md: allow extended partitions on md devices.
[linux-2.6.git] / drivers / md / md.c
1 /*
2    md.c : Multiple Devices driver for Linux
3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5      completely rewritten, based on the MD driver code from Marc Zyngier
6
7    Changes:
8
9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13    - kmod support by: Cyrus Durgin
14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19
20      Neil Brown <neilb@cse.unsw.edu.au>.
21
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25    This program is free software; you can redistribute it and/or modify
26    it under the terms of the GNU General Public License as published by
27    the Free Software Foundation; either version 2, or (at your option)
28    any later version.
29
30    You should have received a copy of the GNU General Public License
31    (for example /usr/src/linux/COPYING); if not, write to the Free
32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/raid/md.h>
37 #include <linux/raid/bitmap.h>
38 #include <linux/sysctl.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
49 #define MAJOR_NR MD_MAJOR
50
51 /* 63 partitions with the alternate major number (mdp) */
52 #define MdpMinorShift 6
53
54 #define DEBUG 0
55 #define dprintk(x...) ((void)(DEBUG && printk(x)))
56
57
58 #ifndef MODULE
59 static void autostart_arrays(int part);
60 #endif
61
62 static LIST_HEAD(pers_list);
63 static DEFINE_SPINLOCK(pers_lock);
64
65 static void md_print_devices(void);
66
67 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
68
69 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
70
71 /*
72  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
73  * is 1000 KB/sec, so the extra system load does not show up that much.
74  * Increase it if you want to have more _guaranteed_ speed. Note that
75  * the RAID driver will use the maximum available bandwidth if the IO
76  * subsystem is idle. There is also an 'absolute maximum' reconstruction
77  * speed limit - in case reconstruction slows down your system despite
78  * idle IO detection.
79  *
80  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
81  * or /sys/block/mdX/md/sync_speed_{min,max}
82  */
83
84 static int sysctl_speed_limit_min = 1000;
85 static int sysctl_speed_limit_max = 200000;
86 static inline int speed_min(mddev_t *mddev)
87 {
88         return mddev->sync_speed_min ?
89                 mddev->sync_speed_min : sysctl_speed_limit_min;
90 }
91
92 static inline int speed_max(mddev_t *mddev)
93 {
94         return mddev->sync_speed_max ?
95                 mddev->sync_speed_max : sysctl_speed_limit_max;
96 }
97
98 static struct ctl_table_header *raid_table_header;
99
100 static ctl_table raid_table[] = {
101         {
102                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
103                 .procname       = "speed_limit_min",
104                 .data           = &sysctl_speed_limit_min,
105                 .maxlen         = sizeof(int),
106                 .mode           = S_IRUGO|S_IWUSR,
107                 .proc_handler   = &proc_dointvec,
108         },
109         {
110                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
111                 .procname       = "speed_limit_max",
112                 .data           = &sysctl_speed_limit_max,
113                 .maxlen         = sizeof(int),
114                 .mode           = S_IRUGO|S_IWUSR,
115                 .proc_handler   = &proc_dointvec,
116         },
117         { .ctl_name = 0 }
118 };
119
120 static ctl_table raid_dir_table[] = {
121         {
122                 .ctl_name       = DEV_RAID,
123                 .procname       = "raid",
124                 .maxlen         = 0,
125                 .mode           = S_IRUGO|S_IXUGO,
126                 .child          = raid_table,
127         },
128         { .ctl_name = 0 }
129 };
130
131 static ctl_table raid_root_table[] = {
132         {
133                 .ctl_name       = CTL_DEV,
134                 .procname       = "dev",
135                 .maxlen         = 0,
136                 .mode           = 0555,
137                 .child          = raid_dir_table,
138         },
139         { .ctl_name = 0 }
140 };
141
142 static struct block_device_operations md_fops;
143
144 static int start_readonly;
145
146 /*
147  * We have a system wide 'event count' that is incremented
148  * on any 'interesting' event, and readers of /proc/mdstat
149  * can use 'poll' or 'select' to find out when the event
150  * count increases.
151  *
152  * Events are:
153  *  start array, stop array, error, add device, remove device,
154  *  start build, activate spare
155  */
156 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
157 static atomic_t md_event_count;
158 void md_new_event(mddev_t *mddev)
159 {
160         atomic_inc(&md_event_count);
161         wake_up(&md_event_waiters);
162 }
163 EXPORT_SYMBOL_GPL(md_new_event);
164
165 /* Alternate version that can be called from interrupts
166  * when calling sysfs_notify isn't needed.
167  */
168 static void md_new_event_inintr(mddev_t *mddev)
169 {
170         atomic_inc(&md_event_count);
171         wake_up(&md_event_waiters);
172 }
173
174 /*
175  * Enables to iterate over all existing md arrays
176  * all_mddevs_lock protects this list.
177  */
178 static LIST_HEAD(all_mddevs);
179 static DEFINE_SPINLOCK(all_mddevs_lock);
180
181
182 /*
183  * iterates through all used mddevs in the system.
184  * We take care to grab the all_mddevs_lock whenever navigating
185  * the list, and to always hold a refcount when unlocked.
186  * Any code which breaks out of this loop while own
187  * a reference to the current mddev and must mddev_put it.
188  */
189 #define for_each_mddev(mddev,tmp)                                       \
190                                                                         \
191         for (({ spin_lock(&all_mddevs_lock);                            \
192                 tmp = all_mddevs.next;                                  \
193                 mddev = NULL;});                                        \
194              ({ if (tmp != &all_mddevs)                                 \
195                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
196                 spin_unlock(&all_mddevs_lock);                          \
197                 if (mddev) mddev_put(mddev);                            \
198                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
199                 tmp != &all_mddevs;});                                  \
200              ({ spin_lock(&all_mddevs_lock);                            \
201                 tmp = tmp->next;})                                      \
202                 )
203
204
205 static int md_fail_request(struct request_queue *q, struct bio *bio)
206 {
207         bio_io_error(bio);
208         return 0;
209 }
210
211 static inline mddev_t *mddev_get(mddev_t *mddev)
212 {
213         atomic_inc(&mddev->active);
214         return mddev;
215 }
216
217 static void mddev_put(mddev_t *mddev)
218 {
219         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
220                 return;
221         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
222                 list_del(&mddev->all_mddevs);
223                 spin_unlock(&all_mddevs_lock);
224                 blk_cleanup_queue(mddev->queue);
225                 if (mddev->sysfs_state)
226                         sysfs_put(mddev->sysfs_state);
227                 mddev->sysfs_state = NULL;
228                 kobject_put(&mddev->kobj);
229         } else
230                 spin_unlock(&all_mddevs_lock);
231 }
232
233 static mddev_t * mddev_find(dev_t unit)
234 {
235         mddev_t *mddev, *new = NULL;
236
237  retry:
238         spin_lock(&all_mddevs_lock);
239         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
240                 if (mddev->unit == unit) {
241                         mddev_get(mddev);
242                         spin_unlock(&all_mddevs_lock);
243                         kfree(new);
244                         return mddev;
245                 }
246
247         if (new) {
248                 list_add(&new->all_mddevs, &all_mddevs);
249                 spin_unlock(&all_mddevs_lock);
250                 return new;
251         }
252         spin_unlock(&all_mddevs_lock);
253
254         new = kzalloc(sizeof(*new), GFP_KERNEL);
255         if (!new)
256                 return NULL;
257
258         new->unit = unit;
259         if (MAJOR(unit) == MD_MAJOR)
260                 new->md_minor = MINOR(unit);
261         else
262                 new->md_minor = MINOR(unit) >> MdpMinorShift;
263
264         mutex_init(&new->reconfig_mutex);
265         INIT_LIST_HEAD(&new->disks);
266         INIT_LIST_HEAD(&new->all_mddevs);
267         init_timer(&new->safemode_timer);
268         atomic_set(&new->active, 1);
269         atomic_set(&new->openers, 0);
270         spin_lock_init(&new->write_lock);
271         init_waitqueue_head(&new->sb_wait);
272         init_waitqueue_head(&new->recovery_wait);
273         new->reshape_position = MaxSector;
274         new->resync_min = 0;
275         new->resync_max = MaxSector;
276         new->level = LEVEL_NONE;
277
278         new->queue = blk_alloc_queue(GFP_KERNEL);
279         if (!new->queue) {
280                 kfree(new);
281                 return NULL;
282         }
283         /* Can be unlocked because the queue is new: no concurrency */
284         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, new->queue);
285
286         blk_queue_make_request(new->queue, md_fail_request);
287
288         goto retry;
289 }
290
291 static inline int mddev_lock(mddev_t * mddev)
292 {
293         return mutex_lock_interruptible(&mddev->reconfig_mutex);
294 }
295
296 static inline int mddev_trylock(mddev_t * mddev)
297 {
298         return mutex_trylock(&mddev->reconfig_mutex);
299 }
300
301 static inline void mddev_unlock(mddev_t * mddev)
302 {
303         mutex_unlock(&mddev->reconfig_mutex);
304
305         md_wakeup_thread(mddev->thread);
306 }
307
308 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
309 {
310         mdk_rdev_t * rdev;
311         struct list_head *tmp;
312
313         rdev_for_each(rdev, tmp, mddev) {
314                 if (rdev->desc_nr == nr)
315                         return rdev;
316         }
317         return NULL;
318 }
319
320 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
321 {
322         struct list_head *tmp;
323         mdk_rdev_t *rdev;
324
325         rdev_for_each(rdev, tmp, mddev) {
326                 if (rdev->bdev->bd_dev == dev)
327                         return rdev;
328         }
329         return NULL;
330 }
331
332 static struct mdk_personality *find_pers(int level, char *clevel)
333 {
334         struct mdk_personality *pers;
335         list_for_each_entry(pers, &pers_list, list) {
336                 if (level != LEVEL_NONE && pers->level == level)
337                         return pers;
338                 if (strcmp(pers->name, clevel)==0)
339                         return pers;
340         }
341         return NULL;
342 }
343
344 /* return the offset of the super block in 512byte sectors */
345 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
346 {
347         sector_t num_sectors = bdev->bd_inode->i_size / 512;
348         return MD_NEW_SIZE_SECTORS(num_sectors);
349 }
350
351 static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
352 {
353         sector_t num_sectors = rdev->sb_start;
354
355         if (chunk_size)
356                 num_sectors &= ~((sector_t)chunk_size/512 - 1);
357         return num_sectors;
358 }
359
360 static int alloc_disk_sb(mdk_rdev_t * rdev)
361 {
362         if (rdev->sb_page)
363                 MD_BUG();
364
365         rdev->sb_page = alloc_page(GFP_KERNEL);
366         if (!rdev->sb_page) {
367                 printk(KERN_ALERT "md: out of memory.\n");
368                 return -ENOMEM;
369         }
370
371         return 0;
372 }
373
374 static void free_disk_sb(mdk_rdev_t * rdev)
375 {
376         if (rdev->sb_page) {
377                 put_page(rdev->sb_page);
378                 rdev->sb_loaded = 0;
379                 rdev->sb_page = NULL;
380                 rdev->sb_start = 0;
381                 rdev->size = 0;
382         }
383 }
384
385
386 static void super_written(struct bio *bio, int error)
387 {
388         mdk_rdev_t *rdev = bio->bi_private;
389         mddev_t *mddev = rdev->mddev;
390
391         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
392                 printk("md: super_written gets error=%d, uptodate=%d\n",
393                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
394                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
395                 md_error(mddev, rdev);
396         }
397
398         if (atomic_dec_and_test(&mddev->pending_writes))
399                 wake_up(&mddev->sb_wait);
400         bio_put(bio);
401 }
402
403 static void super_written_barrier(struct bio *bio, int error)
404 {
405         struct bio *bio2 = bio->bi_private;
406         mdk_rdev_t *rdev = bio2->bi_private;
407         mddev_t *mddev = rdev->mddev;
408
409         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
410             error == -EOPNOTSUPP) {
411                 unsigned long flags;
412                 /* barriers don't appear to be supported :-( */
413                 set_bit(BarriersNotsupp, &rdev->flags);
414                 mddev->barriers_work = 0;
415                 spin_lock_irqsave(&mddev->write_lock, flags);
416                 bio2->bi_next = mddev->biolist;
417                 mddev->biolist = bio2;
418                 spin_unlock_irqrestore(&mddev->write_lock, flags);
419                 wake_up(&mddev->sb_wait);
420                 bio_put(bio);
421         } else {
422                 bio_put(bio2);
423                 bio->bi_private = rdev;
424                 super_written(bio, error);
425         }
426 }
427
428 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
429                    sector_t sector, int size, struct page *page)
430 {
431         /* write first size bytes of page to sector of rdev
432          * Increment mddev->pending_writes before returning
433          * and decrement it on completion, waking up sb_wait
434          * if zero is reached.
435          * If an error occurred, call md_error
436          *
437          * As we might need to resubmit the request if BIO_RW_BARRIER
438          * causes ENOTSUPP, we allocate a spare bio...
439          */
440         struct bio *bio = bio_alloc(GFP_NOIO, 1);
441         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
442
443         bio->bi_bdev = rdev->bdev;
444         bio->bi_sector = sector;
445         bio_add_page(bio, page, size, 0);
446         bio->bi_private = rdev;
447         bio->bi_end_io = super_written;
448         bio->bi_rw = rw;
449
450         atomic_inc(&mddev->pending_writes);
451         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
452                 struct bio *rbio;
453                 rw |= (1<<BIO_RW_BARRIER);
454                 rbio = bio_clone(bio, GFP_NOIO);
455                 rbio->bi_private = bio;
456                 rbio->bi_end_io = super_written_barrier;
457                 submit_bio(rw, rbio);
458         } else
459                 submit_bio(rw, bio);
460 }
461
462 void md_super_wait(mddev_t *mddev)
463 {
464         /* wait for all superblock writes that were scheduled to complete.
465          * if any had to be retried (due to BARRIER problems), retry them
466          */
467         DEFINE_WAIT(wq);
468         for(;;) {
469                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
470                 if (atomic_read(&mddev->pending_writes)==0)
471                         break;
472                 while (mddev->biolist) {
473                         struct bio *bio;
474                         spin_lock_irq(&mddev->write_lock);
475                         bio = mddev->biolist;
476                         mddev->biolist = bio->bi_next ;
477                         bio->bi_next = NULL;
478                         spin_unlock_irq(&mddev->write_lock);
479                         submit_bio(bio->bi_rw, bio);
480                 }
481                 schedule();
482         }
483         finish_wait(&mddev->sb_wait, &wq);
484 }
485
486 static void bi_complete(struct bio *bio, int error)
487 {
488         complete((struct completion*)bio->bi_private);
489 }
490
491 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
492                    struct page *page, int rw)
493 {
494         struct bio *bio = bio_alloc(GFP_NOIO, 1);
495         struct completion event;
496         int ret;
497
498         rw |= (1 << BIO_RW_SYNC);
499
500         bio->bi_bdev = bdev;
501         bio->bi_sector = sector;
502         bio_add_page(bio, page, size, 0);
503         init_completion(&event);
504         bio->bi_private = &event;
505         bio->bi_end_io = bi_complete;
506         submit_bio(rw, bio);
507         wait_for_completion(&event);
508
509         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
510         bio_put(bio);
511         return ret;
512 }
513 EXPORT_SYMBOL_GPL(sync_page_io);
514
515 static int read_disk_sb(mdk_rdev_t * rdev, int size)
516 {
517         char b[BDEVNAME_SIZE];
518         if (!rdev->sb_page) {
519                 MD_BUG();
520                 return -EINVAL;
521         }
522         if (rdev->sb_loaded)
523                 return 0;
524
525
526         if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
527                 goto fail;
528         rdev->sb_loaded = 1;
529         return 0;
530
531 fail:
532         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
533                 bdevname(rdev->bdev,b));
534         return -EINVAL;
535 }
536
537 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
538 {
539         return  sb1->set_uuid0 == sb2->set_uuid0 &&
540                 sb1->set_uuid1 == sb2->set_uuid1 &&
541                 sb1->set_uuid2 == sb2->set_uuid2 &&
542                 sb1->set_uuid3 == sb2->set_uuid3;
543 }
544
545 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
546 {
547         int ret;
548         mdp_super_t *tmp1, *tmp2;
549
550         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
551         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
552
553         if (!tmp1 || !tmp2) {
554                 ret = 0;
555                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
556                 goto abort;
557         }
558
559         *tmp1 = *sb1;
560         *tmp2 = *sb2;
561
562         /*
563          * nr_disks is not constant
564          */
565         tmp1->nr_disks = 0;
566         tmp2->nr_disks = 0;
567
568         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
569 abort:
570         kfree(tmp1);
571         kfree(tmp2);
572         return ret;
573 }
574
575
576 static u32 md_csum_fold(u32 csum)
577 {
578         csum = (csum & 0xffff) + (csum >> 16);
579         return (csum & 0xffff) + (csum >> 16);
580 }
581
582 static unsigned int calc_sb_csum(mdp_super_t * sb)
583 {
584         u64 newcsum = 0;
585         u32 *sb32 = (u32*)sb;
586         int i;
587         unsigned int disk_csum, csum;
588
589         disk_csum = sb->sb_csum;
590         sb->sb_csum = 0;
591
592         for (i = 0; i < MD_SB_BYTES/4 ; i++)
593                 newcsum += sb32[i];
594         csum = (newcsum & 0xffffffff) + (newcsum>>32);
595
596
597 #ifdef CONFIG_ALPHA
598         /* This used to use csum_partial, which was wrong for several
599          * reasons including that different results are returned on
600          * different architectures.  It isn't critical that we get exactly
601          * the same return value as before (we always csum_fold before
602          * testing, and that removes any differences).  However as we
603          * know that csum_partial always returned a 16bit value on
604          * alphas, do a fold to maximise conformity to previous behaviour.
605          */
606         sb->sb_csum = md_csum_fold(disk_csum);
607 #else
608         sb->sb_csum = disk_csum;
609 #endif
610         return csum;
611 }
612
613
614 /*
615  * Handle superblock details.
616  * We want to be able to handle multiple superblock formats
617  * so we have a common interface to them all, and an array of
618  * different handlers.
619  * We rely on user-space to write the initial superblock, and support
620  * reading and updating of superblocks.
621  * Interface methods are:
622  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
623  *      loads and validates a superblock on dev.
624  *      if refdev != NULL, compare superblocks on both devices
625  *    Return:
626  *      0 - dev has a superblock that is compatible with refdev
627  *      1 - dev has a superblock that is compatible and newer than refdev
628  *          so dev should be used as the refdev in future
629  *     -EINVAL superblock incompatible or invalid
630  *     -othererror e.g. -EIO
631  *
632  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
633  *      Verify that dev is acceptable into mddev.
634  *       The first time, mddev->raid_disks will be 0, and data from
635  *       dev should be merged in.  Subsequent calls check that dev
636  *       is new enough.  Return 0 or -EINVAL
637  *
638  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
639  *     Update the superblock for rdev with data in mddev
640  *     This does not write to disc.
641  *
642  */
643
644 struct super_type  {
645         char                *name;
646         struct module       *owner;
647         int                 (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
648                                           int minor_version);
649         int                 (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
650         void                (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
651         unsigned long long  (*rdev_size_change)(mdk_rdev_t *rdev,
652                                                 sector_t num_sectors);
653 };
654
655 /*
656  * load_super for 0.90.0 
657  */
658 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
659 {
660         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
661         mdp_super_t *sb;
662         int ret;
663
664         /*
665          * Calculate the position of the superblock (512byte sectors),
666          * it's at the end of the disk.
667          *
668          * It also happens to be a multiple of 4Kb.
669          */
670         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
671
672         ret = read_disk_sb(rdev, MD_SB_BYTES);
673         if (ret) return ret;
674
675         ret = -EINVAL;
676
677         bdevname(rdev->bdev, b);
678         sb = (mdp_super_t*)page_address(rdev->sb_page);
679
680         if (sb->md_magic != MD_SB_MAGIC) {
681                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
682                        b);
683                 goto abort;
684         }
685
686         if (sb->major_version != 0 ||
687             sb->minor_version < 90 ||
688             sb->minor_version > 91) {
689                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
690                         sb->major_version, sb->minor_version,
691                         b);
692                 goto abort;
693         }
694
695         if (sb->raid_disks <= 0)
696                 goto abort;
697
698         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
699                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
700                         b);
701                 goto abort;
702         }
703
704         rdev->preferred_minor = sb->md_minor;
705         rdev->data_offset = 0;
706         rdev->sb_size = MD_SB_BYTES;
707
708         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
709                 if (sb->level != 1 && sb->level != 4
710                     && sb->level != 5 && sb->level != 6
711                     && sb->level != 10) {
712                         /* FIXME use a better test */
713                         printk(KERN_WARNING
714                                "md: bitmaps not supported for this level.\n");
715                         goto abort;
716                 }
717         }
718
719         if (sb->level == LEVEL_MULTIPATH)
720                 rdev->desc_nr = -1;
721         else
722                 rdev->desc_nr = sb->this_disk.number;
723
724         if (!refdev) {
725                 ret = 1;
726         } else {
727                 __u64 ev1, ev2;
728                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
729                 if (!uuid_equal(refsb, sb)) {
730                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
731                                 b, bdevname(refdev->bdev,b2));
732                         goto abort;
733                 }
734                 if (!sb_equal(refsb, sb)) {
735                         printk(KERN_WARNING "md: %s has same UUID"
736                                " but different superblock to %s\n",
737                                b, bdevname(refdev->bdev, b2));
738                         goto abort;
739                 }
740                 ev1 = md_event(sb);
741                 ev2 = md_event(refsb);
742                 if (ev1 > ev2)
743                         ret = 1;
744                 else 
745                         ret = 0;
746         }
747         rdev->size = calc_num_sectors(rdev, sb->chunk_size) / 2;
748
749         if (rdev->size < sb->size && sb->level > 1)
750                 /* "this cannot possibly happen" ... */
751                 ret = -EINVAL;
752
753  abort:
754         return ret;
755 }
756
757 /*
758  * validate_super for 0.90.0
759  */
760 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
761 {
762         mdp_disk_t *desc;
763         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
764         __u64 ev1 = md_event(sb);
765
766         rdev->raid_disk = -1;
767         clear_bit(Faulty, &rdev->flags);
768         clear_bit(In_sync, &rdev->flags);
769         clear_bit(WriteMostly, &rdev->flags);
770         clear_bit(BarriersNotsupp, &rdev->flags);
771
772         if (mddev->raid_disks == 0) {
773                 mddev->major_version = 0;
774                 mddev->minor_version = sb->minor_version;
775                 mddev->patch_version = sb->patch_version;
776                 mddev->external = 0;
777                 mddev->chunk_size = sb->chunk_size;
778                 mddev->ctime = sb->ctime;
779                 mddev->utime = sb->utime;
780                 mddev->level = sb->level;
781                 mddev->clevel[0] = 0;
782                 mddev->layout = sb->layout;
783                 mddev->raid_disks = sb->raid_disks;
784                 mddev->size = sb->size;
785                 mddev->events = ev1;
786                 mddev->bitmap_offset = 0;
787                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
788
789                 if (mddev->minor_version >= 91) {
790                         mddev->reshape_position = sb->reshape_position;
791                         mddev->delta_disks = sb->delta_disks;
792                         mddev->new_level = sb->new_level;
793                         mddev->new_layout = sb->new_layout;
794                         mddev->new_chunk = sb->new_chunk;
795                 } else {
796                         mddev->reshape_position = MaxSector;
797                         mddev->delta_disks = 0;
798                         mddev->new_level = mddev->level;
799                         mddev->new_layout = mddev->layout;
800                         mddev->new_chunk = mddev->chunk_size;
801                 }
802
803                 if (sb->state & (1<<MD_SB_CLEAN))
804                         mddev->recovery_cp = MaxSector;
805                 else {
806                         if (sb->events_hi == sb->cp_events_hi && 
807                                 sb->events_lo == sb->cp_events_lo) {
808                                 mddev->recovery_cp = sb->recovery_cp;
809                         } else
810                                 mddev->recovery_cp = 0;
811                 }
812
813                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
814                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
815                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
816                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
817
818                 mddev->max_disks = MD_SB_DISKS;
819
820                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
821                     mddev->bitmap_file == NULL)
822                         mddev->bitmap_offset = mddev->default_bitmap_offset;
823
824         } else if (mddev->pers == NULL) {
825                 /* Insist on good event counter while assembling */
826                 ++ev1;
827                 if (ev1 < mddev->events) 
828                         return -EINVAL;
829         } else if (mddev->bitmap) {
830                 /* if adding to array with a bitmap, then we can accept an
831                  * older device ... but not too old.
832                  */
833                 if (ev1 < mddev->bitmap->events_cleared)
834                         return 0;
835         } else {
836                 if (ev1 < mddev->events)
837                         /* just a hot-add of a new device, leave raid_disk at -1 */
838                         return 0;
839         }
840
841         if (mddev->level != LEVEL_MULTIPATH) {
842                 desc = sb->disks + rdev->desc_nr;
843
844                 if (desc->state & (1<<MD_DISK_FAULTY))
845                         set_bit(Faulty, &rdev->flags);
846                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
847                             desc->raid_disk < mddev->raid_disks */) {
848                         set_bit(In_sync, &rdev->flags);
849                         rdev->raid_disk = desc->raid_disk;
850                 }
851                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
852                         set_bit(WriteMostly, &rdev->flags);
853         } else /* MULTIPATH are always insync */
854                 set_bit(In_sync, &rdev->flags);
855         return 0;
856 }
857
858 /*
859  * sync_super for 0.90.0
860  */
861 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
862 {
863         mdp_super_t *sb;
864         struct list_head *tmp;
865         mdk_rdev_t *rdev2;
866         int next_spare = mddev->raid_disks;
867
868
869         /* make rdev->sb match mddev data..
870          *
871          * 1/ zero out disks
872          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
873          * 3/ any empty disks < next_spare become removed
874          *
875          * disks[0] gets initialised to REMOVED because
876          * we cannot be sure from other fields if it has
877          * been initialised or not.
878          */
879         int i;
880         int active=0, working=0,failed=0,spare=0,nr_disks=0;
881
882         rdev->sb_size = MD_SB_BYTES;
883
884         sb = (mdp_super_t*)page_address(rdev->sb_page);
885
886         memset(sb, 0, sizeof(*sb));
887
888         sb->md_magic = MD_SB_MAGIC;
889         sb->major_version = mddev->major_version;
890         sb->patch_version = mddev->patch_version;
891         sb->gvalid_words  = 0; /* ignored */
892         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
893         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
894         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
895         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
896
897         sb->ctime = mddev->ctime;
898         sb->level = mddev->level;
899         sb->size  = mddev->size;
900         sb->raid_disks = mddev->raid_disks;
901         sb->md_minor = mddev->md_minor;
902         sb->not_persistent = 0;
903         sb->utime = mddev->utime;
904         sb->state = 0;
905         sb->events_hi = (mddev->events>>32);
906         sb->events_lo = (u32)mddev->events;
907
908         if (mddev->reshape_position == MaxSector)
909                 sb->minor_version = 90;
910         else {
911                 sb->minor_version = 91;
912                 sb->reshape_position = mddev->reshape_position;
913                 sb->new_level = mddev->new_level;
914                 sb->delta_disks = mddev->delta_disks;
915                 sb->new_layout = mddev->new_layout;
916                 sb->new_chunk = mddev->new_chunk;
917         }
918         mddev->minor_version = sb->minor_version;
919         if (mddev->in_sync)
920         {
921                 sb->recovery_cp = mddev->recovery_cp;
922                 sb->cp_events_hi = (mddev->events>>32);
923                 sb->cp_events_lo = (u32)mddev->events;
924                 if (mddev->recovery_cp == MaxSector)
925                         sb->state = (1<< MD_SB_CLEAN);
926         } else
927                 sb->recovery_cp = 0;
928
929         sb->layout = mddev->layout;
930         sb->chunk_size = mddev->chunk_size;
931
932         if (mddev->bitmap && mddev->bitmap_file == NULL)
933                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
934
935         sb->disks[0].state = (1<<MD_DISK_REMOVED);
936         rdev_for_each(rdev2, tmp, mddev) {
937                 mdp_disk_t *d;
938                 int desc_nr;
939                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
940                     && !test_bit(Faulty, &rdev2->flags))
941                         desc_nr = rdev2->raid_disk;
942                 else
943                         desc_nr = next_spare++;
944                 rdev2->desc_nr = desc_nr;
945                 d = &sb->disks[rdev2->desc_nr];
946                 nr_disks++;
947                 d->number = rdev2->desc_nr;
948                 d->major = MAJOR(rdev2->bdev->bd_dev);
949                 d->minor = MINOR(rdev2->bdev->bd_dev);
950                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
951                     && !test_bit(Faulty, &rdev2->flags))
952                         d->raid_disk = rdev2->raid_disk;
953                 else
954                         d->raid_disk = rdev2->desc_nr; /* compatibility */
955                 if (test_bit(Faulty, &rdev2->flags))
956                         d->state = (1<<MD_DISK_FAULTY);
957                 else if (test_bit(In_sync, &rdev2->flags)) {
958                         d->state = (1<<MD_DISK_ACTIVE);
959                         d->state |= (1<<MD_DISK_SYNC);
960                         active++;
961                         working++;
962                 } else {
963                         d->state = 0;
964                         spare++;
965                         working++;
966                 }
967                 if (test_bit(WriteMostly, &rdev2->flags))
968                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
969         }
970         /* now set the "removed" and "faulty" bits on any missing devices */
971         for (i=0 ; i < mddev->raid_disks ; i++) {
972                 mdp_disk_t *d = &sb->disks[i];
973                 if (d->state == 0 && d->number == 0) {
974                         d->number = i;
975                         d->raid_disk = i;
976                         d->state = (1<<MD_DISK_REMOVED);
977                         d->state |= (1<<MD_DISK_FAULTY);
978                         failed++;
979                 }
980         }
981         sb->nr_disks = nr_disks;
982         sb->active_disks = active;
983         sb->working_disks = working;
984         sb->failed_disks = failed;
985         sb->spare_disks = spare;
986
987         sb->this_disk = sb->disks[rdev->desc_nr];
988         sb->sb_csum = calc_sb_csum(sb);
989 }
990
991 /*
992  * rdev_size_change for 0.90.0
993  */
994 static unsigned long long
995 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
996 {
997         if (num_sectors && num_sectors < rdev->mddev->size * 2)
998                 return 0; /* component must fit device */
999         if (rdev->mddev->bitmap_offset)
1000                 return 0; /* can't move bitmap */
1001         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1002         if (!num_sectors || num_sectors > rdev->sb_start)
1003                 num_sectors = rdev->sb_start;
1004         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1005                        rdev->sb_page);
1006         md_super_wait(rdev->mddev);
1007         return num_sectors / 2; /* kB for sysfs */
1008 }
1009
1010
1011 /*
1012  * version 1 superblock
1013  */
1014
1015 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1016 {
1017         __le32 disk_csum;
1018         u32 csum;
1019         unsigned long long newcsum;
1020         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1021         __le32 *isuper = (__le32*)sb;
1022         int i;
1023
1024         disk_csum = sb->sb_csum;
1025         sb->sb_csum = 0;
1026         newcsum = 0;
1027         for (i=0; size>=4; size -= 4 )
1028                 newcsum += le32_to_cpu(*isuper++);
1029
1030         if (size == 2)
1031                 newcsum += le16_to_cpu(*(__le16*) isuper);
1032
1033         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1034         sb->sb_csum = disk_csum;
1035         return cpu_to_le32(csum);
1036 }
1037
1038 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1039 {
1040         struct mdp_superblock_1 *sb;
1041         int ret;
1042         sector_t sb_start;
1043         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1044         int bmask;
1045
1046         /*
1047          * Calculate the position of the superblock in 512byte sectors.
1048          * It is always aligned to a 4K boundary and
1049          * depeding on minor_version, it can be:
1050          * 0: At least 8K, but less than 12K, from end of device
1051          * 1: At start of device
1052          * 2: 4K from start of device.
1053          */
1054         switch(minor_version) {
1055         case 0:
1056                 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1057                 sb_start -= 8*2;
1058                 sb_start &= ~(sector_t)(4*2-1);
1059                 break;
1060         case 1:
1061                 sb_start = 0;
1062                 break;
1063         case 2:
1064                 sb_start = 8;
1065                 break;
1066         default:
1067                 return -EINVAL;
1068         }
1069         rdev->sb_start = sb_start;
1070
1071         /* superblock is rarely larger than 1K, but it can be larger,
1072          * and it is safe to read 4k, so we do that
1073          */
1074         ret = read_disk_sb(rdev, 4096);
1075         if (ret) return ret;
1076
1077
1078         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1079
1080         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1081             sb->major_version != cpu_to_le32(1) ||
1082             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1083             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1084             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1085                 return -EINVAL;
1086
1087         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1088                 printk("md: invalid superblock checksum on %s\n",
1089                         bdevname(rdev->bdev,b));
1090                 return -EINVAL;
1091         }
1092         if (le64_to_cpu(sb->data_size) < 10) {
1093                 printk("md: data_size too small on %s\n",
1094                        bdevname(rdev->bdev,b));
1095                 return -EINVAL;
1096         }
1097         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1098                 if (sb->level != cpu_to_le32(1) &&
1099                     sb->level != cpu_to_le32(4) &&
1100                     sb->level != cpu_to_le32(5) &&
1101                     sb->level != cpu_to_le32(6) &&
1102                     sb->level != cpu_to_le32(10)) {
1103                         printk(KERN_WARNING
1104                                "md: bitmaps not supported for this level.\n");
1105                         return -EINVAL;
1106                 }
1107         }
1108
1109         rdev->preferred_minor = 0xffff;
1110         rdev->data_offset = le64_to_cpu(sb->data_offset);
1111         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1112
1113         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1114         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1115         if (rdev->sb_size & bmask)
1116                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1117
1118         if (minor_version
1119             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1120                 return -EINVAL;
1121
1122         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1123                 rdev->desc_nr = -1;
1124         else
1125                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1126
1127         if (!refdev) {
1128                 ret = 1;
1129         } else {
1130                 __u64 ev1, ev2;
1131                 struct mdp_superblock_1 *refsb = 
1132                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1133
1134                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1135                     sb->level != refsb->level ||
1136                     sb->layout != refsb->layout ||
1137                     sb->chunksize != refsb->chunksize) {
1138                         printk(KERN_WARNING "md: %s has strangely different"
1139                                 " superblock to %s\n",
1140                                 bdevname(rdev->bdev,b),
1141                                 bdevname(refdev->bdev,b2));
1142                         return -EINVAL;
1143                 }
1144                 ev1 = le64_to_cpu(sb->events);
1145                 ev2 = le64_to_cpu(refsb->events);
1146
1147                 if (ev1 > ev2)
1148                         ret = 1;
1149                 else
1150                         ret = 0;
1151         }
1152         if (minor_version)
1153                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1154         else
1155                 rdev->size = rdev->sb_start / 2;
1156         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1157                 return -EINVAL;
1158         rdev->size = le64_to_cpu(sb->data_size)/2;
1159         if (le32_to_cpu(sb->chunksize))
1160                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1161
1162         if (le64_to_cpu(sb->size) > rdev->size*2)
1163                 return -EINVAL;
1164         return ret;
1165 }
1166
1167 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1168 {
1169         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1170         __u64 ev1 = le64_to_cpu(sb->events);
1171
1172         rdev->raid_disk = -1;
1173         clear_bit(Faulty, &rdev->flags);
1174         clear_bit(In_sync, &rdev->flags);
1175         clear_bit(WriteMostly, &rdev->flags);
1176         clear_bit(BarriersNotsupp, &rdev->flags);
1177
1178         if (mddev->raid_disks == 0) {
1179                 mddev->major_version = 1;
1180                 mddev->patch_version = 0;
1181                 mddev->external = 0;
1182                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1183                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1184                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1185                 mddev->level = le32_to_cpu(sb->level);
1186                 mddev->clevel[0] = 0;
1187                 mddev->layout = le32_to_cpu(sb->layout);
1188                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1189                 mddev->size = le64_to_cpu(sb->size)/2;
1190                 mddev->events = ev1;
1191                 mddev->bitmap_offset = 0;
1192                 mddev->default_bitmap_offset = 1024 >> 9;
1193                 
1194                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1195                 memcpy(mddev->uuid, sb->set_uuid, 16);
1196
1197                 mddev->max_disks =  (4096-256)/2;
1198
1199                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1200                     mddev->bitmap_file == NULL )
1201                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1202
1203                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1204                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1205                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1206                         mddev->new_level = le32_to_cpu(sb->new_level);
1207                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1208                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1209                 } else {
1210                         mddev->reshape_position = MaxSector;
1211                         mddev->delta_disks = 0;
1212                         mddev->new_level = mddev->level;
1213                         mddev->new_layout = mddev->layout;
1214                         mddev->new_chunk = mddev->chunk_size;
1215                 }
1216
1217         } else if (mddev->pers == NULL) {
1218                 /* Insist of good event counter while assembling */
1219                 ++ev1;
1220                 if (ev1 < mddev->events)
1221                         return -EINVAL;
1222         } else if (mddev->bitmap) {
1223                 /* If adding to array with a bitmap, then we can accept an
1224                  * older device, but not too old.
1225                  */
1226                 if (ev1 < mddev->bitmap->events_cleared)
1227                         return 0;
1228         } else {
1229                 if (ev1 < mddev->events)
1230                         /* just a hot-add of a new device, leave raid_disk at -1 */
1231                         return 0;
1232         }
1233         if (mddev->level != LEVEL_MULTIPATH) {
1234                 int role;
1235                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1236                 switch(role) {
1237                 case 0xffff: /* spare */
1238                         break;
1239                 case 0xfffe: /* faulty */
1240                         set_bit(Faulty, &rdev->flags);
1241                         break;
1242                 default:
1243                         if ((le32_to_cpu(sb->feature_map) &
1244                              MD_FEATURE_RECOVERY_OFFSET))
1245                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1246                         else
1247                                 set_bit(In_sync, &rdev->flags);
1248                         rdev->raid_disk = role;
1249                         break;
1250                 }
1251                 if (sb->devflags & WriteMostly1)
1252                         set_bit(WriteMostly, &rdev->flags);
1253         } else /* MULTIPATH are always insync */
1254                 set_bit(In_sync, &rdev->flags);
1255
1256         return 0;
1257 }
1258
1259 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1260 {
1261         struct mdp_superblock_1 *sb;
1262         struct list_head *tmp;
1263         mdk_rdev_t *rdev2;
1264         int max_dev, i;
1265         /* make rdev->sb match mddev and rdev data. */
1266
1267         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1268
1269         sb->feature_map = 0;
1270         sb->pad0 = 0;
1271         sb->recovery_offset = cpu_to_le64(0);
1272         memset(sb->pad1, 0, sizeof(sb->pad1));
1273         memset(sb->pad2, 0, sizeof(sb->pad2));
1274         memset(sb->pad3, 0, sizeof(sb->pad3));
1275
1276         sb->utime = cpu_to_le64((__u64)mddev->utime);
1277         sb->events = cpu_to_le64(mddev->events);
1278         if (mddev->in_sync)
1279                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1280         else
1281                 sb->resync_offset = cpu_to_le64(0);
1282
1283         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1284
1285         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1286         sb->size = cpu_to_le64(mddev->size<<1);
1287
1288         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1289                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1290                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1291         }
1292
1293         if (rdev->raid_disk >= 0 &&
1294             !test_bit(In_sync, &rdev->flags) &&
1295             rdev->recovery_offset > 0) {
1296                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1297                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1298         }
1299
1300         if (mddev->reshape_position != MaxSector) {
1301                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1302                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1303                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1304                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1305                 sb->new_level = cpu_to_le32(mddev->new_level);
1306                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1307         }
1308
1309         max_dev = 0;
1310         rdev_for_each(rdev2, tmp, mddev)
1311                 if (rdev2->desc_nr+1 > max_dev)
1312                         max_dev = rdev2->desc_nr+1;
1313
1314         if (max_dev > le32_to_cpu(sb->max_dev))
1315                 sb->max_dev = cpu_to_le32(max_dev);
1316         for (i=0; i<max_dev;i++)
1317                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1318         
1319         rdev_for_each(rdev2, tmp, mddev) {
1320                 i = rdev2->desc_nr;
1321                 if (test_bit(Faulty, &rdev2->flags))
1322                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1323                 else if (test_bit(In_sync, &rdev2->flags))
1324                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1325                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1326                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1327                 else
1328                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1329         }
1330
1331         sb->sb_csum = calc_sb_1_csum(sb);
1332 }
1333
1334 static unsigned long long
1335 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1336 {
1337         struct mdp_superblock_1 *sb;
1338         sector_t max_sectors;
1339         if (num_sectors && num_sectors < rdev->mddev->size * 2)
1340                 return 0; /* component must fit device */
1341         if (rdev->sb_start < rdev->data_offset) {
1342                 /* minor versions 1 and 2; superblock before data */
1343                 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1344                 max_sectors -= rdev->data_offset;
1345                 if (!num_sectors || num_sectors > max_sectors)
1346                         num_sectors = max_sectors;
1347         } else if (rdev->mddev->bitmap_offset) {
1348                 /* minor version 0 with bitmap we can't move */
1349                 return 0;
1350         } else {
1351                 /* minor version 0; superblock after data */
1352                 sector_t sb_start;
1353                 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1354                 sb_start &= ~(sector_t)(4*2 - 1);
1355                 max_sectors = rdev->size * 2 + sb_start - rdev->sb_start;
1356                 if (!num_sectors || num_sectors > max_sectors)
1357                         num_sectors = max_sectors;
1358                 rdev->sb_start = sb_start;
1359         }
1360         sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1361         sb->data_size = cpu_to_le64(num_sectors);
1362         sb->super_offset = rdev->sb_start;
1363         sb->sb_csum = calc_sb_1_csum(sb);
1364         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1365                        rdev->sb_page);
1366         md_super_wait(rdev->mddev);
1367         return num_sectors / 2; /* kB for sysfs */
1368 }
1369
1370 static struct super_type super_types[] = {
1371         [0] = {
1372                 .name   = "0.90.0",
1373                 .owner  = THIS_MODULE,
1374                 .load_super         = super_90_load,
1375                 .validate_super     = super_90_validate,
1376                 .sync_super         = super_90_sync,
1377                 .rdev_size_change   = super_90_rdev_size_change,
1378         },
1379         [1] = {
1380                 .name   = "md-1",
1381                 .owner  = THIS_MODULE,
1382                 .load_super         = super_1_load,
1383                 .validate_super     = super_1_validate,
1384                 .sync_super         = super_1_sync,
1385                 .rdev_size_change   = super_1_rdev_size_change,
1386         },
1387 };
1388
1389 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1390 {
1391         mdk_rdev_t *rdev, *rdev2;
1392
1393         rcu_read_lock();
1394         rdev_for_each_rcu(rdev, mddev1)
1395                 rdev_for_each_rcu(rdev2, mddev2)
1396                         if (rdev->bdev->bd_contains ==
1397                             rdev2->bdev->bd_contains) {
1398                                 rcu_read_unlock();
1399                                 return 1;
1400                         }
1401         rcu_read_unlock();
1402         return 0;
1403 }
1404
1405 static LIST_HEAD(pending_raid_disks);
1406
1407 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1408 {
1409         char b[BDEVNAME_SIZE];
1410         struct kobject *ko;
1411         char *s;
1412         int err;
1413
1414         if (rdev->mddev) {
1415                 MD_BUG();
1416                 return -EINVAL;
1417         }
1418
1419         /* prevent duplicates */
1420         if (find_rdev(mddev, rdev->bdev->bd_dev))
1421                 return -EEXIST;
1422
1423         /* make sure rdev->size exceeds mddev->size */
1424         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1425                 if (mddev->pers) {
1426                         /* Cannot change size, so fail
1427                          * If mddev->level <= 0, then we don't care
1428                          * about aligning sizes (e.g. linear)
1429                          */
1430                         if (mddev->level > 0)
1431                                 return -ENOSPC;
1432                 } else
1433                         mddev->size = rdev->size;
1434         }
1435
1436         /* Verify rdev->desc_nr is unique.
1437          * If it is -1, assign a free number, else
1438          * check number is not in use
1439          */
1440         if (rdev->desc_nr < 0) {
1441                 int choice = 0;
1442                 if (mddev->pers) choice = mddev->raid_disks;
1443                 while (find_rdev_nr(mddev, choice))
1444                         choice++;
1445                 rdev->desc_nr = choice;
1446         } else {
1447                 if (find_rdev_nr(mddev, rdev->desc_nr))
1448                         return -EBUSY;
1449         }
1450         bdevname(rdev->bdev,b);
1451         while ( (s=strchr(b, '/')) != NULL)
1452                 *s = '!';
1453
1454         rdev->mddev = mddev;
1455         printk(KERN_INFO "md: bind<%s>\n", b);
1456
1457         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1458                 goto fail;
1459
1460         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1461         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1462                 kobject_del(&rdev->kobj);
1463                 goto fail;
1464         }
1465         rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1466
1467         list_add_rcu(&rdev->same_set, &mddev->disks);
1468         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1469         return 0;
1470
1471  fail:
1472         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1473                b, mdname(mddev));
1474         return err;
1475 }
1476
1477 static void md_delayed_delete(struct work_struct *ws)
1478 {
1479         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1480         kobject_del(&rdev->kobj);
1481         kobject_put(&rdev->kobj);
1482 }
1483
1484 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1485 {
1486         char b[BDEVNAME_SIZE];
1487         if (!rdev->mddev) {
1488                 MD_BUG();
1489                 return;
1490         }
1491         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1492         list_del_rcu(&rdev->same_set);
1493         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1494         rdev->mddev = NULL;
1495         sysfs_remove_link(&rdev->kobj, "block");
1496         sysfs_put(rdev->sysfs_state);
1497         rdev->sysfs_state = NULL;
1498         /* We need to delay this, otherwise we can deadlock when
1499          * writing to 'remove' to "dev/state".  We also need
1500          * to delay it due to rcu usage.
1501          */
1502         synchronize_rcu();
1503         INIT_WORK(&rdev->del_work, md_delayed_delete);
1504         kobject_get(&rdev->kobj);
1505         schedule_work(&rdev->del_work);
1506 }
1507
1508 /*
1509  * prevent the device from being mounted, repartitioned or
1510  * otherwise reused by a RAID array (or any other kernel
1511  * subsystem), by bd_claiming the device.
1512  */
1513 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1514 {
1515         int err = 0;
1516         struct block_device *bdev;
1517         char b[BDEVNAME_SIZE];
1518
1519         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1520         if (IS_ERR(bdev)) {
1521                 printk(KERN_ERR "md: could not open %s.\n",
1522                         __bdevname(dev, b));
1523                 return PTR_ERR(bdev);
1524         }
1525         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1526         if (err) {
1527                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1528                         bdevname(bdev, b));
1529                 blkdev_put(bdev);
1530                 return err;
1531         }
1532         if (!shared)
1533                 set_bit(AllReserved, &rdev->flags);
1534         rdev->bdev = bdev;
1535         return err;
1536 }
1537
1538 static void unlock_rdev(mdk_rdev_t *rdev)
1539 {
1540         struct block_device *bdev = rdev->bdev;
1541         rdev->bdev = NULL;
1542         if (!bdev)
1543                 MD_BUG();
1544         bd_release(bdev);
1545         blkdev_put(bdev);
1546 }
1547
1548 void md_autodetect_dev(dev_t dev);
1549
1550 static void export_rdev(mdk_rdev_t * rdev)
1551 {
1552         char b[BDEVNAME_SIZE];
1553         printk(KERN_INFO "md: export_rdev(%s)\n",
1554                 bdevname(rdev->bdev,b));
1555         if (rdev->mddev)
1556                 MD_BUG();
1557         free_disk_sb(rdev);
1558 #ifndef MODULE
1559         if (test_bit(AutoDetected, &rdev->flags))
1560                 md_autodetect_dev(rdev->bdev->bd_dev);
1561 #endif
1562         unlock_rdev(rdev);
1563         kobject_put(&rdev->kobj);
1564 }
1565
1566 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1567 {
1568         unbind_rdev_from_array(rdev);
1569         export_rdev(rdev);
1570 }
1571
1572 static void export_array(mddev_t *mddev)
1573 {
1574         struct list_head *tmp;
1575         mdk_rdev_t *rdev;
1576
1577         rdev_for_each(rdev, tmp, mddev) {
1578                 if (!rdev->mddev) {
1579                         MD_BUG();
1580                         continue;
1581                 }
1582                 kick_rdev_from_array(rdev);
1583         }
1584         if (!list_empty(&mddev->disks))
1585                 MD_BUG();
1586         mddev->raid_disks = 0;
1587         mddev->major_version = 0;
1588 }
1589
1590 static void print_desc(mdp_disk_t *desc)
1591 {
1592         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1593                 desc->major,desc->minor,desc->raid_disk,desc->state);
1594 }
1595
1596 static void print_sb(mdp_super_t *sb)
1597 {
1598         int i;
1599
1600         printk(KERN_INFO 
1601                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1602                 sb->major_version, sb->minor_version, sb->patch_version,
1603                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1604                 sb->ctime);
1605         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1606                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1607                 sb->md_minor, sb->layout, sb->chunk_size);
1608         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1609                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1610                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1611                 sb->failed_disks, sb->spare_disks,
1612                 sb->sb_csum, (unsigned long)sb->events_lo);
1613
1614         printk(KERN_INFO);
1615         for (i = 0; i < MD_SB_DISKS; i++) {
1616                 mdp_disk_t *desc;
1617
1618                 desc = sb->disks + i;
1619                 if (desc->number || desc->major || desc->minor ||
1620                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1621                         printk("     D %2d: ", i);
1622                         print_desc(desc);
1623                 }
1624         }
1625         printk(KERN_INFO "md:     THIS: ");
1626         print_desc(&sb->this_disk);
1627
1628 }
1629
1630 static void print_rdev(mdk_rdev_t *rdev)
1631 {
1632         char b[BDEVNAME_SIZE];
1633         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1634                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1635                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1636                 rdev->desc_nr);
1637         if (rdev->sb_loaded) {
1638                 printk(KERN_INFO "md: rdev superblock:\n");
1639                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1640         } else
1641                 printk(KERN_INFO "md: no rdev superblock!\n");
1642 }
1643
1644 static void md_print_devices(void)
1645 {
1646         struct list_head *tmp, *tmp2;
1647         mdk_rdev_t *rdev;
1648         mddev_t *mddev;
1649         char b[BDEVNAME_SIZE];
1650
1651         printk("\n");
1652         printk("md:     **********************************\n");
1653         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1654         printk("md:     **********************************\n");
1655         for_each_mddev(mddev, tmp) {
1656
1657                 if (mddev->bitmap)
1658                         bitmap_print_sb(mddev->bitmap);
1659                 else
1660                         printk("%s: ", mdname(mddev));
1661                 rdev_for_each(rdev, tmp2, mddev)
1662                         printk("<%s>", bdevname(rdev->bdev,b));
1663                 printk("\n");
1664
1665                 rdev_for_each(rdev, tmp2, mddev)
1666                         print_rdev(rdev);
1667         }
1668         printk("md:     **********************************\n");
1669         printk("\n");
1670 }
1671
1672
1673 static void sync_sbs(mddev_t * mddev, int nospares)
1674 {
1675         /* Update each superblock (in-memory image), but
1676          * if we are allowed to, skip spares which already
1677          * have the right event counter, or have one earlier
1678          * (which would mean they aren't being marked as dirty
1679          * with the rest of the array)
1680          */
1681         mdk_rdev_t *rdev;
1682         struct list_head *tmp;
1683
1684         rdev_for_each(rdev, tmp, mddev) {
1685                 if (rdev->sb_events == mddev->events ||
1686                     (nospares &&
1687                      rdev->raid_disk < 0 &&
1688                      (rdev->sb_events&1)==0 &&
1689                      rdev->sb_events+1 == mddev->events)) {
1690                         /* Don't update this superblock */
1691                         rdev->sb_loaded = 2;
1692                 } else {
1693                         super_types[mddev->major_version].
1694                                 sync_super(mddev, rdev);
1695                         rdev->sb_loaded = 1;
1696                 }
1697         }
1698 }
1699
1700 static void md_update_sb(mddev_t * mddev, int force_change)
1701 {
1702         struct list_head *tmp;
1703         mdk_rdev_t *rdev;
1704         int sync_req;
1705         int nospares = 0;
1706
1707         if (mddev->external)
1708                 return;
1709 repeat:
1710         spin_lock_irq(&mddev->write_lock);
1711
1712         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1713         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1714                 force_change = 1;
1715         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1716                 /* just a clean<-> dirty transition, possibly leave spares alone,
1717                  * though if events isn't the right even/odd, we will have to do
1718                  * spares after all
1719                  */
1720                 nospares = 1;
1721         if (force_change)
1722                 nospares = 0;
1723         if (mddev->degraded)
1724                 /* If the array is degraded, then skipping spares is both
1725                  * dangerous and fairly pointless.
1726                  * Dangerous because a device that was removed from the array
1727                  * might have a event_count that still looks up-to-date,
1728                  * so it can be re-added without a resync.
1729                  * Pointless because if there are any spares to skip,
1730                  * then a recovery will happen and soon that array won't
1731                  * be degraded any more and the spare can go back to sleep then.
1732                  */
1733                 nospares = 0;
1734
1735         sync_req = mddev->in_sync;
1736         mddev->utime = get_seconds();
1737
1738         /* If this is just a dirty<->clean transition, and the array is clean
1739          * and 'events' is odd, we can roll back to the previous clean state */
1740         if (nospares
1741             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1742             && (mddev->events & 1)
1743             && mddev->events != 1)
1744                 mddev->events--;
1745         else {
1746                 /* otherwise we have to go forward and ... */
1747                 mddev->events ++;
1748                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1749                         /* .. if the array isn't clean, insist on an odd 'events' */
1750                         if ((mddev->events&1)==0) {
1751                                 mddev->events++;
1752                                 nospares = 0;
1753                         }
1754                 } else {
1755                         /* otherwise insist on an even 'events' (for clean states) */
1756                         if ((mddev->events&1)) {
1757                                 mddev->events++;
1758                                 nospares = 0;
1759                         }
1760                 }
1761         }
1762
1763         if (!mddev->events) {
1764                 /*
1765                  * oops, this 64-bit counter should never wrap.
1766                  * Either we are in around ~1 trillion A.C., assuming
1767                  * 1 reboot per second, or we have a bug:
1768                  */
1769                 MD_BUG();
1770                 mddev->events --;
1771         }
1772
1773         /*
1774          * do not write anything to disk if using
1775          * nonpersistent superblocks
1776          */
1777         if (!mddev->persistent) {
1778                 if (!mddev->external)
1779                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1780
1781                 spin_unlock_irq(&mddev->write_lock);
1782                 wake_up(&mddev->sb_wait);
1783                 return;
1784         }
1785         sync_sbs(mddev, nospares);
1786         spin_unlock_irq(&mddev->write_lock);
1787
1788         dprintk(KERN_INFO 
1789                 "md: updating %s RAID superblock on device (in sync %d)\n",
1790                 mdname(mddev),mddev->in_sync);
1791
1792         bitmap_update_sb(mddev->bitmap);
1793         rdev_for_each(rdev, tmp, mddev) {
1794                 char b[BDEVNAME_SIZE];
1795                 dprintk(KERN_INFO "md: ");
1796                 if (rdev->sb_loaded != 1)
1797                         continue; /* no noise on spare devices */
1798                 if (test_bit(Faulty, &rdev->flags))
1799                         dprintk("(skipping faulty ");
1800
1801                 dprintk("%s ", bdevname(rdev->bdev,b));
1802                 if (!test_bit(Faulty, &rdev->flags)) {
1803                         md_super_write(mddev,rdev,
1804                                        rdev->sb_start, rdev->sb_size,
1805                                        rdev->sb_page);
1806                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1807                                 bdevname(rdev->bdev,b),
1808                                 (unsigned long long)rdev->sb_start);
1809                         rdev->sb_events = mddev->events;
1810
1811                 } else
1812                         dprintk(")\n");
1813                 if (mddev->level == LEVEL_MULTIPATH)
1814                         /* only need to write one superblock... */
1815                         break;
1816         }
1817         md_super_wait(mddev);
1818         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1819
1820         spin_lock_irq(&mddev->write_lock);
1821         if (mddev->in_sync != sync_req ||
1822             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1823                 /* have to write it out again */
1824                 spin_unlock_irq(&mddev->write_lock);
1825                 goto repeat;
1826         }
1827         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1828         spin_unlock_irq(&mddev->write_lock);
1829         wake_up(&mddev->sb_wait);
1830
1831 }
1832
1833 /* words written to sysfs files may, or may not, be \n terminated.
1834  * We want to accept with case. For this we use cmd_match.
1835  */
1836 static int cmd_match(const char *cmd, const char *str)
1837 {
1838         /* See if cmd, written into a sysfs file, matches
1839          * str.  They must either be the same, or cmd can
1840          * have a trailing newline
1841          */
1842         while (*cmd && *str && *cmd == *str) {
1843                 cmd++;
1844                 str++;
1845         }
1846         if (*cmd == '\n')
1847                 cmd++;
1848         if (*str || *cmd)
1849                 return 0;
1850         return 1;
1851 }
1852
1853 struct rdev_sysfs_entry {
1854         struct attribute attr;
1855         ssize_t (*show)(mdk_rdev_t *, char *);
1856         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1857 };
1858
1859 static ssize_t
1860 state_show(mdk_rdev_t *rdev, char *page)
1861 {
1862         char *sep = "";
1863         size_t len = 0;
1864
1865         if (test_bit(Faulty, &rdev->flags)) {
1866                 len+= sprintf(page+len, "%sfaulty",sep);
1867                 sep = ",";
1868         }
1869         if (test_bit(In_sync, &rdev->flags)) {
1870                 len += sprintf(page+len, "%sin_sync",sep);
1871                 sep = ",";
1872         }
1873         if (test_bit(WriteMostly, &rdev->flags)) {
1874                 len += sprintf(page+len, "%swrite_mostly",sep);
1875                 sep = ",";
1876         }
1877         if (test_bit(Blocked, &rdev->flags)) {
1878                 len += sprintf(page+len, "%sblocked", sep);
1879                 sep = ",";
1880         }
1881         if (!test_bit(Faulty, &rdev->flags) &&
1882             !test_bit(In_sync, &rdev->flags)) {
1883                 len += sprintf(page+len, "%sspare", sep);
1884                 sep = ",";
1885         }
1886         return len+sprintf(page+len, "\n");
1887 }
1888
1889 static ssize_t
1890 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1891 {
1892         /* can write
1893          *  faulty  - simulates and error
1894          *  remove  - disconnects the device
1895          *  writemostly - sets write_mostly
1896          *  -writemostly - clears write_mostly
1897          *  blocked - sets the Blocked flag
1898          *  -blocked - clears the Blocked flag
1899          */
1900         int err = -EINVAL;
1901         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1902                 md_error(rdev->mddev, rdev);
1903                 err = 0;
1904         } else if (cmd_match(buf, "remove")) {
1905                 if (rdev->raid_disk >= 0)
1906                         err = -EBUSY;
1907                 else {
1908                         mddev_t *mddev = rdev->mddev;
1909                         kick_rdev_from_array(rdev);
1910                         if (mddev->pers)
1911                                 md_update_sb(mddev, 1);
1912                         md_new_event(mddev);
1913                         err = 0;
1914                 }
1915         } else if (cmd_match(buf, "writemostly")) {
1916                 set_bit(WriteMostly, &rdev->flags);
1917                 err = 0;
1918         } else if (cmd_match(buf, "-writemostly")) {
1919                 clear_bit(WriteMostly, &rdev->flags);
1920                 err = 0;
1921         } else if (cmd_match(buf, "blocked")) {
1922                 set_bit(Blocked, &rdev->flags);
1923                 err = 0;
1924         } else if (cmd_match(buf, "-blocked")) {
1925                 clear_bit(Blocked, &rdev->flags);
1926                 wake_up(&rdev->blocked_wait);
1927                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1928                 md_wakeup_thread(rdev->mddev->thread);
1929
1930                 err = 0;
1931         }
1932         if (!err && rdev->sysfs_state)
1933                 sysfs_notify_dirent(rdev->sysfs_state);
1934         return err ? err : len;
1935 }
1936 static struct rdev_sysfs_entry rdev_state =
1937 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1938
1939 static ssize_t
1940 errors_show(mdk_rdev_t *rdev, char *page)
1941 {
1942         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1943 }
1944
1945 static ssize_t
1946 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1947 {
1948         char *e;
1949         unsigned long n = simple_strtoul(buf, &e, 10);
1950         if (*buf && (*e == 0 || *e == '\n')) {
1951                 atomic_set(&rdev->corrected_errors, n);
1952                 return len;
1953         }
1954         return -EINVAL;
1955 }
1956 static struct rdev_sysfs_entry rdev_errors =
1957 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1958
1959 static ssize_t
1960 slot_show(mdk_rdev_t *rdev, char *page)
1961 {
1962         if (rdev->raid_disk < 0)
1963                 return sprintf(page, "none\n");
1964         else
1965                 return sprintf(page, "%d\n", rdev->raid_disk);
1966 }
1967
1968 static ssize_t
1969 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1970 {
1971         char *e;
1972         int err;
1973         char nm[20];
1974         int slot = simple_strtoul(buf, &e, 10);
1975         if (strncmp(buf, "none", 4)==0)
1976                 slot = -1;
1977         else if (e==buf || (*e && *e!= '\n'))
1978                 return -EINVAL;
1979         if (rdev->mddev->pers && slot == -1) {
1980                 /* Setting 'slot' on an active array requires also
1981                  * updating the 'rd%d' link, and communicating
1982                  * with the personality with ->hot_*_disk.
1983                  * For now we only support removing
1984                  * failed/spare devices.  This normally happens automatically,
1985                  * but not when the metadata is externally managed.
1986                  */
1987                 if (rdev->raid_disk == -1)
1988                         return -EEXIST;
1989                 /* personality does all needed checks */
1990                 if (rdev->mddev->pers->hot_add_disk == NULL)
1991                         return -EINVAL;
1992                 err = rdev->mddev->pers->
1993                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
1994                 if (err)
1995                         return err;
1996                 sprintf(nm, "rd%d", rdev->raid_disk);
1997                 sysfs_remove_link(&rdev->mddev->kobj, nm);
1998                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
1999                 md_wakeup_thread(rdev->mddev->thread);
2000         } else if (rdev->mddev->pers) {
2001                 mdk_rdev_t *rdev2;
2002                 struct list_head *tmp;
2003                 /* Activating a spare .. or possibly reactivating
2004                  * if we every get bitmaps working here.
2005                  */
2006
2007                 if (rdev->raid_disk != -1)
2008                         return -EBUSY;
2009
2010                 if (rdev->mddev->pers->hot_add_disk == NULL)
2011                         return -EINVAL;
2012
2013                 rdev_for_each(rdev2, tmp, rdev->mddev)
2014                         if (rdev2->raid_disk == slot)
2015                                 return -EEXIST;
2016
2017                 rdev->raid_disk = slot;
2018                 if (test_bit(In_sync, &rdev->flags))
2019                         rdev->saved_raid_disk = slot;
2020                 else
2021                         rdev->saved_raid_disk = -1;
2022                 err = rdev->mddev->pers->
2023                         hot_add_disk(rdev->mddev, rdev);
2024                 if (err) {
2025                         rdev->raid_disk = -1;
2026                         return err;
2027                 } else
2028                         sysfs_notify_dirent(rdev->sysfs_state);
2029                 sprintf(nm, "rd%d", rdev->raid_disk);
2030                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2031                         printk(KERN_WARNING
2032                                "md: cannot register "
2033                                "%s for %s\n",
2034                                nm, mdname(rdev->mddev));
2035
2036                 /* don't wakeup anyone, leave that to userspace. */
2037         } else {
2038                 if (slot >= rdev->mddev->raid_disks)
2039                         return -ENOSPC;
2040                 rdev->raid_disk = slot;
2041                 /* assume it is working */
2042                 clear_bit(Faulty, &rdev->flags);
2043                 clear_bit(WriteMostly, &rdev->flags);
2044                 set_bit(In_sync, &rdev->flags);
2045                 sysfs_notify_dirent(rdev->sysfs_state);
2046         }
2047         return len;
2048 }
2049
2050
2051 static struct rdev_sysfs_entry rdev_slot =
2052 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2053
2054 static ssize_t
2055 offset_show(mdk_rdev_t *rdev, char *page)
2056 {
2057         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2058 }
2059
2060 static ssize_t
2061 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2062 {
2063         char *e;
2064         unsigned long long offset = simple_strtoull(buf, &e, 10);
2065         if (e==buf || (*e && *e != '\n'))
2066                 return -EINVAL;
2067         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2068                 return -EBUSY;
2069         if (rdev->size && rdev->mddev->external)
2070                 /* Must set offset before size, so overlap checks
2071                  * can be sane */
2072                 return -EBUSY;
2073         rdev->data_offset = offset;
2074         return len;
2075 }
2076
2077 static struct rdev_sysfs_entry rdev_offset =
2078 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2079
2080 static ssize_t
2081 rdev_size_show(mdk_rdev_t *rdev, char *page)
2082 {
2083         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
2084 }
2085
2086 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2087 {
2088         /* check if two start/length pairs overlap */
2089         if (s1+l1 <= s2)
2090                 return 0;
2091         if (s2+l2 <= s1)
2092                 return 0;
2093         return 1;
2094 }
2095
2096 static ssize_t
2097 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2098 {
2099         unsigned long long size;
2100         unsigned long long oldsize = rdev->size;
2101         mddev_t *my_mddev = rdev->mddev;
2102
2103         if (strict_strtoull(buf, 10, &size) < 0)
2104                 return -EINVAL;
2105         if (my_mddev->pers && rdev->raid_disk >= 0) {
2106                 if (my_mddev->persistent) {
2107                         size = super_types[my_mddev->major_version].
2108                                 rdev_size_change(rdev, size * 2);
2109                         if (!size)
2110                                 return -EBUSY;
2111                 } else if (!size) {
2112                         size = (rdev->bdev->bd_inode->i_size >> 10);
2113                         size -= rdev->data_offset/2;
2114                 }
2115         }
2116         if (size < my_mddev->size)
2117                 return -EINVAL; /* component must fit device */
2118
2119         rdev->size = size;
2120         if (size > oldsize && my_mddev->external) {
2121                 /* need to check that all other rdevs with the same ->bdev
2122                  * do not overlap.  We need to unlock the mddev to avoid
2123                  * a deadlock.  We have already changed rdev->size, and if
2124                  * we have to change it back, we will have the lock again.
2125                  */
2126                 mddev_t *mddev;
2127                 int overlap = 0;
2128                 struct list_head *tmp, *tmp2;
2129
2130                 mddev_unlock(my_mddev);
2131                 for_each_mddev(mddev, tmp) {
2132                         mdk_rdev_t *rdev2;
2133
2134                         mddev_lock(mddev);
2135                         rdev_for_each(rdev2, tmp2, mddev)
2136                                 if (test_bit(AllReserved, &rdev2->flags) ||
2137                                     (rdev->bdev == rdev2->bdev &&
2138                                      rdev != rdev2 &&
2139                                      overlaps(rdev->data_offset, rdev->size * 2,
2140                                               rdev2->data_offset,
2141                                               rdev2->size * 2))) {
2142                                         overlap = 1;
2143                                         break;
2144                                 }
2145                         mddev_unlock(mddev);
2146                         if (overlap) {
2147                                 mddev_put(mddev);
2148                                 break;
2149                         }
2150                 }
2151                 mddev_lock(my_mddev);
2152                 if (overlap) {
2153                         /* Someone else could have slipped in a size
2154                          * change here, but doing so is just silly.
2155                          * We put oldsize back because we *know* it is
2156                          * safe, and trust userspace not to race with
2157                          * itself
2158                          */
2159                         rdev->size = oldsize;
2160                         return -EBUSY;
2161                 }
2162         }
2163         return len;
2164 }
2165
2166 static struct rdev_sysfs_entry rdev_size =
2167 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2168
2169 static struct attribute *rdev_default_attrs[] = {
2170         &rdev_state.attr,
2171         &rdev_errors.attr,
2172         &rdev_slot.attr,
2173         &rdev_offset.attr,
2174         &rdev_size.attr,
2175         NULL,
2176 };
2177 static ssize_t
2178 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2179 {
2180         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2181         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2182         mddev_t *mddev = rdev->mddev;
2183         ssize_t rv;
2184
2185         if (!entry->show)
2186                 return -EIO;
2187
2188         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2189         if (!rv) {
2190                 if (rdev->mddev == NULL)
2191                         rv = -EBUSY;
2192                 else
2193                         rv = entry->show(rdev, page);
2194                 mddev_unlock(mddev);
2195         }
2196         return rv;
2197 }
2198
2199 static ssize_t
2200 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2201               const char *page, size_t length)
2202 {
2203         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2204         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2205         ssize_t rv;
2206         mddev_t *mddev = rdev->mddev;
2207
2208         if (!entry->store)
2209                 return -EIO;
2210         if (!capable(CAP_SYS_ADMIN))
2211                 return -EACCES;
2212         rv = mddev ? mddev_lock(mddev): -EBUSY;
2213         if (!rv) {
2214                 if (rdev->mddev == NULL)
2215                         rv = -EBUSY;
2216                 else
2217                         rv = entry->store(rdev, page, length);
2218                 mddev_unlock(mddev);
2219         }
2220         return rv;
2221 }
2222
2223 static void rdev_free(struct kobject *ko)
2224 {
2225         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2226         kfree(rdev);
2227 }
2228 static struct sysfs_ops rdev_sysfs_ops = {
2229         .show           = rdev_attr_show,
2230         .store          = rdev_attr_store,
2231 };
2232 static struct kobj_type rdev_ktype = {
2233         .release        = rdev_free,
2234         .sysfs_ops      = &rdev_sysfs_ops,
2235         .default_attrs  = rdev_default_attrs,
2236 };
2237
2238 /*
2239  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2240  *
2241  * mark the device faulty if:
2242  *
2243  *   - the device is nonexistent (zero size)
2244  *   - the device has no valid superblock
2245  *
2246  * a faulty rdev _never_ has rdev->sb set.
2247  */
2248 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2249 {
2250         char b[BDEVNAME_SIZE];
2251         int err;
2252         mdk_rdev_t *rdev;
2253         sector_t size;
2254
2255         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2256         if (!rdev) {
2257                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2258                 return ERR_PTR(-ENOMEM);
2259         }
2260
2261         if ((err = alloc_disk_sb(rdev)))
2262                 goto abort_free;
2263
2264         err = lock_rdev(rdev, newdev, super_format == -2);
2265         if (err)
2266                 goto abort_free;
2267
2268         kobject_init(&rdev->kobj, &rdev_ktype);
2269
2270         rdev->desc_nr = -1;
2271         rdev->saved_raid_disk = -1;
2272         rdev->raid_disk = -1;
2273         rdev->flags = 0;
2274         rdev->data_offset = 0;
2275         rdev->sb_events = 0;
2276         atomic_set(&rdev->nr_pending, 0);
2277         atomic_set(&rdev->read_errors, 0);
2278         atomic_set(&rdev->corrected_errors, 0);
2279
2280         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2281         if (!size) {
2282                 printk(KERN_WARNING 
2283                         "md: %s has zero or unknown size, marking faulty!\n",
2284                         bdevname(rdev->bdev,b));
2285                 err = -EINVAL;
2286                 goto abort_free;
2287         }
2288
2289         if (super_format >= 0) {
2290                 err = super_types[super_format].
2291                         load_super(rdev, NULL, super_minor);
2292                 if (err == -EINVAL) {
2293                         printk(KERN_WARNING
2294                                 "md: %s does not have a valid v%d.%d "
2295                                "superblock, not importing!\n",
2296                                 bdevname(rdev->bdev,b),
2297                                super_format, super_minor);
2298                         goto abort_free;
2299                 }
2300                 if (err < 0) {
2301                         printk(KERN_WARNING 
2302                                 "md: could not read %s's sb, not importing!\n",
2303                                 bdevname(rdev->bdev,b));
2304                         goto abort_free;
2305                 }
2306         }
2307
2308         INIT_LIST_HEAD(&rdev->same_set);
2309         init_waitqueue_head(&rdev->blocked_wait);
2310
2311         return rdev;
2312
2313 abort_free:
2314         if (rdev->sb_page) {
2315                 if (rdev->bdev)
2316                         unlock_rdev(rdev);
2317                 free_disk_sb(rdev);
2318         }
2319         kfree(rdev);
2320         return ERR_PTR(err);
2321 }
2322
2323 /*
2324  * Check a full RAID array for plausibility
2325  */
2326
2327
2328 static void analyze_sbs(mddev_t * mddev)
2329 {
2330         int i;
2331         struct list_head *tmp;
2332         mdk_rdev_t *rdev, *freshest;
2333         char b[BDEVNAME_SIZE];
2334
2335         freshest = NULL;
2336         rdev_for_each(rdev, tmp, mddev)
2337                 switch (super_types[mddev->major_version].
2338                         load_super(rdev, freshest, mddev->minor_version)) {
2339                 case 1:
2340                         freshest = rdev;
2341                         break;
2342                 case 0:
2343                         break;
2344                 default:
2345                         printk( KERN_ERR \
2346                                 "md: fatal superblock inconsistency in %s"
2347                                 " -- removing from array\n", 
2348                                 bdevname(rdev->bdev,b));
2349                         kick_rdev_from_array(rdev);
2350                 }
2351
2352
2353         super_types[mddev->major_version].
2354                 validate_super(mddev, freshest);
2355
2356         i = 0;
2357         rdev_for_each(rdev, tmp, mddev) {
2358                 if (rdev != freshest)
2359                         if (super_types[mddev->major_version].
2360                             validate_super(mddev, rdev)) {
2361                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2362                                         " from array!\n",
2363                                         bdevname(rdev->bdev,b));
2364                                 kick_rdev_from_array(rdev);
2365                                 continue;
2366                         }
2367                 if (mddev->level == LEVEL_MULTIPATH) {
2368                         rdev->desc_nr = i++;
2369                         rdev->raid_disk = rdev->desc_nr;
2370                         set_bit(In_sync, &rdev->flags);
2371                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2372                         rdev->raid_disk = -1;
2373                         clear_bit(In_sync, &rdev->flags);
2374                 }
2375         }
2376
2377
2378
2379         if (mddev->recovery_cp != MaxSector &&
2380             mddev->level >= 1)
2381                 printk(KERN_ERR "md: %s: raid array is not clean"
2382                        " -- starting background reconstruction\n",
2383                        mdname(mddev));
2384
2385 }
2386
2387 static void md_safemode_timeout(unsigned long data);
2388
2389 static ssize_t
2390 safe_delay_show(mddev_t *mddev, char *page)
2391 {
2392         int msec = (mddev->safemode_delay*1000)/HZ;
2393         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2394 }
2395 static ssize_t
2396 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2397 {
2398         int scale=1;
2399         int dot=0;
2400         int i;
2401         unsigned long msec;
2402         char buf[30];
2403
2404         /* remove a period, and count digits after it */
2405         if (len >= sizeof(buf))
2406                 return -EINVAL;
2407         strlcpy(buf, cbuf, sizeof(buf));
2408         for (i=0; i<len; i++) {
2409                 if (dot) {
2410                         if (isdigit(buf[i])) {
2411                                 buf[i-1] = buf[i];
2412                                 scale *= 10;
2413                         }
2414                         buf[i] = 0;
2415                 } else if (buf[i] == '.') {
2416                         dot=1;
2417                         buf[i] = 0;
2418                 }
2419         }
2420         if (strict_strtoul(buf, 10, &msec) < 0)
2421                 return -EINVAL;
2422         msec = (msec * 1000) / scale;
2423         if (msec == 0)
2424                 mddev->safemode_delay = 0;
2425         else {
2426                 unsigned long old_delay = mddev->safemode_delay;
2427                 mddev->safemode_delay = (msec*HZ)/1000;
2428                 if (mddev->safemode_delay == 0)
2429                         mddev->safemode_delay = 1;
2430                 if (mddev->safemode_delay < old_delay)
2431                         md_safemode_timeout((unsigned long)mddev);
2432         }
2433         return len;
2434 }
2435 static struct md_sysfs_entry md_safe_delay =
2436 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2437
2438 static ssize_t
2439 level_show(mddev_t *mddev, char *page)
2440 {
2441         struct mdk_personality *p = mddev->pers;
2442         if (p)
2443                 return sprintf(page, "%s\n", p->name);
2444         else if (mddev->clevel[0])
2445                 return sprintf(page, "%s\n", mddev->clevel);
2446         else if (mddev->level != LEVEL_NONE)
2447                 return sprintf(page, "%d\n", mddev->level);
2448         else
2449                 return 0;
2450 }
2451
2452 static ssize_t
2453 level_store(mddev_t *mddev, const char *buf, size_t len)
2454 {
2455         ssize_t rv = len;
2456         if (mddev->pers)
2457                 return -EBUSY;
2458         if (len == 0)
2459                 return 0;
2460         if (len >= sizeof(mddev->clevel))
2461                 return -ENOSPC;
2462         strncpy(mddev->clevel, buf, len);
2463         if (mddev->clevel[len-1] == '\n')
2464                 len--;
2465         mddev->clevel[len] = 0;
2466         mddev->level = LEVEL_NONE;
2467         return rv;
2468 }
2469
2470 static struct md_sysfs_entry md_level =
2471 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2472
2473
2474 static ssize_t
2475 layout_show(mddev_t *mddev, char *page)
2476 {
2477         /* just a number, not meaningful for all levels */
2478         if (mddev->reshape_position != MaxSector &&
2479             mddev->layout != mddev->new_layout)
2480                 return sprintf(page, "%d (%d)\n",
2481                                mddev->new_layout, mddev->layout);
2482         return sprintf(page, "%d\n", mddev->layout);
2483 }
2484
2485 static ssize_t
2486 layout_store(mddev_t *mddev, const char *buf, size_t len)
2487 {
2488         char *e;
2489         unsigned long n = simple_strtoul(buf, &e, 10);
2490
2491         if (!*buf || (*e && *e != '\n'))
2492                 return -EINVAL;
2493
2494         if (mddev->pers)
2495                 return -EBUSY;
2496         if (mddev->reshape_position != MaxSector)
2497                 mddev->new_layout = n;
2498         else
2499                 mddev->layout = n;
2500         return len;
2501 }
2502 static struct md_sysfs_entry md_layout =
2503 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2504
2505
2506 static ssize_t
2507 raid_disks_show(mddev_t *mddev, char *page)
2508 {
2509         if (mddev->raid_disks == 0)
2510                 return 0;
2511         if (mddev->reshape_position != MaxSector &&
2512             mddev->delta_disks != 0)
2513                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2514                                mddev->raid_disks - mddev->delta_disks);
2515         return sprintf(page, "%d\n", mddev->raid_disks);
2516 }
2517
2518 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2519
2520 static ssize_t
2521 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2522 {
2523         char *e;
2524         int rv = 0;
2525         unsigned long n = simple_strtoul(buf, &e, 10);
2526
2527         if (!*buf || (*e && *e != '\n'))
2528                 return -EINVAL;
2529
2530         if (mddev->pers)
2531                 rv = update_raid_disks(mddev, n);
2532         else if (mddev->reshape_position != MaxSector) {
2533                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2534                 mddev->delta_disks = n - olddisks;
2535                 mddev->raid_disks = n;
2536         } else
2537                 mddev->raid_disks = n;
2538         return rv ? rv : len;
2539 }
2540 static struct md_sysfs_entry md_raid_disks =
2541 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2542
2543 static ssize_t
2544 chunk_size_show(mddev_t *mddev, char *page)
2545 {
2546         if (mddev->reshape_position != MaxSector &&
2547             mddev->chunk_size != mddev->new_chunk)
2548                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2549                                mddev->chunk_size);
2550         return sprintf(page, "%d\n", mddev->chunk_size);
2551 }
2552
2553 static ssize_t
2554 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2555 {
2556         /* can only set chunk_size if array is not yet active */
2557         char *e;
2558         unsigned long n = simple_strtoul(buf, &e, 10);
2559
2560         if (!*buf || (*e && *e != '\n'))
2561                 return -EINVAL;
2562
2563         if (mddev->pers)
2564                 return -EBUSY;
2565         else if (mddev->reshape_position != MaxSector)
2566                 mddev->new_chunk = n;
2567         else
2568                 mddev->chunk_size = n;
2569         return len;
2570 }
2571 static struct md_sysfs_entry md_chunk_size =
2572 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2573
2574 static ssize_t
2575 resync_start_show(mddev_t *mddev, char *page)
2576 {
2577         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2578 }
2579
2580 static ssize_t
2581 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2582 {
2583         char *e;
2584         unsigned long long n = simple_strtoull(buf, &e, 10);
2585
2586         if (mddev->pers)
2587                 return -EBUSY;
2588         if (!*buf || (*e && *e != '\n'))
2589                 return -EINVAL;
2590
2591         mddev->recovery_cp = n;
2592         return len;
2593 }
2594 static struct md_sysfs_entry md_resync_start =
2595 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2596
2597 /*
2598  * The array state can be:
2599  *
2600  * clear
2601  *     No devices, no size, no level
2602  *     Equivalent to STOP_ARRAY ioctl
2603  * inactive
2604  *     May have some settings, but array is not active
2605  *        all IO results in error
2606  *     When written, doesn't tear down array, but just stops it
2607  * suspended (not supported yet)
2608  *     All IO requests will block. The array can be reconfigured.
2609  *     Writing this, if accepted, will block until array is quiescent
2610  * readonly
2611  *     no resync can happen.  no superblocks get written.
2612  *     write requests fail
2613  * read-auto
2614  *     like readonly, but behaves like 'clean' on a write request.
2615  *
2616  * clean - no pending writes, but otherwise active.
2617  *     When written to inactive array, starts without resync
2618  *     If a write request arrives then
2619  *       if metadata is known, mark 'dirty' and switch to 'active'.
2620  *       if not known, block and switch to write-pending
2621  *     If written to an active array that has pending writes, then fails.
2622  * active
2623  *     fully active: IO and resync can be happening.
2624  *     When written to inactive array, starts with resync
2625  *
2626  * write-pending
2627  *     clean, but writes are blocked waiting for 'active' to be written.
2628  *
2629  * active-idle
2630  *     like active, but no writes have been seen for a while (100msec).
2631  *
2632  */
2633 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2634                    write_pending, active_idle, bad_word};
2635 static char *array_states[] = {
2636         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2637         "write-pending", "active-idle", NULL };
2638
2639 static int match_word(const char *word, char **list)
2640 {
2641         int n;
2642         for (n=0; list[n]; n++)
2643                 if (cmd_match(word, list[n]))
2644                         break;
2645         return n;
2646 }
2647
2648 static ssize_t
2649 array_state_show(mddev_t *mddev, char *page)
2650 {
2651         enum array_state st = inactive;
2652
2653         if (mddev->pers)
2654                 switch(mddev->ro) {
2655                 case 1:
2656                         st = readonly;
2657                         break;
2658                 case 2:
2659                         st = read_auto;
2660                         break;
2661                 case 0:
2662                         if (mddev->in_sync)
2663                                 st = clean;
2664                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
2665                                 st = write_pending;
2666                         else if (mddev->safemode)
2667                                 st = active_idle;
2668                         else
2669                                 st = active;
2670                 }
2671         else {
2672                 if (list_empty(&mddev->disks) &&
2673                     mddev->raid_disks == 0 &&
2674                     mddev->size == 0)
2675                         st = clear;
2676                 else
2677                         st = inactive;
2678         }
2679         return sprintf(page, "%s\n", array_states[st]);
2680 }
2681
2682 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
2683 static int do_md_run(mddev_t * mddev);
2684 static int restart_array(mddev_t *mddev);
2685
2686 static ssize_t
2687 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2688 {
2689         int err = -EINVAL;
2690         enum array_state st = match_word(buf, array_states);
2691         switch(st) {
2692         case bad_word:
2693                 break;
2694         case clear:
2695                 /* stopping an active array */
2696                 if (atomic_read(&mddev->openers) > 0)
2697                         return -EBUSY;
2698                 err = do_md_stop(mddev, 0, 0);
2699                 break;
2700         case inactive:
2701                 /* stopping an active array */
2702                 if (mddev->pers) {
2703                         if (atomic_read(&mddev->openers) > 0)
2704                                 return -EBUSY;
2705                         err = do_md_stop(mddev, 2, 0);
2706                 } else
2707                         err = 0; /* already inactive */
2708                 break;
2709         case suspended:
2710                 break; /* not supported yet */
2711         case readonly:
2712                 if (mddev->pers)
2713                         err = do_md_stop(mddev, 1, 0);
2714                 else {
2715                         mddev->ro = 1;
2716                         set_disk_ro(mddev->gendisk, 1);
2717                         err = do_md_run(mddev);
2718                 }
2719                 break;
2720         case read_auto:
2721                 if (mddev->pers) {
2722                         if (mddev->ro == 0)
2723                                 err = do_md_stop(mddev, 1, 0);
2724                         else if (mddev->ro == 1)
2725                                 err = restart_array(mddev);
2726                         if (err == 0) {
2727                                 mddev->ro = 2;
2728                                 set_disk_ro(mddev->gendisk, 0);
2729                         }
2730                 } else {
2731                         mddev->ro = 2;
2732                         err = do_md_run(mddev);
2733                 }
2734                 break;
2735         case clean:
2736                 if (mddev->pers) {
2737                         restart_array(mddev);
2738                         spin_lock_irq(&mddev->write_lock);
2739                         if (atomic_read(&mddev->writes_pending) == 0) {
2740                                 if (mddev->in_sync == 0) {
2741                                         mddev->in_sync = 1;
2742                                         if (mddev->safemode == 1)
2743                                                 mddev->safemode = 0;
2744                                         if (mddev->persistent)
2745                                                 set_bit(MD_CHANGE_CLEAN,
2746                                                         &mddev->flags);
2747                                 }
2748                                 err = 0;
2749                         } else
2750                                 err = -EBUSY;
2751                         spin_unlock_irq(&mddev->write_lock);
2752                 } else {
2753                         mddev->ro = 0;
2754                         mddev->recovery_cp = MaxSector;
2755                         err = do_md_run(mddev);
2756                 }
2757                 break;
2758         case active:
2759                 if (mddev->pers) {
2760                         restart_array(mddev);
2761                         if (mddev->external)
2762                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2763                         wake_up(&mddev->sb_wait);
2764                         err = 0;
2765                 } else {
2766                         mddev->ro = 0;
2767                         set_disk_ro(mddev->gendisk, 0);
2768                         err = do_md_run(mddev);
2769                 }
2770                 break;
2771         case write_pending:
2772         case active_idle:
2773                 /* these cannot be set */
2774                 break;
2775         }
2776         if (err)
2777                 return err;
2778         else {
2779                 sysfs_notify_dirent(mddev->sysfs_state);
2780                 return len;
2781         }
2782 }
2783 static struct md_sysfs_entry md_array_state =
2784 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2785
2786 static ssize_t
2787 null_show(mddev_t *mddev, char *page)
2788 {
2789         return -EINVAL;
2790 }
2791
2792 static ssize_t
2793 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2794 {
2795         /* buf must be %d:%d\n? giving major and minor numbers */
2796         /* The new device is added to the array.
2797          * If the array has a persistent superblock, we read the
2798          * superblock to initialise info and check validity.
2799          * Otherwise, only checking done is that in bind_rdev_to_array,
2800          * which mainly checks size.
2801          */
2802         char *e;
2803         int major = simple_strtoul(buf, &e, 10);
2804         int minor;
2805         dev_t dev;
2806         mdk_rdev_t *rdev;
2807         int err;
2808
2809         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2810                 return -EINVAL;
2811         minor = simple_strtoul(e+1, &e, 10);
2812         if (*e && *e != '\n')
2813                 return -EINVAL;
2814         dev = MKDEV(major, minor);
2815         if (major != MAJOR(dev) ||
2816             minor != MINOR(dev))
2817                 return -EOVERFLOW;
2818
2819
2820         if (mddev->persistent) {
2821                 rdev = md_import_device(dev, mddev->major_version,
2822                                         mddev->minor_version);
2823                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2824                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2825                                                        mdk_rdev_t, same_set);
2826                         err = super_types[mddev->major_version]
2827                                 .load_super(rdev, rdev0, mddev->minor_version);
2828                         if (err < 0)
2829                                 goto out;
2830                 }
2831         } else if (mddev->external)
2832                 rdev = md_import_device(dev, -2, -1);
2833         else
2834                 rdev = md_import_device(dev, -1, -1);
2835
2836         if (IS_ERR(rdev))
2837                 return PTR_ERR(rdev);
2838         err = bind_rdev_to_array(rdev, mddev);
2839  out:
2840         if (err)
2841                 export_rdev(rdev);
2842         return err ? err : len;
2843 }
2844
2845 static struct md_sysfs_entry md_new_device =
2846 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2847
2848 static ssize_t
2849 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2850 {
2851         char *end;
2852         unsigned long chunk, end_chunk;
2853
2854         if (!mddev->bitmap)
2855                 goto out;
2856         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2857         while (*buf) {
2858                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2859                 if (buf == end) break;
2860                 if (*end == '-') { /* range */
2861                         buf = end + 1;
2862                         end_chunk = simple_strtoul(buf, &end, 0);
2863                         if (buf == end) break;
2864                 }
2865                 if (*end && !isspace(*end)) break;
2866                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2867                 buf = end;
2868                 while (isspace(*buf)) buf++;
2869         }
2870         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2871 out:
2872         return len;
2873 }
2874
2875 static struct md_sysfs_entry md_bitmap =
2876 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2877
2878 static ssize_t
2879 size_show(mddev_t *mddev, char *page)
2880 {
2881         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2882 }
2883
2884 static int update_size(mddev_t *mddev, sector_t num_sectors);
2885
2886 static ssize_t
2887 size_store(mddev_t *mddev, const char *buf, size_t len)
2888 {
2889         /* If array is inactive, we can reduce the component size, but
2890          * not increase it (except from 0).
2891          * If array is active, we can try an on-line resize
2892          */
2893         char *e;
2894         int err = 0;
2895         unsigned long long size = simple_strtoull(buf, &e, 10);
2896         if (!*buf || *buf == '\n' ||
2897             (*e && *e != '\n'))
2898                 return -EINVAL;
2899
2900         if (mddev->pers) {
2901                 err = update_size(mddev, size * 2);
2902                 md_update_sb(mddev, 1);
2903         } else {
2904                 if (mddev->size == 0 ||
2905                     mddev->size > size)
2906                         mddev->size = size;
2907                 else
2908                         err = -ENOSPC;
2909         }
2910         return err ? err : len;
2911 }
2912
2913 static struct md_sysfs_entry md_size =
2914 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2915
2916
2917 /* Metdata version.
2918  * This is one of
2919  *   'none' for arrays with no metadata (good luck...)
2920  *   'external' for arrays with externally managed metadata,
2921  * or N.M for internally known formats
2922  */
2923 static ssize_t
2924 metadata_show(mddev_t *mddev, char *page)
2925 {
2926         if (mddev->persistent)
2927                 return sprintf(page, "%d.%d\n",
2928                                mddev->major_version, mddev->minor_version);
2929         else if (mddev->external)
2930                 return sprintf(page, "external:%s\n", mddev->metadata_type);
2931         else
2932                 return sprintf(page, "none\n");
2933 }
2934
2935 static ssize_t
2936 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2937 {
2938         int major, minor;
2939         char *e;
2940         /* Changing the details of 'external' metadata is
2941          * always permitted.  Otherwise there must be
2942          * no devices attached to the array.
2943          */
2944         if (mddev->external && strncmp(buf, "external:", 9) == 0)
2945                 ;
2946         else if (!list_empty(&mddev->disks))
2947                 return -EBUSY;
2948
2949         if (cmd_match(buf, "none")) {
2950                 mddev->persistent = 0;
2951                 mddev->external = 0;
2952                 mddev->major_version = 0;
2953                 mddev->minor_version = 90;
2954                 return len;
2955         }
2956         if (strncmp(buf, "external:", 9) == 0) {
2957                 size_t namelen = len-9;
2958                 if (namelen >= sizeof(mddev->metadata_type))
2959                         namelen = sizeof(mddev->metadata_type)-1;
2960                 strncpy(mddev->metadata_type, buf+9, namelen);
2961                 mddev->metadata_type[namelen] = 0;
2962                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
2963                         mddev->metadata_type[--namelen] = 0;
2964                 mddev->persistent = 0;
2965                 mddev->external = 1;
2966                 mddev->major_version = 0;
2967                 mddev->minor_version = 90;
2968                 return len;
2969         }
2970         major = simple_strtoul(buf, &e, 10);
2971         if (e==buf || *e != '.')
2972                 return -EINVAL;
2973         buf = e+1;
2974         minor = simple_strtoul(buf, &e, 10);
2975         if (e==buf || (*e && *e != '\n') )
2976                 return -EINVAL;
2977         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2978                 return -ENOENT;
2979         mddev->major_version = major;
2980         mddev->minor_version = minor;
2981         mddev->persistent = 1;
2982         mddev->external = 0;
2983         return len;
2984 }
2985
2986 static struct md_sysfs_entry md_metadata =
2987 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2988
2989 static ssize_t
2990 action_show(mddev_t *mddev, char *page)
2991 {
2992         char *type = "idle";
2993         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2994             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
2995                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2996                         type = "reshape";
2997                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2998                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2999                                 type = "resync";
3000                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3001                                 type = "check";
3002                         else
3003                                 type = "repair";
3004                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3005                         type = "recover";
3006         }
3007         return sprintf(page, "%s\n", type);
3008 }
3009
3010 static ssize_t
3011 action_store(mddev_t *mddev, const char *page, size_t len)
3012 {
3013         if (!mddev->pers || !mddev->pers->sync_request)
3014                 return -EINVAL;
3015
3016         if (cmd_match(page, "idle")) {
3017                 if (mddev->sync_thread) {
3018                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3019                         md_unregister_thread(mddev->sync_thread);
3020                         mddev->sync_thread = NULL;
3021                         mddev->recovery = 0;
3022                 }
3023         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3024                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3025                 return -EBUSY;
3026         else if (cmd_match(page, "resync"))
3027                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3028         else if (cmd_match(page, "recover")) {
3029                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3030                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3031         } else if (cmd_match(page, "reshape")) {
3032                 int err;
3033                 if (mddev->pers->start_reshape == NULL)
3034                         return -EINVAL;
3035                 err = mddev->pers->start_reshape(mddev);
3036                 if (err)
3037                         return err;
3038                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3039         } else {
3040                 if (cmd_match(page, "check"))
3041                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3042                 else if (!cmd_match(page, "repair"))
3043                         return -EINVAL;
3044                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3045                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3046         }
3047         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3048         md_wakeup_thread(mddev->thread);
3049         sysfs_notify(&mddev->kobj, NULL, "sync_action");
3050         return len;
3051 }
3052
3053 static ssize_t
3054 mismatch_cnt_show(mddev_t *mddev, char *page)
3055 {
3056         return sprintf(page, "%llu\n",
3057                        (unsigned long long) mddev->resync_mismatches);
3058 }
3059
3060 static struct md_sysfs_entry md_scan_mode =
3061 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3062
3063
3064 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3065
3066 static ssize_t
3067 sync_min_show(mddev_t *mddev, char *page)
3068 {
3069         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3070                        mddev->sync_speed_min ? "local": "system");
3071 }
3072
3073 static ssize_t
3074 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3075 {
3076         int min;
3077         char *e;
3078         if (strncmp(buf, "system", 6)==0) {
3079                 mddev->sync_speed_min = 0;
3080                 return len;
3081         }
3082         min = simple_strtoul(buf, &e, 10);
3083         if (buf == e || (*e && *e != '\n') || min <= 0)
3084                 return -EINVAL;
3085         mddev->sync_speed_min = min;
3086         return len;
3087 }
3088
3089 static struct md_sysfs_entry md_sync_min =
3090 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3091
3092 static ssize_t
3093 sync_max_show(mddev_t *mddev, char *page)
3094 {
3095         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3096                        mddev->sync_speed_max ? "local": "system");
3097 }
3098
3099 static ssize_t
3100 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3101 {
3102         int max;
3103         char *e;
3104         if (strncmp(buf, "system", 6)==0) {
3105                 mddev->sync_speed_max = 0;
3106                 return len;
3107         }
3108         max = simple_strtoul(buf, &e, 10);
3109         if (buf == e || (*e && *e != '\n') || max <= 0)
3110                 return -EINVAL;
3111         mddev->sync_speed_max = max;
3112         return len;
3113 }
3114
3115 static struct md_sysfs_entry md_sync_max =
3116 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3117
3118 static ssize_t
3119 degraded_show(mddev_t *mddev, char *page)
3120 {
3121         return sprintf(page, "%d\n", mddev->degraded);
3122 }
3123 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3124
3125 static ssize_t
3126 sync_force_parallel_show(mddev_t *mddev, char *page)
3127 {
3128         return sprintf(page, "%d\n", mddev->parallel_resync);
3129 }
3130
3131 static ssize_t
3132 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3133 {
3134         long n;
3135
3136         if (strict_strtol(buf, 10, &n))
3137                 return -EINVAL;
3138
3139         if (n != 0 && n != 1)
3140                 return -EINVAL;
3141
3142         mddev->parallel_resync = n;
3143
3144         if (mddev->sync_thread)
3145                 wake_up(&resync_wait);
3146
3147         return len;
3148 }
3149
3150 /* force parallel resync, even with shared block devices */
3151 static struct md_sysfs_entry md_sync_force_parallel =
3152 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3153        sync_force_parallel_show, sync_force_parallel_store);
3154
3155 static ssize_t
3156 sync_speed_show(mddev_t *mddev, char *page)
3157 {
3158         unsigned long resync, dt, db;
3159         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3160         dt = (jiffies - mddev->resync_mark) / HZ;
3161         if (!dt) dt++;
3162         db = resync - mddev->resync_mark_cnt;
3163         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3164 }
3165
3166 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3167
3168 static ssize_t
3169 sync_completed_show(mddev_t *mddev, char *page)
3170 {
3171         unsigned long max_blocks, resync;
3172
3173         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3174                 max_blocks = mddev->resync_max_sectors;
3175         else
3176                 max_blocks = mddev->size << 1;
3177
3178         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
3179         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
3180 }
3181
3182 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3183
3184 static ssize_t
3185 min_sync_show(mddev_t *mddev, char *page)
3186 {
3187         return sprintf(page, "%llu\n",
3188                        (unsigned long long)mddev->resync_min);
3189 }
3190 static ssize_t
3191 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3192 {
3193         unsigned long long min;
3194         if (strict_strtoull(buf, 10, &min))
3195                 return -EINVAL;
3196         if (min > mddev->resync_max)
3197                 return -EINVAL;
3198         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3199                 return -EBUSY;
3200
3201         /* Must be a multiple of chunk_size */
3202         if (mddev->chunk_size) {
3203                 if (min & (sector_t)((mddev->chunk_size>>9)-1))
3204                         return -EINVAL;
3205         }
3206         mddev->resync_min = min;
3207
3208         return len;
3209 }
3210
3211 static struct md_sysfs_entry md_min_sync =
3212 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3213
3214 static ssize_t
3215 max_sync_show(mddev_t *mddev, char *page)
3216 {
3217         if (mddev->resync_max == MaxSector)
3218                 return sprintf(page, "max\n");
3219         else
3220                 return sprintf(page, "%llu\n",
3221                                (unsigned long long)mddev->resync_max);
3222 }
3223 static ssize_t
3224 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3225 {
3226         if (strncmp(buf, "max", 3) == 0)
3227                 mddev->resync_max = MaxSector;
3228         else {
3229                 unsigned long long max;
3230                 if (strict_strtoull(buf, 10, &max))
3231                         return -EINVAL;
3232                 if (max < mddev->resync_min)
3233                         return -EINVAL;
3234                 if (max < mddev->resync_max &&
3235                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3236                         return -EBUSY;
3237
3238                 /* Must be a multiple of chunk_size */
3239                 if (mddev->chunk_size) {
3240                         if (max & (sector_t)((mddev->chunk_size>>9)-1))
3241                                 return -EINVAL;
3242                 }
3243                 mddev->resync_max = max;
3244         }
3245         wake_up(&mddev->recovery_wait);
3246         return len;
3247 }
3248
3249 static struct md_sysfs_entry md_max_sync =
3250 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3251
3252 static ssize_t
3253 suspend_lo_show(mddev_t *mddev, char *page)
3254 {
3255         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3256 }
3257
3258 static ssize_t
3259 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3260 {
3261         char *e;
3262         unsigned long long new = simple_strtoull(buf, &e, 10);
3263
3264         if (mddev->pers->quiesce == NULL)
3265                 return -EINVAL;
3266         if (buf == e || (*e && *e != '\n'))
3267                 return -EINVAL;
3268         if (new >= mddev->suspend_hi ||
3269             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3270                 mddev->suspend_lo = new;
3271                 mddev->pers->quiesce(mddev, 2);
3272                 return len;
3273         } else
3274                 return -EINVAL;
3275 }
3276 static struct md_sysfs_entry md_suspend_lo =
3277 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3278
3279
3280 static ssize_t
3281 suspend_hi_show(mddev_t *mddev, char *page)
3282 {
3283         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3284 }
3285
3286 static ssize_t
3287 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3288 {
3289         char *e;
3290         unsigned long long new = simple_strtoull(buf, &e, 10);
3291
3292         if (mddev->pers->quiesce == NULL)
3293                 return -EINVAL;
3294         if (buf == e || (*e && *e != '\n'))
3295                 return -EINVAL;
3296         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3297             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3298                 mddev->suspend_hi = new;
3299                 mddev->pers->quiesce(mddev, 1);
3300                 mddev->pers->quiesce(mddev, 0);
3301                 return len;
3302         } else
3303                 return -EINVAL;
3304 }
3305 static struct md_sysfs_entry md_suspend_hi =
3306 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3307
3308 static ssize_t
3309 reshape_position_show(mddev_t *mddev, char *page)
3310 {
3311         if (mddev->reshape_position != MaxSector)
3312                 return sprintf(page, "%llu\n",
3313                                (unsigned long long)mddev->reshape_position);
3314         strcpy(page, "none\n");
3315         return 5;
3316 }
3317
3318 static ssize_t
3319 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3320 {
3321         char *e;
3322         unsigned long long new = simple_strtoull(buf, &e, 10);
3323         if (mddev->pers)
3324                 return -EBUSY;
3325         if (buf == e || (*e && *e != '\n'))
3326                 return -EINVAL;
3327         mddev->reshape_position = new;
3328         mddev->delta_disks = 0;
3329         mddev->new_level = mddev->level;
3330         mddev->new_layout = mddev->layout;
3331         mddev->new_chunk = mddev->chunk_size;
3332         return len;
3333 }
3334
3335 static struct md_sysfs_entry md_reshape_position =
3336 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3337        reshape_position_store);
3338
3339
3340 static struct attribute *md_default_attrs[] = {
3341         &md_level.attr,
3342         &md_layout.attr,
3343         &md_raid_disks.attr,
3344         &md_chunk_size.attr,
3345         &md_size.attr,
3346         &md_resync_start.attr,
3347         &md_metadata.attr,
3348         &md_new_device.attr,
3349         &md_safe_delay.attr,
3350         &md_array_state.attr,
3351         &md_reshape_position.attr,
3352         NULL,
3353 };
3354
3355 static struct attribute *md_redundancy_attrs[] = {
3356         &md_scan_mode.attr,
3357         &md_mismatches.attr,
3358         &md_sync_min.attr,
3359         &md_sync_max.attr,
3360         &md_sync_speed.attr,
3361         &md_sync_force_parallel.attr,
3362         &md_sync_completed.attr,
3363         &md_min_sync.attr,
3364         &md_max_sync.attr,
3365         &md_suspend_lo.attr,
3366         &md_suspend_hi.attr,
3367         &md_bitmap.attr,
3368         &md_degraded.attr,
3369         NULL,
3370 };
3371 static struct attribute_group md_redundancy_group = {
3372         .name = NULL,
3373         .attrs = md_redundancy_attrs,
3374 };
3375
3376
3377 static ssize_t
3378 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3379 {
3380         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3381         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3382         ssize_t rv;
3383
3384         if (!entry->show)
3385                 return -EIO;
3386         rv = mddev_lock(mddev);
3387         if (!rv) {
3388                 rv = entry->show(mddev, page);
3389                 mddev_unlock(mddev);
3390         }
3391         return rv;
3392 }
3393
3394 static ssize_t
3395 md_attr_store(struct kobject *kobj, struct attribute *attr,
3396               const char *page, size_t length)
3397 {
3398         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3399         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3400         ssize_t rv;
3401
3402         if (!entry->store)
3403                 return -EIO;
3404         if (!capable(CAP_SYS_ADMIN))
3405                 return -EACCES;
3406         rv = mddev_lock(mddev);
3407         if (!rv) {
3408                 rv = entry->store(mddev, page, length);
3409                 mddev_unlock(mddev);
3410         }
3411         return rv;
3412 }
3413
3414 static void md_free(struct kobject *ko)
3415 {
3416         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3417         kfree(mddev);
3418 }
3419
3420 static struct sysfs_ops md_sysfs_ops = {
3421         .show   = md_attr_show,
3422         .store  = md_attr_store,
3423 };
3424 static struct kobj_type md_ktype = {
3425         .release        = md_free,
3426         .sysfs_ops      = &md_sysfs_ops,
3427         .default_attrs  = md_default_attrs,
3428 };
3429
3430 int mdp_major = 0;
3431
3432 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3433 {
3434         static DEFINE_MUTEX(disks_mutex);
3435         mddev_t *mddev = mddev_find(dev);
3436         struct gendisk *disk;
3437         int partitioned = (MAJOR(dev) != MD_MAJOR);
3438         int shift = partitioned ? MdpMinorShift : 0;
3439         int unit = MINOR(dev) >> shift;
3440         int error;
3441
3442         if (!mddev)
3443                 return NULL;
3444
3445         mutex_lock(&disks_mutex);
3446         if (mddev->gendisk) {
3447                 mutex_unlock(&disks_mutex);
3448                 mddev_put(mddev);
3449                 return NULL;
3450         }
3451         disk = alloc_disk(1 << shift);
3452         if (!disk) {
3453                 mutex_unlock(&disks_mutex);
3454                 mddev_put(mddev);
3455                 return NULL;
3456         }
3457         disk->major = MAJOR(dev);
3458         disk->first_minor = unit << shift;
3459         if (partitioned)
3460                 sprintf(disk->disk_name, "md_d%d", unit);
3461         else
3462                 sprintf(disk->disk_name, "md%d", unit);
3463         disk->fops = &md_fops;
3464         disk->private_data = mddev;
3465         disk->queue = mddev->queue;
3466         /* Allow extended partitions.  This makes the
3467          * 'mdp' device redundant, but we can really
3468          * remove it now.
3469          */
3470         disk->flags |= GENHD_FL_EXT_DEVT;
3471         add_disk(disk);
3472         mddev->gendisk = disk;
3473         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3474                                      &disk_to_dev(disk)->kobj, "%s", "md");
3475         mutex_unlock(&disks_mutex);
3476         if (error)
3477                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3478                        disk->disk_name);
3479         else {
3480                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3481                 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3482         }
3483         return NULL;
3484 }
3485
3486 static void md_safemode_timeout(unsigned long data)
3487 {
3488         mddev_t *mddev = (mddev_t *) data;
3489
3490         if (!atomic_read(&mddev->writes_pending)) {
3491                 mddev->safemode = 1;
3492                 if (mddev->external)
3493                         sysfs_notify_dirent(mddev->sysfs_state);
3494         }
3495         md_wakeup_thread(mddev->thread);
3496 }
3497
3498 static int start_dirty_degraded;
3499
3500 static int do_md_run(mddev_t * mddev)
3501 {
3502         int err;
3503         int chunk_size;
3504         struct list_head *tmp;
3505         mdk_rdev_t *rdev;
3506         struct gendisk *disk;
3507         struct mdk_personality *pers;
3508         char b[BDEVNAME_SIZE];
3509
3510         if (list_empty(&mddev->disks))
3511                 /* cannot run an array with no devices.. */
3512                 return -EINVAL;
3513
3514         if (mddev->pers)
3515                 return -EBUSY;
3516
3517         /*
3518          * Analyze all RAID superblock(s)
3519          */
3520         if (!mddev->raid_disks) {
3521                 if (!mddev->persistent)
3522                         return -EINVAL;
3523                 analyze_sbs(mddev);
3524         }
3525
3526         chunk_size = mddev->chunk_size;
3527
3528         if (chunk_size) {
3529                 if (chunk_size > MAX_CHUNK_SIZE) {
3530                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3531                                 chunk_size, MAX_CHUNK_SIZE);
3532                         return -EINVAL;
3533                 }
3534                 /*
3535                  * chunk-size has to be a power of 2
3536                  */
3537                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3538                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3539                         return -EINVAL;
3540                 }
3541
3542                 /* devices must have minimum size of one chunk */
3543                 rdev_for_each(rdev, tmp, mddev) {
3544                         if (test_bit(Faulty, &rdev->flags))
3545                                 continue;
3546                         if (rdev->size < chunk_size / 1024) {
3547                                 printk(KERN_WARNING
3548                                         "md: Dev %s smaller than chunk_size:"
3549                                         " %lluk < %dk\n",
3550                                         bdevname(rdev->bdev,b),
3551                                         (unsigned long long)rdev->size,
3552                                         chunk_size / 1024);
3553                                 return -EINVAL;
3554                         }
3555                 }
3556         }
3557
3558         if (mddev->level != LEVEL_NONE)
3559                 request_module("md-level-%d", mddev->level);
3560         else if (mddev->clevel[0])
3561                 request_module("md-%s", mddev->clevel);
3562
3563         /*
3564          * Drop all container device buffers, from now on
3565          * the only valid external interface is through the md
3566          * device.
3567          */
3568         rdev_for_each(rdev, tmp, mddev) {
3569                 if (test_bit(Faulty, &rdev->flags))
3570                         continue;
3571                 sync_blockdev(rdev->bdev);
3572                 invalidate_bdev(rdev->bdev);
3573
3574                 /* perform some consistency tests on the device.
3575                  * We don't want the data to overlap the metadata,
3576                  * Internal Bitmap issues has handled elsewhere.
3577                  */
3578                 if (rdev->data_offset < rdev->sb_start) {
3579                         if (mddev->size &&
3580                             rdev->data_offset + mddev->size*2
3581                             > rdev->sb_start) {
3582                                 printk("md: %s: data overlaps metadata\n",
3583                                        mdname(mddev));
3584                                 return -EINVAL;
3585                         }
3586                 } else {
3587                         if (rdev->sb_start + rdev->sb_size/512
3588                             > rdev->data_offset) {
3589                                 printk("md: %s: metadata overlaps data\n",
3590                                        mdname(mddev));
3591                                 return -EINVAL;
3592                         }
3593                 }
3594                 sysfs_notify_dirent(rdev->sysfs_state);
3595         }
3596
3597         md_probe(mddev->unit, NULL, NULL);
3598         disk = mddev->gendisk;
3599         if (!disk)
3600                 return -ENOMEM;
3601
3602         spin_lock(&pers_lock);
3603         pers = find_pers(mddev->level, mddev->clevel);
3604         if (!pers || !try_module_get(pers->owner)) {
3605                 spin_unlock(&pers_lock);
3606                 if (mddev->level != LEVEL_NONE)
3607                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3608                                mddev->level);
3609                 else
3610                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3611                                mddev->clevel);
3612                 return -EINVAL;
3613         }
3614         mddev->pers = pers;
3615         spin_unlock(&pers_lock);
3616         mddev->level = pers->level;
3617         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3618
3619         if (mddev->reshape_position != MaxSector &&
3620             pers->start_reshape == NULL) {
3621                 /* This personality cannot handle reshaping... */
3622                 mddev->pers = NULL;
3623                 module_put(pers->owner);
3624                 return -EINVAL;
3625         }
3626
3627         if (pers->sync_request) {
3628                 /* Warn if this is a potentially silly
3629                  * configuration.
3630                  */
3631                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3632                 mdk_rdev_t *rdev2;
3633                 struct list_head *tmp2;
3634                 int warned = 0;
3635                 rdev_for_each(rdev, tmp, mddev) {
3636                         rdev_for_each(rdev2, tmp2, mddev) {
3637                                 if (rdev < rdev2 &&
3638                                     rdev->bdev->bd_contains ==
3639                                     rdev2->bdev->bd_contains) {
3640                                         printk(KERN_WARNING
3641                                                "%s: WARNING: %s appears to be"
3642                                                " on the same physical disk as"
3643                                                " %s.\n",
3644                                                mdname(mddev),
3645                                                bdevname(rdev->bdev,b),
3646                                                bdevname(rdev2->bdev,b2));
3647                                         warned = 1;
3648                                 }
3649                         }
3650                 }
3651                 if (warned)
3652                         printk(KERN_WARNING
3653                                "True protection against single-disk"
3654                                " failure might be compromised.\n");
3655         }
3656
3657         mddev->recovery = 0;
3658         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3659         mddev->barriers_work = 1;
3660         mddev->ok_start_degraded = start_dirty_degraded;
3661
3662         if (start_readonly)
3663                 mddev->ro = 2; /* read-only, but switch on first write */
3664
3665         err = mddev->pers->run(mddev);
3666         if (err)
3667                 printk(KERN_ERR "md: pers->run() failed ...\n");
3668         else if (mddev->pers->sync_request) {
3669                 err = bitmap_create(mddev);
3670                 if (err) {
3671                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3672                                mdname(mddev), err);
3673                         mddev->pers->stop(mddev);
3674                 }
3675         }
3676         if (err) {
3677                 module_put(mddev->pers->owner);
3678                 mddev->pers = NULL;
3679                 bitmap_destroy(mddev);
3680                 return err;
3681         }
3682         if (mddev->pers->sync_request) {
3683                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3684                         printk(KERN_WARNING
3685                                "md: cannot register extra attributes for %s\n",
3686                                mdname(mddev));
3687         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3688                 mddev->ro = 0;
3689
3690         atomic_set(&mddev->writes_pending,0);
3691         mddev->safemode = 0;
3692         mddev->safemode_timer.function = md_safemode_timeout;
3693         mddev->safemode_timer.data = (unsigned long) mddev;
3694         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3695         mddev->in_sync = 1;
3696
3697         rdev_for_each(rdev, tmp, mddev)
3698                 if (rdev->raid_disk >= 0) {
3699                         char nm[20];
3700                         sprintf(nm, "rd%d", rdev->raid_disk);
3701                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3702                                 printk("md: cannot register %s for %s\n",
3703                                        nm, mdname(mddev));
3704                 }
3705         
3706         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3707         
3708         if (mddev->flags)
3709                 md_update_sb(mddev, 0);
3710
3711         set_capacity(disk, mddev->array_sectors);
3712
3713         /* If we call blk_queue_make_request here, it will
3714          * re-initialise max_sectors etc which may have been
3715          * refined inside -> run.  So just set the bits we need to set.
3716          * Most initialisation happended when we called
3717          * blk_queue_make_request(..., md_fail_request)
3718          * earlier.
3719          */
3720         mddev->queue->queuedata = mddev;
3721         mddev->queue->make_request_fn = mddev->pers->make_request;
3722
3723         /* If there is a partially-recovered drive we need to
3724          * start recovery here.  If we leave it to md_check_recovery,
3725          * it will remove the drives and not do the right thing
3726          */
3727         if (mddev->degraded && !mddev->sync_thread) {
3728                 struct list_head *rtmp;
3729                 int spares = 0;
3730                 rdev_for_each(rdev, rtmp, mddev)
3731                         if (rdev->raid_disk >= 0 &&
3732                             !test_bit(In_sync, &rdev->flags) &&
3733                             !test_bit(Faulty, &rdev->flags))
3734                                 /* complete an interrupted recovery */
3735                                 spares++;
3736                 if (spares && mddev->pers->sync_request) {
3737                         mddev->recovery = 0;
3738                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3739                         mddev->sync_thread = md_register_thread(md_do_sync,
3740                                                                 mddev,
3741                                                                 "%s_resync");
3742                         if (!mddev->sync_thread) {
3743                                 printk(KERN_ERR "%s: could not start resync"
3744                                        " thread...\n",
3745                                        mdname(mddev));
3746                                 /* leave the spares where they are, it shouldn't hurt */
3747                                 mddev->recovery = 0;
3748                         }
3749                 }
3750         }
3751         md_wakeup_thread(mddev->thread);
3752         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3753
3754         mddev->changed = 1;
3755         md_new_event(mddev);
3756         sysfs_notify_dirent(mddev->sysfs_state);
3757         sysfs_notify(&mddev->kobj, NULL, "sync_action");
3758         sysfs_notify(&mddev->kobj, NULL, "degraded");
3759         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
3760         return 0;
3761 }
3762
3763 static int restart_array(mddev_t *mddev)
3764 {
3765         struct gendisk *disk = mddev->gendisk;
3766
3767         /* Complain if it has no devices */
3768         if (list_empty(&mddev->disks))
3769                 return -ENXIO;
3770         if (!mddev->pers)
3771                 return -EINVAL;
3772         if (!mddev->ro)
3773                 return -EBUSY;
3774         mddev->safemode = 0;
3775         mddev->ro = 0;
3776         set_disk_ro(disk, 0);
3777         printk(KERN_INFO "md: %s switched to read-write mode.\n",
3778                 mdname(mddev));
3779         /* Kick recovery or resync if necessary */
3780         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3781         md_wakeup_thread(mddev->thread);
3782         md_wakeup_thread(mddev->sync_thread);
3783         sysfs_notify_dirent(mddev->sysfs_state);
3784         return 0;
3785 }
3786
3787 /* similar to deny_write_access, but accounts for our holding a reference
3788  * to the file ourselves */
3789 static int deny_bitmap_write_access(struct file * file)
3790 {
3791         struct inode *inode = file->f_mapping->host;
3792
3793         spin_lock(&inode->i_lock);
3794         if (atomic_read(&inode->i_writecount) > 1) {
3795                 spin_unlock(&inode->i_lock);
3796                 return -ETXTBSY;
3797         }
3798         atomic_set(&inode->i_writecount, -1);
3799         spin_unlock(&inode->i_lock);
3800
3801         return 0;
3802 }
3803
3804 static void restore_bitmap_write_access(struct file *file)
3805 {
3806         struct inode *inode = file->f_mapping->host;
3807
3808         spin_lock(&inode->i_lock);
3809         atomic_set(&inode->i_writecount, 1);
3810         spin_unlock(&inode->i_lock);
3811 }
3812
3813 /* mode:
3814  *   0 - completely stop and dis-assemble array
3815  *   1 - switch to readonly
3816  *   2 - stop but do not disassemble array
3817  */
3818 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
3819 {
3820         int err = 0;
3821         struct gendisk *disk = mddev->gendisk;
3822
3823         if (atomic_read(&mddev->openers) > is_open) {
3824                 printk("md: %s still in use.\n",mdname(mddev));
3825                 return -EBUSY;
3826         }
3827
3828         if (mddev->pers) {
3829
3830                 if (mddev->sync_thread) {
3831                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3832                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3833                         md_unregister_thread(mddev->sync_thread);
3834                         mddev->sync_thread = NULL;
3835                 }
3836
3837                 del_timer_sync(&mddev->safemode_timer);
3838
3839                 switch(mode) {
3840                 case 1: /* readonly */
3841                         err  = -ENXIO;
3842                         if (mddev->ro==1)
3843                                 goto out;
3844                         mddev->ro = 1;
3845                         break;
3846                 case 0: /* disassemble */
3847                 case 2: /* stop */
3848                         bitmap_flush(mddev);
3849                         md_super_wait(mddev);
3850                         if (mddev->ro)
3851                                 set_disk_ro(disk, 0);
3852                         blk_queue_make_request(mddev->queue, md_fail_request);
3853                         mddev->pers->stop(mddev);
3854                         mddev->queue->merge_bvec_fn = NULL;
3855                         mddev->queue->unplug_fn = NULL;
3856                         mddev->queue->backing_dev_info.congested_fn = NULL;
3857                         if (mddev->pers->sync_request)
3858                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3859
3860                         module_put(mddev->pers->owner);
3861                         mddev->pers = NULL;
3862                         /* tell userspace to handle 'inactive' */
3863                         sysfs_notify_dirent(mddev->sysfs_state);
3864
3865                         set_capacity(disk, 0);
3866                         mddev->changed = 1;
3867
3868                         if (mddev->ro)
3869                                 mddev->ro = 0;
3870                 }
3871                 if (!mddev->in_sync || mddev->flags) {
3872                         /* mark array as shutdown cleanly */
3873                         mddev->in_sync = 1;
3874                         md_update_sb(mddev, 1);
3875                 }
3876                 if (mode == 1)
3877                         set_disk_ro(disk, 1);
3878                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3879         }
3880
3881         /*
3882          * Free resources if final stop
3883          */
3884         if (mode == 0) {
3885                 mdk_rdev_t *rdev;
3886                 struct list_head *tmp;
3887
3888                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3889
3890                 bitmap_destroy(mddev);
3891                 if (mddev->bitmap_file) {
3892                         restore_bitmap_write_access(mddev->bitmap_file);
3893                         fput(mddev->bitmap_file);
3894                         mddev->bitmap_file = NULL;
3895                 }
3896                 mddev->bitmap_offset = 0;
3897
3898                 rdev_for_each(rdev, tmp, mddev)
3899                         if (rdev->raid_disk >= 0) {
3900                                 char nm[20];
3901                                 sprintf(nm, "rd%d", rdev->raid_disk);
3902                                 sysfs_remove_link(&mddev->kobj, nm);
3903                         }
3904
3905                 /* make sure all md_delayed_delete calls have finished */
3906                 flush_scheduled_work();
3907
3908                 export_array(mddev);
3909
3910                 mddev->array_sectors = 0;
3911                 mddev->size = 0;
3912                 mddev->raid_disks = 0;
3913                 mddev->recovery_cp = 0;
3914                 mddev->resync_min = 0;
3915                 mddev->resync_max = MaxSector;
3916                 mddev->reshape_position = MaxSector;
3917                 mddev->external = 0;
3918                 mddev->persistent = 0;
3919                 mddev->level = LEVEL_NONE;
3920                 mddev->clevel[0] = 0;
3921                 mddev->flags = 0;
3922                 mddev->ro = 0;
3923                 mddev->metadata_type[0] = 0;
3924                 mddev->chunk_size = 0;
3925                 mddev->ctime = mddev->utime = 0;
3926                 mddev->layout = 0;
3927                 mddev->max_disks = 0;
3928                 mddev->events = 0;
3929                 mddev->delta_disks = 0;
3930                 mddev->new_level = LEVEL_NONE;
3931                 mddev->new_layout = 0;
3932                 mddev->new_chunk = 0;
3933                 mddev->curr_resync = 0;
3934                 mddev->resync_mismatches = 0;
3935                 mddev->suspend_lo = mddev->suspend_hi = 0;
3936                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
3937                 mddev->recovery = 0;
3938                 mddev->in_sync = 0;
3939                 mddev->changed = 0;
3940                 mddev->degraded = 0;
3941                 mddev->barriers_work = 0;
3942                 mddev->safemode = 0;
3943
3944         } else if (mddev->pers)
3945                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3946                         mdname(mddev));
3947         err = 0;
3948         md_new_event(mddev);
3949         sysfs_notify_dirent(mddev->sysfs_state);
3950 out:
3951         return err;
3952 }
3953
3954 #ifndef MODULE
3955 static void autorun_array(mddev_t *mddev)
3956 {
3957         mdk_rdev_t *rdev;
3958         struct list_head *tmp;
3959         int err;
3960
3961         if (list_empty(&mddev->disks))
3962                 return;
3963
3964         printk(KERN_INFO "md: running: ");
3965
3966         rdev_for_each(rdev, tmp, mddev) {
3967                 char b[BDEVNAME_SIZE];
3968                 printk("<%s>", bdevname(rdev->bdev,b));
3969         }
3970         printk("\n");
3971
3972         err = do_md_run(mddev);
3973         if (err) {
3974                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3975                 do_md_stop(mddev, 0, 0);
3976         }
3977 }
3978
3979 /*
3980  * lets try to run arrays based on all disks that have arrived
3981  * until now. (those are in pending_raid_disks)
3982  *
3983  * the method: pick the first pending disk, collect all disks with
3984  * the same UUID, remove all from the pending list and put them into
3985  * the 'same_array' list. Then order this list based on superblock
3986  * update time (freshest comes first), kick out 'old' disks and
3987  * compare superblocks. If everything's fine then run it.
3988  *
3989  * If "unit" is allocated, then bump its reference count
3990  */
3991 static void autorun_devices(int part)
3992 {
3993         struct list_head *tmp;
3994         mdk_rdev_t *rdev0, *rdev;
3995         mddev_t *mddev;
3996         char b[BDEVNAME_SIZE];
3997
3998         printk(KERN_INFO "md: autorun ...\n");
3999         while (!list_empty(&pending_raid_disks)) {
4000                 int unit;
4001                 dev_t dev;
4002                 LIST_HEAD(candidates);
4003                 rdev0 = list_entry(pending_raid_disks.next,
4004                                          mdk_rdev_t, same_set);
4005
4006                 printk(KERN_INFO "md: considering %s ...\n",
4007                         bdevname(rdev0->bdev,b));
4008                 INIT_LIST_HEAD(&candidates);
4009                 rdev_for_each_list(rdev, tmp, pending_raid_disks)
4010                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4011                                 printk(KERN_INFO "md:  adding %s ...\n",
4012                                         bdevname(rdev->bdev,b));
4013                                 list_move(&rdev->same_set, &candidates);
4014                         }
4015                 /*
4016                  * now we have a set of devices, with all of them having
4017                  * mostly sane superblocks. It's time to allocate the
4018                  * mddev.
4019                  */
4020                 if (part) {
4021                         dev = MKDEV(mdp_major,
4022                                     rdev0->preferred_minor << MdpMinorShift);
4023                         unit = MINOR(dev) >> MdpMinorShift;
4024                 } else {
4025                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4026                         unit = MINOR(dev);
4027                 }
4028                 if (rdev0->preferred_minor != unit) {
4029                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4030                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4031                         break;
4032                 }
4033
4034                 md_probe(dev, NULL, NULL);
4035                 mddev = mddev_find(dev);
4036                 if (!mddev || !mddev->gendisk) {
4037                         if (mddev)
4038                                 mddev_put(mddev);
4039                         printk(KERN_ERR
4040                                 "md: cannot allocate memory for md drive.\n");
4041                         break;
4042                 }
4043                 if (mddev_lock(mddev)) 
4044                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4045                                mdname(mddev));
4046                 else if (mddev->raid_disks || mddev->major_version
4047                          || !list_empty(&mddev->disks)) {
4048                         printk(KERN_WARNING 
4049                                 "md: %s already running, cannot run %s\n",
4050                                 mdname(mddev), bdevname(rdev0->bdev,b));
4051                         mddev_unlock(mddev);
4052                 } else {
4053                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4054                         mddev->persistent = 1;
4055                         rdev_for_each_list(rdev, tmp, candidates) {
4056                                 list_del_init(&rdev->same_set);
4057                                 if (bind_rdev_to_array(rdev, mddev))
4058                                         export_rdev(rdev);
4059                         }
4060                         autorun_array(mddev);
4061                         mddev_unlock(mddev);
4062                 }
4063                 /* on success, candidates will be empty, on error
4064                  * it won't...
4065                  */
4066                 rdev_for_each_list(rdev, tmp, candidates) {
4067                         list_del_init(&rdev->same_set);
4068                         export_rdev(rdev);
4069                 }
4070                 mddev_put(mddev);
4071         }
4072         printk(KERN_INFO "md: ... autorun DONE.\n");
4073 }
4074 #endif /* !MODULE */
4075
4076 static int get_version(void __user * arg)
4077 {
4078         mdu_version_t ver;
4079
4080         ver.major = MD_MAJOR_VERSION;
4081         ver.minor = MD_MINOR_VERSION;
4082         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4083
4084         if (copy_to_user(arg, &ver, sizeof(ver)))
4085                 return -EFAULT;
4086
4087         return 0;
4088 }
4089
4090 static int get_array_info(mddev_t * mddev, void __user * arg)
4091 {
4092         mdu_array_info_t info;
4093         int nr,working,active,failed,spare;
4094         mdk_rdev_t *rdev;
4095         struct list_head *tmp;
4096
4097         nr=working=active=failed=spare=0;
4098         rdev_for_each(rdev, tmp, mddev) {
4099                 nr++;
4100                 if (test_bit(Faulty, &rdev->flags))
4101                         failed++;
4102                 else {
4103                         working++;
4104                         if (test_bit(In_sync, &rdev->flags))
4105                                 active++;       
4106                         else
4107                                 spare++;
4108                 }
4109         }
4110
4111         info.major_version = mddev->major_version;
4112         info.minor_version = mddev->minor_version;
4113         info.patch_version = MD_PATCHLEVEL_VERSION;
4114         info.ctime         = mddev->ctime;
4115         info.level         = mddev->level;
4116         info.size          = mddev->size;
4117         if (info.size != mddev->size) /* overflow */
4118                 info.size = -1;
4119         info.nr_disks      = nr;
4120         info.raid_disks    = mddev->raid_disks;
4121         info.md_minor      = mddev->md_minor;
4122         info.not_persistent= !mddev->persistent;
4123
4124         info.utime         = mddev->utime;
4125         info.state         = 0;
4126         if (mddev->in_sync)
4127                 info.state = (1<<MD_SB_CLEAN);
4128         if (mddev->bitmap && mddev->bitmap_offset)
4129                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4130         info.active_disks  = active;
4131         info.working_disks = working;
4132         info.failed_disks  = failed;
4133         info.spare_disks   = spare;
4134
4135         info.layout        = mddev->layout;
4136         info.chunk_size    = mddev->chunk_size;
4137
4138         if (copy_to_user(arg, &info, sizeof(info)))
4139                 return -EFAULT;
4140
4141         return 0;
4142 }
4143
4144 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4145 {
4146         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4147         char *ptr, *buf = NULL;
4148         int err = -ENOMEM;
4149
4150         if (md_allow_write(mddev))
4151                 file = kmalloc(sizeof(*file), GFP_NOIO);
4152         else
4153                 file = kmalloc(sizeof(*file), GFP_KERNEL);
4154
4155         if (!file)
4156                 goto out;
4157
4158         /* bitmap disabled, zero the first byte and copy out */
4159         if (!mddev->bitmap || !mddev->bitmap->file) {
4160                 file->pathname[0] = '\0';
4161                 goto copy_out;
4162         }
4163
4164         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4165         if (!buf)
4166                 goto out;
4167
4168         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4169         if (IS_ERR(ptr))
4170                 goto out;
4171
4172         strcpy(file->pathname, ptr);
4173
4174 copy_out:
4175         err = 0;
4176         if (copy_to_user(arg, file, sizeof(*file)))
4177                 err = -EFAULT;
4178 out:
4179         kfree(buf);
4180         kfree(file);
4181         return err;
4182 }
4183
4184 static int get_disk_info(mddev_t * mddev, void __user * arg)
4185 {
4186         mdu_disk_info_t info;
4187         mdk_rdev_t *rdev;
4188
4189         if (copy_from_user(&info, arg, sizeof(info)))
4190                 return -EFAULT;
4191
4192         rdev = find_rdev_nr(mddev, info.number);
4193         if (rdev) {
4194                 info.major = MAJOR(rdev->bdev->bd_dev);
4195                 info.minor = MINOR(rdev->bdev->bd_dev);
4196                 info.raid_disk = rdev->raid_disk;
4197                 info.state = 0;
4198                 if (test_bit(Faulty, &rdev->flags))
4199                         info.state |= (1<<MD_DISK_FAULTY);
4200                 else if (test_bit(In_sync, &rdev->flags)) {
4201                         info.state |= (1<<MD_DISK_ACTIVE);
4202                         info.state |= (1<<MD_DISK_SYNC);
4203                 }
4204                 if (test_bit(WriteMostly, &rdev->flags))
4205                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4206         } else {
4207                 info.major = info.minor = 0;
4208                 info.raid_disk = -1;
4209                 info.state = (1<<MD_DISK_REMOVED);
4210         }
4211
4212         if (copy_to_user(arg, &info, sizeof(info)))
4213                 return -EFAULT;
4214
4215         return 0;
4216 }
4217
4218 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4219 {
4220         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4221         mdk_rdev_t *rdev;
4222         dev_t dev = MKDEV(info->major,info->minor);
4223
4224         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4225                 return -EOVERFLOW;
4226
4227         if (!mddev->raid_disks) {
4228                 int err;
4229                 /* expecting a device which has a superblock */
4230                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4231                 if (IS_ERR(rdev)) {
4232                         printk(KERN_WARNING 
4233                                 "md: md_import_device returned %ld\n",
4234                                 PTR_ERR(rdev));
4235                         return PTR_ERR(rdev);
4236                 }
4237                 if (!list_empty(&mddev->disks)) {
4238                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4239                                                         mdk_rdev_t, same_set);
4240                         int err = super_types[mddev->major_version]
4241                                 .load_super(rdev, rdev0, mddev->minor_version);
4242                         if (err < 0) {
4243                                 printk(KERN_WARNING 
4244                                         "md: %s has different UUID to %s\n",
4245                                         bdevname(rdev->bdev,b), 
4246                                         bdevname(rdev0->bdev,b2));
4247                                 export_rdev(rdev);
4248                                 return -EINVAL;
4249                         }
4250                 }
4251                 err = bind_rdev_to_array(rdev, mddev);
4252                 if (err)
4253                         export_rdev(rdev);
4254                 return err;
4255         }
4256
4257         /*
4258          * add_new_disk can be used once the array is assembled
4259          * to add "hot spares".  They must already have a superblock
4260          * written
4261          */
4262         if (mddev->pers) {
4263                 int err;
4264                 if (!mddev->pers->hot_add_disk) {
4265                         printk(KERN_WARNING 
4266                                 "%s: personality does not support diskops!\n",
4267                                mdname(mddev));
4268                         return -EINVAL;
4269                 }
4270                 if (mddev->persistent)
4271                         rdev = md_import_device(dev, mddev->major_version,
4272                                                 mddev->minor_version);
4273                 else
4274                         rdev = md_import_device(dev, -1, -1);
4275                 if (IS_ERR(rdev)) {
4276                         printk(KERN_WARNING 
4277                                 "md: md_import_device returned %ld\n",
4278                                 PTR_ERR(rdev));
4279                         return PTR_ERR(rdev);
4280                 }
4281                 /* set save_raid_disk if appropriate */
4282                 if (!mddev->persistent) {
4283                         if (info->state & (1<<MD_DISK_SYNC)  &&
4284                             info->raid_disk < mddev->raid_disks)
4285                                 rdev->raid_disk = info->raid_disk;
4286                         else
4287                                 rdev->raid_disk = -1;
4288                 } else
4289                         super_types[mddev->major_version].
4290                                 validate_super(mddev, rdev);
4291                 rdev->saved_raid_disk = rdev->raid_disk;
4292
4293                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4294                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4295                         set_bit(WriteMostly, &rdev->flags);
4296
4297                 rdev->raid_disk = -1;
4298                 err = bind_rdev_to_array(rdev, mddev);
4299                 if (!err && !mddev->pers->hot_remove_disk) {
4300                         /* If there is hot_add_disk but no hot_remove_disk
4301                          * then added disks for geometry changes,
4302                          * and should be added immediately.
4303                          */
4304                         super_types[mddev->major_version].
4305                                 validate_super(mddev, rdev);
4306                         err = mddev->pers->hot_add_disk(mddev, rdev);
4307                         if (err)
4308                                 unbind_rdev_from_array(rdev);
4309                 }
4310                 if (err)
4311                         export_rdev(rdev);
4312                 else
4313                         sysfs_notify_dirent(rdev->sysfs_state);
4314
4315                 md_update_sb(mddev, 1);
4316                 if (mddev->degraded)
4317                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4318                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4319                 md_wakeup_thread(mddev->thread);
4320                 return err;
4321         }
4322
4323         /* otherwise, add_new_disk is only allowed
4324          * for major_version==0 superblocks
4325          */
4326         if (mddev->major_version != 0) {
4327                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4328                        mdname(mddev));
4329                 return -EINVAL;
4330         }
4331
4332         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4333                 int err;
4334                 rdev = md_import_device(dev, -1, 0);
4335                 if (IS_ERR(rdev)) {
4336                         printk(KERN_WARNING 
4337                                 "md: error, md_import_device() returned %ld\n",
4338                                 PTR_ERR(rdev));
4339                         return PTR_ERR(rdev);
4340                 }
4341                 rdev->desc_nr = info->number;
4342                 if (info->raid_disk < mddev->raid_disks)
4343                         rdev->raid_disk = info->raid_disk;
4344                 else
4345                         rdev->raid_disk = -1;
4346
4347                 if (rdev->raid_disk < mddev->raid_disks)
4348                         if (info->state & (1<<MD_DISK_SYNC))
4349                                 set_bit(In_sync, &rdev->flags);
4350
4351                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4352                         set_bit(WriteMostly, &rdev->flags);
4353
4354                 if (!mddev->persistent) {
4355                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4356                         rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4357                 } else 
4358                         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4359                 rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4360
4361                 err = bind_rdev_to_array(rdev, mddev);
4362                 if (err) {
4363                         export_rdev(rdev);
4364                         return err;
4365                 }
4366         }
4367
4368         return 0;
4369 }
4370
4371 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4372 {
4373         char b[BDEVNAME_SIZE];
4374         mdk_rdev_t *rdev;
4375
4376         rdev = find_rdev(mddev, dev);
4377         if (!rdev)
4378                 return -ENXIO;
4379
4380         if (rdev->raid_disk >= 0)
4381                 goto busy;
4382
4383         kick_rdev_from_array(rdev);
4384         md_update_sb(mddev, 1);
4385         md_new_event(mddev);
4386
4387         return 0;
4388 busy:
4389         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4390                 bdevname(rdev->bdev,b), mdname(mddev));
4391         return -EBUSY;
4392 }
4393
4394 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4395 {
4396         char b[BDEVNAME_SIZE];
4397         int err;
4398         mdk_rdev_t *rdev;
4399
4400         if (!mddev->pers)
4401                 return -ENODEV;
4402
4403         if (mddev->major_version != 0) {
4404                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4405                         " version-0 superblocks.\n",
4406                         mdname(mddev));
4407                 return -EINVAL;
4408         }
4409         if (!mddev->pers->hot_add_disk) {
4410                 printk(KERN_WARNING 
4411                         "%s: personality does not support diskops!\n",
4412                         mdname(mddev));
4413                 return -EINVAL;
4414         }
4415
4416         rdev = md_import_device(dev, -1, 0);
4417         if (IS_ERR(rdev)) {
4418                 printk(KERN_WARNING 
4419                         "md: error, md_import_device() returned %ld\n",
4420                         PTR_ERR(rdev));
4421                 return -EINVAL;
4422         }
4423
4424         if (mddev->persistent)
4425                 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4426         else
4427                 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4428
4429         rdev->size = calc_num_sectors(rdev, mddev->chunk_size) / 2;
4430
4431         if (test_bit(Faulty, &rdev->flags)) {
4432                 printk(KERN_WARNING 
4433                         "md: can not hot-add faulty %s disk to %s!\n",
4434                         bdevname(rdev->bdev,b), mdname(mddev));
4435                 err = -EINVAL;
4436                 goto abort_export;
4437         }
4438         clear_bit(In_sync, &rdev->flags);
4439         rdev->desc_nr = -1;
4440         rdev->saved_raid_disk = -1;
4441         err = bind_rdev_to_array(rdev, mddev);
4442         if (err)
4443                 goto abort_export;
4444
4445         /*
4446          * The rest should better be atomic, we can have disk failures
4447          * noticed in interrupt contexts ...
4448          */
4449
4450         if (rdev->desc_nr == mddev->max_disks) {
4451                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4452                         mdname(mddev));
4453                 err = -EBUSY;
4454                 goto abort_unbind_export;
4455         }
4456
4457         rdev->raid_disk = -1;
4458
4459         md_update_sb(mddev, 1);
4460
4461         /*
4462          * Kick recovery, maybe this spare has to be added to the
4463          * array immediately.
4464          */
4465         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4466         md_wakeup_thread(mddev->thread);
4467         md_new_event(mddev);
4468         return 0;
4469
4470 abort_unbind_export:
4471         unbind_rdev_from_array(rdev);
4472
4473 abort_export:
4474         export_rdev(rdev);
4475         return err;
4476 }
4477
4478 static int set_bitmap_file(mddev_t *mddev, int fd)
4479 {
4480         int err;
4481
4482         if (mddev->pers) {
4483                 if (!mddev->pers->quiesce)
4484                         return -EBUSY;
4485                 if (mddev->recovery || mddev->sync_thread)
4486                         return -EBUSY;
4487                 /* we should be able to change the bitmap.. */
4488         }
4489
4490
4491         if (fd >= 0) {
4492                 if (mddev->bitmap)
4493                         return -EEXIST; /* cannot add when bitmap is present */
4494                 mddev->bitmap_file = fget(fd);
4495
4496                 if (mddev->bitmap_file == NULL) {
4497                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4498                                mdname(mddev));
4499                         return -EBADF;
4500                 }
4501
4502                 err = deny_bitmap_write_access(mddev->bitmap_file);
4503                 if (err) {
4504                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4505                                mdname(mddev));
4506                         fput(mddev->bitmap_file);
4507                         mddev->bitmap_file = NULL;
4508                         return err;
4509                 }
4510                 mddev->bitmap_offset = 0; /* file overrides offset */
4511         } else if (mddev->bitmap == NULL)
4512                 return -ENOENT; /* cannot remove what isn't there */
4513         err = 0;
4514         if (mddev->pers) {
4515                 mddev->pers->quiesce(mddev, 1);
4516                 if (fd >= 0)
4517                         err = bitmap_create(mddev);
4518                 if (fd < 0 || err) {
4519                         bitmap_destroy(mddev);
4520                         fd = -1; /* make sure to put the file */
4521                 }
4522                 mddev->pers->quiesce(mddev, 0);
4523         }
4524         if (fd < 0) {
4525                 if (mddev->bitmap_file) {
4526                         restore_bitmap_write_access(mddev->bitmap_file);
4527                         fput(mddev->bitmap_file);
4528                 }
4529                 mddev->bitmap_file = NULL;
4530         }
4531
4532         return err;
4533 }
4534
4535 /*
4536  * set_array_info is used two different ways
4537  * The original usage is when creating a new array.
4538  * In this usage, raid_disks is > 0 and it together with
4539  *  level, size, not_persistent,layout,chunksize determine the
4540  *  shape of the array.
4541  *  This will always create an array with a type-0.90.0 superblock.
4542  * The newer usage is when assembling an array.
4543  *  In this case raid_disks will be 0, and the major_version field is
4544  *  use to determine which style super-blocks are to be found on the devices.
4545  *  The minor and patch _version numbers are also kept incase the
4546  *  super_block handler wishes to interpret them.
4547  */
4548 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4549 {
4550
4551         if (info->raid_disks == 0) {
4552                 /* just setting version number for superblock loading */
4553                 if (info->major_version < 0 ||
4554                     info->major_version >= ARRAY_SIZE(super_types) ||
4555                     super_types[info->major_version].name == NULL) {
4556                         /* maybe try to auto-load a module? */
4557                         printk(KERN_INFO 
4558                                 "md: superblock version %d not known\n",
4559                                 info->major_version);
4560                         return -EINVAL;
4561                 }
4562                 mddev->major_version = info->major_version;
4563                 mddev->minor_version = info->minor_version;
4564                 mddev->patch_version = info->patch_version;
4565                 mddev->persistent = !info->not_persistent;
4566                 return 0;
4567         }
4568         mddev->major_version = MD_MAJOR_VERSION;
4569         mddev->minor_version = MD_MINOR_VERSION;
4570         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4571         mddev->ctime         = get_seconds();
4572
4573         mddev->level         = info->level;
4574         mddev->clevel[0]     = 0;
4575         mddev->size          = info->size;
4576         mddev->raid_disks    = info->raid_disks;
4577         /* don't set md_minor, it is determined by which /dev/md* was
4578          * openned
4579          */
4580         if (info->state & (1<<MD_SB_CLEAN))
4581                 mddev->recovery_cp = MaxSector;
4582         else
4583                 mddev->recovery_cp = 0;
4584         mddev->persistent    = ! info->not_persistent;
4585         mddev->external      = 0;
4586
4587         mddev->layout        = info->layout;
4588         mddev->chunk_size    = info->chunk_size;
4589
4590         mddev->max_disks     = MD_SB_DISKS;
4591
4592         if (mddev->persistent)
4593                 mddev->flags         = 0;
4594         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4595
4596         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4597         mddev->bitmap_offset = 0;
4598
4599         mddev->reshape_position = MaxSector;
4600
4601         /*
4602          * Generate a 128 bit UUID
4603          */
4604         get_random_bytes(mddev->uuid, 16);
4605
4606         mddev->new_level = mddev->level;
4607         mddev->new_chunk = mddev->chunk_size;
4608         mddev->new_layout = mddev->layout;
4609         mddev->delta_disks = 0;
4610
4611         return 0;
4612 }
4613
4614 static int update_size(mddev_t *mddev, sector_t num_sectors)
4615 {
4616         mdk_rdev_t * rdev;
4617         int rv;
4618         struct list_head *tmp;
4619         int fit = (num_sectors == 0);
4620
4621         if (mddev->pers->resize == NULL)
4622                 return -EINVAL;
4623         /* The "num_sectors" is the number of sectors of each device that
4624          * is used.  This can only make sense for arrays with redundancy.
4625          * linear and raid0 always use whatever space is available. We can only
4626          * consider changing this number if no resync or reconstruction is
4627          * happening, and if the new size is acceptable. It must fit before the
4628          * sb_start or, if that is <data_offset, it must fit before the size
4629          * of each device.  If num_sectors is zero, we find the largest size
4630          * that fits.
4631
4632          */
4633         if (mddev->sync_thread)
4634                 return -EBUSY;
4635         if (mddev->bitmap)
4636                 /* Sorry, cannot grow a bitmap yet, just remove it,
4637                  * grow, and re-add.
4638                  */
4639                 return -EBUSY;
4640         rdev_for_each(rdev, tmp, mddev) {
4641                 sector_t avail;
4642                 avail = rdev->size * 2;
4643
4644                 if (fit && (num_sectors == 0 || num_sectors > avail))
4645                         num_sectors = avail;
4646                 if (avail < num_sectors)
4647                         return -ENOSPC;
4648         }
4649         rv = mddev->pers->resize(mddev, num_sectors);
4650         if (!rv) {
4651                 struct block_device *bdev;
4652
4653                 bdev = bdget_disk(mddev->gendisk, 0);
4654                 if (bdev) {
4655                         mutex_lock(&bdev->bd_inode->i_mutex);
4656                         i_size_write(bdev->bd_inode,
4657                                      (loff_t)mddev->array_sectors << 9);
4658                         mutex_unlock(&bdev->bd_inode->i_mutex);
4659                         bdput(bdev);
4660                 }
4661         }
4662         return rv;
4663 }
4664
4665 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4666 {
4667         int rv;
4668         /* change the number of raid disks */
4669         if (mddev->pers->check_reshape == NULL)
4670                 return -EINVAL;
4671         if (raid_disks <= 0 ||
4672             raid_disks >= mddev->max_disks)
4673                 return -EINVAL;
4674         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4675                 return -EBUSY;
4676         mddev->delta_disks = raid_disks - mddev->raid_disks;
4677
4678         rv = mddev->pers->check_reshape(mddev);
4679         return rv;
4680 }
4681
4682
4683 /*
4684  * update_array_info is used to change the configuration of an
4685  * on-line array.
4686  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4687  * fields in the info are checked against the array.
4688  * Any differences that cannot be handled will cause an error.
4689  * Normally, only one change can be managed at a time.
4690  */
4691 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4692 {
4693         int rv = 0;
4694         int cnt = 0;
4695         int state = 0;
4696
4697         /* calculate expected state,ignoring low bits */
4698         if (mddev->bitmap && mddev->bitmap_offset)
4699                 state |= (1 << MD_SB_BITMAP_PRESENT);
4700
4701         if (mddev->major_version != info->major_version ||
4702             mddev->minor_version != info->minor_version ||
4703 /*          mddev->patch_version != info->patch_version || */
4704             mddev->ctime         != info->ctime         ||
4705             mddev->level         != info->level         ||
4706 /*          mddev->layout        != info->layout        || */
4707             !mddev->persistent   != info->not_persistent||
4708             mddev->chunk_size    != info->chunk_size    ||
4709             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4710             ((state^info->state) & 0xfffffe00)
4711                 )
4712                 return -EINVAL;
4713         /* Check there is only one change */
4714         if (info->size >= 0 && mddev->size != info->size) cnt++;
4715         if (mddev->raid_disks != info->raid_disks) cnt++;
4716         if (mddev->layout != info->layout) cnt++;
4717         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4718         if (cnt == 0) return 0;
4719         if (cnt > 1) return -EINVAL;
4720
4721         if (mddev->layout != info->layout) {
4722                 /* Change layout
4723                  * we don't need to do anything at the md level, the
4724                  * personality will take care of it all.
4725                  */
4726                 if (mddev->pers->reconfig == NULL)
4727                         return -EINVAL;
4728                 else
4729                         return mddev->pers->reconfig(mddev, info->layout, -1);
4730         }
4731         if (info->size >= 0 && mddev->size != info->size)
4732                 rv = update_size(mddev, (sector_t)info->size * 2);
4733
4734         if (mddev->raid_disks    != info->raid_disks)
4735                 rv = update_raid_disks(mddev, info->raid_disks);
4736
4737         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4738                 if (mddev->pers->quiesce == NULL)
4739                         return -EINVAL;
4740                 if (mddev->recovery || mddev->sync_thread)
4741                         return -EBUSY;
4742                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4743                         /* add the bitmap */
4744                         if (mddev->bitmap)
4745                                 return -EEXIST;
4746                         if (mddev->default_bitmap_offset == 0)
4747                                 return -EINVAL;
4748                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4749                         mddev->pers->quiesce(mddev, 1);
4750                         rv = bitmap_create(mddev);
4751                         if (rv)
4752                                 bitmap_destroy(mddev);
4753                         mddev->pers->quiesce(mddev, 0);
4754                 } else {
4755                         /* remove the bitmap */
4756                         if (!mddev->bitmap)
4757                                 return -ENOENT;
4758                         if (mddev->bitmap->file)
4759                                 return -EINVAL;
4760                         mddev->pers->quiesce(mddev, 1);
4761                         bitmap_destroy(mddev);
4762                         mddev->pers->quiesce(mddev, 0);
4763                         mddev->bitmap_offset = 0;
4764                 }
4765         }
4766         md_update_sb(mddev, 1);
4767         return rv;
4768 }
4769
4770 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4771 {
4772         mdk_rdev_t *rdev;
4773
4774         if (mddev->pers == NULL)
4775                 return -ENODEV;
4776
4777         rdev = find_rdev(mddev, dev);
4778         if (!rdev)
4779                 return -ENODEV;
4780
4781         md_error(mddev, rdev);
4782         return 0;
4783 }
4784
4785 /*
4786  * We have a problem here : there is no easy way to give a CHS
4787  * virtual geometry. We currently pretend that we have a 2 heads
4788  * 4 sectors (with a BIG number of cylinders...). This drives
4789  * dosfs just mad... ;-)
4790  */
4791 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4792 {
4793         mddev_t *mddev = bdev->bd_disk->private_data;
4794
4795         geo->heads = 2;
4796         geo->sectors = 4;
4797         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4798         return 0;
4799 }
4800
4801 static int md_ioctl(struct inode *inode, struct file *file,
4802                         unsigned int cmd, unsigned long arg)
4803 {
4804         int err = 0;
4805         void __user *argp = (void __user *)arg;
4806         mddev_t *mddev = NULL;
4807
4808         if (!capable(CAP_SYS_ADMIN))
4809                 return -EACCES;
4810
4811         /*
4812          * Commands dealing with the RAID driver but not any
4813          * particular array:
4814          */
4815         switch (cmd)
4816         {
4817                 case RAID_VERSION:
4818                         err = get_version(argp);
4819                         goto done;
4820
4821                 case PRINT_RAID_DEBUG:
4822                         err = 0;
4823                         md_print_devices();
4824                         goto done;
4825
4826 #ifndef MODULE
4827                 case RAID_AUTORUN:
4828                         err = 0;
4829                         autostart_arrays(arg);
4830                         goto done;
4831 #endif
4832                 default:;
4833         }
4834
4835         /*
4836          * Commands creating/starting a new array:
4837          */
4838
4839         mddev = inode->i_bdev->bd_disk->private_data;
4840
4841         if (!mddev) {
4842                 BUG();
4843                 goto abort;
4844         }
4845
4846         err = mddev_lock(mddev);
4847         if (err) {
4848                 printk(KERN_INFO 
4849                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4850                         err, cmd);
4851                 goto abort;
4852         }
4853
4854         switch (cmd)
4855         {
4856                 case SET_ARRAY_INFO:
4857                         {
4858                                 mdu_array_info_t info;
4859                                 if (!arg)
4860                                         memset(&info, 0, sizeof(info));
4861                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4862                                         err = -EFAULT;
4863                                         goto abort_unlock;
4864                                 }
4865                                 if (mddev->pers) {
4866                                         err = update_array_info(mddev, &info);
4867                                         if (err) {
4868                                                 printk(KERN_WARNING "md: couldn't update"
4869                                                        " array info. %d\n", err);
4870                                                 goto abort_unlock;
4871                                         }
4872                                         goto done_unlock;
4873                                 }
4874                                 if (!list_empty(&mddev->disks)) {
4875                                         printk(KERN_WARNING
4876                                                "md: array %s already has disks!\n",
4877                                                mdname(mddev));
4878                                         err = -EBUSY;
4879                                         goto abort_unlock;
4880                                 }
4881                                 if (mddev->raid_disks) {
4882                                         printk(KERN_WARNING
4883                                                "md: array %s already initialised!\n",
4884                                                mdname(mddev));
4885                                         err = -EBUSY;
4886                                         goto abort_unlock;
4887                                 }
4888                                 err = set_array_info(mddev, &info);
4889                                 if (err) {
4890                                         printk(KERN_WARNING "md: couldn't set"
4891                                                " array info. %d\n", err);
4892                                         goto abort_unlock;
4893                                 }
4894                         }
4895                         goto done_unlock;
4896
4897                 default:;
4898         }
4899
4900         /*
4901          * Commands querying/configuring an existing array:
4902          */
4903         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4904          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4905         if ((!mddev->raid_disks && !mddev->external)
4906             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4907             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4908             && cmd != GET_BITMAP_FILE) {
4909                 err = -ENODEV;
4910                 goto abort_unlock;
4911         }
4912
4913         /*
4914          * Commands even a read-only array can execute:
4915          */
4916         switch (cmd)
4917         {
4918                 case GET_ARRAY_INFO:
4919                         err = get_array_info(mddev, argp);
4920                         goto done_unlock;
4921
4922                 case GET_BITMAP_FILE:
4923                         err = get_bitmap_file(mddev, argp);
4924                         goto done_unlock;
4925
4926                 case GET_DISK_INFO:
4927                         err = get_disk_info(mddev, argp);
4928                         goto done_unlock;
4929
4930                 case RESTART_ARRAY_RW:
4931                         err = restart_array(mddev);
4932                         goto done_unlock;
4933
4934                 case STOP_ARRAY:
4935                         err = do_md_stop(mddev, 0, 1);
4936                         goto done_unlock;
4937
4938                 case STOP_ARRAY_RO:
4939                         err = do_md_stop(mddev, 1, 1);
4940                         goto done_unlock;
4941
4942         }
4943
4944         /*
4945          * The remaining ioctls are changing the state of the
4946          * superblock, so we do not allow them on read-only arrays.
4947          * However non-MD ioctls (e.g. get-size) will still come through
4948          * here and hit the 'default' below, so only disallow
4949          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4950          */
4951         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
4952                 if (mddev->ro == 2) {
4953                         mddev->ro = 0;
4954                         sysfs_notify_dirent(mddev->sysfs_state);
4955                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4956                         md_wakeup_thread(mddev->thread);
4957                 } else {
4958                         err = -EROFS;
4959                         goto abort_unlock;
4960                 }
4961         }
4962
4963         switch (cmd)
4964         {
4965                 case ADD_NEW_DISK:
4966                 {
4967                         mdu_disk_info_t info;
4968                         if (copy_from_user(&info, argp, sizeof(info)))
4969                                 err = -EFAULT;
4970                         else
4971                                 err = add_new_disk(mddev, &info);
4972                         goto done_unlock;
4973                 }
4974
4975                 case HOT_REMOVE_DISK:
4976                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4977                         goto done_unlock;
4978
4979                 case HOT_ADD_DISK:
4980                         err = hot_add_disk(mddev, new_decode_dev(arg));
4981                         goto done_unlock;
4982
4983                 case SET_DISK_FAULTY:
4984                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4985                         goto done_unlock;
4986
4987                 case RUN_ARRAY:
4988                         err = do_md_run(mddev);
4989                         goto done_unlock;
4990
4991                 case SET_BITMAP_FILE:
4992                         err = set_bitmap_file(mddev, (int)arg);
4993                         goto done_unlock;
4994
4995                 default:
4996                         err = -EINVAL;
4997                         goto abort_unlock;
4998         }
4999
5000 done_unlock:
5001 abort_unlock:
5002         mddev_unlock(mddev);
5003
5004         return err;
5005 done:
5006         if (err)
5007                 MD_BUG();
5008 abort:
5009         return err;
5010 }
5011
5012 static int md_open(struct inode *inode, struct file *file)
5013 {
5014         /*
5015          * Succeed if we can lock the mddev, which confirms that
5016          * it isn't being stopped right now.
5017          */
5018         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5019         int err;
5020
5021         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5022                 goto out;
5023
5024         err = 0;
5025         mddev_get(mddev);
5026         atomic_inc(&mddev->openers);
5027         mddev_unlock(mddev);
5028
5029         check_disk_change(inode->i_bdev);
5030  out:
5031         return err;
5032 }
5033
5034 static int md_release(struct inode *inode, struct file * file)
5035 {
5036         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
5037
5038         BUG_ON(!mddev);
5039         atomic_dec(&mddev->openers);
5040         mddev_put(mddev);
5041
5042         return 0;
5043 }
5044
5045 static int md_media_changed(struct gendisk *disk)
5046 {
5047         mddev_t *mddev = disk->private_data;
5048
5049         return mddev->changed;
5050 }
5051
5052 static int md_revalidate(struct gendisk *disk)
5053 {
5054         mddev_t *mddev = disk->private_data;
5055
5056         mddev->changed = 0;
5057         return 0;
5058 }
5059 static struct block_device_operations md_fops =
5060 {
5061         .owner          = THIS_MODULE,
5062         .open           = md_open,
5063         .release        = md_release,
5064         .ioctl          = md_ioctl,
5065         .getgeo         = md_getgeo,
5066         .media_changed  = md_media_changed,
5067         .revalidate_disk= md_revalidate,
5068 };
5069
5070 static int md_thread(void * arg)
5071 {
5072         mdk_thread_t *thread = arg;
5073
5074         /*
5075          * md_thread is a 'system-thread', it's priority should be very
5076          * high. We avoid resource deadlocks individually in each
5077          * raid personality. (RAID5 does preallocation) We also use RR and
5078          * the very same RT priority as kswapd, thus we will never get
5079          * into a priority inversion deadlock.
5080          *
5081          * we definitely have to have equal or higher priority than
5082          * bdflush, otherwise bdflush will deadlock if there are too
5083          * many dirty RAID5 blocks.
5084          */
5085
5086         allow_signal(SIGKILL);
5087         while (!kthread_should_stop()) {
5088
5089                 /* We need to wait INTERRUPTIBLE so that
5090                  * we don't add to the load-average.
5091                  * That means we need to be sure no signals are
5092                  * pending
5093                  */
5094                 if (signal_pending(current))
5095                         flush_signals(current);
5096
5097                 wait_event_interruptible_timeout
5098                         (thread->wqueue,
5099                          test_bit(THREAD_WAKEUP, &thread->flags)
5100                          || kthread_should_stop(),
5101                          thread->timeout);
5102
5103                 clear_bit(THREAD_WAKEUP, &thread->flags);
5104
5105                 thread->run(thread->mddev);
5106         }
5107
5108         return 0;
5109 }
5110
5111 void md_wakeup_thread(mdk_thread_t *thread)
5112 {
5113         if (thread) {
5114                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5115                 set_bit(THREAD_WAKEUP, &thread->flags);
5116                 wake_up(&thread->wqueue);
5117         }
5118 }
5119
5120 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5121                                  const char *name)
5122 {
5123         mdk_thread_t *thread;
5124
5125         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5126         if (!thread)
5127                 return NULL;
5128
5129         init_waitqueue_head(&thread->wqueue);
5130
5131         thread->run = run;
5132         thread->mddev = mddev;
5133         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5134         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5135         if (IS_ERR(thread->tsk)) {
5136                 kfree(thread);
5137                 return NULL;
5138         }
5139         return thread;
5140 }
5141
5142 void md_unregister_thread(mdk_thread_t *thread)
5143 {
5144         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5145
5146         kthread_stop(thread->tsk);
5147         kfree(thread);
5148 }
5149
5150 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5151 {
5152         if (!mddev) {
5153                 MD_BUG();
5154                 return;
5155         }
5156
5157         if (!rdev || test_bit(Faulty, &rdev->flags))
5158                 return;
5159
5160         if (mddev->external)
5161                 set_bit(Blocked, &rdev->flags);
5162 /*
5163         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5164                 mdname(mddev),
5165                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5166                 __builtin_return_address(0),__builtin_return_address(1),
5167                 __builtin_return_address(2),__builtin_return_address(3));
5168 */
5169         if (!mddev->pers)
5170                 return;
5171         if (!mddev->pers->error_handler)
5172                 return;
5173         mddev->pers->error_handler(mddev,rdev);
5174         if (mddev->degraded)
5175                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5176         set_bit(StateChanged, &rdev->flags);
5177         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5178         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5179         md_wakeup_thread(mddev->thread);
5180         md_new_event_inintr(mddev);
5181 }
5182
5183 /* seq_file implementation /proc/mdstat */
5184
5185 static void status_unused(struct seq_file *seq)
5186 {
5187         int i = 0;
5188         mdk_rdev_t *rdev;
5189         struct list_head *tmp;
5190
5191         seq_printf(seq, "unused devices: ");
5192
5193         rdev_for_each_list(rdev, tmp, pending_raid_disks) {
5194                 char b[BDEVNAME_SIZE];
5195                 i++;
5196                 seq_printf(seq, "%s ",
5197                               bdevname(rdev->bdev,b));
5198         }
5199         if (!i)
5200                 seq_printf(seq, "<none>");
5201
5202         seq_printf(seq, "\n");
5203 }
5204
5205
5206 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5207 {
5208         sector_t max_blocks, resync, res;
5209         unsigned long dt, db, rt;
5210         int scale;
5211         unsigned int per_milli;
5212
5213         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
5214
5215         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5216                 max_blocks = mddev->resync_max_sectors >> 1;
5217         else
5218                 max_blocks = mddev->size;
5219
5220         /*
5221          * Should not happen.
5222          */
5223         if (!max_blocks) {
5224                 MD_BUG();
5225                 return;
5226         }
5227         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5228          * in a sector_t, and (max_blocks>>scale) will fit in a
5229          * u32, as those are the requirements for sector_div.
5230          * Thus 'scale' must be at least 10
5231          */
5232         scale = 10;
5233         if (sizeof(sector_t) > sizeof(unsigned long)) {
5234                 while ( max_blocks/2 > (1ULL<<(scale+32)))
5235                         scale++;
5236         }
5237         res = (resync>>scale)*1000;
5238         sector_div(res, (u32)((max_blocks>>scale)+1));
5239
5240         per_milli = res;
5241         {
5242                 int i, x = per_milli/50, y = 20-x;
5243                 seq_printf(seq, "[");
5244                 for (i = 0; i < x; i++)
5245                         seq_printf(seq, "=");
5246                 seq_printf(seq, ">");
5247                 for (i = 0; i < y; i++)
5248                         seq_printf(seq, ".");
5249                 seq_printf(seq, "] ");
5250         }
5251         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5252                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5253                     "reshape" :
5254                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5255                      "check" :
5256                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5257                       "resync" : "recovery"))),
5258                    per_milli/10, per_milli % 10,
5259                    (unsigned long long) resync,
5260                    (unsigned long long) max_blocks);
5261
5262         /*
5263          * We do not want to overflow, so the order of operands and
5264          * the * 100 / 100 trick are important. We do a +1 to be
5265          * safe against division by zero. We only estimate anyway.
5266          *
5267          * dt: time from mark until now
5268          * db: blocks written from mark until now
5269          * rt: remaining time
5270          */
5271         dt = ((jiffies - mddev->resync_mark) / HZ);
5272         if (!dt) dt++;
5273         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5274                 - mddev->resync_mark_cnt;
5275         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
5276
5277         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
5278
5279         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5280 }
5281
5282 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5283 {
5284         struct list_head *tmp;
5285         loff_t l = *pos;
5286         mddev_t *mddev;
5287
5288         if (l >= 0x10000)
5289                 return NULL;
5290         if (!l--)
5291                 /* header */
5292                 return (void*)1;
5293
5294         spin_lock(&all_mddevs_lock);
5295         list_for_each(tmp,&all_mddevs)
5296                 if (!l--) {
5297                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5298                         mddev_get(mddev);
5299                         spin_unlock(&all_mddevs_lock);
5300                         return mddev;
5301                 }
5302         spin_unlock(&all_mddevs_lock);
5303         if (!l--)
5304                 return (void*)2;/* tail */
5305         return NULL;
5306 }
5307
5308 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5309 {
5310         struct list_head *tmp;
5311         mddev_t *next_mddev, *mddev = v;
5312         
5313         ++*pos;
5314         if (v == (void*)2)
5315                 return NULL;
5316
5317         spin_lock(&all_mddevs_lock);
5318         if (v == (void*)1)
5319                 tmp = all_mddevs.next;
5320         else
5321                 tmp = mddev->all_mddevs.next;
5322         if (tmp != &all_mddevs)
5323                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5324         else {
5325                 next_mddev = (void*)2;
5326                 *pos = 0x10000;
5327         }               
5328         spin_unlock(&all_mddevs_lock);
5329
5330         if (v != (void*)1)
5331                 mddev_put(mddev);
5332         return next_mddev;
5333
5334 }
5335
5336 static void md_seq_stop(struct seq_file *seq, void *v)
5337 {
5338         mddev_t *mddev = v;
5339
5340         if (mddev && v != (void*)1 && v != (void*)2)
5341                 mddev_put(mddev);
5342 }
5343
5344 struct mdstat_info {
5345         int event;
5346 };
5347
5348 static int md_seq_show(struct seq_file *seq, void *v)
5349 {
5350         mddev_t *mddev = v;
5351         sector_t size;
5352         struct list_head *tmp2;
5353         mdk_rdev_t *rdev;
5354         struct mdstat_info *mi = seq->private;
5355         struct bitmap *bitmap;
5356
5357         if (v == (void*)1) {
5358                 struct mdk_personality *pers;
5359                 seq_printf(seq, "Personalities : ");
5360                 spin_lock(&pers_lock);
5361                 list_for_each_entry(pers, &pers_list, list)
5362                         seq_printf(seq, "[%s] ", pers->name);
5363
5364                 spin_unlock(&pers_lock);
5365                 seq_printf(seq, "\n");
5366                 mi->event = atomic_read(&md_event_count);
5367                 return 0;
5368         }
5369         if (v == (void*)2) {
5370                 status_unused(seq);
5371                 return 0;
5372         }
5373
5374         if (mddev_lock(mddev) < 0)
5375                 return -EINTR;
5376
5377         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5378                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5379                                                 mddev->pers ? "" : "in");
5380                 if (mddev->pers) {
5381                         if (mddev->ro==1)
5382                                 seq_printf(seq, " (read-only)");
5383                         if (mddev->ro==2)
5384                                 seq_printf(seq, " (auto-read-only)");
5385                         seq_printf(seq, " %s", mddev->pers->name);
5386                 }
5387
5388                 size = 0;
5389                 rdev_for_each(rdev, tmp2, mddev) {
5390                         char b[BDEVNAME_SIZE];
5391                         seq_printf(seq, " %s[%d]",
5392                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5393                         if (test_bit(WriteMostly, &rdev->flags))
5394                                 seq_printf(seq, "(W)");
5395                         if (test_bit(Faulty, &rdev->flags)) {
5396                                 seq_printf(seq, "(F)");
5397                                 continue;
5398                         } else if (rdev->raid_disk < 0)
5399                                 seq_printf(seq, "(S)"); /* spare */
5400                         size += rdev->size;
5401                 }
5402
5403                 if (!list_empty(&mddev->disks)) {
5404                         if (mddev->pers)
5405                                 seq_printf(seq, "\n      %llu blocks",
5406                                            (unsigned long long)
5407                                            mddev->array_sectors / 2);
5408                         else
5409                                 seq_printf(seq, "\n      %llu blocks",
5410                                            (unsigned long long)size);
5411                 }
5412                 if (mddev->persistent) {
5413                         if (mddev->major_version != 0 ||
5414                             mddev->minor_version != 90) {
5415                                 seq_printf(seq," super %d.%d",
5416                                            mddev->major_version,
5417                                            mddev->minor_version);
5418                         }
5419                 } else if (mddev->external)
5420                         seq_printf(seq, " super external:%s",
5421                                    mddev->metadata_type);
5422                 else
5423                         seq_printf(seq, " super non-persistent");
5424
5425                 if (mddev->pers) {
5426                         mddev->pers->status(seq, mddev);
5427                         seq_printf(seq, "\n      ");
5428                         if (mddev->pers->sync_request) {
5429                                 if (mddev->curr_resync > 2) {
5430                                         status_resync(seq, mddev);
5431                                         seq_printf(seq, "\n      ");
5432                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5433                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5434                                 else if (mddev->recovery_cp < MaxSector)
5435                                         seq_printf(seq, "\tresync=PENDING\n      ");
5436                         }
5437                 } else
5438                         seq_printf(seq, "\n       ");
5439
5440                 if ((bitmap = mddev->bitmap)) {
5441                         unsigned long chunk_kb;
5442                         unsigned long flags;
5443                         spin_lock_irqsave(&bitmap->lock, flags);
5444                         chunk_kb = bitmap->chunksize >> 10;
5445                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5446                                 "%lu%s chunk",
5447                                 bitmap->pages - bitmap->missing_pages,
5448                                 bitmap->pages,
5449                                 (bitmap->pages - bitmap->missing_pages)
5450                                         << (PAGE_SHIFT - 10),
5451                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5452                                 chunk_kb ? "KB" : "B");
5453                         if (bitmap->file) {
5454                                 seq_printf(seq, ", file: ");
5455                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5456                         }
5457
5458                         seq_printf(seq, "\n");
5459                         spin_unlock_irqrestore(&bitmap->lock, flags);
5460                 }
5461
5462                 seq_printf(seq, "\n");
5463         }
5464         mddev_unlock(mddev);
5465         
5466         return 0;
5467 }
5468
5469 static struct seq_operations md_seq_ops = {
5470         .start  = md_seq_start,
5471         .next   = md_seq_next,
5472         .stop   = md_seq_stop,
5473         .show   = md_seq_show,
5474 };
5475
5476 static int md_seq_open(struct inode *inode, struct file *file)
5477 {
5478         int error;
5479         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5480         if (mi == NULL)
5481                 return -ENOMEM;
5482
5483         error = seq_open(file, &md_seq_ops);
5484         if (error)
5485                 kfree(mi);
5486         else {
5487                 struct seq_file *p = file->private_data;
5488                 p->private = mi;
5489                 mi->event = atomic_read(&md_event_count);
5490         }
5491         return error;
5492 }
5493
5494 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5495 {
5496         struct seq_file *m = filp->private_data;
5497         struct mdstat_info *mi = m->private;
5498         int mask;
5499
5500         poll_wait(filp, &md_event_waiters, wait);
5501
5502         /* always allow read */
5503         mask = POLLIN | POLLRDNORM;
5504
5505         if (mi->event != atomic_read(&md_event_count))
5506                 mask |= POLLERR | POLLPRI;
5507         return mask;
5508 }
5509
5510 static const struct file_operations md_seq_fops = {
5511         .owner          = THIS_MODULE,
5512         .open           = md_seq_open,
5513         .read           = seq_read,
5514         .llseek         = seq_lseek,
5515         .release        = seq_release_private,
5516         .poll           = mdstat_poll,
5517 };
5518
5519 int register_md_personality(struct mdk_personality *p)
5520 {
5521         spin_lock(&pers_lock);
5522         list_add_tail(&p->list, &pers_list);
5523         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5524         spin_unlock(&pers_lock);
5525         return 0;
5526 }
5527
5528 int unregister_md_personality(struct mdk_personality *p)
5529 {
5530         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5531         spin_lock(&pers_lock);
5532         list_del_init(&p->list);
5533         spin_unlock(&pers_lock);
5534         return 0;
5535 }
5536
5537 static int is_mddev_idle(mddev_t *mddev)
5538 {
5539         mdk_rdev_t * rdev;
5540         int idle;
5541         long curr_events;
5542
5543         idle = 1;
5544         rcu_read_lock();
5545         rdev_for_each_rcu(rdev, mddev) {
5546                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5547                 curr_events = part_stat_read(&disk->part0, sectors[0]) +
5548                                 part_stat_read(&disk->part0, sectors[1]) -
5549                                 atomic_read(&disk->sync_io);
5550                 /* sync IO will cause sync_io to increase before the disk_stats
5551                  * as sync_io is counted when a request starts, and
5552                  * disk_stats is counted when it completes.
5553                  * So resync activity will cause curr_events to be smaller than
5554                  * when there was no such activity.
5555                  * non-sync IO will cause disk_stat to increase without
5556                  * increasing sync_io so curr_events will (eventually)
5557                  * be larger than it was before.  Once it becomes
5558                  * substantially larger, the test below will cause
5559                  * the array to appear non-idle, and resync will slow
5560                  * down.
5561                  * If there is a lot of outstanding resync activity when
5562                  * we set last_event to curr_events, then all that activity
5563                  * completing might cause the array to appear non-idle
5564                  * and resync will be slowed down even though there might
5565                  * not have been non-resync activity.  This will only
5566                  * happen once though.  'last_events' will soon reflect
5567                  * the state where there is little or no outstanding
5568                  * resync requests, and further resync activity will
5569                  * always make curr_events less than last_events.
5570                  *
5571                  */
5572                 if (curr_events - rdev->last_events > 4096) {
5573                         rdev->last_events = curr_events;
5574                         idle = 0;
5575                 }
5576         }
5577         rcu_read_unlock();
5578         return idle;
5579 }
5580
5581 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5582 {
5583         /* another "blocks" (512byte) blocks have been synced */
5584         atomic_sub(blocks, &mddev->recovery_active);
5585         wake_up(&mddev->recovery_wait);
5586         if (!ok) {
5587                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5588                 md_wakeup_thread(mddev->thread);
5589                 // stop recovery, signal do_sync ....
5590         }
5591 }
5592
5593
5594 /* md_write_start(mddev, bi)
5595  * If we need to update some array metadata (e.g. 'active' flag
5596  * in superblock) before writing, schedule a superblock update
5597  * and wait for it to complete.
5598  */
5599 void md_write_start(mddev_t *mddev, struct bio *bi)
5600 {
5601         int did_change = 0;
5602         if (bio_data_dir(bi) != WRITE)
5603                 return;
5604
5605         BUG_ON(mddev->ro == 1);
5606         if (mddev->ro == 2) {
5607                 /* need to switch to read/write */
5608                 mddev->ro = 0;
5609                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5610                 md_wakeup_thread(mddev->thread);
5611                 md_wakeup_thread(mddev->sync_thread);
5612                 did_change = 1;
5613         }
5614         atomic_inc(&mddev->writes_pending);
5615         if (mddev->safemode == 1)
5616                 mddev->safemode = 0;
5617         if (mddev->in_sync) {
5618                 spin_lock_irq(&mddev->write_lock);
5619                 if (mddev->in_sync) {
5620                         mddev->in_sync = 0;
5621                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5622                         md_wakeup_thread(mddev->thread);
5623                         did_change = 1;
5624                 }
5625                 spin_unlock_irq(&mddev->write_lock);
5626         }
5627         if (did_change)
5628                 sysfs_notify_dirent(mddev->sysfs_state);
5629         wait_event(mddev->sb_wait,
5630                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
5631                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
5632 }
5633
5634 void md_write_end(mddev_t *mddev)
5635 {
5636         if (atomic_dec_and_test(&mddev->writes_pending)) {
5637                 if (mddev->safemode == 2)
5638                         md_wakeup_thread(mddev->thread);
5639                 else if (mddev->safemode_delay)
5640                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5641         }
5642 }
5643
5644 /* md_allow_write(mddev)
5645  * Calling this ensures that the array is marked 'active' so that writes
5646  * may proceed without blocking.  It is important to call this before
5647  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5648  * Must be called with mddev_lock held.
5649  *
5650  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
5651  * is dropped, so return -EAGAIN after notifying userspace.
5652  */
5653 int md_allow_write(mddev_t *mddev)
5654 {
5655         if (!mddev->pers)
5656                 return 0;
5657         if (mddev->ro)
5658                 return 0;
5659         if (!mddev->pers->sync_request)
5660                 return 0;
5661
5662         spin_lock_irq(&mddev->write_lock);
5663         if (mddev->in_sync) {
5664                 mddev->in_sync = 0;
5665                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5666                 if (mddev->safemode_delay &&
5667                     mddev->safemode == 0)
5668                         mddev->safemode = 1;
5669                 spin_unlock_irq(&mddev->write_lock);
5670                 md_update_sb(mddev, 0);
5671                 sysfs_notify_dirent(mddev->sysfs_state);
5672         } else
5673                 spin_unlock_irq(&mddev->write_lock);
5674
5675         if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
5676                 return -EAGAIN;
5677         else
5678                 return 0;
5679 }
5680 EXPORT_SYMBOL_GPL(md_allow_write);
5681
5682 #define SYNC_MARKS      10
5683 #define SYNC_MARK_STEP  (3*HZ)
5684 void md_do_sync(mddev_t *mddev)
5685 {
5686         mddev_t *mddev2;
5687         unsigned int currspeed = 0,
5688                  window;
5689         sector_t max_sectors,j, io_sectors;
5690         unsigned long mark[SYNC_MARKS];
5691         sector_t mark_cnt[SYNC_MARKS];
5692         int last_mark,m;
5693         struct list_head *tmp;
5694         sector_t last_check;
5695         int skipped = 0;
5696         struct list_head *rtmp;
5697         mdk_rdev_t *rdev;
5698         char *desc;
5699
5700         /* just incase thread restarts... */
5701         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5702                 return;
5703         if (mddev->ro) /* never try to sync a read-only array */
5704                 return;
5705
5706         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5707                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5708                         desc = "data-check";
5709                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5710                         desc = "requested-resync";
5711                 else
5712                         desc = "resync";
5713         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5714                 desc = "reshape";
5715         else
5716                 desc = "recovery";
5717
5718         /* we overload curr_resync somewhat here.
5719          * 0 == not engaged in resync at all
5720          * 2 == checking that there is no conflict with another sync
5721          * 1 == like 2, but have yielded to allow conflicting resync to
5722          *              commense
5723          * other == active in resync - this many blocks
5724          *
5725          * Before starting a resync we must have set curr_resync to
5726          * 2, and then checked that every "conflicting" array has curr_resync
5727          * less than ours.  When we find one that is the same or higher
5728          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5729          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5730          * This will mean we have to start checking from the beginning again.
5731          *
5732          */
5733
5734         do {
5735                 mddev->curr_resync = 2;
5736
5737         try_again:
5738                 if (kthread_should_stop()) {
5739                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5740                         goto skip;
5741                 }
5742                 for_each_mddev(mddev2, tmp) {
5743                         if (mddev2 == mddev)
5744                                 continue;
5745                         if (!mddev->parallel_resync
5746                         &&  mddev2->curr_resync
5747                         &&  match_mddev_units(mddev, mddev2)) {
5748                                 DEFINE_WAIT(wq);
5749                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5750                                         /* arbitrarily yield */
5751                                         mddev->curr_resync = 1;
5752                                         wake_up(&resync_wait);
5753                                 }
5754                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5755                                         /* no need to wait here, we can wait the next
5756                                          * time 'round when curr_resync == 2
5757                                          */
5758                                         continue;
5759                                 /* We need to wait 'interruptible' so as not to
5760                                  * contribute to the load average, and not to
5761                                  * be caught by 'softlockup'
5762                                  */
5763                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
5764                                 if (!kthread_should_stop() &&
5765                                     mddev2->curr_resync >= mddev->curr_resync) {
5766                                         printk(KERN_INFO "md: delaying %s of %s"
5767                                                " until %s has finished (they"
5768                                                " share one or more physical units)\n",
5769                                                desc, mdname(mddev), mdname(mddev2));
5770                                         mddev_put(mddev2);
5771                                         if (signal_pending(current))
5772                                                 flush_signals(current);
5773                                         schedule();
5774                                         finish_wait(&resync_wait, &wq);
5775                                         goto try_again;
5776                                 }
5777                                 finish_wait(&resync_wait, &wq);
5778                         }
5779                 }
5780         } while (mddev->curr_resync < 2);
5781
5782         j = 0;
5783         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5784                 /* resync follows the size requested by the personality,
5785                  * which defaults to physical size, but can be virtual size
5786                  */
5787                 max_sectors = mddev->resync_max_sectors;
5788                 mddev->resync_mismatches = 0;
5789                 /* we don't use the checkpoint if there's a bitmap */
5790                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5791                         j = mddev->resync_min;
5792                 else if (!mddev->bitmap)
5793                         j = mddev->recovery_cp;
5794
5795         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5796                 max_sectors = mddev->size << 1;
5797         else {
5798                 /* recovery follows the physical size of devices */
5799                 max_sectors = mddev->size << 1;
5800                 j = MaxSector;
5801                 rdev_for_each(rdev, rtmp, mddev)
5802                         if (rdev->raid_disk >= 0 &&
5803                             !test_bit(Faulty, &rdev->flags) &&
5804                             !test_bit(In_sync, &rdev->flags) &&
5805                             rdev->recovery_offset < j)
5806                                 j = rdev->recovery_offset;
5807         }
5808
5809         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5810         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5811                 " %d KB/sec/disk.\n", speed_min(mddev));
5812         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5813                "(but not more than %d KB/sec) for %s.\n",
5814                speed_max(mddev), desc);
5815
5816         is_mddev_idle(mddev); /* this also initializes IO event counters */
5817
5818         io_sectors = 0;
5819         for (m = 0; m < SYNC_MARKS; m++) {
5820                 mark[m] = jiffies;
5821                 mark_cnt[m] = io_sectors;
5822         }
5823         last_mark = 0;
5824         mddev->resync_mark = mark[last_mark];
5825         mddev->resync_mark_cnt = mark_cnt[last_mark];
5826
5827         /*
5828          * Tune reconstruction:
5829          */
5830         window = 32*(PAGE_SIZE/512);
5831         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5832                 window/2,(unsigned long long) max_sectors/2);
5833
5834         atomic_set(&mddev->recovery_active, 0);
5835         last_check = 0;
5836
5837         if (j>2) {
5838                 printk(KERN_INFO 
5839                        "md: resuming %s of %s from checkpoint.\n",
5840                        desc, mdname(mddev));
5841                 mddev->curr_resync = j;
5842         }
5843
5844         while (j < max_sectors) {
5845                 sector_t sectors;
5846
5847                 skipped = 0;
5848                 if (j >= mddev->resync_max) {
5849                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
5850                         wait_event(mddev->recovery_wait,
5851                                    mddev->resync_max > j
5852                                    || kthread_should_stop());
5853                 }
5854                 if (kthread_should_stop())
5855                         goto interrupted;
5856                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5857                                                   currspeed < speed_min(mddev));
5858                 if (sectors == 0) {
5859                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5860                         goto out;
5861                 }
5862
5863                 if (!skipped) { /* actual IO requested */
5864                         io_sectors += sectors;
5865                         atomic_add(sectors, &mddev->recovery_active);
5866         &n