ec3d8e8a0bd33422545fd15a9d4df4ec12fdf258
[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/module.h>
36 #include <linux/kthread.h>
37 #include <linux/linkage.h>
38 #include <linux/raid/md.h>
39 #include <linux/raid/bitmap.h>
40 #include <linux/sysctl.h>
41 #include <linux/buffer_head.h> /* for invalidate_bdev */
42 #include <linux/poll.h>
43 #include <linux/mutex.h>
44 #include <linux/ctype.h>
45 #include <linux/freezer.h>
46
47 #include <linux/init.h>
48
49 #include <linux/file.h>
50
51 #ifdef CONFIG_KMOD
52 #include <linux/kmod.h>
53 #endif
54
55 #include <asm/unaligned.h>
56
57 #define MAJOR_NR MD_MAJOR
58 #define MD_DRIVER
59
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
62
63 #define DEBUG 0
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
65
66
67 #ifndef MODULE
68 static void autostart_arrays (int part);
69 #endif
70
71 static LIST_HEAD(pers_list);
72 static DEFINE_SPINLOCK(pers_lock);
73
74 static void md_print_devices(void);
75
76 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
77
78 /*
79  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
80  * is 1000 KB/sec, so the extra system load does not show up that much.
81  * Increase it if you want to have more _guaranteed_ speed. Note that
82  * the RAID driver will use the maximum available bandwidth if the IO
83  * subsystem is idle. There is also an 'absolute maximum' reconstruction
84  * speed limit - in case reconstruction slows down your system despite
85  * idle IO detection.
86  *
87  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88  * or /sys/block/mdX/md/sync_speed_{min,max}
89  */
90
91 static int sysctl_speed_limit_min = 1000;
92 static int sysctl_speed_limit_max = 200000;
93 static inline int speed_min(mddev_t *mddev)
94 {
95         return mddev->sync_speed_min ?
96                 mddev->sync_speed_min : sysctl_speed_limit_min;
97 }
98
99 static inline int speed_max(mddev_t *mddev)
100 {
101         return mddev->sync_speed_max ?
102                 mddev->sync_speed_max : sysctl_speed_limit_max;
103 }
104
105 static struct ctl_table_header *raid_table_header;
106
107 static ctl_table raid_table[] = {
108         {
109                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
110                 .procname       = "speed_limit_min",
111                 .data           = &sysctl_speed_limit_min,
112                 .maxlen         = sizeof(int),
113                 .mode           = S_IRUGO|S_IWUSR,
114                 .proc_handler   = &proc_dointvec,
115         },
116         {
117                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
118                 .procname       = "speed_limit_max",
119                 .data           = &sysctl_speed_limit_max,
120                 .maxlen         = sizeof(int),
121                 .mode           = S_IRUGO|S_IWUSR,
122                 .proc_handler   = &proc_dointvec,
123         },
124         { .ctl_name = 0 }
125 };
126
127 static ctl_table raid_dir_table[] = {
128         {
129                 .ctl_name       = DEV_RAID,
130                 .procname       = "raid",
131                 .maxlen         = 0,
132                 .mode           = S_IRUGO|S_IXUGO,
133                 .child          = raid_table,
134         },
135         { .ctl_name = 0 }
136 };
137
138 static ctl_table raid_root_table[] = {
139         {
140                 .ctl_name       = CTL_DEV,
141                 .procname       = "dev",
142                 .maxlen         = 0,
143                 .mode           = 0555,
144                 .child          = raid_dir_table,
145         },
146         { .ctl_name = 0 }
147 };
148
149 static struct block_device_operations md_fops;
150
151 static int start_readonly;
152
153 /*
154  * We have a system wide 'event count' that is incremented
155  * on any 'interesting' event, and readers of /proc/mdstat
156  * can use 'poll' or 'select' to find out when the event
157  * count increases.
158  *
159  * Events are:
160  *  start array, stop array, error, add device, remove device,
161  *  start build, activate spare
162  */
163 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
164 static atomic_t md_event_count;
165 void md_new_event(mddev_t *mddev)
166 {
167         atomic_inc(&md_event_count);
168         wake_up(&md_event_waiters);
169         sysfs_notify(&mddev->kobj, NULL, "sync_action");
170 }
171 EXPORT_SYMBOL_GPL(md_new_event);
172
173 /* Alternate version that can be called from interrupts
174  * when calling sysfs_notify isn't needed.
175  */
176 static void md_new_event_inintr(mddev_t *mddev)
177 {
178         atomic_inc(&md_event_count);
179         wake_up(&md_event_waiters);
180 }
181
182 /*
183  * Enables to iterate over all existing md arrays
184  * all_mddevs_lock protects this list.
185  */
186 static LIST_HEAD(all_mddevs);
187 static DEFINE_SPINLOCK(all_mddevs_lock);
188
189
190 /*
191  * iterates through all used mddevs in the system.
192  * We take care to grab the all_mddevs_lock whenever navigating
193  * the list, and to always hold a refcount when unlocked.
194  * Any code which breaks out of this loop while own
195  * a reference to the current mddev and must mddev_put it.
196  */
197 #define ITERATE_MDDEV(mddev,tmp)                                        \
198                                                                         \
199         for (({ spin_lock(&all_mddevs_lock);                            \
200                 tmp = all_mddevs.next;                                  \
201                 mddev = NULL;});                                        \
202              ({ if (tmp != &all_mddevs)                                 \
203                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
204                 spin_unlock(&all_mddevs_lock);                          \
205                 if (mddev) mddev_put(mddev);                            \
206                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
207                 tmp != &all_mddevs;});                                  \
208              ({ spin_lock(&all_mddevs_lock);                            \
209                 tmp = tmp->next;})                                      \
210                 )
211
212
213 static int md_fail_request (request_queue_t *q, struct bio *bio)
214 {
215         bio_io_error(bio, bio->bi_size);
216         return 0;
217 }
218
219 static inline mddev_t *mddev_get(mddev_t *mddev)
220 {
221         atomic_inc(&mddev->active);
222         return mddev;
223 }
224
225 static void mddev_put(mddev_t *mddev)
226 {
227         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
228                 return;
229         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
230                 list_del(&mddev->all_mddevs);
231                 spin_unlock(&all_mddevs_lock);
232                 blk_cleanup_queue(mddev->queue);
233                 kobject_unregister(&mddev->kobj);
234         } else
235                 spin_unlock(&all_mddevs_lock);
236 }
237
238 static mddev_t * mddev_find(dev_t unit)
239 {
240         mddev_t *mddev, *new = NULL;
241
242  retry:
243         spin_lock(&all_mddevs_lock);
244         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
245                 if (mddev->unit == unit) {
246                         mddev_get(mddev);
247                         spin_unlock(&all_mddevs_lock);
248                         kfree(new);
249                         return mddev;
250                 }
251
252         if (new) {
253                 list_add(&new->all_mddevs, &all_mddevs);
254                 spin_unlock(&all_mddevs_lock);
255                 return new;
256         }
257         spin_unlock(&all_mddevs_lock);
258
259         new = kzalloc(sizeof(*new), GFP_KERNEL);
260         if (!new)
261                 return NULL;
262
263         new->unit = unit;
264         if (MAJOR(unit) == MD_MAJOR)
265                 new->md_minor = MINOR(unit);
266         else
267                 new->md_minor = MINOR(unit) >> MdpMinorShift;
268
269         mutex_init(&new->reconfig_mutex);
270         INIT_LIST_HEAD(&new->disks);
271         INIT_LIST_HEAD(&new->all_mddevs);
272         init_timer(&new->safemode_timer);
273         atomic_set(&new->active, 1);
274         spin_lock_init(&new->write_lock);
275         init_waitqueue_head(&new->sb_wait);
276
277         new->queue = blk_alloc_queue(GFP_KERNEL);
278         if (!new->queue) {
279                 kfree(new);
280                 return NULL;
281         }
282         set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
283
284         blk_queue_make_request(new->queue, md_fail_request);
285
286         goto retry;
287 }
288
289 static inline int mddev_lock(mddev_t * mddev)
290 {
291         return mutex_lock_interruptible(&mddev->reconfig_mutex);
292 }
293
294 static inline int mddev_trylock(mddev_t * mddev)
295 {
296         return mutex_trylock(&mddev->reconfig_mutex);
297 }
298
299 static inline void mddev_unlock(mddev_t * mddev)
300 {
301         mutex_unlock(&mddev->reconfig_mutex);
302
303         md_wakeup_thread(mddev->thread);
304 }
305
306 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
307 {
308         mdk_rdev_t * rdev;
309         struct list_head *tmp;
310
311         ITERATE_RDEV(mddev,rdev,tmp) {
312                 if (rdev->desc_nr == nr)
313                         return rdev;
314         }
315         return NULL;
316 }
317
318 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
319 {
320         struct list_head *tmp;
321         mdk_rdev_t *rdev;
322
323         ITERATE_RDEV(mddev,rdev,tmp) {
324                 if (rdev->bdev->bd_dev == dev)
325                         return rdev;
326         }
327         return NULL;
328 }
329
330 static struct mdk_personality *find_pers(int level, char *clevel)
331 {
332         struct mdk_personality *pers;
333         list_for_each_entry(pers, &pers_list, list) {
334                 if (level != LEVEL_NONE && pers->level == level)
335                         return pers;
336                 if (strcmp(pers->name, clevel)==0)
337                         return pers;
338         }
339         return NULL;
340 }
341
342 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
343 {
344         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
345         return MD_NEW_SIZE_BLOCKS(size);
346 }
347
348 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
349 {
350         sector_t size;
351
352         size = rdev->sb_offset;
353
354         if (chunk_size)
355                 size &= ~((sector_t)chunk_size/1024 - 1);
356         return size;
357 }
358
359 static int alloc_disk_sb(mdk_rdev_t * rdev)
360 {
361         if (rdev->sb_page)
362                 MD_BUG();
363
364         rdev->sb_page = alloc_page(GFP_KERNEL);
365         if (!rdev->sb_page) {
366                 printk(KERN_ALERT "md: out of memory.\n");
367                 return -EINVAL;
368         }
369
370         return 0;
371 }
372
373 static void free_disk_sb(mdk_rdev_t * rdev)
374 {
375         if (rdev->sb_page) {
376                 put_page(rdev->sb_page);
377                 rdev->sb_loaded = 0;
378                 rdev->sb_page = NULL;
379                 rdev->sb_offset = 0;
380                 rdev->size = 0;
381         }
382 }
383
384
385 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
386 {
387         mdk_rdev_t *rdev = bio->bi_private;
388         mddev_t *mddev = rdev->mddev;
389         if (bio->bi_size)
390                 return 1;
391
392         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
393                 printk("md: super_written gets error=%d, uptodate=%d\n",
394                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
395                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
396                 md_error(mddev, rdev);
397         }
398
399         if (atomic_dec_and_test(&mddev->pending_writes))
400                 wake_up(&mddev->sb_wait);
401         bio_put(bio);
402         return 0;
403 }
404
405 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
406 {
407         struct bio *bio2 = bio->bi_private;
408         mdk_rdev_t *rdev = bio2->bi_private;
409         mddev_t *mddev = rdev->mddev;
410         if (bio->bi_size)
411                 return 1;
412
413         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
414             error == -EOPNOTSUPP) {
415                 unsigned long flags;
416                 /* barriers don't appear to be supported :-( */
417                 set_bit(BarriersNotsupp, &rdev->flags);
418                 mddev->barriers_work = 0;
419                 spin_lock_irqsave(&mddev->write_lock, flags);
420                 bio2->bi_next = mddev->biolist;
421                 mddev->biolist = bio2;
422                 spin_unlock_irqrestore(&mddev->write_lock, flags);
423                 wake_up(&mddev->sb_wait);
424                 bio_put(bio);
425                 return 0;
426         }
427         bio_put(bio2);
428         bio->bi_private = rdev;
429         return super_written(bio, bytes_done, error);
430 }
431
432 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
433                    sector_t sector, int size, struct page *page)
434 {
435         /* write first size bytes of page to sector of rdev
436          * Increment mddev->pending_writes before returning
437          * and decrement it on completion, waking up sb_wait
438          * if zero is reached.
439          * If an error occurred, call md_error
440          *
441          * As we might need to resubmit the request if BIO_RW_BARRIER
442          * causes ENOTSUPP, we allocate a spare bio...
443          */
444         struct bio *bio = bio_alloc(GFP_NOIO, 1);
445         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
446
447         bio->bi_bdev = rdev->bdev;
448         bio->bi_sector = sector;
449         bio_add_page(bio, page, size, 0);
450         bio->bi_private = rdev;
451         bio->bi_end_io = super_written;
452         bio->bi_rw = rw;
453
454         atomic_inc(&mddev->pending_writes);
455         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
456                 struct bio *rbio;
457                 rw |= (1<<BIO_RW_BARRIER);
458                 rbio = bio_clone(bio, GFP_NOIO);
459                 rbio->bi_private = bio;
460                 rbio->bi_end_io = super_written_barrier;
461                 submit_bio(rw, rbio);
462         } else
463                 submit_bio(rw, bio);
464 }
465
466 void md_super_wait(mddev_t *mddev)
467 {
468         /* wait for all superblock writes that were scheduled to complete.
469          * if any had to be retried (due to BARRIER problems), retry them
470          */
471         DEFINE_WAIT(wq);
472         for(;;) {
473                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
474                 if (atomic_read(&mddev->pending_writes)==0)
475                         break;
476                 while (mddev->biolist) {
477                         struct bio *bio;
478                         spin_lock_irq(&mddev->write_lock);
479                         bio = mddev->biolist;
480                         mddev->biolist = bio->bi_next ;
481                         bio->bi_next = NULL;
482                         spin_unlock_irq(&mddev->write_lock);
483                         submit_bio(bio->bi_rw, bio);
484                 }
485                 schedule();
486         }
487         finish_wait(&mddev->sb_wait, &wq);
488 }
489
490 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
491 {
492         if (bio->bi_size)
493                 return 1;
494
495         complete((struct completion*)bio->bi_private);
496         return 0;
497 }
498
499 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
500                    struct page *page, int rw)
501 {
502         struct bio *bio = bio_alloc(GFP_NOIO, 1);
503         struct completion event;
504         int ret;
505
506         rw |= (1 << BIO_RW_SYNC);
507
508         bio->bi_bdev = bdev;
509         bio->bi_sector = sector;
510         bio_add_page(bio, page, size, 0);
511         init_completion(&event);
512         bio->bi_private = &event;
513         bio->bi_end_io = bi_complete;
514         submit_bio(rw, bio);
515         wait_for_completion(&event);
516
517         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
518         bio_put(bio);
519         return ret;
520 }
521 EXPORT_SYMBOL_GPL(sync_page_io);
522
523 static int read_disk_sb(mdk_rdev_t * rdev, int size)
524 {
525         char b[BDEVNAME_SIZE];
526         if (!rdev->sb_page) {
527                 MD_BUG();
528                 return -EINVAL;
529         }
530         if (rdev->sb_loaded)
531                 return 0;
532
533
534         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
535                 goto fail;
536         rdev->sb_loaded = 1;
537         return 0;
538
539 fail:
540         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
541                 bdevname(rdev->bdev,b));
542         return -EINVAL;
543 }
544
545 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
546 {
547         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
548                 (sb1->set_uuid1 == sb2->set_uuid1) &&
549                 (sb1->set_uuid2 == sb2->set_uuid2) &&
550                 (sb1->set_uuid3 == sb2->set_uuid3))
551
552                 return 1;
553
554         return 0;
555 }
556
557
558 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
559 {
560         int ret;
561         mdp_super_t *tmp1, *tmp2;
562
563         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
564         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
565
566         if (!tmp1 || !tmp2) {
567                 ret = 0;
568                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
569                 goto abort;
570         }
571
572         *tmp1 = *sb1;
573         *tmp2 = *sb2;
574
575         /*
576          * nr_disks is not constant
577          */
578         tmp1->nr_disks = 0;
579         tmp2->nr_disks = 0;
580
581         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
582                 ret = 0;
583         else
584                 ret = 1;
585
586 abort:
587         kfree(tmp1);
588         kfree(tmp2);
589         return ret;
590 }
591
592 static unsigned int calc_sb_csum(mdp_super_t * sb)
593 {
594         unsigned int disk_csum, csum;
595
596         disk_csum = sb->sb_csum;
597         sb->sb_csum = 0;
598         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
599         sb->sb_csum = disk_csum;
600         return csum;
601 }
602
603
604 /*
605  * Handle superblock details.
606  * We want to be able to handle multiple superblock formats
607  * so we have a common interface to them all, and an array of
608  * different handlers.
609  * We rely on user-space to write the initial superblock, and support
610  * reading and updating of superblocks.
611  * Interface methods are:
612  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
613  *      loads and validates a superblock on dev.
614  *      if refdev != NULL, compare superblocks on both devices
615  *    Return:
616  *      0 - dev has a superblock that is compatible with refdev
617  *      1 - dev has a superblock that is compatible and newer than refdev
618  *          so dev should be used as the refdev in future
619  *     -EINVAL superblock incompatible or invalid
620  *     -othererror e.g. -EIO
621  *
622  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
623  *      Verify that dev is acceptable into mddev.
624  *       The first time, mddev->raid_disks will be 0, and data from
625  *       dev should be merged in.  Subsequent calls check that dev
626  *       is new enough.  Return 0 or -EINVAL
627  *
628  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
629  *     Update the superblock for rdev with data in mddev
630  *     This does not write to disc.
631  *
632  */
633
634 struct super_type  {
635         char            *name;
636         struct module   *owner;
637         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
638         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
639         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
640 };
641
642 /*
643  * load_super for 0.90.0 
644  */
645 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
646 {
647         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
648         mdp_super_t *sb;
649         int ret;
650         sector_t sb_offset;
651
652         /*
653          * Calculate the position of the superblock,
654          * it's at the end of the disk.
655          *
656          * It also happens to be a multiple of 4Kb.
657          */
658         sb_offset = calc_dev_sboffset(rdev->bdev);
659         rdev->sb_offset = sb_offset;
660
661         ret = read_disk_sb(rdev, MD_SB_BYTES);
662         if (ret) return ret;
663
664         ret = -EINVAL;
665
666         bdevname(rdev->bdev, b);
667         sb = (mdp_super_t*)page_address(rdev->sb_page);
668
669         if (sb->md_magic != MD_SB_MAGIC) {
670                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
671                        b);
672                 goto abort;
673         }
674
675         if (sb->major_version != 0 ||
676             sb->minor_version < 90 ||
677             sb->minor_version > 91) {
678                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
679                         sb->major_version, sb->minor_version,
680                         b);
681                 goto abort;
682         }
683
684         if (sb->raid_disks <= 0)
685                 goto abort;
686
687         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
688                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
689                         b);
690                 goto abort;
691         }
692
693         rdev->preferred_minor = sb->md_minor;
694         rdev->data_offset = 0;
695         rdev->sb_size = MD_SB_BYTES;
696
697         if (sb->level == LEVEL_MULTIPATH)
698                 rdev->desc_nr = -1;
699         else
700                 rdev->desc_nr = sb->this_disk.number;
701
702         if (refdev == 0)
703                 ret = 1;
704         else {
705                 __u64 ev1, ev2;
706                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
707                 if (!uuid_equal(refsb, sb)) {
708                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
709                                 b, bdevname(refdev->bdev,b2));
710                         goto abort;
711                 }
712                 if (!sb_equal(refsb, sb)) {
713                         printk(KERN_WARNING "md: %s has same UUID"
714                                " but different superblock to %s\n",
715                                b, bdevname(refdev->bdev, b2));
716                         goto abort;
717                 }
718                 ev1 = md_event(sb);
719                 ev2 = md_event(refsb);
720                 if (ev1 > ev2)
721                         ret = 1;
722                 else 
723                         ret = 0;
724         }
725         rdev->size = calc_dev_size(rdev, sb->chunk_size);
726
727         if (rdev->size < sb->size && sb->level > 1)
728                 /* "this cannot possibly happen" ... */
729                 ret = -EINVAL;
730
731  abort:
732         return ret;
733 }
734
735 /*
736  * validate_super for 0.90.0
737  */
738 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
739 {
740         mdp_disk_t *desc;
741         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
742         __u64 ev1 = md_event(sb);
743
744         rdev->raid_disk = -1;
745         rdev->flags = 0;
746         if (mddev->raid_disks == 0) {
747                 mddev->major_version = 0;
748                 mddev->minor_version = sb->minor_version;
749                 mddev->patch_version = sb->patch_version;
750                 mddev->persistent = ! sb->not_persistent;
751                 mddev->chunk_size = sb->chunk_size;
752                 mddev->ctime = sb->ctime;
753                 mddev->utime = sb->utime;
754                 mddev->level = sb->level;
755                 mddev->clevel[0] = 0;
756                 mddev->layout = sb->layout;
757                 mddev->raid_disks = sb->raid_disks;
758                 mddev->size = sb->size;
759                 mddev->events = ev1;
760                 mddev->bitmap_offset = 0;
761                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
762
763                 if (mddev->minor_version >= 91) {
764                         mddev->reshape_position = sb->reshape_position;
765                         mddev->delta_disks = sb->delta_disks;
766                         mddev->new_level = sb->new_level;
767                         mddev->new_layout = sb->new_layout;
768                         mddev->new_chunk = sb->new_chunk;
769                 } else {
770                         mddev->reshape_position = MaxSector;
771                         mddev->delta_disks = 0;
772                         mddev->new_level = mddev->level;
773                         mddev->new_layout = mddev->layout;
774                         mddev->new_chunk = mddev->chunk_size;
775                 }
776
777                 if (sb->state & (1<<MD_SB_CLEAN))
778                         mddev->recovery_cp = MaxSector;
779                 else {
780                         if (sb->events_hi == sb->cp_events_hi && 
781                                 sb->events_lo == sb->cp_events_lo) {
782                                 mddev->recovery_cp = sb->recovery_cp;
783                         } else
784                                 mddev->recovery_cp = 0;
785                 }
786
787                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
788                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
789                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
790                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
791
792                 mddev->max_disks = MD_SB_DISKS;
793
794                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
795                     mddev->bitmap_file == NULL) {
796                         if (mddev->level != 1 && mddev->level != 4
797                             && mddev->level != 5 && mddev->level != 6
798                             && mddev->level != 10) {
799                                 /* FIXME use a better test */
800                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
801                                 return -EINVAL;
802                         }
803                         mddev->bitmap_offset = mddev->default_bitmap_offset;
804                 }
805
806         } else if (mddev->pers == NULL) {
807                 /* Insist on good event counter while assembling */
808                 ++ev1;
809                 if (ev1 < mddev->events) 
810                         return -EINVAL;
811         } else if (mddev->bitmap) {
812                 /* if adding to array with a bitmap, then we can accept an
813                  * older device ... but not too old.
814                  */
815                 if (ev1 < mddev->bitmap->events_cleared)
816                         return 0;
817         } else {
818                 if (ev1 < mddev->events)
819                         /* just a hot-add of a new device, leave raid_disk at -1 */
820                         return 0;
821         }
822
823         if (mddev->level != LEVEL_MULTIPATH) {
824                 desc = sb->disks + rdev->desc_nr;
825
826                 if (desc->state & (1<<MD_DISK_FAULTY))
827                         set_bit(Faulty, &rdev->flags);
828                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
829                             desc->raid_disk < mddev->raid_disks */) {
830                         set_bit(In_sync, &rdev->flags);
831                         rdev->raid_disk = desc->raid_disk;
832                 }
833                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
834                         set_bit(WriteMostly, &rdev->flags);
835         } else /* MULTIPATH are always insync */
836                 set_bit(In_sync, &rdev->flags);
837         return 0;
838 }
839
840 /*
841  * sync_super for 0.90.0
842  */
843 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
844 {
845         mdp_super_t *sb;
846         struct list_head *tmp;
847         mdk_rdev_t *rdev2;
848         int next_spare = mddev->raid_disks;
849
850
851         /* make rdev->sb match mddev data..
852          *
853          * 1/ zero out disks
854          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
855          * 3/ any empty disks < next_spare become removed
856          *
857          * disks[0] gets initialised to REMOVED because
858          * we cannot be sure from other fields if it has
859          * been initialised or not.
860          */
861         int i;
862         int active=0, working=0,failed=0,spare=0,nr_disks=0;
863
864         rdev->sb_size = MD_SB_BYTES;
865
866         sb = (mdp_super_t*)page_address(rdev->sb_page);
867
868         memset(sb, 0, sizeof(*sb));
869
870         sb->md_magic = MD_SB_MAGIC;
871         sb->major_version = mddev->major_version;
872         sb->patch_version = mddev->patch_version;
873         sb->gvalid_words  = 0; /* ignored */
874         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
875         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
876         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
877         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
878
879         sb->ctime = mddev->ctime;
880         sb->level = mddev->level;
881         sb->size  = mddev->size;
882         sb->raid_disks = mddev->raid_disks;
883         sb->md_minor = mddev->md_minor;
884         sb->not_persistent = !mddev->persistent;
885         sb->utime = mddev->utime;
886         sb->state = 0;
887         sb->events_hi = (mddev->events>>32);
888         sb->events_lo = (u32)mddev->events;
889
890         if (mddev->reshape_position == MaxSector)
891                 sb->minor_version = 90;
892         else {
893                 sb->minor_version = 91;
894                 sb->reshape_position = mddev->reshape_position;
895                 sb->new_level = mddev->new_level;
896                 sb->delta_disks = mddev->delta_disks;
897                 sb->new_layout = mddev->new_layout;
898                 sb->new_chunk = mddev->new_chunk;
899         }
900         mddev->minor_version = sb->minor_version;
901         if (mddev->in_sync)
902         {
903                 sb->recovery_cp = mddev->recovery_cp;
904                 sb->cp_events_hi = (mddev->events>>32);
905                 sb->cp_events_lo = (u32)mddev->events;
906                 if (mddev->recovery_cp == MaxSector)
907                         sb->state = (1<< MD_SB_CLEAN);
908         } else
909                 sb->recovery_cp = 0;
910
911         sb->layout = mddev->layout;
912         sb->chunk_size = mddev->chunk_size;
913
914         if (mddev->bitmap && mddev->bitmap_file == NULL)
915                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
916
917         sb->disks[0].state = (1<<MD_DISK_REMOVED);
918         ITERATE_RDEV(mddev,rdev2,tmp) {
919                 mdp_disk_t *d;
920                 int desc_nr;
921                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
922                     && !test_bit(Faulty, &rdev2->flags))
923                         desc_nr = rdev2->raid_disk;
924                 else
925                         desc_nr = next_spare++;
926                 rdev2->desc_nr = desc_nr;
927                 d = &sb->disks[rdev2->desc_nr];
928                 nr_disks++;
929                 d->number = rdev2->desc_nr;
930                 d->major = MAJOR(rdev2->bdev->bd_dev);
931                 d->minor = MINOR(rdev2->bdev->bd_dev);
932                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
933                     && !test_bit(Faulty, &rdev2->flags))
934                         d->raid_disk = rdev2->raid_disk;
935                 else
936                         d->raid_disk = rdev2->desc_nr; /* compatibility */
937                 if (test_bit(Faulty, &rdev2->flags))
938                         d->state = (1<<MD_DISK_FAULTY);
939                 else if (test_bit(In_sync, &rdev2->flags)) {
940                         d->state = (1<<MD_DISK_ACTIVE);
941                         d->state |= (1<<MD_DISK_SYNC);
942                         active++;
943                         working++;
944                 } else {
945                         d->state = 0;
946                         spare++;
947                         working++;
948                 }
949                 if (test_bit(WriteMostly, &rdev2->flags))
950                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
951         }
952         /* now set the "removed" and "faulty" bits on any missing devices */
953         for (i=0 ; i < mddev->raid_disks ; i++) {
954                 mdp_disk_t *d = &sb->disks[i];
955                 if (d->state == 0 && d->number == 0) {
956                         d->number = i;
957                         d->raid_disk = i;
958                         d->state = (1<<MD_DISK_REMOVED);
959                         d->state |= (1<<MD_DISK_FAULTY);
960                         failed++;
961                 }
962         }
963         sb->nr_disks = nr_disks;
964         sb->active_disks = active;
965         sb->working_disks = working;
966         sb->failed_disks = failed;
967         sb->spare_disks = spare;
968
969         sb->this_disk = sb->disks[rdev->desc_nr];
970         sb->sb_csum = calc_sb_csum(sb);
971 }
972
973 /*
974  * version 1 superblock
975  */
976
977 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
978 {
979         __le32 disk_csum;
980         u32 csum;
981         unsigned long long newcsum;
982         int size = 256 + le32_to_cpu(sb->max_dev)*2;
983         __le32 *isuper = (__le32*)sb;
984         int i;
985
986         disk_csum = sb->sb_csum;
987         sb->sb_csum = 0;
988         newcsum = 0;
989         for (i=0; size>=4; size -= 4 )
990                 newcsum += le32_to_cpu(*isuper++);
991
992         if (size == 2)
993                 newcsum += le16_to_cpu(*(__le16*) isuper);
994
995         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
996         sb->sb_csum = disk_csum;
997         return cpu_to_le32(csum);
998 }
999
1000 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1001 {
1002         struct mdp_superblock_1 *sb;
1003         int ret;
1004         sector_t sb_offset;
1005         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1006         int bmask;
1007
1008         /*
1009          * Calculate the position of the superblock.
1010          * It is always aligned to a 4K boundary and
1011          * depeding on minor_version, it can be:
1012          * 0: At least 8K, but less than 12K, from end of device
1013          * 1: At start of device
1014          * 2: 4K from start of device.
1015          */
1016         switch(minor_version) {
1017         case 0:
1018                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1019                 sb_offset -= 8*2;
1020                 sb_offset &= ~(sector_t)(4*2-1);
1021                 /* convert from sectors to K */
1022                 sb_offset /= 2;
1023                 break;
1024         case 1:
1025                 sb_offset = 0;
1026                 break;
1027         case 2:
1028                 sb_offset = 4;
1029                 break;
1030         default:
1031                 return -EINVAL;
1032         }
1033         rdev->sb_offset = sb_offset;
1034
1035         /* superblock is rarely larger than 1K, but it can be larger,
1036          * and it is safe to read 4k, so we do that
1037          */
1038         ret = read_disk_sb(rdev, 4096);
1039         if (ret) return ret;
1040
1041
1042         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1043
1044         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1045             sb->major_version != cpu_to_le32(1) ||
1046             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1047             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1048             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1049                 return -EINVAL;
1050
1051         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1052                 printk("md: invalid superblock checksum on %s\n",
1053                         bdevname(rdev->bdev,b));
1054                 return -EINVAL;
1055         }
1056         if (le64_to_cpu(sb->data_size) < 10) {
1057                 printk("md: data_size too small on %s\n",
1058                        bdevname(rdev->bdev,b));
1059                 return -EINVAL;
1060         }
1061         rdev->preferred_minor = 0xffff;
1062         rdev->data_offset = le64_to_cpu(sb->data_offset);
1063         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1064
1065         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1066         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1067         if (rdev->sb_size & bmask)
1068                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1069
1070         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1071                 rdev->desc_nr = -1;
1072         else
1073                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1074
1075         if (refdev == 0)
1076                 ret = 1;
1077         else {
1078                 __u64 ev1, ev2;
1079                 struct mdp_superblock_1 *refsb = 
1080                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1081
1082                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1083                     sb->level != refsb->level ||
1084                     sb->layout != refsb->layout ||
1085                     sb->chunksize != refsb->chunksize) {
1086                         printk(KERN_WARNING "md: %s has strangely different"
1087                                 " superblock to %s\n",
1088                                 bdevname(rdev->bdev,b),
1089                                 bdevname(refdev->bdev,b2));
1090                         return -EINVAL;
1091                 }
1092                 ev1 = le64_to_cpu(sb->events);
1093                 ev2 = le64_to_cpu(refsb->events);
1094
1095                 if (ev1 > ev2)
1096                         ret = 1;
1097                 else
1098                         ret = 0;
1099         }
1100         if (minor_version) 
1101                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1102         else
1103                 rdev->size = rdev->sb_offset;
1104         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1105                 return -EINVAL;
1106         rdev->size = le64_to_cpu(sb->data_size)/2;
1107         if (le32_to_cpu(sb->chunksize))
1108                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1109
1110         if (le64_to_cpu(sb->size) > rdev->size*2)
1111                 return -EINVAL;
1112         return ret;
1113 }
1114
1115 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1116 {
1117         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1118         __u64 ev1 = le64_to_cpu(sb->events);
1119
1120         rdev->raid_disk = -1;
1121         rdev->flags = 0;
1122         if (mddev->raid_disks == 0) {
1123                 mddev->major_version = 1;
1124                 mddev->patch_version = 0;
1125                 mddev->persistent = 1;
1126                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1127                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1128                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1129                 mddev->level = le32_to_cpu(sb->level);
1130                 mddev->clevel[0] = 0;
1131                 mddev->layout = le32_to_cpu(sb->layout);
1132                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1133                 mddev->size = le64_to_cpu(sb->size)/2;
1134                 mddev->events = ev1;
1135                 mddev->bitmap_offset = 0;
1136                 mddev->default_bitmap_offset = 1024 >> 9;
1137                 
1138                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1139                 memcpy(mddev->uuid, sb->set_uuid, 16);
1140
1141                 mddev->max_disks =  (4096-256)/2;
1142
1143                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1144                     mddev->bitmap_file == NULL ) {
1145                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1146                             && mddev->level != 10) {
1147                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1148                                 return -EINVAL;
1149                         }
1150                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1151                 }
1152                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1153                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1154                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1155                         mddev->new_level = le32_to_cpu(sb->new_level);
1156                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1157                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1158                 } else {
1159                         mddev->reshape_position = MaxSector;
1160                         mddev->delta_disks = 0;
1161                         mddev->new_level = mddev->level;
1162                         mddev->new_layout = mddev->layout;
1163                         mddev->new_chunk = mddev->chunk_size;
1164                 }
1165
1166         } else if (mddev->pers == NULL) {
1167                 /* Insist of good event counter while assembling */
1168                 ++ev1;
1169                 if (ev1 < mddev->events)
1170                         return -EINVAL;
1171         } else if (mddev->bitmap) {
1172                 /* If adding to array with a bitmap, then we can accept an
1173                  * older device, but not too old.
1174                  */
1175                 if (ev1 < mddev->bitmap->events_cleared)
1176                         return 0;
1177         } else {
1178                 if (ev1 < mddev->events)
1179                         /* just a hot-add of a new device, leave raid_disk at -1 */
1180                         return 0;
1181         }
1182         if (mddev->level != LEVEL_MULTIPATH) {
1183                 int role;
1184                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1185                 switch(role) {
1186                 case 0xffff: /* spare */
1187                         break;
1188                 case 0xfffe: /* faulty */
1189                         set_bit(Faulty, &rdev->flags);
1190                         break;
1191                 default:
1192                         if ((le32_to_cpu(sb->feature_map) &
1193                              MD_FEATURE_RECOVERY_OFFSET))
1194                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1195                         else
1196                                 set_bit(In_sync, &rdev->flags);
1197                         rdev->raid_disk = role;
1198                         break;
1199                 }
1200                 if (sb->devflags & WriteMostly1)
1201                         set_bit(WriteMostly, &rdev->flags);
1202         } else /* MULTIPATH are always insync */
1203                 set_bit(In_sync, &rdev->flags);
1204
1205         return 0;
1206 }
1207
1208 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1209 {
1210         struct mdp_superblock_1 *sb;
1211         struct list_head *tmp;
1212         mdk_rdev_t *rdev2;
1213         int max_dev, i;
1214         /* make rdev->sb match mddev and rdev data. */
1215
1216         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1217
1218         sb->feature_map = 0;
1219         sb->pad0 = 0;
1220         sb->recovery_offset = cpu_to_le64(0);
1221         memset(sb->pad1, 0, sizeof(sb->pad1));
1222         memset(sb->pad2, 0, sizeof(sb->pad2));
1223         memset(sb->pad3, 0, sizeof(sb->pad3));
1224
1225         sb->utime = cpu_to_le64((__u64)mddev->utime);
1226         sb->events = cpu_to_le64(mddev->events);
1227         if (mddev->in_sync)
1228                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1229         else
1230                 sb->resync_offset = cpu_to_le64(0);
1231
1232         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1233
1234         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1235         sb->size = cpu_to_le64(mddev->size<<1);
1236
1237         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1238                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1239                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1240         }
1241
1242         if (rdev->raid_disk >= 0 &&
1243             !test_bit(In_sync, &rdev->flags) &&
1244             rdev->recovery_offset > 0) {
1245                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1246                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1247         }
1248
1249         if (mddev->reshape_position != MaxSector) {
1250                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1251                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1252                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1253                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1254                 sb->new_level = cpu_to_le32(mddev->new_level);
1255                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1256         }
1257
1258         max_dev = 0;
1259         ITERATE_RDEV(mddev,rdev2,tmp)
1260                 if (rdev2->desc_nr+1 > max_dev)
1261                         max_dev = rdev2->desc_nr+1;
1262         
1263         sb->max_dev = cpu_to_le32(max_dev);
1264         for (i=0; i<max_dev;i++)
1265                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1266         
1267         ITERATE_RDEV(mddev,rdev2,tmp) {
1268                 i = rdev2->desc_nr;
1269                 if (test_bit(Faulty, &rdev2->flags))
1270                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1271                 else if (test_bit(In_sync, &rdev2->flags))
1272                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1273                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1274                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1275                 else
1276                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1277         }
1278
1279         sb->sb_csum = calc_sb_1_csum(sb);
1280 }
1281
1282
1283 static struct super_type super_types[] = {
1284         [0] = {
1285                 .name   = "0.90.0",
1286                 .owner  = THIS_MODULE,
1287                 .load_super     = super_90_load,
1288                 .validate_super = super_90_validate,
1289                 .sync_super     = super_90_sync,
1290         },
1291         [1] = {
1292                 .name   = "md-1",
1293                 .owner  = THIS_MODULE,
1294                 .load_super     = super_1_load,
1295                 .validate_super = super_1_validate,
1296                 .sync_super     = super_1_sync,
1297         },
1298 };
1299         
1300 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1301 {
1302         struct list_head *tmp;
1303         mdk_rdev_t *rdev;
1304
1305         ITERATE_RDEV(mddev,rdev,tmp)
1306                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1307                         return rdev;
1308
1309         return NULL;
1310 }
1311
1312 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1313 {
1314         struct list_head *tmp;
1315         mdk_rdev_t *rdev;
1316
1317         ITERATE_RDEV(mddev1,rdev,tmp)
1318                 if (match_dev_unit(mddev2, rdev))
1319                         return 1;
1320
1321         return 0;
1322 }
1323
1324 static LIST_HEAD(pending_raid_disks);
1325
1326 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1327 {
1328         mdk_rdev_t *same_pdev;
1329         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1330         struct kobject *ko;
1331         char *s;
1332
1333         if (rdev->mddev) {
1334                 MD_BUG();
1335                 return -EINVAL;
1336         }
1337         /* make sure rdev->size exceeds mddev->size */
1338         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1339                 if (mddev->pers)
1340                         /* Cannot change size, so fail */
1341                         return -ENOSPC;
1342                 else
1343                         mddev->size = rdev->size;
1344         }
1345         same_pdev = match_dev_unit(mddev, rdev);
1346         if (same_pdev)
1347                 printk(KERN_WARNING
1348                         "%s: WARNING: %s appears to be on the same physical"
1349                         " disk as %s. True\n     protection against single-disk"
1350                         " failure might be compromised.\n",
1351                         mdname(mddev), bdevname(rdev->bdev,b),
1352                         bdevname(same_pdev->bdev,b2));
1353
1354         /* Verify rdev->desc_nr is unique.
1355          * If it is -1, assign a free number, else
1356          * check number is not in use
1357          */
1358         if (rdev->desc_nr < 0) {
1359                 int choice = 0;
1360                 if (mddev->pers) choice = mddev->raid_disks;
1361                 while (find_rdev_nr(mddev, choice))
1362                         choice++;
1363                 rdev->desc_nr = choice;
1364         } else {
1365                 if (find_rdev_nr(mddev, rdev->desc_nr))
1366                         return -EBUSY;
1367         }
1368         bdevname(rdev->bdev,b);
1369         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1370                 return -ENOMEM;
1371         while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1372                 *s = '!';
1373                         
1374         list_add(&rdev->same_set, &mddev->disks);
1375         rdev->mddev = mddev;
1376         printk(KERN_INFO "md: bind<%s>\n", b);
1377
1378         rdev->kobj.parent = &mddev->kobj;
1379         kobject_add(&rdev->kobj);
1380
1381         if (rdev->bdev->bd_part)
1382                 ko = &rdev->bdev->bd_part->kobj;
1383         else
1384                 ko = &rdev->bdev->bd_disk->kobj;
1385         sysfs_create_link(&rdev->kobj, ko, "block");
1386         bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1387         return 0;
1388 }
1389
1390 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1391 {
1392         char b[BDEVNAME_SIZE];
1393         if (!rdev->mddev) {
1394                 MD_BUG();
1395                 return;
1396         }
1397         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1398         list_del_init(&rdev->same_set);
1399         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1400         rdev->mddev = NULL;
1401         sysfs_remove_link(&rdev->kobj, "block");
1402         kobject_del(&rdev->kobj);
1403 }
1404
1405 /*
1406  * prevent the device from being mounted, repartitioned or
1407  * otherwise reused by a RAID array (or any other kernel
1408  * subsystem), by bd_claiming the device.
1409  */
1410 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1411 {
1412         int err = 0;
1413         struct block_device *bdev;
1414         char b[BDEVNAME_SIZE];
1415
1416         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1417         if (IS_ERR(bdev)) {
1418                 printk(KERN_ERR "md: could not open %s.\n",
1419                         __bdevname(dev, b));
1420                 return PTR_ERR(bdev);
1421         }
1422         err = bd_claim(bdev, rdev);
1423         if (err) {
1424                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1425                         bdevname(bdev, b));
1426                 blkdev_put(bdev);
1427                 return err;
1428         }
1429         rdev->bdev = bdev;
1430         return err;
1431 }
1432
1433 static void unlock_rdev(mdk_rdev_t *rdev)
1434 {
1435         struct block_device *bdev = rdev->bdev;
1436         rdev->bdev = NULL;
1437         if (!bdev)
1438                 MD_BUG();
1439         bd_release(bdev);
1440         blkdev_put(bdev);
1441 }
1442
1443 void md_autodetect_dev(dev_t dev);
1444
1445 static void export_rdev(mdk_rdev_t * rdev)
1446 {
1447         char b[BDEVNAME_SIZE];
1448         printk(KERN_INFO "md: export_rdev(%s)\n",
1449                 bdevname(rdev->bdev,b));
1450         if (rdev->mddev)
1451                 MD_BUG();
1452         free_disk_sb(rdev);
1453         list_del_init(&rdev->same_set);
1454 #ifndef MODULE
1455         md_autodetect_dev(rdev->bdev->bd_dev);
1456 #endif
1457         unlock_rdev(rdev);
1458         kobject_put(&rdev->kobj);
1459 }
1460
1461 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1462 {
1463         unbind_rdev_from_array(rdev);
1464         export_rdev(rdev);
1465 }
1466
1467 static void export_array(mddev_t *mddev)
1468 {
1469         struct list_head *tmp;
1470         mdk_rdev_t *rdev;
1471
1472         ITERATE_RDEV(mddev,rdev,tmp) {
1473                 if (!rdev->mddev) {
1474                         MD_BUG();
1475                         continue;
1476                 }
1477                 kick_rdev_from_array(rdev);
1478         }
1479         if (!list_empty(&mddev->disks))
1480                 MD_BUG();
1481         mddev->raid_disks = 0;
1482         mddev->major_version = 0;
1483 }
1484
1485 static void print_desc(mdp_disk_t *desc)
1486 {
1487         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1488                 desc->major,desc->minor,desc->raid_disk,desc->state);
1489 }
1490
1491 static void print_sb(mdp_super_t *sb)
1492 {
1493         int i;
1494
1495         printk(KERN_INFO 
1496                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1497                 sb->major_version, sb->minor_version, sb->patch_version,
1498                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1499                 sb->ctime);
1500         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1501                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1502                 sb->md_minor, sb->layout, sb->chunk_size);
1503         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1504                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1505                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1506                 sb->failed_disks, sb->spare_disks,
1507                 sb->sb_csum, (unsigned long)sb->events_lo);
1508
1509         printk(KERN_INFO);
1510         for (i = 0; i < MD_SB_DISKS; i++) {
1511                 mdp_disk_t *desc;
1512
1513                 desc = sb->disks + i;
1514                 if (desc->number || desc->major || desc->minor ||
1515                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1516                         printk("     D %2d: ", i);
1517                         print_desc(desc);
1518                 }
1519         }
1520         printk(KERN_INFO "md:     THIS: ");
1521         print_desc(&sb->this_disk);
1522
1523 }
1524
1525 static void print_rdev(mdk_rdev_t *rdev)
1526 {
1527         char b[BDEVNAME_SIZE];
1528         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1529                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1530                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1531                 rdev->desc_nr);
1532         if (rdev->sb_loaded) {
1533                 printk(KERN_INFO "md: rdev superblock:\n");
1534                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1535         } else
1536                 printk(KERN_INFO "md: no rdev superblock!\n");
1537 }
1538
1539 static void md_print_devices(void)
1540 {
1541         struct list_head *tmp, *tmp2;
1542         mdk_rdev_t *rdev;
1543         mddev_t *mddev;
1544         char b[BDEVNAME_SIZE];
1545
1546         printk("\n");
1547         printk("md:     **********************************\n");
1548         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1549         printk("md:     **********************************\n");
1550         ITERATE_MDDEV(mddev,tmp) {
1551
1552                 if (mddev->bitmap)
1553                         bitmap_print_sb(mddev->bitmap);
1554                 else
1555                         printk("%s: ", mdname(mddev));
1556                 ITERATE_RDEV(mddev,rdev,tmp2)
1557                         printk("<%s>", bdevname(rdev->bdev,b));
1558                 printk("\n");
1559
1560                 ITERATE_RDEV(mddev,rdev,tmp2)
1561                         print_rdev(rdev);
1562         }
1563         printk("md:     **********************************\n");
1564         printk("\n");
1565 }
1566
1567
1568 static void sync_sbs(mddev_t * mddev, int nospares)
1569 {
1570         /* Update each superblock (in-memory image), but
1571          * if we are allowed to, skip spares which already
1572          * have the right event counter, or have one earlier
1573          * (which would mean they aren't being marked as dirty
1574          * with the rest of the array)
1575          */
1576         mdk_rdev_t *rdev;
1577         struct list_head *tmp;
1578
1579         ITERATE_RDEV(mddev,rdev,tmp) {
1580                 if (rdev->sb_events == mddev->events ||
1581                     (nospares &&
1582                      rdev->raid_disk < 0 &&
1583                      (rdev->sb_events&1)==0 &&
1584                      rdev->sb_events+1 == mddev->events)) {
1585                         /* Don't update this superblock */
1586                         rdev->sb_loaded = 2;
1587                 } else {
1588                         super_types[mddev->major_version].
1589                                 sync_super(mddev, rdev);
1590                         rdev->sb_loaded = 1;
1591                 }
1592         }
1593 }
1594
1595 static void md_update_sb(mddev_t * mddev, int force_change)
1596 {
1597         int err;
1598         struct list_head *tmp;
1599         mdk_rdev_t *rdev;
1600         int sync_req;
1601         int nospares = 0;
1602
1603 repeat:
1604         spin_lock_irq(&mddev->write_lock);
1605
1606         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1607         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1608                 force_change = 1;
1609         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1610                 /* just a clean<-> dirty transition, possibly leave spares alone,
1611                  * though if events isn't the right even/odd, we will have to do
1612                  * spares after all
1613                  */
1614                 nospares = 1;
1615         if (force_change)
1616                 nospares = 0;
1617         if (mddev->degraded)
1618                 /* If the array is degraded, then skipping spares is both
1619                  * dangerous and fairly pointless.
1620                  * Dangerous because a device that was removed from the array
1621                  * might have a event_count that still looks up-to-date,
1622                  * so it can be re-added without a resync.
1623                  * Pointless because if there are any spares to skip,
1624                  * then a recovery will happen and soon that array won't
1625                  * be degraded any more and the spare can go back to sleep then.
1626                  */
1627                 nospares = 0;
1628
1629         sync_req = mddev->in_sync;
1630         mddev->utime = get_seconds();
1631
1632         /* If this is just a dirty<->clean transition, and the array is clean
1633          * and 'events' is odd, we can roll back to the previous clean state */
1634         if (nospares
1635             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1636             && (mddev->events & 1)
1637             && mddev->events != 1)
1638                 mddev->events--;
1639         else {
1640                 /* otherwise we have to go forward and ... */
1641                 mddev->events ++;
1642                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1643                         /* .. if the array isn't clean, insist on an odd 'events' */
1644                         if ((mddev->events&1)==0) {
1645                                 mddev->events++;
1646                                 nospares = 0;
1647                         }
1648                 } else {
1649                         /* otherwise insist on an even 'events' (for clean states) */
1650                         if ((mddev->events&1)) {
1651                                 mddev->events++;
1652                                 nospares = 0;
1653                         }
1654                 }
1655         }
1656
1657         if (!mddev->events) {
1658                 /*
1659                  * oops, this 64-bit counter should never wrap.
1660                  * Either we are in around ~1 trillion A.C., assuming
1661                  * 1 reboot per second, or we have a bug:
1662                  */
1663                 MD_BUG();
1664                 mddev->events --;
1665         }
1666         sync_sbs(mddev, nospares);
1667
1668         /*
1669          * do not write anything to disk if using
1670          * nonpersistent superblocks
1671          */
1672         if (!mddev->persistent) {
1673                 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1674                 spin_unlock_irq(&mddev->write_lock);
1675                 wake_up(&mddev->sb_wait);
1676                 return;
1677         }
1678         spin_unlock_irq(&mddev->write_lock);
1679
1680         dprintk(KERN_INFO 
1681                 "md: updating %s RAID superblock on device (in sync %d)\n",
1682                 mdname(mddev),mddev->in_sync);
1683
1684         err = bitmap_update_sb(mddev->bitmap);
1685         ITERATE_RDEV(mddev,rdev,tmp) {
1686                 char b[BDEVNAME_SIZE];
1687                 dprintk(KERN_INFO "md: ");
1688                 if (rdev->sb_loaded != 1)
1689                         continue; /* no noise on spare devices */
1690                 if (test_bit(Faulty, &rdev->flags))
1691                         dprintk("(skipping faulty ");
1692
1693                 dprintk("%s ", bdevname(rdev->bdev,b));
1694                 if (!test_bit(Faulty, &rdev->flags)) {
1695                         md_super_write(mddev,rdev,
1696                                        rdev->sb_offset<<1, rdev->sb_size,
1697                                        rdev->sb_page);
1698                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1699                                 bdevname(rdev->bdev,b),
1700                                 (unsigned long long)rdev->sb_offset);
1701                         rdev->sb_events = mddev->events;
1702
1703                 } else
1704                         dprintk(")\n");
1705                 if (mddev->level == LEVEL_MULTIPATH)
1706                         /* only need to write one superblock... */
1707                         break;
1708         }
1709         md_super_wait(mddev);
1710         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1711
1712         spin_lock_irq(&mddev->write_lock);
1713         if (mddev->in_sync != sync_req ||
1714             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1715                 /* have to write it out again */
1716                 spin_unlock_irq(&mddev->write_lock);
1717                 goto repeat;
1718         }
1719         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1720         spin_unlock_irq(&mddev->write_lock);
1721         wake_up(&mddev->sb_wait);
1722
1723 }
1724
1725 /* words written to sysfs files may, or my not, be \n terminated.
1726  * We want to accept with case. For this we use cmd_match.
1727  */
1728 static int cmd_match(const char *cmd, const char *str)
1729 {
1730         /* See if cmd, written into a sysfs file, matches
1731          * str.  They must either be the same, or cmd can
1732          * have a trailing newline
1733          */
1734         while (*cmd && *str && *cmd == *str) {
1735                 cmd++;
1736                 str++;
1737         }
1738         if (*cmd == '\n')
1739                 cmd++;
1740         if (*str || *cmd)
1741                 return 0;
1742         return 1;
1743 }
1744
1745 struct rdev_sysfs_entry {
1746         struct attribute attr;
1747         ssize_t (*show)(mdk_rdev_t *, char *);
1748         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1749 };
1750
1751 static ssize_t
1752 state_show(mdk_rdev_t *rdev, char *page)
1753 {
1754         char *sep = "";
1755         int len=0;
1756
1757         if (test_bit(Faulty, &rdev->flags)) {
1758                 len+= sprintf(page+len, "%sfaulty",sep);
1759                 sep = ",";
1760         }
1761         if (test_bit(In_sync, &rdev->flags)) {
1762                 len += sprintf(page+len, "%sin_sync",sep);
1763                 sep = ",";
1764         }
1765         if (test_bit(WriteMostly, &rdev->flags)) {
1766                 len += sprintf(page+len, "%swrite_mostly",sep);
1767                 sep = ",";
1768         }
1769         if (!test_bit(Faulty, &rdev->flags) &&
1770             !test_bit(In_sync, &rdev->flags)) {
1771                 len += sprintf(page+len, "%sspare", sep);
1772                 sep = ",";
1773         }
1774         return len+sprintf(page+len, "\n");
1775 }
1776
1777 static ssize_t
1778 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1779 {
1780         /* can write
1781          *  faulty  - simulates and error
1782          *  remove  - disconnects the device
1783          *  writemostly - sets write_mostly
1784          *  -writemostly - clears write_mostly
1785          */
1786         int err = -EINVAL;
1787         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1788                 md_error(rdev->mddev, rdev);
1789                 err = 0;
1790         } else if (cmd_match(buf, "remove")) {
1791                 if (rdev->raid_disk >= 0)
1792                         err = -EBUSY;
1793                 else {
1794                         mddev_t *mddev = rdev->mddev;
1795                         kick_rdev_from_array(rdev);
1796                         if (mddev->pers)
1797                                 md_update_sb(mddev, 1);
1798                         md_new_event(mddev);
1799                         err = 0;
1800                 }
1801         } else if (cmd_match(buf, "writemostly")) {
1802                 set_bit(WriteMostly, &rdev->flags);
1803                 err = 0;
1804         } else if (cmd_match(buf, "-writemostly")) {
1805                 clear_bit(WriteMostly, &rdev->flags);
1806                 err = 0;
1807         }
1808         return err ? err : len;
1809 }
1810 static struct rdev_sysfs_entry rdev_state =
1811 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1812
1813 static ssize_t
1814 super_show(mdk_rdev_t *rdev, char *page)
1815 {
1816         if (rdev->sb_loaded && rdev->sb_size) {
1817                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1818                 return rdev->sb_size;
1819         } else
1820                 return 0;
1821 }
1822 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1823
1824 static ssize_t
1825 errors_show(mdk_rdev_t *rdev, char *page)
1826 {
1827         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1828 }
1829
1830 static ssize_t
1831 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1832 {
1833         char *e;
1834         unsigned long n = simple_strtoul(buf, &e, 10);
1835         if (*buf && (*e == 0 || *e == '\n')) {
1836                 atomic_set(&rdev->corrected_errors, n);
1837                 return len;
1838         }
1839         return -EINVAL;
1840 }
1841 static struct rdev_sysfs_entry rdev_errors =
1842 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1843
1844 static ssize_t
1845 slot_show(mdk_rdev_t *rdev, char *page)
1846 {
1847         if (rdev->raid_disk < 0)
1848                 return sprintf(page, "none\n");
1849         else
1850                 return sprintf(page, "%d\n", rdev->raid_disk);
1851 }
1852
1853 static ssize_t
1854 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1855 {
1856         char *e;
1857         int slot = simple_strtoul(buf, &e, 10);
1858         if (strncmp(buf, "none", 4)==0)
1859                 slot = -1;
1860         else if (e==buf || (*e && *e!= '\n'))
1861                 return -EINVAL;
1862         if (rdev->mddev->pers)
1863                 /* Cannot set slot in active array (yet) */
1864                 return -EBUSY;
1865         if (slot >= rdev->mddev->raid_disks)
1866                 return -ENOSPC;
1867         rdev->raid_disk = slot;
1868         /* assume it is working */
1869         rdev->flags = 0;
1870         set_bit(In_sync, &rdev->flags);
1871         return len;
1872 }
1873
1874
1875 static struct rdev_sysfs_entry rdev_slot =
1876 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1877
1878 static ssize_t
1879 offset_show(mdk_rdev_t *rdev, char *page)
1880 {
1881         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1882 }
1883
1884 static ssize_t
1885 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1886 {
1887         char *e;
1888         unsigned long long offset = simple_strtoull(buf, &e, 10);
1889         if (e==buf || (*e && *e != '\n'))
1890                 return -EINVAL;
1891         if (rdev->mddev->pers)
1892                 return -EBUSY;
1893         rdev->data_offset = offset;
1894         return len;
1895 }
1896
1897 static struct rdev_sysfs_entry rdev_offset =
1898 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1899
1900 static ssize_t
1901 rdev_size_show(mdk_rdev_t *rdev, char *page)
1902 {
1903         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1904 }
1905
1906 static ssize_t
1907 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1908 {
1909         char *e;
1910         unsigned long long size = simple_strtoull(buf, &e, 10);
1911         if (e==buf || (*e && *e != '\n'))
1912                 return -EINVAL;
1913         if (rdev->mddev->pers)
1914                 return -EBUSY;
1915         rdev->size = size;
1916         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1917                 rdev->mddev->size = size;
1918         return len;
1919 }
1920
1921 static struct rdev_sysfs_entry rdev_size =
1922 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1923
1924 static struct attribute *rdev_default_attrs[] = {
1925         &rdev_state.attr,
1926         &rdev_super.attr,
1927         &rdev_errors.attr,
1928         &rdev_slot.attr,
1929         &rdev_offset.attr,
1930         &rdev_size.attr,
1931         NULL,
1932 };
1933 static ssize_t
1934 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1935 {
1936         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1937         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1938
1939         if (!entry->show)
1940                 return -EIO;
1941         return entry->show(rdev, page);
1942 }
1943
1944 static ssize_t
1945 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1946               const char *page, size_t length)
1947 {
1948         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1949         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1950
1951         if (!entry->store)
1952                 return -EIO;
1953         if (!capable(CAP_SYS_ADMIN))
1954                 return -EACCES;
1955         return entry->store(rdev, page, length);
1956 }
1957
1958 static void rdev_free(struct kobject *ko)
1959 {
1960         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1961         kfree(rdev);
1962 }
1963 static struct sysfs_ops rdev_sysfs_ops = {
1964         .show           = rdev_attr_show,
1965         .store          = rdev_attr_store,
1966 };
1967 static struct kobj_type rdev_ktype = {
1968         .release        = rdev_free,
1969         .sysfs_ops      = &rdev_sysfs_ops,
1970         .default_attrs  = rdev_default_attrs,
1971 };
1972
1973 /*
1974  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1975  *
1976  * mark the device faulty if:
1977  *
1978  *   - the device is nonexistent (zero size)
1979  *   - the device has no valid superblock
1980  *
1981  * a faulty rdev _never_ has rdev->sb set.
1982  */
1983 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1984 {
1985         char b[BDEVNAME_SIZE];
1986         int err;
1987         mdk_rdev_t *rdev;
1988         sector_t size;
1989
1990         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1991         if (!rdev) {
1992                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1993                 return ERR_PTR(-ENOMEM);
1994         }
1995
1996         if ((err = alloc_disk_sb(rdev)))
1997                 goto abort_free;
1998
1999         err = lock_rdev(rdev, newdev);
2000         if (err)
2001                 goto abort_free;
2002
2003         rdev->kobj.parent = NULL;
2004         rdev->kobj.ktype = &rdev_ktype;
2005         kobject_init(&rdev->kobj);
2006
2007         rdev->desc_nr = -1;
2008         rdev->saved_raid_disk = -1;
2009         rdev->raid_disk = -1;
2010         rdev->flags = 0;
2011         rdev->data_offset = 0;
2012         rdev->sb_events = 0;
2013         atomic_set(&rdev->nr_pending, 0);
2014         atomic_set(&rdev->read_errors, 0);
2015         atomic_set(&rdev->corrected_errors, 0);
2016
2017         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2018         if (!size) {
2019                 printk(KERN_WARNING 
2020                         "md: %s has zero or unknown size, marking faulty!\n",
2021                         bdevname(rdev->bdev,b));
2022                 err = -EINVAL;
2023                 goto abort_free;
2024         }
2025
2026         if (super_format >= 0) {
2027                 err = super_types[super_format].
2028                         load_super(rdev, NULL, super_minor);
2029                 if (err == -EINVAL) {
2030                         printk(KERN_WARNING 
2031                                 "md: %s has invalid sb, not importing!\n",
2032                                 bdevname(rdev->bdev,b));
2033                         goto abort_free;
2034                 }
2035                 if (err < 0) {
2036                         printk(KERN_WARNING 
2037                                 "md: could not read %s's sb, not importing!\n",
2038                                 bdevname(rdev->bdev,b));
2039                         goto abort_free;
2040                 }
2041         }
2042         INIT_LIST_HEAD(&rdev->same_set);
2043
2044         return rdev;
2045
2046 abort_free:
2047         if (rdev->sb_page) {
2048                 if (rdev->bdev)
2049                         unlock_rdev(rdev);
2050                 free_disk_sb(rdev);
2051         }
2052         kfree(rdev);
2053         return ERR_PTR(err);
2054 }
2055
2056 /*
2057  * Check a full RAID array for plausibility
2058  */
2059
2060
2061 static void analyze_sbs(mddev_t * mddev)
2062 {
2063         int i;
2064         struct list_head *tmp;
2065         mdk_rdev_t *rdev, *freshest;
2066         char b[BDEVNAME_SIZE];
2067
2068         freshest = NULL;
2069         ITERATE_RDEV(mddev,rdev,tmp)
2070                 switch (super_types[mddev->major_version].
2071                         load_super(rdev, freshest, mddev->minor_version)) {
2072                 case 1:
2073                         freshest = rdev;
2074                         break;
2075                 case 0:
2076                         break;
2077                 default:
2078                         printk( KERN_ERR \
2079                                 "md: fatal superblock inconsistency in %s"
2080                                 " -- removing from array\n", 
2081                                 bdevname(rdev->bdev,b));
2082                         kick_rdev_from_array(rdev);
2083                 }
2084
2085
2086         super_types[mddev->major_version].
2087                 validate_super(mddev, freshest);
2088
2089         i = 0;
2090         ITERATE_RDEV(mddev,rdev,tmp) {
2091                 if (rdev != freshest)
2092                         if (super_types[mddev->major_version].
2093                             validate_super(mddev, rdev)) {
2094                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2095                                         " from array!\n",
2096                                         bdevname(rdev->bdev,b));
2097                                 kick_rdev_from_array(rdev);
2098                                 continue;
2099                         }
2100                 if (mddev->level == LEVEL_MULTIPATH) {
2101                         rdev->desc_nr = i++;
2102                         rdev->raid_disk = rdev->desc_nr;
2103                         set_bit(In_sync, &rdev->flags);
2104                 }
2105         }
2106
2107
2108
2109         if (mddev->recovery_cp != MaxSector &&
2110             mddev->level >= 1)
2111                 printk(KERN_ERR "md: %s: raid array is not clean"
2112                        " -- starting background reconstruction\n",
2113                        mdname(mddev));
2114
2115 }
2116
2117 static ssize_t
2118 safe_delay_show(mddev_t *mddev, char *page)
2119 {
2120         int msec = (mddev->safemode_delay*1000)/HZ;
2121         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2122 }
2123 static ssize_t
2124 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2125 {
2126         int scale=1;
2127         int dot=0;
2128         int i;
2129         unsigned long msec;
2130         char buf[30];
2131         char *e;
2132         /* remove a period, and count digits after it */
2133         if (len >= sizeof(buf))
2134                 return -EINVAL;
2135         strlcpy(buf, cbuf, len);
2136         buf[len] = 0;
2137         for (i=0; i<len; i++) {
2138                 if (dot) {
2139                         if (isdigit(buf[i])) {
2140                                 buf[i-1] = buf[i];
2141                                 scale *= 10;
2142                         }
2143                         buf[i] = 0;
2144                 } else if (buf[i] == '.') {
2145                         dot=1;
2146                         buf[i] = 0;
2147                 }
2148         }
2149         msec = simple_strtoul(buf, &e, 10);
2150         if (e == buf || (*e && *e != '\n'))
2151                 return -EINVAL;
2152         msec = (msec * 1000) / scale;
2153         if (msec == 0)
2154                 mddev->safemode_delay = 0;
2155         else {
2156                 mddev->safemode_delay = (msec*HZ)/1000;
2157                 if (mddev->safemode_delay == 0)
2158                         mddev->safemode_delay = 1;
2159         }
2160         return len;
2161 }
2162 static struct md_sysfs_entry md_safe_delay =
2163 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2164
2165 static ssize_t
2166 level_show(mddev_t *mddev, char *page)
2167 {
2168         struct mdk_personality *p = mddev->pers;
2169         if (p)
2170                 return sprintf(page, "%s\n", p->name);
2171         else if (mddev->clevel[0])
2172                 return sprintf(page, "%s\n", mddev->clevel);
2173         else if (mddev->level != LEVEL_NONE)
2174                 return sprintf(page, "%d\n", mddev->level);
2175         else
2176                 return 0;
2177 }
2178
2179 static ssize_t
2180 level_store(mddev_t *mddev, const char *buf, size_t len)
2181 {
2182         int rv = len;
2183         if (mddev->pers)
2184                 return -EBUSY;
2185         if (len == 0)
2186                 return 0;
2187         if (len >= sizeof(mddev->clevel))
2188                 return -ENOSPC;
2189         strncpy(mddev->clevel, buf, len);
2190         if (mddev->clevel[len-1] == '\n')
2191                 len--;
2192         mddev->clevel[len] = 0;
2193         mddev->level = LEVEL_NONE;
2194         return rv;
2195 }
2196
2197 static struct md_sysfs_entry md_level =
2198 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2199
2200
2201 static ssize_t
2202 layout_show(mddev_t *mddev, char *page)
2203 {
2204         /* just a number, not meaningful for all levels */
2205         return sprintf(page, "%d\n", mddev->layout);
2206 }
2207
2208 static ssize_t
2209 layout_store(mddev_t *mddev, const char *buf, size_t len)
2210 {
2211         char *e;
2212         unsigned long n = simple_strtoul(buf, &e, 10);
2213         if (mddev->pers)
2214                 return -EBUSY;
2215
2216         if (!*buf || (*e && *e != '\n'))
2217                 return -EINVAL;
2218
2219         mddev->layout = n;
2220         return len;
2221 }
2222 static struct md_sysfs_entry md_layout =
2223 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2224
2225
2226 static ssize_t
2227 raid_disks_show(mddev_t *mddev, char *page)
2228 {
2229         if (mddev->raid_disks == 0)
2230                 return 0;
2231         return sprintf(page, "%d\n", mddev->raid_disks);
2232 }
2233
2234 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2235
2236 static ssize_t
2237 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2238 {
2239         char *e;
2240         int rv = 0;
2241         unsigned long n = simple_strtoul(buf, &e, 10);
2242
2243         if (!*buf || (*e && *e != '\n'))
2244                 return -EINVAL;
2245
2246         if (mddev->pers)
2247                 rv = update_raid_disks(mddev, n);
2248         else
2249                 mddev->raid_disks = n;
2250         return rv ? rv : len;
2251 }
2252 static struct md_sysfs_entry md_raid_disks =
2253 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2254
2255 static ssize_t
2256 chunk_size_show(mddev_t *mddev, char *page)
2257 {
2258         return sprintf(page, "%d\n", mddev->chunk_size);
2259 }
2260
2261 static ssize_t
2262 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2263 {
2264         /* can only set chunk_size if array is not yet active */
2265         char *e;
2266         unsigned long n = simple_strtoul(buf, &e, 10);
2267
2268         if (mddev->pers)
2269                 return -EBUSY;
2270         if (!*buf || (*e && *e != '\n'))
2271                 return -EINVAL;
2272
2273         mddev->chunk_size = n;
2274         return len;
2275 }
2276 static struct md_sysfs_entry md_chunk_size =
2277 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2278
2279 static ssize_t
2280 resync_start_show(mddev_t *mddev, char *page)
2281 {
2282         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2283 }
2284
2285 static ssize_t
2286 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2287 {
2288         /* can only set chunk_size if array is not yet active */
2289         char *e;
2290         unsigned long long n = simple_strtoull(buf, &e, 10);
2291
2292         if (mddev->pers)
2293                 return -EBUSY;
2294         if (!*buf || (*e && *e != '\n'))
2295                 return -EINVAL;
2296
2297         mddev->recovery_cp = n;
2298         return len;
2299 }
2300 static struct md_sysfs_entry md_resync_start =
2301 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2302
2303 /*
2304  * The array state can be:
2305  *
2306  * clear
2307  *     No devices, no size, no level
2308  *     Equivalent to STOP_ARRAY ioctl
2309  * inactive
2310  *     May have some settings, but array is not active
2311  *        all IO results in error
2312  *     When written, doesn't tear down array, but just stops it
2313  * suspended (not supported yet)
2314  *     All IO requests will block. The array can be reconfigured.
2315  *     Writing this, if accepted, will block until array is quiessent
2316  * readonly
2317  *     no resync can happen.  no superblocks get written.
2318  *     write requests fail
2319  * read-auto
2320  *     like readonly, but behaves like 'clean' on a write request.
2321  *
2322  * clean - no pending writes, but otherwise active.
2323  *     When written to inactive array, starts without resync
2324  *     If a write request arrives then
2325  *       if metadata is known, mark 'dirty' and switch to 'active'.
2326  *       if not known, block and switch to write-pending
2327  *     If written to an active array that has pending writes, then fails.
2328  * active
2329  *     fully active: IO and resync can be happening.
2330  *     When written to inactive array, starts with resync
2331  *
2332  * write-pending
2333  *     clean, but writes are blocked waiting for 'active' to be written.
2334  *
2335  * active-idle
2336  *     like active, but no writes have been seen for a while (100msec).
2337  *
2338  */
2339 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2340                    write_pending, active_idle, bad_word};
2341 static char *array_states[] = {
2342         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2343         "write-pending", "active-idle", NULL };
2344
2345 static int match_word(const char *word, char **list)
2346 {
2347         int n;
2348         for (n=0; list[n]; n++)
2349                 if (cmd_match(word, list[n]))
2350                         break;
2351         return n;
2352 }
2353
2354 static ssize_t
2355 array_state_show(mddev_t *mddev, char *page)
2356 {
2357         enum array_state st = inactive;
2358
2359         if (mddev->pers)
2360                 switch(mddev->ro) {
2361                 case 1:
2362                         st = readonly;
2363                         break;
2364                 case 2:
2365                         st = read_auto;
2366                         break;
2367                 case 0:
2368                         if (mddev->in_sync)
2369                                 st = clean;
2370                         else if (mddev->safemode)
2371                                 st = active_idle;
2372                         else
2373                                 st = active;
2374                 }
2375         else {
2376                 if (list_empty(&mddev->disks) &&
2377                     mddev->raid_disks == 0 &&
2378                     mddev->size == 0)
2379                         st = clear;
2380                 else
2381                         st = inactive;
2382         }
2383         return sprintf(page, "%s\n", array_states[st]);
2384 }
2385
2386 static int do_md_stop(mddev_t * mddev, int ro);
2387 static int do_md_run(mddev_t * mddev);
2388 static int restart_array(mddev_t *mddev);
2389
2390 static ssize_t
2391 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2392 {
2393         int err = -EINVAL;
2394         enum array_state st = match_word(buf, array_states);
2395         switch(st) {
2396         case bad_word:
2397                 break;
2398         case clear:
2399                 /* stopping an active array */
2400                 if (mddev->pers) {
2401                         if (atomic_read(&mddev->active) > 1)
2402                                 return -EBUSY;
2403                         err = do_md_stop(mddev, 0);
2404                 }
2405                 break;
2406         case inactive:
2407                 /* stopping an active array */
2408                 if (mddev->pers) {
2409                         if (atomic_read(&mddev->active) > 1)
2410                                 return -EBUSY;
2411                         err = do_md_stop(mddev, 2);
2412                 }
2413                 break;
2414         case suspended:
2415                 break; /* not supported yet */
2416         case readonly:
2417                 if (mddev->pers)
2418                         err = do_md_stop(mddev, 1);
2419                 else {
2420                         mddev->ro = 1;
2421                         err = do_md_run(mddev);
2422                 }
2423                 break;
2424         case read_auto:
2425                 /* stopping an active array */
2426                 if (mddev->pers) {
2427                         err = do_md_stop(mddev, 1);
2428                         if (err == 0)
2429                                 mddev->ro = 2; /* FIXME mark devices writable */
2430                 } else {
2431                         mddev->ro = 2;
2432                         err = do_md_run(mddev);
2433                 }
2434                 break;
2435         case clean:
2436                 if (mddev->pers) {
2437                         restart_array(mddev);
2438                         spin_lock_irq(&mddev->write_lock);
2439                         if (atomic_read(&mddev->writes_pending) == 0) {
2440                                 mddev->in_sync = 1;
2441                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
2442                         }
2443                         spin_unlock_irq(&mddev->write_lock);
2444                 } else {
2445                         mddev->ro = 0;
2446                         mddev->recovery_cp = MaxSector;
2447                         err = do_md_run(mddev);
2448                 }
2449                 break;
2450         case active:
2451                 if (mddev->pers) {
2452                         restart_array(mddev);
2453                         clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2454                         wake_up(&mddev->sb_wait);
2455                         err = 0;
2456                 } else {
2457                         mddev->ro = 0;
2458                         err = do_md_run(mddev);
2459                 }
2460                 break;
2461         case write_pending:
2462         case active_idle:
2463                 /* these cannot be set */
2464                 break;
2465         }
2466         if (err)
2467                 return err;
2468         else
2469                 return len;
2470 }
2471 static struct md_sysfs_entry md_array_state =
2472 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2473
2474 static ssize_t
2475 null_show(mddev_t *mddev, char *page)
2476 {
2477         return -EINVAL;
2478 }
2479
2480 static ssize_t
2481 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2482 {
2483         /* buf must be %d:%d\n? giving major and minor numbers */
2484         /* The new device is added to the array.
2485          * If the array has a persistent superblock, we read the
2486          * superblock to initialise info and check validity.
2487          * Otherwise, only checking done is that in bind_rdev_to_array,
2488          * which mainly checks size.
2489          */
2490         char *e;
2491         int major = simple_strtoul(buf, &e, 10);
2492         int minor;
2493         dev_t dev;
2494         mdk_rdev_t *rdev;
2495         int err;
2496
2497         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2498                 return -EINVAL;
2499         minor = simple_strtoul(e+1, &e, 10);
2500         if (*e && *e != '\n')
2501                 return -EINVAL;
2502         dev = MKDEV(major, minor);
2503         if (major != MAJOR(dev) ||
2504             minor != MINOR(dev))
2505                 return -EOVERFLOW;
2506
2507
2508         if (mddev->persistent) {
2509                 rdev = md_import_device(dev, mddev->major_version,
2510                                         mddev->minor_version);
2511                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2512                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2513                                                        mdk_rdev_t, same_set);
2514                         err = super_types[mddev->major_version]
2515                                 .load_super(rdev, rdev0, mddev->minor_version);
2516                         if (err < 0)
2517                                 goto out;
2518                 }
2519         } else
2520                 rdev = md_import_device(dev, -1, -1);
2521
2522         if (IS_ERR(rdev))
2523                 return PTR_ERR(rdev);
2524         err = bind_rdev_to_array(rdev, mddev);
2525  out:
2526         if (err)
2527                 export_rdev(rdev);
2528         return err ? err : len;
2529 }
2530
2531 static struct md_sysfs_entry md_new_device =
2532 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2533
2534 static ssize_t
2535 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2536 {
2537         char *end;
2538         unsigned long chunk, end_chunk;
2539
2540         if (!mddev->bitmap)
2541                 goto out;
2542         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2543         while (*buf) {
2544                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2545                 if (buf == end) break;
2546                 if (*end == '-') { /* range */
2547                         buf = end + 1;
2548                         end_chunk = simple_strtoul(buf, &end, 0);
2549                         if (buf == end) break;
2550                 }
2551                 if (*end && !isspace(*end)) break;
2552                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2553                 buf = end;
2554                 while (isspace(*buf)) buf++;
2555         }
2556         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2557 out:
2558         return len;
2559 }
2560
2561 static struct md_sysfs_entry md_bitmap =
2562 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2563
2564 static ssize_t
2565 size_show(mddev_t *mddev, char *page)
2566 {
2567         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2568 }
2569
2570 static int update_size(mddev_t *mddev, unsigned long size);
2571
2572 static ssize_t
2573 size_store(mddev_t *mddev, const char *buf, size_t len)
2574 {
2575         /* If array is inactive, we can reduce the component size, but
2576          * not increase it (except from 0).
2577          * If array is active, we can try an on-line resize
2578          */
2579         char *e;
2580         int err = 0;
2581         unsigned long long size = simple_strtoull(buf, &e, 10);
2582         if (!*buf || *buf == '\n' ||
2583             (*e && *e != '\n'))
2584                 return -EINVAL;
2585
2586         if (mddev->pers) {
2587                 err = update_size(mddev, size);
2588                 md_update_sb(mddev, 1);
2589         } else {
2590                 if (mddev->size == 0 ||
2591                     mddev->size > size)
2592                         mddev->size = size;
2593                 else
2594                         err = -ENOSPC;
2595         }
2596         return err ? err : len;
2597 }
2598
2599 static struct md_sysfs_entry md_size =
2600 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2601
2602
2603 /* Metdata version.
2604  * This is either 'none' for arrays with externally managed metadata,
2605  * or N.M for internally known formats
2606  */
2607 static ssize_t
2608 metadata_show(mddev_t *mddev, char *page)
2609 {
2610         if (mddev->persistent)
2611                 return sprintf(page, "%d.%d\n",
2612                                mddev->major_version, mddev->minor_version);
2613         else
2614                 return sprintf(page, "none\n");
2615 }
2616
2617 static ssize_t
2618 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2619 {
2620         int major, minor;
2621         char *e;
2622         if (!list_empty(&mddev->disks))
2623                 return -EBUSY;
2624
2625         if (cmd_match(buf, "none")) {
2626                 mddev->persistent = 0;
2627                 mddev->major_version = 0;
2628                 mddev->minor_version = 90;
2629                 return len;
2630         }
2631         major = simple_strtoul(buf, &e, 10);
2632         if (e==buf || *e != '.')
2633                 return -EINVAL;
2634         buf = e+1;
2635         minor = simple_strtoul(buf, &e, 10);
2636         if (e==buf || (*e && *e != '\n') )
2637                 return -EINVAL;
2638         if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2639             super_types[major].name == NULL)
2640                 return -ENOENT;
2641         mddev->major_version = major;
2642         mddev->minor_version = minor;
2643         mddev->persistent = 1;
2644         return len;
2645 }
2646
2647 static struct md_sysfs_entry md_metadata =
2648 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2649
2650 static ssize_t
2651 action_show(mddev_t *mddev, char *page)
2652 {
2653         char *type = "idle";
2654         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2655             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2656                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2657                         type = "reshape";
2658                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2659                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2660                                 type = "resync";
2661                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2662                                 type = "check";
2663                         else
2664                                 type = "repair";
2665                 } else
2666                         type = "recover";
2667         }
2668         return sprintf(page, "%s\n", type);
2669 }
2670
2671 static ssize_t
2672 action_store(mddev_t *mddev, const char *page, size_t len)
2673 {
2674         if (!mddev->pers || !mddev->pers->sync_request)
2675                 return -EINVAL;
2676
2677         if (cmd_match(page, "idle")) {
2678                 if (mddev->sync_thread) {
2679                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2680                         md_unregister_thread(mddev->sync_thread);
2681                         mddev->sync_thread = NULL;
2682                         mddev->recovery = 0;
2683                 }
2684         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2685                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2686                 return -EBUSY;
2687         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2688                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2689         else if (cmd_match(page, "reshape")) {
2690                 int err;
2691                 if (mddev->pers->start_reshape == NULL)
2692                         return -EINVAL;
2693                 err = mddev->pers->start_reshape(mddev);
2694                 if (err)
2695                         return err;
2696         } else {
2697                 if (cmd_match(page, "check"))
2698                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2699                 else if (!cmd_match(page, "repair"))
2700                         return -EINVAL;
2701                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2702                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2703         }
2704         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2705         md_wakeup_thread(mddev->thread);
2706         return len;
2707 }
2708
2709 static ssize_t
2710 mismatch_cnt_show(mddev_t *mddev, char *page)
2711 {
2712         return sprintf(page, "%llu\n",
2713                        (unsigned long long) mddev->resync_mismatches);
2714 }
2715
2716 static struct md_sysfs_entry md_scan_mode =
2717 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2718
2719
2720 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2721
2722 static ssize_t
2723 sync_min_show(mddev_t *mddev, char *page)
2724 {
2725         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2726                        mddev->sync_speed_min ? "local": "system");
2727 }
2728
2729 static ssize_t
2730 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2731 {
2732         int min;
2733         char *e;
2734         if (strncmp(buf, "system", 6)==0) {
2735                 mddev->sync_speed_min = 0;
2736                 return len;
2737         }
2738         min = simple_strtoul(buf, &e, 10);
2739         if (buf == e || (*e && *e != '\n') || min <= 0)
2740                 return -EINVAL;
2741         mddev->sync_speed_min = min;
2742         return len;
2743 }
2744
2745 static struct md_sysfs_entry md_sync_min =
2746 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2747
2748 static ssize_t
2749 sync_max_show(mddev_t *mddev, char *page)
2750 {
2751         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2752                        mddev->sync_speed_max ? "local": "system");
2753 }
2754
2755 static ssize_t
2756 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2757 {
2758         int max;
2759         char *e;
2760         if (strncmp(buf, "system", 6)==0) {
2761                 mddev->sync_speed_max = 0;
2762                 return len;
2763         }
2764         max = simple_strtoul(buf, &e, 10);
2765         if (buf == e || (*e && *e != '\n') || max <= 0)
2766                 return -EINVAL;
2767         mddev->sync_speed_max = max;
2768         return len;
2769 }
2770
2771 static struct md_sysfs_entry md_sync_max =
2772 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2773
2774
2775 static ssize_t
2776 sync_speed_show(mddev_t *mddev, char *page)
2777 {
2778         unsigned long resync, dt, db;
2779         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2780         dt = ((jiffies - mddev->resync_mark) / HZ);
2781         if (!dt) dt++;
2782         db = resync - (mddev->resync_mark_cnt);
2783         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2784 }
2785
2786 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2787
2788 static ssize_t
2789 sync_completed_show(mddev_t *mddev, char *page)
2790 {
2791         unsigned long max_blocks, resync;
2792
2793         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2794                 max_blocks = mddev->resync_max_sectors;
2795         else
2796                 max_blocks = mddev->size << 1;
2797
2798         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2799         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2800 }
2801
2802 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2803
2804 static ssize_t
2805 suspend_lo_show(mddev_t *mddev, char *page)
2806 {
2807         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2808 }
2809
2810 static ssize_t
2811 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2812 {
2813         char *e;
2814         unsigned long long new = simple_strtoull(buf, &e, 10);
2815
2816         if (mddev->pers->quiesce == NULL)
2817                 return -EINVAL;
2818         if (buf == e || (*e && *e != '\n'))
2819                 return -EINVAL;
2820         if (new >= mddev->suspend_hi ||
2821             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2822                 mddev->suspend_lo = new;
2823                 mddev->pers->quiesce(mddev, 2);
2824                 return len;
2825         } else
2826                 return -EINVAL;
2827 }
2828 static struct md_sysfs_entry md_suspend_lo =
2829 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2830
2831
2832 static ssize_t
2833 suspend_hi_show(mddev_t *mddev, char *page)
2834 {
2835         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2836 }
2837
2838 static ssize_t
2839 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2840 {
2841         char *e;
2842         unsigned long long new = simple_strtoull(buf, &e, 10);
2843
2844         if (mddev->pers->quiesce == NULL)
2845                 return -EINVAL;
2846         if (buf == e || (*e && *e != '\n'))
2847                 return -EINVAL;
2848         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2849             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2850                 mddev->suspend_hi = new;
2851                 mddev->pers->quiesce(mddev, 1);
2852                 mddev->pers->quiesce(mddev, 0);
2853                 return len;
2854         } else
2855                 return -EINVAL;
2856 }
2857 static struct md_sysfs_entry md_suspend_hi =
2858 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2859
2860
2861 static struct attribute *md_default_attrs[] = {
2862         &md_level.attr,
2863         &md_layout.attr,
2864         &md_raid_disks.attr,
2865         &md_chunk_size.attr,
2866         &md_size.attr,
2867         &md_resync_start.attr,
2868         &md_metadata.attr,
2869         &md_new_device.attr,
2870         &md_safe_delay.attr,
2871         &md_array_state.attr,
2872         NULL,
2873 };
2874
2875 static struct attribute *md_redundancy_attrs[] = {
2876         &md_scan_mode.attr,
2877         &md_mismatches.attr,
2878         &md_sync_min.attr,
2879         &md_sync_max.attr,
2880         &md_sync_speed.attr,
2881         &md_sync_completed.attr,
2882         &md_suspend_lo.attr,
2883         &md_suspend_hi.attr,
2884         &md_bitmap.attr,
2885         NULL,
2886 };
2887 static struct attribute_group md_redundancy_group = {
2888         .name = NULL,
2889         .attrs = md_redundancy_attrs,
2890 };
2891
2892
2893 static ssize_t
2894 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2895 {
2896         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2897         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2898         ssize_t rv;
2899
2900         if (!entry->show)
2901                 return -EIO;
2902         rv = mddev_lock(mddev);
2903         if (!rv) {
2904                 rv = entry->show(mddev, page);
2905                 mddev_unlock(mddev);
2906         }
2907         return rv;
2908 }
2909
2910 static ssize_t
2911 md_attr_store(struct kobject *kobj, struct attribute *attr,
2912               const char *page, size_t length)
2913 {
2914         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2915         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2916         ssize_t rv;
2917
2918         if (!entry->store)
2919                 return -EIO;
2920         if (!capable(CAP_SYS_ADMIN))
2921                 return -EACCES;
2922         rv = mddev_lock(mddev);
2923         if (!rv) {
2924                 rv = entry->store(mddev, page, length);
2925                 mddev_unlock(mddev);
2926         }
2927         return rv;
2928 }
2929
2930 static void md_free(struct kobject *ko)
2931 {
2932         mddev_t *mddev = container_of(ko, mddev_t, kobj);
2933         kfree(mddev);
2934 }
2935
2936 static struct sysfs_ops md_sysfs_ops = {
2937         .show   = md_attr_show,
2938         .store  = md_attr_store,
2939 };
2940 static struct kobj_type md_ktype = {
2941         .release        = md_free,
2942         .sysfs_ops      = &md_sysfs_ops,
2943         .default_attrs  = md_default_attrs,
2944 };
2945
2946 int mdp_major = 0;
2947
2948 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2949 {
2950         static DEFINE_MUTEX(disks_mutex);
2951         mddev_t *mddev = mddev_find(dev);
2952         struct gendisk *disk;
2953         int partitioned = (MAJOR(dev) != MD_MAJOR);
2954         int shift = partitioned ? MdpMinorShift : 0;
2955         int unit = MINOR(dev) >> shift;
2956
2957         if (!mddev)
2958                 return NULL;
2959
2960         mutex_lock(&disks_mutex);
2961         if (mddev->gendisk) {
2962                 mutex_unlock(&disks_mutex);
2963                 mddev_put(mddev);
2964                 return NULL;
2965         }
2966         disk = alloc_disk(1 << shift);
2967         if (!disk) {
2968                 mutex_unlock(&disks_mutex);
2969                 mddev_put(mddev);
2970                 return NULL;
2971         }
2972         disk->major = MAJOR(dev);
2973         disk->first_minor = unit << shift;
2974         if (partitioned)
2975                 sprintf(disk->disk_name, "md_d%d", unit);
2976         else
2977                 sprintf(disk->disk_name, "md%d", unit);
2978         disk->fops = &md_fops;
2979         disk->private_data = mddev;
2980         disk->queue = mddev->queue;
2981         add_disk(disk);
2982         mddev->gendisk = disk;
2983         mutex_unlock(&disks_mutex);
2984         mddev->kobj.parent = &disk->kobj;
2985         mddev->kobj.k_name = NULL;
2986         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2987         mddev->kobj.ktype = &md_ktype;
2988         kobject_register(&mddev->kobj);
2989         return NULL;
2990 }
2991
2992 static void md_safemode_timeout(unsigned long data)
2993 {
2994         mddev_t *mddev = (mddev_t *) data;
2995
2996         mddev->safemode = 1;
2997         md_wakeup_thread(mddev->thread);
2998 }
2999
3000 static int start_dirty_degraded;
3001
3002 static int do_md_run(mddev_t * mddev)
3003 {
3004         int err;
3005         int chunk_size;
3006         struct list_head *tmp;
3007         mdk_rdev_t *rdev;
3008         struct gendisk *disk;
3009         struct mdk_personality *pers;
3010         char b[BDEVNAME_SIZE];
3011
3012         if (list_empty(&mddev->disks))
3013                 /* cannot run an array with no devices.. */
3014                 return -EINVAL;
3015
3016         if (mddev->pers)
3017                 return -EBUSY;
3018
3019         /*
3020          * Analyze all RAID superblock(s)
3021          */
3022         if (!mddev->raid_disks)
3023                 analyze_sbs(mddev);
3024
3025         chunk_size = mddev->chunk_size;
3026
3027         if (chunk_size) {
3028                 if (chunk_size > MAX_CHUNK_SIZE) {
3029                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3030                                 chunk_size, MAX_CHUNK_SIZE);
3031                         return -EINVAL;
3032                 }
3033                 /*
3034                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3035                  */
3036                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3037                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3038                         return -EINVAL;
3039                 }
3040                 if (chunk_size < PAGE_SIZE) {
3041                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3042                                 chunk_size, PAGE_SIZE);
3043                         return -EINVAL;
3044                 }
3045
3046                 /* devices must have minimum size of one chunk */
3047                 ITERATE_RDEV(mddev,rdev,tmp) {
3048                         if (test_bit(Faulty, &rdev->flags))
3049                                 continue;
3050                         if (rdev->size < chunk_size / 1024) {
3051                                 printk(KERN_WARNING
3052                                         "md: Dev %s smaller than chunk_size:"
3053                                         " %lluk < %dk\n",
3054                                         bdevname(rdev->bdev,b),
3055                                         (unsigned long long)rdev->size,
3056                                         chunk_size / 1024);
3057                                 return -EINVAL;
3058                         }
3059                 }
3060         }
3061
3062 #ifdef CONFIG_KMOD
3063         if (mddev->level != LEVEL_NONE)
3064                 request_module("md-level-%d", mddev->level);
3065         else if (mddev->clevel[0])
3066                 request_module("md-%s", mddev->clevel);
3067 #endif
3068
3069         /*
3070          * Drop all container device buffers, from now on
3071          * the only valid external interface is through the md
3072          * device.
3073          * Also find largest hardsector size
3074          */
3075         ITERATE_RDEV(mddev,rdev,tmp) {
3076                 if (test_bit(Faulty, &rdev->flags))
3077                         continue;
3078                 sync_blockdev(rdev->bdev);
3079                 invalidate_bdev(rdev->bdev, 0);
3080         }
3081
3082         md_probe(mddev->unit, NULL, NULL);
3083         disk = mddev->gendisk;
3084         if (!disk)
3085                 return -ENOMEM;
3086
3087         spin_lock(&pers_lock);
3088         pers = find_pers(mddev->level, mddev->clevel);
3089         if (!pers || !try_module_get(pers->owner)) {
3090                 spin_unlock(&pers_lock);
3091                 if (mddev->level != LEVEL_NONE)
3092                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3093                                mddev->level);
3094                 else
3095                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3096                                mddev->clevel);
3097                 return -EINVAL;
3098         }
3099         mddev->pers = pers;
3100         spin_unlock(&pers_lock);
3101         mddev->level = pers->level;
3102         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3103
3104         if (mddev->reshape_position != MaxSector &&
3105             pers->start_reshape == NULL) {
3106                 /* This personality cannot handle reshaping... */
3107                 mddev->pers = NULL;
3108                 module_put(pers->owner);
3109                 return -EINVAL;
3110         }
3111
3112         mddev->recovery = 0;
3113         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3114         mddev->barriers_work = 1;
3115         mddev->ok_start_degraded = start_dirty_degraded;
3116
3117         if (start_readonly)
3118                 mddev->ro = 2; /* read-only, but switch on first write */
3119
3120         err = mddev->pers->run(mddev);
3121         if (!err && mddev->pers->sync_request) {
3122                 err = bitmap_create(mddev);
3123                 if (err) {
3124                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3125                                mdname(mddev), err);
3126                         mddev->pers->stop(mddev);
3127                 }
3128         }
3129         if (err) {
3130                 printk(KERN_ERR "md: pers->run() failed ...\n");
3131                 module_put(mddev->pers->owner);
3132                 mddev->pers = NULL;
3133                 bitmap_destroy(mddev);
3134                 return err;
3135         }
3136         if (mddev->pers->sync_request)
3137                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
3138         else if (mddev->ro == 2) /* auto-readonly not meaningful */
3139                 mddev->ro = 0;
3140
3141         atomic_set(&mddev->writes_pending,0);
3142         mddev->safemode = 0;
3143         mddev->safemode_timer.function = md_safemode_timeout;
3144         mddev->safemode_timer.data = (unsigned long) mddev;
3145         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3146         mddev->in_sync = 1;
3147
3148         ITERATE_RDEV(mddev,rdev,tmp)
3149                 if (rdev->raid_disk >= 0) {
3150                         char nm[20];
3151                         sprintf(nm, "rd%d", rdev->raid_disk);
3152                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
3153                 }
3154         
3155         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3156         
3157         if (mddev->flags)
3158                 md_update_sb(mddev, 0);
3159
3160         set_capacity(disk, mddev->array_size<<1);
3161
3162         /* If we call blk_queue_make_request here, it will
3163          * re-initialise max_sectors etc which may have been
3164          * refined inside -> run.  So just set the bits we need to set.
3165          * Most initialisation happended when we called
3166          * blk_queue_make_request(..., md_fail_request)
3167          * earlier.
3168          */
3169         mddev->queue->queuedata = mddev;
3170         mddev->queue->make_request_fn = mddev->pers->make_request;
3171
3172         /* If there is a partially-recovered drive we need to
3173          * start recovery here.  If we leave it to md_check_recovery,
3174          * it will remove the drives and not do the right thing
3175          */
3176         if (mddev->degraded && !mddev->sync_thread) {
3177                 struct list_head *rtmp;
3178                 int spares = 0;
3179                 ITERATE_RDEV(mddev,rdev,rtmp)
3180                         if (rdev->raid_disk >= 0 &&
3181                             !test_bit(In_sync, &rdev->flags) &&
3182                             !test_bit(Faulty, &rdev->flags))
3183                                 /* complete an interrupted recovery */
3184                                 spares++;
3185                 if (spares && mddev->pers->sync_request) {
3186                         mddev->recovery = 0;
3187                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3188                         mddev->sync_thread = md_register_thread(md_do_sync,
3189                                                                 mddev,
3190                                                                 "%s_resync");
3191                         if (!mddev->sync_thread) {
3192                                 printk(KERN_ERR "%s: could not start resync"
3193                                        " thread...\n",
3194                                        mdname(mddev));
3195                                 /* leave the spares where they are, it shouldn't hurt */
3196                                 mddev->recovery = 0;
3197                         }
3198                 }
3199         }
3200         md_wakeup_thread(mddev->thread);
3201         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3202
3203         mddev->changed = 1;
3204         md_new_event(mddev);
3205         kobject_uevent(&mddev->gendisk->kobj, KOBJ_CHANGE);
3206         return 0;
3207 }
3208
3209 static int restart_array(mddev_t *mddev)
3210 {
3211         struct gendisk *disk = mddev->gendisk;
3212         int err;
3213
3214         /*
3215          * Complain if it has no devices
3216          */
3217         err = -ENXIO;
3218         if (list_empty(&mddev->disks))
3219                 goto out;
3220
3221         if (mddev->pers) {
3222                 err = -EBUSY;
3223                 if (!mddev->ro)
3224                         goto out;
3225
3226                 mddev->safemode = 0;
3227                 mddev->ro = 0;
3228                 set_disk_ro(disk, 0);
3229
3230                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3231                         mdname(mddev));
3232                 /*
3233                  * Kick recovery or resync if necessary
3234                  */
3235                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3236                 md_wakeup_thread(mddev->thread);
3237                 md_wakeup_thread(mddev->sync_thread);
3238                 err = 0;
3239         } else
3240                 err = -EINVAL;
3241
3242 out:
3243         return err;
3244 }
3245
3246 /* similar to deny_write_access, but accounts for our holding a reference
3247  * to the file ourselves */
3248 static int deny_bitmap_write_access(struct file * file)
3249 {
3250         struct inode *inode = file->f_mapping->host;
3251
3252         spin_lock(&inode->i_lock);
3253         if (atomic_read(&inode->i_writecount) > 1) {
3254                 spin_unlock(&inode->i_lock);
3255                 return -ETXTBSY;
3256         }
3257         atomic_set(&inode->i_writecount, -1);
3258         spin_unlock(&inode->i_lock);
3259
3260         return 0;
3261 }
3262
3263 static void restore_bitmap_write_access(struct file *file)
3264 {
3265         struct inode *inode = file->f_mapping->host;
3266
3267         spin_lock(&inode->i_lock);
3268         atomic_set(&inode->i_writecount, 1);
3269         spin_unlock(&inode->i_lock);
3270 }
3271
3272 /* mode:
3273  *   0 - completely stop and dis-assemble array
3274  *   1 - switch to readonly
3275  *   2 - stop but do not disassemble array
3276  */
3277 static int do_md_stop(mddev_t * mddev, int mode)
3278 {
3279         int err = 0;
3280         struct gendisk *disk = mddev->gendisk;
3281
3282         if (mddev->pers) {
3283                 if (atomic_read(&mddev->active)>2) {
3284                         printk("md: %s still in use.\n",mdname(mddev));
3285                         return -EBUSY;
3286                 }
3287
3288                 if (mddev->sync_thread) {
3289                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3290                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3291                         md_unregister_thread(mddev->sync_thread);
3292                         mddev->sync_thread = NULL;
3293                 }
3294
3295                 del_timer_sync(&mddev->safemode_timer);
3296
3297                 invalidate_partition(disk, 0);
3298
3299                 switch(mode) {
3300                 case 1: /* readonly */
3301                         err  = -ENXIO;
3302                         if (mddev->ro==1)
3303                                 goto out;
3304                         mddev->ro = 1;
3305                         break;
3306                 case 0: /* disassemble */
3307                 case 2: /* stop */
3308                         bitmap_flush(mddev);
3309                         md_super_wait(mddev);
3310                         if (mddev->ro)
3311                                 set_disk_ro(disk, 0);
3312                         blk_queue_make_request(mddev->queue, md_fail_request);
3313                         mddev->pers->stop(mddev);
3314                         if (mddev->pers->sync_request)
3315                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3316
3317                         module_put(mddev->pers->owner);
3318                         mddev->pers = NULL;
3319
3320                         set_capacity(disk, 0);
3321                         mddev->changed = 1;
3322
3323                         if (mddev->ro)
3324                                 mddev->ro = 0;
3325                 }
3326                 if (!mddev->in_sync || mddev->flags) {
3327                         /* mark array as shutdown cleanly */
3328                         mddev->in_sync = 1;
3329                         md_update_sb(mddev, 1);
3330                 }
3331                 if (mode == 1)
3332                         set_disk_ro(disk, 1);
3333                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3334         }
3335
3336         /*
3337          * Free resources if final stop
3338          */
3339         if (mode == 0) {
3340                 mdk_rdev_t *rdev;
3341                 struct list_head *tmp;
3342
3343                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3344
3345                 bitmap_destroy(mddev);
3346                 if (mddev->bitmap_file) {
3347                         restore_bitmap_write_access(mddev->bitmap_file);
3348                         fput(mddev->bitmap_file);
3349                         mddev->bitmap_file = NULL;
3350                 }
3351                 mddev->bitmap_offset = 0;
3352
3353                 ITERATE_RDEV(mddev,rdev,tmp)
3354                         if (rdev->raid_disk >= 0) {
3355                                 char nm[20];
3356                                 sprintf(nm, "rd%d", rdev->raid_disk);
3357                                 sysfs_remove_link(&mddev->kobj, nm);
3358                         }
3359
3360                 export_array(mddev);
3361
3362                 mddev->array_size = 0;
3363                 mddev->size = 0;
3364                 mddev->raid_disks = 0;
3365                 mddev->recovery_cp = 0;
3366
3367         } else if (mddev->pers)
3368                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3369                         mdname(mddev));
3370         err = 0;
3371         md_new_event(mddev);
3372 out:
3373         return err;
3374 }
3375
3376 #ifndef MODULE
3377 static void autorun_array(mddev_t *mddev)
3378 {
3379         mdk_rdev_t *rdev;
3380         struct list_head *tmp;
3381         int err;
3382
3383         if (list_empty(&mddev->disks))
3384                 return;
3385
3386         printk(KERN_INFO "md: running: ");
3387
3388         ITERATE_RDEV(mddev,rdev,tmp) {
3389                 char b[BDEVNAME_SIZE];
3390                 printk("<%s>", bdevname(rdev->bdev,b));
3391         }
3392         printk("\n");
3393
3394         err = do_md_run (mddev);
3395         if (err) {
3396                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3397                 do_md_stop (mddev, 0);
3398         }
3399 }
3400
3401 /*
3402  * lets try to run arrays based on all disks that have arrived
3403  * until now. (those are in pending_raid_disks)
3404  *
3405  * the method: pick the first pending disk, collect all disks with
3406  * the same UUID, remove all from the pending list and put them into
3407  * the 'same_array' list. Then order this list based on superblock
3408  * update time (freshest comes first), kick out 'old' disks and
3409  * compare superblocks. If everything's fine then run it.
3410  *
3411  * If "unit" is allocated, then bump its reference count
3412  */
3413 static void autorun_devices(int part)
3414 {
3415         struct list_head *tmp;
3416         mdk_rdev_t *rdev0, *rdev;
3417         mddev_t *mddev;
3418         char b[BDEVNAME_SIZE];
3419
3420         printk(KERN_INFO "md: autorun ...\n");
3421         while (!list_empty(&pending_raid_disks)) {
3422                 int unit;
3423                 dev_t dev;
3424                 LIST_HEAD(candidates);
3425                 rdev0 = list_entry(pending_raid_disks.next,
3426                                          mdk_rdev_t, same_set);
3427
3428                 printk(KERN_INFO "md: considering %s ...\n",
3429                         bdevname(rdev0->bdev,b));
3430                 INIT_LIST_HEAD(&candidates);
3431                 ITERATE_RDEV_PENDING(rdev,tmp)
3432                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3433                                 printk(KERN_INFO "md:  adding %s ...\n",
3434                                         bdevname(rdev->bdev,b));
3435                                 list_move(&rdev->same_set, &candidates);
3436                         }
3437                 /*
3438                  * now we have a set of devices, with all of them having
3439                  * mostly sane superblocks. It's time to allocate the
3440                  * mddev.
3441                  */
3442                 if (part) {
3443                         dev = MKDEV(mdp_major,
3444                                     rdev0->preferred_minor << MdpMinorShift);
3445                         unit = MINOR(dev) >> MdpMinorShift;
3446                 } else {
3447                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3448                         unit = MINOR(dev);
3449                 }
3450                 if (rdev0->preferred_minor != unit) {
3451                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3452                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3453                         break;
3454                 }
3455
3456                 md_probe(dev, NULL, NULL);
3457                 mddev = mddev_find(dev);
3458                 if (!mddev) {
3459                         printk(KERN_ERR 
3460                                 "md: cannot allocate memory for md drive.\n");
3461                         break;
3462                 }
3463                 if (mddev_lock(mddev)) 
3464                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3465                                mdname(mddev));
3466                 else if (mddev->raid_disks || mddev->major_version
3467                          || !list_empty(&mddev->disks)) {
3468                         printk(KERN_WARNING 
3469                                 "md: %s already running, cannot run %s\n",
3470                                 mdname(mddev), bdevname(rdev0->bdev,b));
3471                         mddev_unlock(mddev);
3472                 } else {
3473                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3474                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3475                                 list_del_init(&rdev->same_set);
3476                                 if (bind_rdev_to_array(rdev, mddev))
3477                                         export_rdev(rdev);
3478                         }
3479                         autorun_array(mddev);
3480                         mddev_unlock(mddev);
3481                 }
3482                 /* on success, candidates will be empty, on error
3483                  * it won't...
3484                  */
3485                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3486                         export_rdev(rdev);
3487                 mddev_put(mddev);
3488         }
3489         printk(KERN_INFO "md: ... autorun DONE.\n");
3490 }
3491 #endif /* !MODULE */
3492
3493 static int get_version(void __user * arg)
3494 {
3495         mdu_version_t ver;
3496
3497         ver.major = MD_MAJOR_VERSION;
3498         ver.minor = MD_MINOR_VERSION;
3499         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3500
3501         if (copy_to_user(arg, &ver, sizeof(ver)))
3502                 return -EFAULT;
3503
3504         return 0;
3505 }
3506
3507 static int get_array_info(mddev_t * mddev, void __user * arg)
3508 {
3509         mdu_array_info_t info;
3510         int nr,working,active,failed,spare;
3511         mdk_rdev_t *rdev;
3512         struct list_head *tmp;
3513
3514         nr=working=active=failed=spare=0;
3515         ITERATE_RDEV(mddev,rdev,tmp) {
3516                 nr++;
3517                 if (test_bit(Faulty, &rdev->flags))
3518                         failed++;
3519                 else {
3520                         working++;
3521                         if (test_bit(In_sync, &rdev->flags))
3522                                 active++;       
3523                         else
3524                                 spare++;
3525                 }
3526         }
3527
3528         info.major_version = mddev->major_version;
3529         info.minor_version = mddev->minor_version;
3530         info.patch_version = MD_PATCHLEVEL_VERSION;
3531         info.ctime         = mddev->ctime;
3532         info.level         = mddev->level;
3533         info.size          = mddev->size;
3534         if (info.size != mddev->size) /* overflow */
3535                 info.size = -1;
3536         info.nr_disks      = nr;
3537         info.raid_disks    = mddev->raid_disks;
3538         info.md_minor      = mddev->md_minor;
3539         info.not_persistent= !mddev->persistent;
3540
3541         info.utime         = mddev->utime;
3542         info.state         = 0;
3543         if (mddev->in_sync)
3544                 info.state = (1<<MD_SB_CLEAN);
3545         if (mddev->bitmap && mddev->bitmap_offset)
3546                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3547         info.active_disks  = active;
3548         info.working_disks = working;
3549         info.failed_disks  = failed;
3550         info.spare_disks   = spare;
3551
3552         info.layout        = mddev->layout;
3553         info.chunk_size    = mddev->chunk_size;
3554
3555         if (copy_to_user(arg, &info, sizeof(info)))
3556                 return -EFAULT;
3557
3558         return 0;
3559 }
3560
3561 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3562 {
3563         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3564         char *ptr, *buf = NULL;
3565         int err = -ENOMEM;
3566
3567         file = kmalloc(sizeof(*file), GFP_KERNEL);
3568         if (!file)
3569                 goto out;
3570
3571         /* bitmap disabled, zero the first byte and copy out */
3572         if (!mddev->bitmap || !mddev->bitmap->file) {
3573                 file->pathname[0] = '\0';
3574                 goto copy_out;
3575         }
3576
3577         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3578         if (!buf)
3579                 goto out;
3580
3581         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3582         if (!ptr)
3583                 goto out;
3584
3585         strcpy(file->pathname, ptr);
3586
3587 copy_out:
3588         err = 0;
3589         if (copy_to_user(arg, file, sizeof(*file)))
3590                 err = -EFAULT;
3591 out:
3592         kfree(buf);
3593         kfree(file);
3594         return err;
3595 }
3596
3597 static int get_disk_info(mddev_t * mddev, void __user * arg)
3598 {
3599         mdu_disk_info_t info;
3600         unsigned int nr;
3601         mdk_rdev_t *rdev;
3602
3603         if (copy_from_user(&info, arg, sizeof(info)))
3604                 return -EFAULT;
3605
3606         nr = info.number;
3607
3608         rdev = find_rdev_nr(mddev, nr);
3609         if (rdev) {
3610                 info.major = MAJOR(rdev->bdev->bd_dev);
3611                 info.minor = MINOR(rdev->bdev->bd_dev);
3612                 info.raid_disk = rdev->raid_disk;
3613                 info.state = 0;
3614                 if (test_bit(Faulty, &rdev->flags))
3615                         info.state |= (1<<MD_DISK_FAULTY);
3616                 else if (test_bit(In_sync, &rdev->flags)) {
3617                         info.state |= (1<<MD_DISK_ACTIVE);
3618                         info.state |= (1<<MD_DISK_SYNC);
3619                 }
3620                 if (test_bit(WriteMostly, &rdev->flags))
3621                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3622         } else {
3623                 info.major = info.minor = 0;
3624                 info.raid_disk = -1;
3625                 info.state = (1<<MD_DISK_REMOVED);
3626         }
3627
3628         if (copy_to_user(arg, &info, sizeof(info)))
3629                 return -EFAULT;
3630
3631         return 0;
3632 }
3633
3634 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3635 {
3636         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3637         mdk_rdev_t *rdev;
3638         dev_t dev = MKDEV(info->major,info->minor);
3639
3640         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3641                 return -EOVERFLOW;
3642
3643         if (!mddev->raid_disks) {
3644                 int err;
3645                 /* expecting a device which has a superblock */
3646                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3647                 if (IS_ERR(rdev)) {
3648                         printk(KERN_WARNING 
3649                                 "md: md_import_device returned %ld\n",
3650                                 PTR_ERR(rdev));
3651                         return PTR_ERR(rdev);
3652                 }
3653                 if (!list_empty(&mddev->disks)) {
3654                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3655                                                         mdk_rdev_t, same_set);
3656                         int err = super_types[mddev->major_version]
3657                                 .load_super(rdev, rdev0, mddev->minor_version);
3658                         if (err < 0) {
3659                                 printk(KERN_WARNING 
3660                                         "md: %s has different UUID to %s\n",
3661                                         bdevname(rdev->bdev,b), 
3662                                         bdevname(rdev0->bdev,b2));
3663                                 export_rdev(rdev);
3664                                 return -EINVAL;
3665                         }
3666                 }
3667                 err = bind_rdev_to_array(rdev, mddev);
3668                 if (err)
3669                         export_rdev(rdev);
3670                 return err;
3671         }
3672
3673         /*
3674          * add_new_disk can be used once the array is assembled
3675          * to add "hot spares".  They must already have a superblock
3676          * written
3677          */
3678         if (mddev->pers) {
3679                 int err;
3680                 if (!mddev->pers->hot_add_disk) {
3681                         printk(KERN_WARNING 
3682                                 "%s: personality does not support diskops!\n",
3683                                mdname(mddev));
3684                         return -EINVAL;
3685                 }
3686                 if (mddev->persistent)
3687                         rdev = md_import_device(dev, mddev->major_version,
3688                                                 mddev->minor_version);
3689                 else
3690                         rdev = md_import_device(dev, -1, -1);
3691                 if (IS_ERR(rdev)) {
3692                         printk(KERN_WARNING 
3693                                 "md: md_import_device returned %ld\n",
3694                                 PTR_ERR(rdev));
3695                         return PTR_ERR(rdev);
3696                 }
3697                 /* set save_raid_disk if appropriate */
3698                 if (!mddev->persistent) {
3699                         if (info->state & (1<<MD_DISK_SYNC)  &&
3700                             info->raid_disk < mddev->raid_disks)
3701                                 rdev->raid_disk = info->raid_disk;
3702                         else
3703                                 rdev->raid_disk = -1;
3704                 } else
3705                         super_types[mddev->major_version].
3706                                 validate_super(mddev, rdev);
3707                 rdev->saved_raid_disk = rdev->raid_disk;
3708
3709                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3710                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3711                         set_bit(WriteMostly, &rdev->flags);
3712
3713                 rdev->raid_disk = -1;
3714                 err = bind_rdev_to_array(rdev, mddev);
3715                 if (!err && !mddev->pers->hot_remove_disk) {
3716                         /* If there is hot_add_disk but no hot_remove_disk
3717                          * then added disks for geometry changes,
3718                          * and should be added immediately.
3719                          */
3720                         super_types[mddev->major_version].
3721                                 validate_super(mddev, rdev);
3722                         err = mddev->pers->hot_add_disk(mddev, rdev);
3723                         if (err)
3724                                 unbind_rdev_from_array(rdev);
3725                 }
3726                 if (err)
3727                         export_rdev(rdev);
3728
3729                 md_update_sb(mddev, 1);
3730                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3731                 md_wakeup_thread(mddev->thread);
3732                 return err;
3733         }
3734
3735         /* otherwise, add_new_disk is only allowed
3736          * for major_version==0 superblocks
3737          */
3738         if (mddev->major_version != 0) {
3739                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3740                        mdname(mddev));
3741                 return -EINVAL;
3742         }
3743
3744         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3745                 int err;
3746                 rdev = md_import_device (dev, -1, 0);
3747                 if (IS_ERR(rdev)) {
3748                         printk(KERN_WARNING 
3749                                 "md: error, md_import_device() returned %ld\n",
3750                                 PTR_ERR(rdev));
3751                         return PTR_ERR(rdev);
3752                 }
3753                 rdev->desc_nr = info->number;
3754                 if (info->raid_disk < mddev->raid_disks)
3755                         rdev->raid_disk = info->raid_disk;
3756                 else
3757                         rdev->raid_disk = -1;
3758
3759                 rdev->flags = 0;
3760
3761                 if (rdev->raid_disk < mddev->raid_disks)
3762                         if (info->state & (1<<MD_DISK_SYNC))
3763                                 set_bit(In_sync, &rdev->flags);
3764
3765                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3766                         set_bit(WriteMostly, &rdev->flags);
3767
3768                 if (!mddev->persistent) {
3769                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3770                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3771                 } else 
3772                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3773                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3774
3775                 err = bind_rdev_to_array(rdev, mddev);
3776                 if (err) {
3777                         export_rdev(rdev);
3778                         return err;
3779                 }
3780         }
3781
3782         return 0;
3783 }
3784
3785 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3786 {
3787         char b[BDEVNAME_SIZE];
3788         mdk_rdev_t *rdev;
3789
3790         if (!mddev->pers)
3791                 return -ENODEV;
3792
3793         rdev = find_rdev(mddev, dev);
3794         if (!rdev)
3795                 return -ENXIO;
3796
3797         if (rdev->raid_disk >= 0)
3798                 goto busy;
3799
3800         kick_rdev_from_array(rdev);
3801         md_update_sb(mddev, 1);
3802         md_new_event(mddev);
3803
3804         return 0;
3805 busy:
3806         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3807                 bdevname(rdev->bdev,b), mdname(mddev));
3808         return -EBUSY;
3809 }
3810
3811 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3812 {
3813         char b[BDEVNAME_SIZE];
3814         int err;
3815         unsigned int size;
3816         mdk_rdev_t *rdev;
3817
3818         if (!mddev->pers)
3819                 return -ENODEV;
3820
3821         if (mddev->major_version != 0) {
3822                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3823                         " version-0 superblocks.\n",
3824                         mdname(mddev));
3825                 return -EINVAL;
3826         }
3827         if (!mddev->pers->hot_add_disk) {
3828                 printk(KERN_WARNING 
3829                         "%s: personality does not support diskops!\n",
3830                         mdname(mddev));
3831                 return -EINVAL;
3832         }
3833
3834         rdev = md_import_device (dev, -1, 0);
3835         if (IS_ERR(rdev)) {
3836                 printk(KERN_WARNING 
3837                         "md: error, md_import_device() returned %ld\n",
3838                         PTR_ERR(rdev));
3839                 return -EINVAL;
3840         }
3841
3842         if (mddev->persistent)
3843                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3844         else
3845                 rdev->sb_offset =
3846                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3847
3848         size = calc_dev_size(rdev, mddev->chunk_size);
3849         rdev->size = size;
3850
3851         if (test_bit(Faulty, &rdev->flags)) {
3852                 printk(KERN_WARNING 
3853                         "md: can not hot-add faulty %s disk to %s!\n",
3854                         bdevname(rdev->bdev,b), mdname(mddev));
3855                 err = -EINVAL;
3856                 goto abort_export;
3857         }
3858         clear_bit(In_sync, &rdev->flags);
3859         rdev->desc_nr = -1;
3860         rdev->saved_raid_disk = -1;
3861         err = bind_rdev_to_array(rdev, mddev);
3862         if (err)
3863                 goto abort_export;
3864
3865         /*
3866          * The rest should better be atomic, we can have disk failures
3867          * noticed in interrupt contexts ...
3868          */
3869
3870         if (rdev->desc_nr == mddev->max_disks) {
3871                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3872                         mdname(mddev));
3873                 err = -EBUSY;
3874                 goto abort_unbind_export;
3875         }
3876
3877         rdev->raid_disk = -1;
3878
3879         md_update_sb(mddev, 1);
3880
3881         /*
3882          * Kick recovery, maybe this spare has to be added to the
3883          * array immediately.
3884          */
3885         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3886         md_wakeup_thread(mddev->thread);
3887         md_new_event(mddev);
3888         return 0;
3889
3890 abort_unbind_export:
3891         unbind_rdev_from_array(rdev);
3892
3893 abort_export:
3894         export_rdev(rdev);
3895         return err;
3896 }
3897
3898 static int set_bitmap_file(mddev_t *mddev, int fd)
3899 {
3900         int err;
3901
3902         if (mddev->pers) {
3903                 if (!mddev->pers->quiesce)
3904                         return -EBUSY;
3905                 if (mddev->recovery || mddev->sync_thread)
3906                         return -EBUSY;
3907                 /* we should be able to change the bitmap.. */
3908         }
3909
3910
3911         if (fd >= 0) {
3912                 if (mddev->bitmap)
3913                         return -EEXIST; /* cannot add when bitmap is present */
3914                 mddev->bitmap_file = fget(fd);
3915
3916                 if (mddev->bitmap_file == NULL) {
3917                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3918                                mdname(mddev));
3919                         return -EBADF;
3920                 }
3921
3922                 err = deny_bitmap_write_access(mddev->bitmap_file);
3923                 if (err) {
3924                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3925                                mdname(mddev));
3926                         fput(mddev->bitmap_file);
3927                         mddev->bitmap_file = NULL;
3928                         return err;
3929                 }
3930                 mddev->bitmap_offset = 0; /* file overrides offset */
3931         } else if (mddev->bitmap == NULL)
3932                 return -ENOENT; /* cannot remove what isn't there */
3933         err = 0;
3934         if (mddev->pers) {
3935                 mddev->pers->quiesce(mddev, 1);
3936                 if (fd >= 0)
3937                         err = bitmap_create(mddev);
3938                 if (fd < 0 || err) {
3939                         bitmap_destroy(mddev);
3940                         fd = -1; /* make sure to put the file */
3941                 }
3942                 mddev->pers->quiesce(mddev, 0);
3943         }
3944         if (fd < 0) {
3945                 if (mddev->bitmap_file) {
3946                         restore_bitmap_write_access(mddev->bitmap_file);
3947                         fput(mddev->bitmap_file);
3948                 }
3949                 mddev->bitmap_file = NULL;
3950         }
3951
3952         return err;
3953 }
3954
3955 /*
3956  * set_array_info is used two different ways
3957  * The original usage is when creating a new array.
3958  * In this usage, raid_disks is > 0 and it together with
3959  *  level, size, not_persistent,layout,chunksize determine the
3960  *  shape of the array.
3961  *  This will always create an array with a type-0.90.0 superblock.
3962  * The newer usage is when assembling an array.
3963  *  In this case raid_disks will be 0, and the major_version field is
3964  *  use to determine which style super-blocks are to be found on the devices.
3965  *  The minor and patch _version numbers are also kept incase the
3966  *  super_block handler wishes to interpret them.
3967  */
3968 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3969 {
3970
3971         if (info->raid_disks == 0) {
3972                 /* just setting version number for superblock loading */
3973                 if (info->major_version < 0 ||
3974                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3975                     super_types[info->major_version].name == NULL) {
3976                         /* maybe try to auto-load a module? */
3977                         printk(KERN_INFO 
3978                                 "md: superblock version %d not known\n",
3979                                 info->major_version);
3980                         return -EINVAL;
3981                 }
3982                 mddev->major_version = info->major_version;
3983                 mddev->minor_version = info->minor_version;
3984                 mddev->patch_version = info->patch_version;
3985                 mddev->persistent = !info->not_persistent;
3986                 return 0;
3987         }
3988         mddev->major_version = MD_MAJOR_VERSION;
3989         mddev->minor_version = MD_MINOR_VERSION;
3990         mddev->patch_version = MD_PATCHLEVEL_VERSION;
3991         mddev->ctime         = get_seconds();
3992
3993         mddev->level         = info->level;
3994         mddev->clevel[0]     = 0;
3995         mddev->size          = info->size;
3996         mddev->raid_disks    = info->raid_disks;
3997         /* don't set md_minor, it is determined by which /dev/md* was
3998          * openned
3999          */
4000         if (info->state & (1<<MD_SB_CLEAN))
4001                 mddev->recovery_cp = MaxSector;
4002         else
4003                 mddev->recovery_cp = 0;
4004         mddev->persistent    = ! info->not_persistent;
4005
4006         mddev->layout        = info->layout;
4007         mddev->chunk_size    = info->chunk_size;
4008
4009         mddev->max_disks     = MD_SB_DISKS;
4010
4011         mddev->flags         = 0;
4012         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4013
4014         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4015         mddev->bitmap_offset = 0;
4016
4017         mddev->reshape_position = MaxSector;
4018
4019         /*
4020          * Generate a 128 bit UUID
4021          */
4022         get_random_bytes(mddev->uuid, 16);
4023
4024         mddev->new_level = mddev->level;
4025         mddev->new_chunk = mddev->chunk_size;
4026         mddev->new_layout = mddev->layout;
4027         mddev->delta_disks = 0;
4028
4029         return 0;
4030 }
4031
4032 static int update_size(mddev_t *mddev, unsigned long size)
4033 {
4034         mdk_rdev_t * rdev;
4035         int rv;
4036         struct list_head *tmp;
4037         int fit = (size == 0);
4038
4039         if (mddev->pers->resize == NULL)
4040                 return -EINVAL;
4041         /* The "size" is the amount of each device that is used.
4042          * This can only make sense for arrays with redundancy.
4043          * linear and raid0 always use whatever space is available
4044          * We can only consider changing the size if no resync
4045          * or reconstruction is happening, and if the new size
4046          * is acceptable. It must fit before the sb_offset or,
4047          * if that is <data_offset, it must fit before the
4048          * size of each device.
4049          * If size is zero, we find the largest size that fits.
4050          */
4051         if (mddev->sync_thread)
4052                 return -EBUSY;
4053         ITERATE_RDEV(mddev,rdev,tmp) {
4054                 sector_t avail;
4055                 avail = rdev->size * 2;
4056
4057                 if (fit && (size == 0 || size > avail/2))
4058                         size = avail/2;
4059                 if (avail < ((sector_t)size << 1))
4060                         return -ENOSPC;
4061         }
4062         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4063         if (!rv) {
4064                 struct block_device *bdev;
4065
4066                 bdev = bdget_disk(mddev->gendisk, 0);
4067                 if (bdev) {
4068                         mutex_lock(&bdev->bd_inode->i_mutex);
4069                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4070                         mutex_unlock(&bdev->bd_inode->i_mutex);
4071                         bdput(bdev);
4072                 }
4073         }
4074         return rv;
4075 }
4076
4077 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4078 {
4079         int rv;
4080         /* change the number of raid disks */
4081         if (mddev->pers->check_reshape == NULL)
4082                 return -EINVAL;
4083         if (raid_disks <= 0 ||
4084             raid_disks >= mddev->max_disks)
4085                 return -EINVAL;
4086         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4087                 return -EBUSY;
4088         mddev->delta_disks = raid_disks - mddev->raid_disks;
4089
4090         rv = mddev->pers->check_reshape(mddev);
4091         return rv;
4092 }
4093
4094
4095 /*
4096  * update_array_info is used to change the configuration of an
4097  * on-line array.
4098  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4099  * fields in the info are checked against the array.
4100  * Any differences that cannot be handled will cause an error.
4101  * Normally, only one change can be managed at a time.
4102  */
4103 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4104 {
4105         int rv = 0;
4106         int cnt = 0;
4107         int state = 0;
4108
4109         /* calculate expected state,ignoring low bits */
4110         if (mddev->bitmap && mddev->bitmap_offset)
4111                 state |= (1 << MD_SB_BITMAP_PRESENT);
4112
4113         if (mddev->major_version != info->major_version ||
4114             mddev->minor_version != info->minor_version ||
4115 /*          mddev->patch_version != info->patch_version || */
4116             mddev->ctime         != info->ctime         ||
4117             mddev->level         != info->level         ||
4118 /*          mddev->layout        != info->layout        || */
4119             !mddev->persistent   != info->not_persistent||
4120             mddev->chunk_size    != info->chunk_size    ||
4121             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4122             ((state^info->state) & 0xfffffe00)
4123                 )
4124                 return -EINVAL;
4125         /* Check there is only one change */
4126         if (info->size >= 0 && mddev->size != info->size) cnt++;
4127         if (mddev->raid_disks != info->raid_disks) cnt++;
4128         if (mddev->layout != info->layout) cnt++;
4129         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4130         if (cnt == 0) return 0;
4131         if (cnt > 1) return -EINVAL;
4132
4133         if (mddev->layout != info->layout) {
4134                 /* Change layout
4135                  * we don't need to do anything at the md level, the
4136                  * personality will take care of it all.
4137                  */
4138                 if (mddev->pers->reconfig == NULL)
4139                         return -EINVAL;
4140                 else
4141                         return mddev->pers->reconfig(mddev, info->layout, -1);
4142         }
4143         if (info->size >= 0 && mddev->size != info->size)
4144                 rv = update_size(mddev, info->size);
4145
4146         if (mddev->raid_disks    != info->raid_disks)
4147                 rv = update_raid_disks(mddev, info->raid_disks);
4148
4149         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4150                 if (mddev->pers->quiesce == NULL)
4151                         return -EINVAL;
4152                 if (mddev->recovery || mddev->sync_thread)
4153                         return -EBUSY;
4154                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4155                         /* add the bitmap */
4156                         if (mddev->bitmap)
4157                                 return -EEXIST;
4158                         if (mddev->default_bitmap_offset == 0)
4159                                 return -EINVAL;
4160                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4161                         mddev->pers->quiesce(mddev, 1);
4162                         rv = bitmap_create(mddev);
4163                         if (rv)
4164                                 bitmap_destroy(mddev);
4165                         mddev->pers->quiesce(mddev, 0);
4166                 } else {
4167                         /* remove the bitmap */
4168                         if (!mddev->bitmap)
4169                                 return -ENOENT;
4170                         if (mddev->bitmap->file)
4171                                 return -EINVAL;
4172                         mddev->pers->quiesce(mddev, 1);
4173                         bitmap_destroy(mddev);
4174                         mddev->pers->quiesce(mddev, 0);
4175                         mddev->bitmap_offset = 0;
4176                 }
4177         }
4178         md_update_sb(mddev, 1);
4179         return rv;
4180 }
4181
4182 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4183 {
4184         mdk_rdev_t *rdev;
4185
4186         if (mddev->pers == NULL)
4187                 return -ENODEV;
4188
4189         rdev = find_rdev(mddev, dev);
4190         if (!rdev)
4191                 return -ENODEV;
4192
4193         md_error(mddev, rdev);
4194         return 0;
4195 }
4196
4197 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4198 {
4199         mddev_t *mddev = bdev->bd_disk->private_data;
4200
4201         geo->heads = 2;
4202         geo->sectors = 4;
4203         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4204         return 0;
4205 }
4206
4207 static int md_ioctl(struct inode *inode, struct file *file,
4208                         unsigned int cmd, unsigned long arg)
4209 {
4210         int err = 0;
4211         void __user *argp = (void __user *)arg;
4212         mddev_t *mddev = NULL;
4213
4214         if (!capable(CAP_SYS_ADMIN))
4215                 return -EACCES;
4216
4217         /*
4218          * Commands dealing with the RAID driver but not any
4219          * particular array:
4220          */
4221         switch (cmd)
4222         {
4223                 case RAID_VERSION:
4224                         err = get_version(argp);
4225                         goto done;
4226
4227                 case PRINT_RAID_DEBUG:
4228                         err = 0;
4229                         md_print_devices();
4230                         goto done;
4231
4232 #ifndef MODULE
4233                 case RAID_AUTORUN:
4234                         err = 0;
4235                         autostart_arrays(arg);
4236                         goto done;
4237 #endif
4238                 default:;
4239         }
4240
4241         /*
4242          * Commands creating/starting a new array:
4243          */
4244
4245         mddev = inode->i_bdev->bd_disk->private_data;
4246
4247         if (!mddev) {
4248                 BUG();
4249                 goto abort;
4250         }
4251
4252         err = mddev_lock(mddev);
4253         if (err) {
4254                 printk(KERN_INFO 
4255                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4256                         err, cmd);
4257                 goto abort;
4258         }
4259
4260         switch (cmd)
4261         {
4262                 case SET_ARRAY_INFO:
4263                         {
4264                                 mdu_array_info_t info;
4265                                 if (!arg)
4266                                         memset(&info, 0, sizeof(info));
4267                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4268                                         err = -EFAULT;
4269                                         goto abort_unlock;
4270                                 }
4271                                 if (mddev->pers) {
4272                                         err = update_array_info(mddev, &info);
4273                                         if (err) {
4274                                                 printk(KERN_WARNING "md: couldn't update"
4275                                                        " array info. %d\n", err);
4276                                                 goto abort_unlock;
4277                                         }
4278                                         goto done_unlock;
4279                                 }
4280                                 if (!list_empty(&mddev->disks)) {
4281                                         printk(KERN_WARNING
4282                                                "md: array %s already has disks!\n",
4283                                                mdname(mddev));
4284                                         err = -EBUSY;
4285                                         goto abort_unlock;
4286                                 }
4287                                 if (mddev->raid_disks) {
4288                                         printk(KERN_WARNING
4289                                                "md: array %s already initialised!\n",
4290                                                mdname(mddev));
4291                                         err = -EBUSY;
4292                                         goto abort_unlock;
4293                                 }
4294                                 err = set_array_info(mddev, &info);
4295                                 if (err) {
4296                                         printk(KERN_WARNING "md: couldn't set"
4297                                                " array info. %d\n", err);
4298                                         goto abort_unlock;
4299                                 }
4300                         }
4301                         goto done_unlock;
4302
4303                 default:;
4304         }
4305
4306         /*
4307          * Commands querying/configuring an existing array:
4308          */
4309         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4310          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4311         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4312                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4313                         && cmd != GET_BITMAP_FILE) {
4314                 err = -ENODEV;
4315                 goto abort_unlock;
4316         }
4317
4318         /*
4319          * Commands even a read-only array can execute:
4320          */
4321         switch (cmd)
4322         {
4323                 case GET_ARRAY_INFO:
4324                         err = get_array_info(mddev, argp);
4325                         goto done_unlock;
4326
4327                 case GET_BITMAP_FILE:
4328                         err = get_bitmap_file(mddev, argp);
4329                         goto done_unlock;
4330
4331                 case GET_DISK_INFO:
4332                         err = get_disk_info(mddev, argp);
4333                         goto done_unlock;
4334
4335                 case RESTART_ARRAY_RW:
4336                         err = restart_array(mddev);
4337                         goto done_unlock;
4338
4339                 case STOP_ARRAY:
4340                         err = do_md_stop (mddev, 0);
4341                         goto done_unlock;
4342
4343                 case STOP_ARRAY_RO:
4344                         err = do_md_stop (mddev, 1);
4345                         goto done_unlock;
4346
4347         /*
4348          * We have a problem here : there is no easy way to give a CHS
4349          * virtual geometry. We currently pretend that we have a 2 heads
4350          * 4 sectors (with a BIG number of cylinders...). This drives
4351          * dosfs just mad... ;-)
4352          */
4353         }
4354
4355         /*
4356          * The remaining ioctls are changing the state of the
4357          * superblock, so we do not allow them on read-only arrays.
4358          * However non-MD ioctls (e.g. get-size) will still come through
4359          * here and hit the 'default' below, so only disallow
4360          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4361          */
4362         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4363             mddev->ro && mddev->pers) {
4364                 if (mddev->ro == 2) {
4365                         mddev->ro = 0;
4366                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4367                 md_wakeup_thread(mddev->thread);
4368
4369                 } else {
4370                         err = -EROFS;
4371                         goto abort_unlock;
4372                 }
4373         }
4374
4375         switch (cmd)
4376         {
4377                 case ADD_NEW_DISK:
4378                 {
4379                         mdu_disk_info_t info;
4380                         if (copy_from_user(&info, argp, sizeof(info)))
4381                                 err = -EFAULT;
4382                         else
4383                                 err = add_new_disk(mddev, &info);
4384                         goto done_unlock;
4385                 }
4386
4387                 case HOT_REMOVE_DISK:
4388                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4389                         goto done_unlock;
4390
4391                 case HOT_ADD_DISK:
4392                         err = hot_add_disk(mddev, new_decode_dev(arg));
4393                         goto done_unlock;
4394
4395                 case SET_DISK_FAULTY:
4396                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4397                         goto done_unlock;
4398
4399                 case RUN_ARRAY:
4400                         err = do_md_run (mddev);
4401                         goto done_unlock;
4402
4403                 case SET_BITMAP_FILE:
4404                         err = set_bitmap_file(mddev, (int)arg);
4405                         goto done_unlock;
4406
4407                 default:
4408                         err = -EINVAL;
4409                         goto abort_unlock;
4410         }
4411
4412 done_unlock:
4413 abort_unlock:
4414         mddev_unlock(mddev);
4415
4416         return err;
4417 done:
4418         if (err)
4419                 MD_BUG();
4420 abort:
4421         return err;
4422 }
4423
4424 static int md_open(struct inode *inode, struct file *file)
4425 {
4426         /*
4427          * Succeed if we can lock the mddev, which confirms that
4428          * it isn't being stopped right now.
4429          */
4430         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4431         int err;
4432
4433         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4434                 goto out;
4435
4436         err = 0;
4437         mddev_get(mddev);
4438         mddev_unlock(mddev);
4439
4440         check_disk_change(inode->i_bdev);
4441  out:
4442         return err;
4443 }
4444
4445 static int md_release(struct inode *inode, struct file * file)
4446 {
4447         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4448
4449         BUG_ON(!mddev);
4450         mddev_put(mddev);
4451
4452         return 0;
4453 }
4454
4455 static int md_media_changed(struct gendisk *disk)
4456 {
4457         mddev_t *mddev = disk->private_data;
4458
4459         return mddev->changed;
4460 }
4461
4462 static int md_revalidate(struct gendisk *disk)
4463 {
4464         mddev_t *mddev = disk->private_data;
4465
4466         mddev->changed = 0;
4467         return 0;
4468 }
4469 static struct block_device_operations md_fops =
4470 {
4471         .owner          = THIS_MODULE,
4472         .open           = md_open,
4473         .release        = md_release,
4474         .ioctl          = md_ioctl,
4475         .getgeo         = md_getgeo,
4476         .media_changed  = md_media_changed,
4477         .revalidate_disk= md_revalidate,
4478 };
4479
4480 static int md_thread(void * arg)
4481 {
4482         mdk_thread_t *thread = arg;
4483
4484         /*
4485          * md_thread is a 'system-thread', it's priority should be very
4486          * high. We avoid resource deadlocks individually in each
4487          * raid personality. (RAID5 does preallocation) We also use RR and
4488          * the very same RT priority as kswapd, thus we will never get
4489          * into a priority inversion deadlock.
4490          *
4491          * we definitely have to have equal or higher priority than
4492          * bdflush, otherwise bdflush will deadlock if there are too
4493          * many dirty RAID5 blocks.
4494          */
4495
4496         current->flags |= PF_NOFREEZE;
4497         allow_signal(SIGKILL);
4498         while (!kthread_should_stop()) {
4499
4500                 /* We need to wait INTERRUPTIBLE so that
4501                  * we don't add to the load-average.
4502                  * That means we need to be sure no signals are
4503                  * pending
4504                  */
4505                 if (signal_pending(current))
4506                         flush_signals(current);
4507
4508                 wait_event_interruptible_timeout
4509                         (thread->wqueue,
4510                          test_bit(THREAD_WAKEUP, &thread->flags)
4511                          || kthread_should_stop(),
4512                          thread->timeout);
4513
4514                 clear_bit(THREAD_WAKEUP, &thread->flags);
4515
4516                 thread->run(thread->mddev);
4517         }
4518
4519         return 0;
4520 }
4521
4522 void md_wakeup_thread(mdk_thread_t *thread)
4523 {
4524         if (thread) {
4525                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4526                 set_bit(THREAD_WAKEUP, &thread->flags);
4527                 wake_up(&thread->wqueue);
4528         }
4529 }
4530
4531 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4532                                  const char *name)
4533 {
4534         mdk_thread_t *thread;
4535
4536         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4537         if (!thread)
4538                 return NULL;
4539
4540         init_waitqueue_head(&thread->wqueue);
4541
4542         thread->run = run;
4543         thread->mddev = mddev;
4544         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4545         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4546         if (IS_ERR(thread->tsk)) {
4547                 kfree(thread);
4548                 return NULL;
4549         }
4550         return thread;
4551 }
4552
4553 void md_unregister_thread(mdk_thread_t *thread)
4554 {
4555         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4556
4557         kthread_stop(thread->tsk);
4558         kfree(thread);
4559 }
4560
4561 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4562 {
4563         if (!mddev) {
4564                 MD_BUG();
4565                 return;
4566         }
4567
4568         if (!rdev || test_bit(Faulty, &rdev->flags))
4569                 return;
4570 /*
4571         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4572                 mdname(mddev),
4573                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4574                 __builtin_return_address(0),__builtin_return_address(1),
4575                 __builtin_return_address(2),__builtin_return_address(3));
4576 */
4577         if (!mddev->pers)
4578                 return;
4579         if (!mddev->pers->error_handler)
4580                 return;
4581         mddev->pers->error_handler(mddev,rdev);
4582         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4583         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4584         md_wakeup_thread(mddev->thread);
4585         md_new_event_inintr(mddev);
4586 }
4587
4588 /* seq_file implementation /proc/mdstat */
4589
4590 static void status_unused(struct seq_file *seq)
4591 {
4592         int i = 0;
4593         mdk_rdev_t *rdev;
4594         struct list_head *tmp;
4595
4596         seq_printf(seq, "unused devices: ");
4597
4598         ITERATE_RDEV_PENDING(rdev,tmp) {
4599                 char b[BDEVNAME_SIZE];
4600                 i++;
4601                 seq_printf(seq, "%s ",
4602                               bdevname(rdev->bdev,b));
4603         }
4604         if (!i)
4605                 seq_printf(seq, "<none>");
4606
4607         seq_printf(seq, "\n");
4608 }
4609
4610
4611 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4612 {
4613         sector_t max_blocks, resync, res;
4614         unsigned long dt, db, rt;
4615         int scale;
4616         unsigned int per_milli;
4617
4618         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4619
4620         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4621                 max_blocks = mddev->resync_max_sectors >> 1;
4622         else
4623                 max_blocks = mddev->size;
4624
4625         /*
4626          * Should not happen.
4627          */
4628         if (!max_blocks) {
4629                 MD_BUG();
4630                 return;
4631         }
4632         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4633          * in a sector_t, and (max_blocks>>scale) will fit in a
4634          * u32, as those are the requirements for sector_div.
4635          * Thus 'scale' must be at least 10
4636          */
4637         scale = 10;
4638         if (sizeof(sector_t) > sizeof(unsigned long)) {
4639                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4640                         scale++;
4641         }
4642         res = (resync>>scale)*1000;
4643         sector_div(res, (u32)((max_blocks>>scale)+1));
4644
4645         per_milli = res;
4646         {
4647                 int i, x = per_milli/50, y = 20-x;
4648                 seq_printf(seq, "[");
4649                 for (i = 0; i < x; i++)
4650                         seq_printf(seq, "=");
4651                 seq_printf(seq, ">");
4652                 for (i = 0; i < y; i++)
4653                         seq_printf(seq, ".");
4654                 seq_printf(seq, "] ");
4655         }
4656         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4657                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4658                     "reshape" :
4659                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4660                      "check" :
4661                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4662                       "resync" : "recovery"))),
4663                    per_milli/10, per_milli % 10,
4664                    (unsigned long long) resync,
4665                    (unsigned long long) max_blocks);
4666
4667         /*
4668          * We do not want to overflow, so the order of operands and
4669          * the * 100 / 100 trick are important. We do a +1 to be
4670          * safe against division by zero. We only estimate anyway.
4671          *
4672          * dt: time from mark until now
4673          * db: blocks written from mark until now
4674          * rt: remaining time
4675          */
4676         dt = ((jiffies - mddev->resync_mark) / HZ);
4677         if (!dt) dt++;
4678         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4679                 - mddev->resync_mark_cnt;
4680         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4681
4682         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4683
4684         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4685 }
4686
4687 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4688 {
4689         struct list_head *tmp;
4690         loff_t l = *pos;
4691         mddev_t *mddev;
4692
4693         if (l >= 0x10000)
4694                 return NULL;
4695         if (!l--)
4696                 /* header */
4697                 return (void*)1;
4698
4699         spin_lock(&all_mddevs_lock);
4700         list_for_each(tmp,&all_mddevs)
4701                 if (!l--) {
4702                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4703                         mddev_get(mddev);
4704                         spin_unlock(&all_mddevs_lock);
4705                         return mddev;
4706                 }
4707         spin_unlock(&all_mddevs_lock);
4708         if (!l--)
4709                 return (void*)2;/* tail */
4710         return NULL;
4711 }
4712
4713 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4714 {
4715         struct list_head *tmp;
4716         mddev_t *next_mddev, *mddev = v;
4717         
4718         ++*pos;
4719         if (v == (void*)2)
4720                 return NULL;
4721
4722         spin_lock(&all_mddevs_lock);
4723         if (v == (void*)1)
4724                 tmp = all_mddevs.next;
4725         else
4726                 tmp = mddev->all_mddevs.next;
4727         if (tmp != &all_mddevs)
4728                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4729         else {
4730                 next_mddev = (void*)2;
4731                 *pos = 0x10000;
4732         }               
4733         spin_unlock(&all_mddevs_lock);
4734
4735         if (v != (void*)1)
4736                 mddev_put(mddev);
4737         return next_mddev;
4738
4739 }
4740
4741 static void md_seq_stop(struct seq_file *seq, void *v)
4742 {
4743         mddev_t *mddev = v;
4744
4745         if (mddev && v != (void*)1 && v != (void*)2)
4746                 mddev_put(mddev);
4747 }
4748
4749 struct mdstat_info {
4750         int event;
4751 };
4752
4753 static int md_seq_show(struct seq_file *seq, void *v)
4754 {
4755         mddev_t *mddev = v;
4756         sector_t size;
4757         struct list_head *tmp2;
4758         mdk_rdev_t *rdev;
4759         struct mdstat_info *mi = seq->private;
4760         struct bitmap *bitmap;
4761
4762         if (v == (void*)1) {
4763                 struct mdk_personality *pers;
4764                 seq_printf(seq, "Personalities : ");
4765                 spin_lock(&pers_lock);
4766                 list_for_each_entry(pers, &pers_list, list)
4767                         seq_printf(seq, "[%s] ", pers->name);
4768
4769                 spin_unlock(&pers_lock);
4770                 seq_printf(seq, "\n");
4771                 mi->event = atomic_read(&md_event_count);
4772                 return 0;
4773         }
4774         if (v == (void*)2) {
4775                 status_unused(seq);
4776                 return 0;
4777         }
4778
4779         if (mddev_lock(mddev) < 0)
4780                 return -EINTR;
4781
4782         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4783                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4784                                                 mddev->pers ? "" : "in");
4785                 if (mddev->pers) {
4786                         if (mddev->ro==1)
4787                                 seq_printf(seq, " (read-only)");
4788                         if (mddev->ro==2)
4789                                 seq_printf(seq, "(auto-read-only)");
4790                         seq_printf(seq, " %s", mddev->pers->name);
4791                 }
4792
4793                 size = 0;
4794                 ITERATE_RDEV(mddev,rdev,tmp2) {
4795                         char b[BDEVNAME_SIZE];
4796                         seq_printf(seq, " %s[%d]",
4797                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4798                         if (test_bit(WriteMostly, &rdev->flags))
4799                                 seq_printf(seq, "(W)");
4800                         if (test_bit(Faulty, &rdev->flags)) {
4801                                 seq_printf(seq, "(F)");
4802                                 continue;
4803                         } else if (rdev->raid_disk < 0)
4804                                 seq_printf(seq, "(S)"); /* spare */
4805                         size += rdev->size;
4806                 }
4807
4808                 if (!list_empty(&mddev->disks)) {
4809                         if (mddev->pers)
4810                                 seq_printf(seq, "\n      %llu blocks",
4811                                         (unsigned long long)mddev->array_size);
4812                         else
4813                                 seq_printf(seq, "\n      %llu blocks",
4814                                         (unsigned long long)size);
4815                 }
4816                 if (mddev->persistent) {
4817                         if (mddev->major_version != 0 ||
4818                             mddev->minor_version != 90) {
4819                                 seq_printf(seq," super %d.%d",
4820                                            mddev->major_version,
4821                                            mddev->minor_version);
4822                         }
4823                 } else
4824                         seq_printf(seq, " super non-persistent");
4825
4826                 if (mddev->pers) {
4827                         mddev->pers->status (seq, mddev);
4828                         seq_printf(seq, "\n      ");
4829                         if (mddev->pers->sync_request) {
4830                                 if (mddev->curr_resync > 2) {
4831                                         status_resync (seq, mddev);
4832                                         seq_printf(seq, "\n      ");
4833                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4834                                         seq_printf(seq, "\tresync=DELAYED\n      ");
4835                                 else if (mddev->recovery_cp < MaxSector)
4836                                         seq_printf(seq, "\tresync=PENDING\n      ");
4837                         }
4838                 } else
4839                         seq_printf(seq, "\n       ");
4840
4841                 if ((bitmap = mddev->bitmap)) {
4842                         unsigned long chunk_kb;
4843                         unsigned long flags;
4844                         spin_lock_irqsave(&bitmap->lock, flags);
4845                         chunk_kb = bitmap->chunksize >> 10;
4846                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4847                                 "%lu%s chunk",
4848                                 bitmap->pages - bitmap->missing_pages,
4849                                 bitmap->pages,
4850                                 (bitmap->pages - bitmap->missing_pages)
4851                                         << (PAGE_SHIFT - 10),
4852                                 chunk_kb ? chunk_kb : bitmap->chunksize,
4853                                 chunk_kb ? "KB" : "B");
4854                         if (bitmap->file) {
4855                                 seq_printf(seq, ", file: ");
4856                                 seq_path(seq, bitmap->file->f_path.mnt,
4857                                          bitmap->file->f_path.dentry," \t\n");
4858                         }
4859
4860                         seq_printf(seq, "\n");
4861                         spin_unlock_irqrestore(&bitmap->lock, flags);
4862                 }
4863
4864                 seq_printf(seq, "\n");
4865         }
4866         mddev_unlock(mddev);
4867         
4868         return 0;
4869 }
4870
4871 static struct seq_operations md_seq_ops = {
4872         .start  = md_seq_start,
4873         .next   = md_seq_next,
4874         .stop   = md_seq_stop,
4875         .show   = md_seq_show,
4876 };
4877
4878 static int md_seq_open(struct inode *inode, struct file *file)
4879 {
4880         int error;
4881         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4882         if (mi == NULL)
4883                 return -ENOMEM;
4884
4885         error = seq_open(file, &md_seq_ops);
4886         if (error)
4887                 kfree(mi);
4888         else {
4889                 struct seq_file *p = file->private_data;
4890                 p->private = mi;
4891                 mi->event = atomic_read(&md_event_count);
4892         }
4893         return error;
4894 }
4895
4896 static int md_seq_release(struct inode *inode, struct file *file)
4897 {
4898         struct seq_file *m = file->private_data;
4899         struct mdstat_info *mi = m->private;
4900         m->private = NULL;
4901         kfree(mi);
4902         return seq_release(inode, file);
4903 }
4904
4905 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4906 {
4907         struct seq_file *m = filp->private_data;
4908         struct mdstat_info *mi = m->private;
4909         int mask;
4910
4911         poll_wait(filp, &md_event_waiters, wait);
4912
4913         /* always allow read */
4914         mask = POLLIN | POLLRDNORM;
4915
4916         if (mi->event != atomic_read(&md_event_count))
4917                 mask |= POLLERR | POLLPRI;
4918         return mask;
4919 }
4920
4921 static struct file_operations md_seq_fops = {
4922         .owner          = THIS_MODULE,
4923         .open           = md_seq_open,
4924         .read           = seq_read,
4925         .llseek         = seq_lseek,
4926         .release        = md_seq_release,
4927         .poll           = mdstat_poll,
4928 };
4929
4930 int register_md_personality(struct mdk_personality *p)
4931 {
4932         spin_lock(&pers_lock);
4933         list_add_tail(&p->list, &pers_list);
4934         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4935         spin_unlock(&pers_lock);
4936         return 0;
4937 }
4938
4939 int unregister_md_personality(struct mdk_personality *p)
4940 {
4941         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4942         spin_lock(&pers_lock);
4943         list_del_init(&p->list);
4944         spin_unlock(&pers_lock);
4945         return 0;
4946 }
4947
4948 static int is_mddev_idle(mddev_t *mddev)
4949 {
4950         mdk_rdev_t * rdev;
4951         struct list_head *tmp;
4952         int idle;
4953         unsigned long curr_events;
4954
4955         idle = 1;
4956         ITERATE_RDEV(mddev,rdev,tmp) {
4957                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4958                 curr_events = disk_stat_read(disk, sectors[0]) + 
4959                                 disk_stat_read(disk, sectors[1]) - 
4960                                 atomic_read(&disk->sync_io);
4961                 /* The difference between curr_events and last_events
4962                  * will be affected by any new non-sync IO (making
4963                  * curr_events bigger) and any difference in the amount of
4964                  * in-flight syncio (making current_events bigger or smaller)
4965                  * The amount in-flight is currently limited to
4966                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4967                  * which is at most 4096 sectors.
4968                  * These numbers are fairly fragile and should be made
4969                  * more robust, probably by enforcing the
4970                  * 'window size' that md_do_sync sort-of uses.
4971                  *
4972                  * Note: the following is an unsigned comparison.
4973                  */
4974                 if ((curr_events - rdev->last_events + 4096) > 8192) {
4975                         rdev->last_events = curr_events;
4976                         idle = 0;
4977                 }
4978         }
4979         return idle;
4980 }
4981
4982 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4983 {
4984         /* another "blocks" (512byte) blocks have been synced */
4985         atomic_sub(blocks, &mddev->recovery_active);
4986         wake_up(&mddev->recovery_wait);
4987         if (!ok) {
4988                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4989                 md_wakeup_thread(mddev->thread);
4990                 // stop recovery, signal do_sync ....
4991         }
4992 }
4993
4994
4995 /* md_write_start(mddev, bi)
4996  * If we need to update some array metadata (e.g. 'active' flag
4997  * in superblock) before writing, schedule a superblock update
4998  * and wait for it to complete.
4999  */
5000 void md_write_start(mddev_t *mddev, struct bio *bi)
5001 {
5002         if (bio_data_dir(bi) != WRITE)
5003                 return;
5004
5005         BUG_ON(mddev->ro == 1);
5006         if (mddev->ro == 2) {
5007                 /* need to switch to read/write */
5008                 mddev->ro = 0;
5009                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5010                 md_wakeup_thread(mddev->thread);
5011         }
5012         atomic_inc(&mddev->writes_pending);
5013         if (mddev->in_sync) {
5014                 spin_lock_irq(&mddev->write_lock);
5015                 if (mddev->in_sync) {
5016                         mddev->in_sync = 0;
5017                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5018                         md_wakeup_thread(mddev->thread);
5019                 }
5020                 spin_unlock_irq(&mddev->write_lock);
5021         }
5022         wait_event(mddev->sb_wait, mddev->flags==0);
5023 }
5024
5025 void md_write_end(mddev_t *mddev)
5026 {
5027         if (atomic_dec_and_test(&mddev->writes_pending)) {
5028                 if (mddev->safemode == 2)
5029                         md_wakeup_thread(mddev->thread);
5030                 else if (mddev->safemode_delay)
5031                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5032         }
5033 }
5034
5035 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5036
5037 #define SYNC_MARKS      10
5038 #define SYNC_MARK_STEP  (3*HZ)
5039 void md_do_sync(mddev_t *mddev)
5040 {
5041         mddev_t *mddev2;
5042         unsigned int currspeed = 0,
5043                  window;
5044         sector_t max_sectors,j, io_sectors;
5045         unsigned long mark[SYNC_MARKS];
5046         sector_t mark_cnt[SYNC_MARKS];
5047         int last_mark,m;
5048         struct list_head *tmp;
5049         sector_t last_check;
5050         int skipped = 0;
5051         struct list_head *rtmp;
5052         mdk_rdev_t *rdev;
5053         char *desc;
5054
5055         /* just incase thread restarts... */
5056         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5057                 return;
5058         if (mddev->ro) /* never try to sync a read-only array */
5059                 return;
5060
5061         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5062                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5063                         desc = "data-check";
5064                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5065                         desc = "requested-resync";
5066                 else
5067                         desc = "resync";
5068         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5069                 desc = "reshape";
5070         else
5071                 desc = "recovery";
5072
5073         /* we overload curr_resync somewhat here.
5074          * 0 == not engaged in resync at all
5075          * 2 == checking that there is no conflict with another sync
5076          * 1 == like 2, but have yielded to allow conflicting resync to
5077          *              commense
5078          * other == active in resync - this many blocks
5079          *
5080          * Before starting a resync we must have set curr_resync to
5081          * 2, and then checked that every "conflicting" array has curr_resync
5082          * less than ours.  When we find one that is the same or higher
5083          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5084          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5085          * This will mean we have to start checking from the beginning again.
5086          *
5087          */
5088
5089         do {
5090                 mddev->curr_resync = 2;
5091
5092         try_again:
5093                 if (kthread_should_stop()) {
5094                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5095                         goto skip;
5096                 }
5097                 ITERATE_MDDEV(mddev2,tmp) {
5098                         if (mddev2 == mddev)
5099                                 continue;
5100                         if (mddev2->curr_resync && 
5101                             match_mddev_units(mddev,mddev2)) {
5102                                 DEFINE_WAIT(wq);
5103                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5104                                         /* arbitrarily yield */
5105                                         mddev->curr_resync = 1;
5106                                         wake_up(&resync_wait);
5107                                 }
5108                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5109                                         /* no need to wait here, we can wait the next
5110                                          * time 'round when curr_resync == 2
5111                                          */
5112                                         continue;
5113                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5114                                 if (!kthread_should_stop() &&
5115                                     mddev2->curr_resync >= mddev->curr_resync) {
5116                                         printk(KERN_INFO "md: delaying %s of %s"
5117                                                " until %s has finished (they"
5118                                                " share one or more physical units)\n",
5119                                                desc, mdname(mddev), mdname(mddev2));
5120                                         mddev_put(mddev2);
5121                                         schedule();
5122                                         finish_wait(&resync_wait, &wq);
5123                                         goto try_again;
5124                                 }
5125                                 finish_wait(&resync_wait, &wq);
5126                         }
5127                 }
5128         } while (mddev->curr_resync < 2);
5129
5130         j = 0;
5131         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5132                 /* resync follows the size requested by the personality,
5133                  * which defaults to physical size, but can be virtual size
5134                  */
5135                 max_sectors = mddev->resync_max_sectors;
5136                 mddev->resync_mismatches = 0;
5137                 /* we don't use the checkpoint if there's a bitmap */
5138                 if (!mddev->bitmap &&
5139                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5140                         j = mddev->recovery_cp;
5141         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5142                 max_sectors = mddev->size << 1;
5143         else {
5144                 /* recovery follows the physical size of devices */
5145                 max_sectors = mddev->size << 1;
5146                 j = MaxSector;
5147                 ITERATE_RDEV(mddev,rdev,rtmp)
5148                         if (rdev->raid_disk >= 0 &&
5149                             !test_bit(Faulty, &rdev->flags) &&
5150                             !test_bit(In_sync, &rdev->flags) &&
5151                             rdev->recovery_offset < j)
5152                                 j = rdev->recovery_offset;
5153         }
5154
5155         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5156         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5157                 " %d KB/sec/disk.\n", speed_min(mddev));
5158         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5159                "(but not more than %d KB/sec) for %s.\n",
5160                speed_max(mddev), desc);
5161
5162         is_mddev_idle(mddev); /* this also initializes IO event counters */
5163
5164         io_sectors = 0;
5165         for (m = 0; m < SYNC_MARKS; m++) {
5166                 mark[m] = jiffies;
5167                 mark_cnt[m] = io_sectors;
5168         }
5169         last_mark = 0;
5170         mddev->resync_mark = mark[last_mark];
5171         mddev->resync_mark_cnt = mark_cnt[last_mark];
5172
5173         /*
5174          * Tune reconstruction:
5175          */
5176         window = 32*(PAGE_SIZE/512);
5177         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5178                 window/2,(unsigned long long) max_sectors/2);
5179
5180         atomic_set(&mddev->recovery_active, 0);
5181         init_waitqueue_head(&mddev->recovery_wait);
5182         last_check = 0;
5183
5184         if (j>2) {
5185                 printk(KERN_INFO 
5186                        "md: resuming %s of %s from checkpoint.\n",
5187                        desc, mdname(mddev));
5188                 mddev->curr_resync = j;
5189         }
5190
5191         while (j < max_sectors) {
5192                 sector_t sectors;
5193
5194                 skipped = 0;
5195                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5196                                             currspeed < speed_min(mddev));
5197                 if (sectors == 0) {
5198                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5199                         goto out;
5200                 }
5201
5202                 if (!skipped) { /* actual IO requested */
5203                         io_sectors += sectors;
5204                         atomic_add(sectors, &mddev->recovery_active);
5205                 }
5206
5207                 j += sectors;
5208                 if (j>1) mddev->curr_resync = j;
5209                 mddev->curr_mark_cnt = io_sectors;
5210                 if (last_check == 0)
5211                         /* this is the earliers that rebuilt will be
5212                          * visible in /proc/mdstat
5213                          */
5214                         md_new_event(mddev);
5215
5216                 if (last_check + window > io_sectors || j == max_sectors)
5217                         continue;
5218
5219                 last_check = io_sectors;
5220
5221                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5222                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5223                         break;
5224
5225         repeat:
5226                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5227                         /* step marks */
5228                         int next = (last_mark+1) % SYNC_MARKS;
5229
5230                         mddev->resync_mark = mark[next];
5231                         mddev->resync_mark_cnt = mark_cnt[next];
5232                         mark[next] = jiffies;
5233                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5234                         last_mark = next;
5235                 }
5236
5237
5238                 if (kthread_should_stop()) {
5239                         /*
5240                          * got a signal, exit.
5241                          */
5242                         printk(KERN_INFO 
5243                                 "md: md_do_sync() got signal ... exiting\n");
5244                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5245                         goto out;
5246                 }
5247
5248                 /*
5249                  * this loop exits only if either when we are slower than
5250                  * the 'hard' speed limit, or the system was IO-idle for
5251                  * a jiffy.
5252                  * the system might be non-idle CPU-wise, but we only care
5253                  * about not overloading the IO subsystem. (things like an
5254                  * e2fsck being done on the RAID array should execute fast)
5255                  */
5256                 mddev->queue->unplug_fn(mddev->queue);
5257                 cond_resched();
5258
5259                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5260                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5261
5262                 if (currspeed > speed_min(mddev)) {
5263                         if ((currspeed > speed_max(mddev)) ||
5264                                         !is_mddev_idle(mddev)) {
5265                                 msleep(500);
5266                                 goto repeat;
5267                         }
5268                 }
5269         }
5270         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5271         /*
5272          * this also signals 'finished resyncing' to md_stop
5273          */
5274  out:
5275         mddev->queue->unplug_fn(mddev->queue);
5276
5277         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5278
5279         /* tell personality that we are finished */
5280         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5281
5282         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5283             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5284             mddev->curr_resync > 2) {
5285                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5286                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5287                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5288                                         printk(KERN_INFO
5289                                                "md: checkpointing %s of %s.\n",
5290                                                desc, mdname(mddev));
5291                                         mddev->recovery_cp = mddev->curr_resync;
5292                                 }
5293                         } else
5294                                 mddev->recovery_cp = MaxSector;
5295                 } else {
5296                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5297                                 mddev->curr_resync = MaxSector;
5298                         ITERATE_RDEV(mddev,rdev,rtmp)
5299                                 if (rdev->raid_disk >= 0 &&
5300                                     !test_bit(Faulty, &rdev->flags) &&
5301                                     !test_bit(In_sync, &rdev->flags) &&
5302                                     rdev->recovery_offset < mddev->curr_resync)
5303                                         rdev->recovery_offset = mddev->curr_resync;
5304                 }
5305         }
5306         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5307
5308  skip:
5309         mddev->curr_resync = 0;
5310         wake_up(&resync_wait);
5311         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5312         md_wakeup_thread(mddev->thread);
5313 }
5314 EXPORT_SYMBOL_GPL(md_do_sync);
5315
5316
5317 /*
5318  * This routine is regularly called by all per-raid-array threads to
5319  * deal with generic issues like resync and super-block update.
5320  * Raid personalities that don't have a thread (linear/raid0) do not
5321  * need this as they never do any recovery or update the superblock.
5322  *
5323  * It does not do any resync itself, but rather "forks" off other threads
5324  * to do that as needed.
5325  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5326  * "->recovery" and create a thread at ->sync_thread.
5327  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5328  * and wakeups up this thread which will reap the thread and finish up.
5329  * This thread also removes any faulty devices (with nr_pending == 0).
5330  *
5331  * The overall approach is:
5332  *  1/ if the superblock needs updating, update it.
5333  *  2/ If a recovery thread is running, don't do anything else.
5334  *  3/ If recovery has finished, clean up, possibly marking spares active.
5335  *  4/ If there are any faulty devices, remove them.
5336  *  5/ If array is degraded, try to add spares devices
5337  *  6/ If array has spares or is not in-sync, start a resync thread.
5338  */
5339 void md_check_recovery(mddev_t *mddev)
5340 {
5341         mdk_rdev_t *rdev;
5342         struct list_head *rtmp;
5343
5344
5345         if (mddev->bitmap)
5346                 bitmap_daemon_work(mddev->bitmap);
5347
5348         if (mddev->ro)
5349                 return;
5350
5351         if (signal_pending(current)) {
5352                 if (mddev->pers->sync_request) {
5353                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5354                                mdname(mddev));
5355                         mddev->safemode = 2;
5356                 }
5357                 flush_signals(current);
5358         }
5359
5360         if ( ! (
5361                 mddev->flags ||
5362                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5363                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5364                 (mddev->safemode == 1) ||
5365                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5366                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5367                 ))
5368                 return;
5369
5370         if (mddev_trylock(mddev)) {
5371                 int spares =0;
5372
5373                 spin_lock_irq(&mddev->write_lock);
5374                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5375                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5376                         mddev->in_sync = 1;
5377                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5378                 }
5379                 if (mddev->safemode == 1)
5380                         mddev->safemode = 0;
5381                 spin_unlock_irq(&mddev->write_lock);
5382
5383                 if (mddev->flags)
5384                         md_update_sb(mddev, 0);
5385
5386
5387                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5388                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5389                         /* resync/recovery still happening */
5390                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5391                         goto unlock;
5392                 }
5393                 if (mddev->sync_thread) {
5394                         /* resync has finished, collect result */
5395                         md_unregister_thread(mddev->sync_thread);
5396                         mddev->sync_thread = NULL;
5397                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5398                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5399                                 /* success...*/
5400                                 /* activate any spares */
5401                                 mddev->pers->spare_active(mddev);
5402                         }
5403                         md_update_sb(mddev, 1);
5404
5405                         /* if array is no-longer degraded, then any saved_raid_disk
5406                          * information must be scrapped
5407                          */
5408                         if (!mddev->degraded)
5409                                 ITERATE_RDEV(mddev,rdev,rtmp)
5410                                         rdev->saved_raid_disk = -1;
5411
5412                         mddev->recovery = 0;
5413                         /* flag recovery needed just to double check */
5414                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5415                         md_new_event(mddev);
5416                         goto unlock;
5417                 }
5418                 /* Clear some bits that don't mean anything, but
5419                  * might be left set
5420                  */
5421                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5422                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5423                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5424                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5425
5426                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5427                         goto unlock;
5428                 /* no recovery is running.
5429                  * remove any failed drives, then
5430                  * add spares if possible.
5431                  * Spare are also removed and re-added, to allow
5432                  * the personality to fail the re-add.
5433                  */
5434                 ITERATE_RDEV(mddev,rdev,rtmp)
5435                         if (rdev->raid_disk >= 0 &&
5436                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
5437                             atomic_read(&rdev->nr_pending)==0) {
5438                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
5439                                         char nm[20];
5440                                         sprintf(nm,"rd%d", rdev->raid_disk);
5441                                         sysfs_remove_link(&mddev->kobj, nm);
5442                                         rdev->raid_disk = -1;
5443                                 }
5444                         }
5445
5446                 if (mddev->degraded) {
5447                         ITERATE_RDEV(mddev,rdev,rtmp)
5448                                 if (rdev->raid_disk < 0
5449                                     && !test_bit(Faulty, &rdev->flags)) {
5450                                         rdev->recovery_offset = 0;
5451                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
5452                                                 char nm[20];
5453                                                 sprintf(nm, "rd%d", rdev->raid_disk);
5454                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
5455                                                 spares++;
5456                                                 md_new_event(mddev);
5457                                         } else
5458                                                 break;
5459                                 }
5460                 }
5461
5462                 if (spares) {
5463                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5464                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5465                 } else if (mddev->recovery_cp < MaxSector) {
5466                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5467                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5468                         /* nothing to be done ... */
5469                         goto unlock;
5470
5471                 if (mddev->pers->sync_request) {
5472                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5473                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5474                                 /* We are adding a device or devices to an array
5475                                  * which has the bitmap stored on all devices.
5476                                  * So make sure all bitmap pages get written
5477                                  */
5478                                 bitmap_write_all(mddev->bitmap);
5479                         }
5480                         mddev->sync_thread = md_register_thread(md_do_sync,
5481                                                                 mddev,
5482                                                                 "%s_resync");
5483                         if (!mddev->sync_thread) {
5484                                 printk(KERN_ERR "%s: could not start resync"
5485                                         " thread...\n", 
5486                                         mdname(mddev));
5487                                 /* leave the spares where they are, it shouldn't hurt */
5488                                 mddev->recovery = 0;
5489                         } else
5490                                 md_wakeup_thread(mddev->sync_thread);
5491                         md_new_event(mddev);
5492                 }
5493         unlock:
5494                 mddev_unlock(mddev);
5495         }
5496 }
5497
5498 static int md_notify_reboot(struct notifier_block *this,
5499                             unsigned long code, void *x)
5500 {
5501         struct list_head *tmp;
5502         mddev_t *mddev;
5503
5504         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5505
5506                 printk(KERN_INFO "md: stopping all md devices.\n");
5507
5508                 ITERATE_MDDEV(mddev,tmp)
5509                         if (mddev_trylock(mddev)) {
5510                                 do_md_stop (mddev, 1);
5511                                 mddev_unlock(mddev);
5512                         }
5513                 /*
5514                  * certain more exotic SCSI devices are known to be
5515                  * volatile wrt too early system reboots. While the
5516                  * right place to handle this issue is the given
5517                  * driver, we do want to have a safe RAID driver ...
5518                  */
5519                 mdelay(1000*1);
5520         }
5521         return NOTIFY_DONE;
5522 }
5523
5524 static struct notifier_block md_notifier = {
5525         .notifier_call  = md_notify_reboot,
5526         .next           = NULL,
5527         .priority       = INT_MAX, /* before any real devices */
5528 };
5529
5530 static void md_geninit(void)
5531 {
5532         struct proc_dir_entry *p;
5533
5534         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5535
5536         p = create_proc_entry("mdstat", S_IRUGO, NULL);
5537         if (p)
5538                 p->proc_fops = &md_seq_fops;
5539 }
5540
5541 static int __init md_init(void)
5542 {
5543         if (register_blkdev(MAJOR_NR, "md"))
5544                 return -1;
5545         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5546                 unregister_blkdev(MAJOR_NR, "md");
5547                 return -1;
5548         }
5549         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5550                             md_probe, NULL, NULL);
5551         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5552                             md_probe, NULL, NULL);
5553
5554         register_reboot_notifier(&md_notifier);
5555         raid_table_header = register_sysctl_table(raid_root_table, 1);
5556
5557         md_geninit();
5558         return (0);
5559 }
5560
5561
5562 #ifndef MODULE
5563
5564 /*
5565  * Searches all registered partitions for autorun RAID arrays
5566  * at boot time.
5567  */
5568 static dev_t detected_devices[128];
5569 static int dev_cnt;
5570
5571 void md_autodetect_dev(dev_t dev)
5572 {
5573         if (dev_cnt >= 0 && dev_cnt < 127)
5574                 detected_devices[dev_cnt++] = dev;
5575 }
5576
5577
5578 static void autostart_arrays(int part)
5579 {
5580         mdk_rdev_t *rdev;
5581         int i;
5582
5583         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5584
5585         for (i = 0; i < dev_cnt; i++) {
5586                 dev_t dev = detected_devices[i];
5587
5588                 rdev = md_import_device(dev,0, 0);
5589                 if (IS_ERR(rdev))
5590                         continue;
5591
5592                 if (test_bit(Faulty, &rdev->flags)) {
5593                         MD_BUG();
5594                         continue;
5595                 }
5596                 list_add(&rdev->same_set, &pending_raid_disks);
5597         }
5598         dev_cnt = 0;
5599
5600         autorun_devices(part);
5601 }
5602
5603 #endif /* !MODULE */
5604
5605 static __exit void md_exit(void)
5606 {
5607         mddev_t *mddev;
5608         struct list_head *tmp;
5609
5610         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5611         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5612
5613         unregister_blkdev(MAJOR_NR,"md");
5614         unregister_blkdev(mdp_major, "mdp");
5615         unregister_reboot_notifier(&md_notifier);
5616         unregister_sysctl_table(raid_table_header);
5617         remove_proc_entry("mdstat", NULL);
5618         ITERATE_MDDEV(mddev,tmp) {
5619                 struct gendisk *disk = mddev->gendisk;
5620                 if (!disk)
5621                         continue;
5622                 export_array(mddev);
5623                 del_gendisk(disk);
5624                 put_disk(disk);
5625                 mddev->gendisk = NULL;
5626                 mddev_put(mddev);
5627         }
5628 }
5629
5630 module_init(md_init)
5631 module_exit(md_exit)
5632
5633 static int get_ro(char *buffer, struct kernel_param *kp)
5634 {
5635         return sprintf(buffer, "%d", start_readonly);
5636 }
5637 static int set_ro(const char *val, struct kernel_param *kp)
5638 {
5639         char *e;
5640         int num = simple_strtoul(val, &e, 10);
5641         if (*val && (*e == '\0' || *e == '\n')) {
5642                 start_readonly = num;
5643                 return 0;
5644         }
5645         return -EINVAL;
5646 }
5647
5648 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5649 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5650
5651
5652 EXPORT_SYMBOL(register_md_personality);
5653 EXPORT_SYMBOL(unregister_md_personality);
5654 EXPORT_SYMBOL(md_error);
5655 EXPORT_SYMBOL(md_done_sync);
5656 EXPORT_SYMBOL(md_write_start);
5657 EXPORT_SYMBOL(md_write_end);
5658 EXPORT_SYMBOL(md_register_thread);
5659 EXPORT_SYMBOL(md_unregister_thread);
5660 EXPORT_SYMBOL(md_wakeup_thread);
5661 EXPORT_SYMBOL(md_check_recovery);
5662 MODULE_LICENSE("GPL");
5663 MODULE_ALIAS("md");
5664 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);