
6640                                 sysfs_notify_dirent(rdev->sysfs_state);
6641
6642
6643                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6644                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6645                         /* resync/recovery still happening */
6646                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6647                         goto unlock;
6648                 }
6649                 if (mddev->sync_thread) {
6650                         /* resync has finished, collect result */
6651                         md_unregister_thread(mddev->sync_thread);
6652                         mddev->sync_thread = NULL;
6653                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6654                             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6655                                 /* success...*/
6656                                 /* activate any spares */
6657                                 if (mddev->pers->spare_active(mddev))
6658                                         sysfs_notify(&mddev->kobj, NULL,
6659                                                      "degraded");
6660                         }
6661                         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6662                             mddev->pers->finish_reshape)
6663                                 mddev->pers->finish_reshape(mddev);
6664                         md_update_sb(mddev, 1);
6665
6666                         /* if array is no-longer degraded, then any saved_raid_disk
6667                          * information must be scrapped
6668                          */
6669                         if (!mddev->degraded)
6670                                 list_for_each_entry(rdev, &mddev->disks, same_set)
6671                                         rdev->saved_raid_disk = -1;
6672
6673                         mddev->recovery = 0;
6674                         /* flag recovery needed just to double check */
6675                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6676                         sysfs_notify_dirent(mddev->sysfs_action);
6677                         md_new_event(mddev);
6678                         goto unlock;
6679                 }
6680                 /* Set RUNNING before clearing NEEDED to avoid
6681                  * any transients in the value of "sync_action".
6682                  */
6683                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6684                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6685                 /* Clear some bits that don't mean anything, but
6686                  * might be left set
6687                  */
6688                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6689                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6690
6691                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6692                         goto unlock;
6693                 /* no recovery is running.
6694                  * remove any failed drives, then
6695                  * add spares if possible.
6696                  * Spare are also removed and re-added, to allow
6697                  * the personality to fail the re-add.
6698                  */
6699
6700                 if (mddev->reshape_position != MaxSector) {
6701                         if (mddev->pers->check_reshape(mddev) != 0)
6702                                 /* Cannot proceed */
6703                                 goto unlock;
6704                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6705                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6706                 } else if ((spares = remove_and_add_spares(mddev))) {
6707                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6708                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6709                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6710                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6711                 } else if (mddev->recovery_cp < MaxSector) {
6712                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6713                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6714                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6715                         /* nothing to be done ... */
6716                         goto unlock;
6717
6718                 if (mddev->pers->sync_request) {
6719                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6720                                 /* We are adding a device or devices to an array
6721                                  * which has the bitmap stored on all devices.
6722                                  * So make sure all bitmap pages get written
6723                                  */
6724                                 bitmap_write_all(mddev->bitmap);
6725                         }
6726                         mddev->sync_thread = md_register_thread(md_do_sync,
6727                                                                 mddev,
6728                                                                 "%s_resync");
6729                         if (!mddev->sync_thread) {
6730                                 printk(KERN_ERR "%s: could not start resync"
6731                                         " thread...\n", 
6732                                         mdname(mddev));
6733                                 /* leave the spares where they are, it shouldn't hurt */
6734                                 mddev->recovery = 0;
6735                         } else
6736                                 md_wakeup_thread(mddev->sync_thread);
6737                         sysfs_notify_dirent(mddev->sysfs_action);
6738                         md_new_event(mddev);
6739                 }
6740         unlock:
6741                 if (!mddev->sync_thread) {
6742                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6743                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6744                                                &mddev->recovery))
6745                                 if (mddev->sysfs_action)
6746                                         sysfs_notify_dirent(mddev->sysfs_action);
6747                 }
6748                 mddev_unlock(mddev);
6749         }
6750 }
6751
6752 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6753 {
6754         sysfs_notify_dirent(rdev->sysfs_state);
6755         wait_event_timeout(rdev->blocked_wait,
6756                            !test_bit(Blocked, &rdev->flags),
6757                            msecs_to_jiffies(5000));
6758         rdev_dec_pending(rdev, mddev);
6759 }
6760 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6761
6762 static int md_notify_reboot(struct notifier_block *this,
6763                             unsigned long code, void *x)
6764 {
6765         struct list_head *tmp;
6766         mddev_t *mddev;
6767
6768         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6769
6770                 printk(KERN_INFO "md: stopping all md devices.\n");
6771
6772                 for_each_mddev(mddev, tmp)
6773                         if (mddev_trylock(mddev)) {
6774                                 /* Force a switch to readonly even array
6775                                  * appears to still be in use.  Hence
6776                                  * the '100'.
6777                                  */
6778                                 do_md_stop(mddev, 1, 100);
6779                                 mddev_unlock(mddev);
6780                         }
6781                 /*
6782                  * certain more exotic SCSI devices are known to be
6783                  * volatile wrt too early system reboots. While the
6784                  * right place to handle this issue is the given
6785                  * driver, we do want to have a safe RAID driver ...
6786                  */
6787                 mdelay(1000*1);
6788         }
6789         return NOTIFY_DONE;
6790 }
6791
6792 static struct notifier_block md_notifier = {
6793         .notifier_call  = md_notify_reboot,
6794         .next           = NULL,
6795         .priority       = INT_MAX, /* before any real devices */
6796 };
6797
6798 static void md_geninit(void)
6799 {
6800         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6801
6802         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6803 }
6804
6805 static int __init md_init(void)
6806 {
6807         if (register_blkdev(MD_MAJOR, "md"))
6808                 return -1;
6809         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6810                 unregister_blkdev(MD_MAJOR, "md");
6811                 return -1;
6812         }
6813         blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6814                             md_probe, NULL, NULL);
6815         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6816                             md_probe, NULL, NULL);
6817
6818         register_reboot_notifier(&md_notifier);
6819         raid_table_header = register_sysctl_table(raid_root_table);
6820
6821         md_geninit();
6822         return 0;
6823 }
6824
6825
6826 #ifndef MODULE
6827
6828 /*
6829  * Searches all registered partitions for autorun RAID arrays
6830  * at boot time.
6831  */
6832
6833 static LIST_HEAD(all_detected_devices);
6834 struct detected_devices_node {
6835         struct list_head list;
6836         dev_t dev;
6837 };
6838
6839 void md_autodetect_dev(dev_t dev)
6840 {
6841         struct detected_devices_node *node_detected_dev;
6842
6843         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6844         if (node_detected_dev) {
6845                 node_detected_dev->dev = dev;
6846                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6847         } else {
6848                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6849                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6850         }
6851 }
6852
6853
6854 static void autostart_arrays(int part)
6855 {
6856         mdk_rdev_t *rdev;
6857         struct detected_devices_node *node_detected_dev;
6858         dev_t dev;
6859         int i_scanned, i_passed;
6860
6861         i_scanned = 0;
6862         i_passed = 0;
6863
6864         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6865
6866         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6867                 i_scanned++;
6868                 node_detected_dev = list_entry(all_detected_devices.next,
6869                                         struct detected_devices_node, list);
6870                 list_del(&node_detected_dev->list);
6871                 dev = node_detected_dev->dev;
6872                 kfree(node_detected_dev);
6873                 rdev = md_import_device(dev,0, 90);
6874                 if (IS_ERR(rdev))
6875                         continue;
6876
6877                 if (test_bit(Faulty, &rdev->flags)) {
6878                         MD_BUG();
6879                         continue;
6880                 }
6881                 set_bit(AutoDetected, &rdev->flags);
6882                 list_add(&rdev->same_set, &pending_raid_disks);
6883                 i_passed++;
6884         }
6885
6886         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6887                                                 i_scanned, i_passed);
6888
6889         autorun_devices(part);
6890 }
6891
6892 #endif /* !MODULE */
6893
6894 static __exit void md_exit(void)
6895 {
6896         mddev_t *mddev;
6897         struct list_head *tmp;
6898
6899         blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6900         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6901
6902         unregister_blkdev(MD_MAJOR,"md");
6903         unregister_blkdev(mdp_major, "mdp");
6904         unregister_reboot_notifier(&md_notifier);
6905         unregister_sysctl_table(raid_table_header);
6906         remove_proc_entry("mdstat", NULL);
6907         for_each_mddev(mddev, tmp) {
6908                 export_array(mddev);
6909                 mddev->hold_active = 0;
6910         }
6911 }
6912
6913 subsys_initcall(md_init);
6914 module_exit(md_exit)
6915
6916 static int get_ro(char *buffer, struct kernel_param *kp)
6917 {
6918         return sprintf(buffer, "%d", start_readonly);
6919 }
6920 static int set_ro(const char *val, struct kernel_param *kp)
6921 {
6922         char *e;
6923         int num = simple_strtoul(val, &e, 10);
6924         if (*val && (*e == '\0' || *e == '\n')) {
6925                 start_readonly = num;
6926                 return 0;
6927         }
6928         return -EINVAL;
6929 }
6930
6931 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6932 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6933
6934 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6935
6936 EXPORT_SYMBOL(register_md_personality);
6937 EXPORT_SYMBOL(unregister_md_personality);
6938 EXPORT_SYMBOL(md_error);
6939 EXPORT_SYMBOL(md_done_sync);
6940 EXPORT_SYMBOL(md_write_start);
6941 EXPORT_SYMBOL(md_write_end);
6942 EXPORT_SYMBOL(md_register_thread);
6943 EXPORT_SYMBOL(md_unregister_thread);
6944 EXPORT_SYMBOL(md_wakeup_thread);
6945 EXPORT_SYMBOL(md_check_recovery);
6946 MODULE_LICENSE("GPL");
6947 MODULE_ALIAS("md");
6948 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);