[PATCH] md: split off some md attributes in sysfs to a separate group
[linux-2.6] / drivers / md / md.c
1 /*
2    md.c : Multiple Devices driver for Linux
3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5      completely rewritten, based on the MD driver code from Marc Zyngier
6
7    Changes:
8
9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13    - kmod support by: Cyrus Durgin
14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19
20      Neil Brown <neilb@cse.unsw.edu.au>.
21
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25    This program is free software; you can redistribute it and/or modify
26    it under the terms of the GNU General Public License as published by
27    the Free Software Foundation; either version 2, or (at your option)
28    any later version.
29
30    You should have received a copy of the GNU General Public License
31    (for example /usr/src/linux/COPYING); if not, write to the Free
32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/module.h>
36 #include <linux/config.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/devfs_fs_kernel.h>
43 #include <linux/buffer_head.h> /* for invalidate_bdev */
44 #include <linux/suspend.h>
45
46 #include <linux/init.h>
47
48 #include <linux/file.h>
49
50 #ifdef CONFIG_KMOD
51 #include <linux/kmod.h>
52 #endif
53
54 #include <asm/unaligned.h>
55
56 #define MAJOR_NR MD_MAJOR
57 #define MD_DRIVER
58
59 /* 63 partitions with the alternate major number (mdp) */
60 #define MdpMinorShift 6
61
62 #define DEBUG 0
63 #define dprintk(x...) ((void)(DEBUG && printk(x)))
64
65
66 #ifndef MODULE
67 static void autostart_arrays (int part);
68 #endif
69
70 static mdk_personality_t *pers[MAX_PERSONALITY];
71 static DEFINE_SPINLOCK(pers_lock);
72
73 /*
74  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
75  * is 1000 KB/sec, so the extra system load does not show up that much.
76  * Increase it if you want to have more _guaranteed_ speed. Note that
77  * the RAID driver will use the maximum available bandwidth if the IO
78  * subsystem is idle. There is also an 'absolute maximum' reconstruction
79  * speed limit - in case reconstruction slows down your system despite
80  * idle IO detection.
81  *
82  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
83  */
84
85 static int sysctl_speed_limit_min = 1000;
86 static int sysctl_speed_limit_max = 200000;
87
88 static struct ctl_table_header *raid_table_header;
89
90 static ctl_table raid_table[] = {
91         {
92                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
93                 .procname       = "speed_limit_min",
94                 .data           = &sysctl_speed_limit_min,
95                 .maxlen         = sizeof(int),
96                 .mode           = 0644,
97                 .proc_handler   = &proc_dointvec,
98         },
99         {
100                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
101                 .procname       = "speed_limit_max",
102                 .data           = &sysctl_speed_limit_max,
103                 .maxlen         = sizeof(int),
104                 .mode           = 0644,
105                 .proc_handler   = &proc_dointvec,
106         },
107         { .ctl_name = 0 }
108 };
109
110 static ctl_table raid_dir_table[] = {
111         {
112                 .ctl_name       = DEV_RAID,
113                 .procname       = "raid",
114                 .maxlen         = 0,
115                 .mode           = 0555,
116                 .child          = raid_table,
117         },
118         { .ctl_name = 0 }
119 };
120
121 static ctl_table raid_root_table[] = {
122         {
123                 .ctl_name       = CTL_DEV,
124                 .procname       = "dev",
125                 .maxlen         = 0,
126                 .mode           = 0555,
127                 .child          = raid_dir_table,
128         },
129         { .ctl_name = 0 }
130 };
131
132 static struct block_device_operations md_fops;
133
134 static int start_readonly;
135
136 /*
137  * Enables to iterate over all existing md arrays
138  * all_mddevs_lock protects this list.
139  */
140 static LIST_HEAD(all_mddevs);
141 static DEFINE_SPINLOCK(all_mddevs_lock);
142
143
144 /*
145  * iterates through all used mddevs in the system.
146  * We take care to grab the all_mddevs_lock whenever navigating
147  * the list, and to always hold a refcount when unlocked.
148  * Any code which breaks out of this loop while own
149  * a reference to the current mddev and must mddev_put it.
150  */
151 #define ITERATE_MDDEV(mddev,tmp)                                        \
152                                                                         \
153         for (({ spin_lock(&all_mddevs_lock);                            \
154                 tmp = all_mddevs.next;                                  \
155                 mddev = NULL;});                                        \
156              ({ if (tmp != &all_mddevs)                                 \
157                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
158                 spin_unlock(&all_mddevs_lock);                          \
159                 if (mddev) mddev_put(mddev);                            \
160                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
161                 tmp != &all_mddevs;});                                  \
162              ({ spin_lock(&all_mddevs_lock);                            \
163                 tmp = tmp->next;})                                      \
164                 )
165
166
167 static int md_fail_request (request_queue_t *q, struct bio *bio)
168 {
169         bio_io_error(bio, bio->bi_size);
170         return 0;
171 }
172
173 static inline mddev_t *mddev_get(mddev_t *mddev)
174 {
175         atomic_inc(&mddev->active);
176         return mddev;
177 }
178
179 static void mddev_put(mddev_t *mddev)
180 {
181         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
182                 return;
183         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
184                 list_del(&mddev->all_mddevs);
185                 blk_put_queue(mddev->queue);
186                 kobject_unregister(&mddev->kobj);
187         }
188         spin_unlock(&all_mddevs_lock);
189 }
190
191 static mddev_t * mddev_find(dev_t unit)
192 {
193         mddev_t *mddev, *new = NULL;
194
195  retry:
196         spin_lock(&all_mddevs_lock);
197         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
198                 if (mddev->unit == unit) {
199                         mddev_get(mddev);
200                         spin_unlock(&all_mddevs_lock);
201                         kfree(new);
202                         return mddev;
203                 }
204
205         if (new) {
206                 list_add(&new->all_mddevs, &all_mddevs);
207                 spin_unlock(&all_mddevs_lock);
208                 return new;
209         }
210         spin_unlock(&all_mddevs_lock);
211
212         new = (mddev_t *) kmalloc(sizeof(*new), GFP_KERNEL);
213         if (!new)
214                 return NULL;
215
216         memset(new, 0, sizeof(*new));
217
218         new->unit = unit;
219         if (MAJOR(unit) == MD_MAJOR)
220                 new->md_minor = MINOR(unit);
221         else
222                 new->md_minor = MINOR(unit) >> MdpMinorShift;
223
224         init_MUTEX(&new->reconfig_sem);
225         INIT_LIST_HEAD(&new->disks);
226         INIT_LIST_HEAD(&new->all_mddevs);
227         init_timer(&new->safemode_timer);
228         atomic_set(&new->active, 1);
229         spin_lock_init(&new->write_lock);
230         init_waitqueue_head(&new->sb_wait);
231
232         new->queue = blk_alloc_queue(GFP_KERNEL);
233         if (!new->queue) {
234                 kfree(new);
235                 return NULL;
236         }
237
238         blk_queue_make_request(new->queue, md_fail_request);
239
240         goto retry;
241 }
242
243 static inline int mddev_lock(mddev_t * mddev)
244 {
245         return down_interruptible(&mddev->reconfig_sem);
246 }
247
248 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
249 {
250         down(&mddev->reconfig_sem);
251 }
252
253 static inline int mddev_trylock(mddev_t * mddev)
254 {
255         return down_trylock(&mddev->reconfig_sem);
256 }
257
258 static inline void mddev_unlock(mddev_t * mddev)
259 {
260         up(&mddev->reconfig_sem);
261
262         md_wakeup_thread(mddev->thread);
263 }
264
265 mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
266 {
267         mdk_rdev_t * rdev;
268         struct list_head *tmp;
269
270         ITERATE_RDEV(mddev,rdev,tmp) {
271                 if (rdev->desc_nr == nr)
272                         return rdev;
273         }
274         return NULL;
275 }
276
277 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
278 {
279         struct list_head *tmp;
280         mdk_rdev_t *rdev;
281
282         ITERATE_RDEV(mddev,rdev,tmp) {
283                 if (rdev->bdev->bd_dev == dev)
284                         return rdev;
285         }
286         return NULL;
287 }
288
289 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
290 {
291         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
292         return MD_NEW_SIZE_BLOCKS(size);
293 }
294
295 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
296 {
297         sector_t size;
298
299         size = rdev->sb_offset;
300
301         if (chunk_size)
302                 size &= ~((sector_t)chunk_size/1024 - 1);
303         return size;
304 }
305
306 static int alloc_disk_sb(mdk_rdev_t * rdev)
307 {
308         if (rdev->sb_page)
309                 MD_BUG();
310
311         rdev->sb_page = alloc_page(GFP_KERNEL);
312         if (!rdev->sb_page) {
313                 printk(KERN_ALERT "md: out of memory.\n");
314                 return -EINVAL;
315         }
316
317         return 0;
318 }
319
320 static void free_disk_sb(mdk_rdev_t * rdev)
321 {
322         if (rdev->sb_page) {
323                 page_cache_release(rdev->sb_page);
324                 rdev->sb_loaded = 0;
325                 rdev->sb_page = NULL;
326                 rdev->sb_offset = 0;
327                 rdev->size = 0;
328         }
329 }
330
331
332 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
333 {
334         mdk_rdev_t *rdev = bio->bi_private;
335         mddev_t *mddev = rdev->mddev;
336         if (bio->bi_size)
337                 return 1;
338
339         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
340                 md_error(mddev, rdev);
341
342         if (atomic_dec_and_test(&mddev->pending_writes))
343                 wake_up(&mddev->sb_wait);
344         bio_put(bio);
345         return 0;
346 }
347
348 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
349 {
350         struct bio *bio2 = bio->bi_private;
351         mdk_rdev_t *rdev = bio2->bi_private;
352         mddev_t *mddev = rdev->mddev;
353         if (bio->bi_size)
354                 return 1;
355
356         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
357             error == -EOPNOTSUPP) {
358                 unsigned long flags;
359                 /* barriers don't appear to be supported :-( */
360                 set_bit(BarriersNotsupp, &rdev->flags);
361                 mddev->barriers_work = 0;
362                 spin_lock_irqsave(&mddev->write_lock, flags);
363                 bio2->bi_next = mddev->biolist;
364                 mddev->biolist = bio2;
365                 spin_unlock_irqrestore(&mddev->write_lock, flags);
366                 wake_up(&mddev->sb_wait);
367                 bio_put(bio);
368                 return 0;
369         }
370         bio_put(bio2);
371         bio->bi_private = rdev;
372         return super_written(bio, bytes_done, error);
373 }
374
375 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
376                    sector_t sector, int size, struct page *page)
377 {
378         /* write first size bytes of page to sector of rdev
379          * Increment mddev->pending_writes before returning
380          * and decrement it on completion, waking up sb_wait
381          * if zero is reached.
382          * If an error occurred, call md_error
383          *
384          * As we might need to resubmit the request if BIO_RW_BARRIER
385          * causes ENOTSUPP, we allocate a spare bio...
386          */
387         struct bio *bio = bio_alloc(GFP_NOIO, 1);
388         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
389
390         bio->bi_bdev = rdev->bdev;
391         bio->bi_sector = sector;
392         bio_add_page(bio, page, size, 0);
393         bio->bi_private = rdev;
394         bio->bi_end_io = super_written;
395         bio->bi_rw = rw;
396
397         atomic_inc(&mddev->pending_writes);
398         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
399                 struct bio *rbio;
400                 rw |= (1<<BIO_RW_BARRIER);
401                 rbio = bio_clone(bio, GFP_NOIO);
402                 rbio->bi_private = bio;
403                 rbio->bi_end_io = super_written_barrier;
404                 submit_bio(rw, rbio);
405         } else
406                 submit_bio(rw, bio);
407 }
408
409 void md_super_wait(mddev_t *mddev)
410 {
411         /* wait for all superblock writes that were scheduled to complete.
412          * if any had to be retried (due to BARRIER problems), retry them
413          */
414         DEFINE_WAIT(wq);
415         for(;;) {
416                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
417                 if (atomic_read(&mddev->pending_writes)==0)
418                         break;
419                 while (mddev->biolist) {
420                         struct bio *bio;
421                         spin_lock_irq(&mddev->write_lock);
422                         bio = mddev->biolist;
423                         mddev->biolist = bio->bi_next ;
424                         bio->bi_next = NULL;
425                         spin_unlock_irq(&mddev->write_lock);
426                         submit_bio(bio->bi_rw, bio);
427                 }
428                 schedule();
429         }
430         finish_wait(&mddev->sb_wait, &wq);
431 }
432
433 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
434 {
435         if (bio->bi_size)
436                 return 1;
437
438         complete((struct completion*)bio->bi_private);
439         return 0;
440 }
441
442 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
443                    struct page *page, int rw)
444 {
445         struct bio *bio = bio_alloc(GFP_NOIO, 1);
446         struct completion event;
447         int ret;
448
449         rw |= (1 << BIO_RW_SYNC);
450
451         bio->bi_bdev = bdev;
452         bio->bi_sector = sector;
453         bio_add_page(bio, page, size, 0);
454         init_completion(&event);
455         bio->bi_private = &event;
456         bio->bi_end_io = bi_complete;
457         submit_bio(rw, bio);
458         wait_for_completion(&event);
459
460         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
461         bio_put(bio);
462         return ret;
463 }
464
465 static int read_disk_sb(mdk_rdev_t * rdev, int size)
466 {
467         char b[BDEVNAME_SIZE];
468         if (!rdev->sb_page) {
469                 MD_BUG();
470                 return -EINVAL;
471         }
472         if (rdev->sb_loaded)
473                 return 0;
474
475
476         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
477                 goto fail;
478         rdev->sb_loaded = 1;
479         return 0;
480
481 fail:
482         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
483                 bdevname(rdev->bdev,b));
484         return -EINVAL;
485 }
486
487 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
488 {
489         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
490                 (sb1->set_uuid1 == sb2->set_uuid1) &&
491                 (sb1->set_uuid2 == sb2->set_uuid2) &&
492                 (sb1->set_uuid3 == sb2->set_uuid3))
493
494                 return 1;
495
496         return 0;
497 }
498
499
500 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
501 {
502         int ret;
503         mdp_super_t *tmp1, *tmp2;
504
505         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
506         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
507
508         if (!tmp1 || !tmp2) {
509                 ret = 0;
510                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
511                 goto abort;
512         }
513
514         *tmp1 = *sb1;
515         *tmp2 = *sb2;
516
517         /*
518          * nr_disks is not constant
519          */
520         tmp1->nr_disks = 0;
521         tmp2->nr_disks = 0;
522
523         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
524                 ret = 0;
525         else
526                 ret = 1;
527
528 abort:
529         kfree(tmp1);
530         kfree(tmp2);
531         return ret;
532 }
533
534 static unsigned int calc_sb_csum(mdp_super_t * sb)
535 {
536         unsigned int disk_csum, csum;
537
538         disk_csum = sb->sb_csum;
539         sb->sb_csum = 0;
540         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
541         sb->sb_csum = disk_csum;
542         return csum;
543 }
544
545
546 /*
547  * Handle superblock details.
548  * We want to be able to handle multiple superblock formats
549  * so we have a common interface to them all, and an array of
550  * different handlers.
551  * We rely on user-space to write the initial superblock, and support
552  * reading and updating of superblocks.
553  * Interface methods are:
554  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
555  *      loads and validates a superblock on dev.
556  *      if refdev != NULL, compare superblocks on both devices
557  *    Return:
558  *      0 - dev has a superblock that is compatible with refdev
559  *      1 - dev has a superblock that is compatible and newer than refdev
560  *          so dev should be used as the refdev in future
561  *     -EINVAL superblock incompatible or invalid
562  *     -othererror e.g. -EIO
563  *
564  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
565  *      Verify that dev is acceptable into mddev.
566  *       The first time, mddev->raid_disks will be 0, and data from
567  *       dev should be merged in.  Subsequent calls check that dev
568  *       is new enough.  Return 0 or -EINVAL
569  *
570  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
571  *     Update the superblock for rdev with data in mddev
572  *     This does not write to disc.
573  *
574  */
575
576 struct super_type  {
577         char            *name;
578         struct module   *owner;
579         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
580         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
581         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
582 };
583
584 /*
585  * load_super for 0.90.0 
586  */
587 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
588 {
589         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
590         mdp_super_t *sb;
591         int ret;
592         sector_t sb_offset;
593
594         /*
595          * Calculate the position of the superblock,
596          * it's at the end of the disk.
597          *
598          * It also happens to be a multiple of 4Kb.
599          */
600         sb_offset = calc_dev_sboffset(rdev->bdev);
601         rdev->sb_offset = sb_offset;
602
603         ret = read_disk_sb(rdev, MD_SB_BYTES);
604         if (ret) return ret;
605
606         ret = -EINVAL;
607
608         bdevname(rdev->bdev, b);
609         sb = (mdp_super_t*)page_address(rdev->sb_page);
610
611         if (sb->md_magic != MD_SB_MAGIC) {
612                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
613                        b);
614                 goto abort;
615         }
616
617         if (sb->major_version != 0 ||
618             sb->minor_version != 90) {
619                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
620                         sb->major_version, sb->minor_version,
621                         b);
622                 goto abort;
623         }
624
625         if (sb->raid_disks <= 0)
626                 goto abort;
627
628         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
629                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
630                         b);
631                 goto abort;
632         }
633
634         rdev->preferred_minor = sb->md_minor;
635         rdev->data_offset = 0;
636         rdev->sb_size = MD_SB_BYTES;
637
638         if (sb->level == LEVEL_MULTIPATH)
639                 rdev->desc_nr = -1;
640         else
641                 rdev->desc_nr = sb->this_disk.number;
642
643         if (refdev == 0)
644                 ret = 1;
645         else {
646                 __u64 ev1, ev2;
647                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
648                 if (!uuid_equal(refsb, sb)) {
649                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
650                                 b, bdevname(refdev->bdev,b2));
651                         goto abort;
652                 }
653                 if (!sb_equal(refsb, sb)) {
654                         printk(KERN_WARNING "md: %s has same UUID"
655                                " but different superblock to %s\n",
656                                b, bdevname(refdev->bdev, b2));
657                         goto abort;
658                 }
659                 ev1 = md_event(sb);
660                 ev2 = md_event(refsb);
661                 if (ev1 > ev2)
662                         ret = 1;
663                 else 
664                         ret = 0;
665         }
666         rdev->size = calc_dev_size(rdev, sb->chunk_size);
667
668  abort:
669         return ret;
670 }
671
672 /*
673  * validate_super for 0.90.0
674  */
675 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
676 {
677         mdp_disk_t *desc;
678         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
679
680         rdev->raid_disk = -1;
681         rdev->flags = 0;
682         if (mddev->raid_disks == 0) {
683                 mddev->major_version = 0;
684                 mddev->minor_version = sb->minor_version;
685                 mddev->patch_version = sb->patch_version;
686                 mddev->persistent = ! sb->not_persistent;
687                 mddev->chunk_size = sb->chunk_size;
688                 mddev->ctime = sb->ctime;
689                 mddev->utime = sb->utime;
690                 mddev->level = sb->level;
691                 mddev->layout = sb->layout;
692                 mddev->raid_disks = sb->raid_disks;
693                 mddev->size = sb->size;
694                 mddev->events = md_event(sb);
695                 mddev->bitmap_offset = 0;
696                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
697
698                 if (sb->state & (1<<MD_SB_CLEAN))
699                         mddev->recovery_cp = MaxSector;
700                 else {
701                         if (sb->events_hi == sb->cp_events_hi && 
702                                 sb->events_lo == sb->cp_events_lo) {
703                                 mddev->recovery_cp = sb->recovery_cp;
704                         } else
705                                 mddev->recovery_cp = 0;
706                 }
707
708                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
709                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
710                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
711                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
712
713                 mddev->max_disks = MD_SB_DISKS;
714
715                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
716                     mddev->bitmap_file == NULL) {
717                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6) {
718                                 /* FIXME use a better test */
719                                 printk(KERN_WARNING "md: bitmaps only support for raid1\n");
720                                 return -EINVAL;
721                         }
722                         mddev->bitmap_offset = mddev->default_bitmap_offset;
723                 }
724
725         } else if (mddev->pers == NULL) {
726                 /* Insist on good event counter while assembling */
727                 __u64 ev1 = md_event(sb);
728                 ++ev1;
729                 if (ev1 < mddev->events) 
730                         return -EINVAL;
731         } else if (mddev->bitmap) {
732                 /* if adding to array with a bitmap, then we can accept an
733                  * older device ... but not too old.
734                  */
735                 __u64 ev1 = md_event(sb);
736                 if (ev1 < mddev->bitmap->events_cleared)
737                         return 0;
738         } else /* just a hot-add of a new device, leave raid_disk at -1 */
739                 return 0;
740
741         if (mddev->level != LEVEL_MULTIPATH) {
742                 desc = sb->disks + rdev->desc_nr;
743
744                 if (desc->state & (1<<MD_DISK_FAULTY))
745                         set_bit(Faulty, &rdev->flags);
746                 else if (desc->state & (1<<MD_DISK_SYNC) &&
747                          desc->raid_disk < mddev->raid_disks) {
748                         set_bit(In_sync, &rdev->flags);
749                         rdev->raid_disk = desc->raid_disk;
750                 }
751                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
752                         set_bit(WriteMostly, &rdev->flags);
753         } else /* MULTIPATH are always insync */
754                 set_bit(In_sync, &rdev->flags);
755         return 0;
756 }
757
758 /*
759  * sync_super for 0.90.0
760  */
761 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
762 {
763         mdp_super_t *sb;
764         struct list_head *tmp;
765         mdk_rdev_t *rdev2;
766         int next_spare = mddev->raid_disks;
767
768
769         /* make rdev->sb match mddev data..
770          *
771          * 1/ zero out disks
772          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
773          * 3/ any empty disks < next_spare become removed
774          *
775          * disks[0] gets initialised to REMOVED because
776          * we cannot be sure from other fields if it has
777          * been initialised or not.
778          */
779         int i;
780         int active=0, working=0,failed=0,spare=0,nr_disks=0;
781
782         rdev->sb_size = MD_SB_BYTES;
783
784         sb = (mdp_super_t*)page_address(rdev->sb_page);
785
786         memset(sb, 0, sizeof(*sb));
787
788         sb->md_magic = MD_SB_MAGIC;
789         sb->major_version = mddev->major_version;
790         sb->minor_version = mddev->minor_version;
791         sb->patch_version = mddev->patch_version;
792         sb->gvalid_words  = 0; /* ignored */
793         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
794         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
795         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
796         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
797
798         sb->ctime = mddev->ctime;
799         sb->level = mddev->level;
800         sb->size  = mddev->size;
801         sb->raid_disks = mddev->raid_disks;
802         sb->md_minor = mddev->md_minor;
803         sb->not_persistent = !mddev->persistent;
804         sb->utime = mddev->utime;
805         sb->state = 0;
806         sb->events_hi = (mddev->events>>32);
807         sb->events_lo = (u32)mddev->events;
808
809         if (mddev->in_sync)
810         {
811                 sb->recovery_cp = mddev->recovery_cp;
812                 sb->cp_events_hi = (mddev->events>>32);
813                 sb->cp_events_lo = (u32)mddev->events;
814                 if (mddev->recovery_cp == MaxSector)
815                         sb->state = (1<< MD_SB_CLEAN);
816         } else
817                 sb->recovery_cp = 0;
818
819         sb->layout = mddev->layout;
820         sb->chunk_size = mddev->chunk_size;
821
822         if (mddev->bitmap && mddev->bitmap_file == NULL)
823                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
824
825         sb->disks[0].state = (1<<MD_DISK_REMOVED);
826         ITERATE_RDEV(mddev,rdev2,tmp) {
827                 mdp_disk_t *d;
828                 int desc_nr;
829                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
830                     && !test_bit(Faulty, &rdev2->flags))
831                         desc_nr = rdev2->raid_disk;
832                 else
833                         desc_nr = next_spare++;
834                 rdev2->desc_nr = desc_nr;
835                 d = &sb->disks[rdev2->desc_nr];
836                 nr_disks++;
837                 d->number = rdev2->desc_nr;
838                 d->major = MAJOR(rdev2->bdev->bd_dev);
839                 d->minor = MINOR(rdev2->bdev->bd_dev);
840                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
841                     && !test_bit(Faulty, &rdev2->flags))
842                         d->raid_disk = rdev2->raid_disk;
843                 else
844                         d->raid_disk = rdev2->desc_nr; /* compatibility */
845                 if (test_bit(Faulty, &rdev2->flags)) {
846                         d->state = (1<<MD_DISK_FAULTY);
847                         failed++;
848                 } else if (test_bit(In_sync, &rdev2->flags)) {
849                         d->state = (1<<MD_DISK_ACTIVE);
850                         d->state |= (1<<MD_DISK_SYNC);
851                         active++;
852                         working++;
853                 } else {
854                         d->state = 0;
855                         spare++;
856                         working++;
857                 }
858                 if (test_bit(WriteMostly, &rdev2->flags))
859                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
860         }
861         /* now set the "removed" and "faulty" bits on any missing devices */
862         for (i=0 ; i < mddev->raid_disks ; i++) {
863                 mdp_disk_t *d = &sb->disks[i];
864                 if (d->state == 0 && d->number == 0) {
865                         d->number = i;
866                         d->raid_disk = i;
867                         d->state = (1<<MD_DISK_REMOVED);
868                         d->state |= (1<<MD_DISK_FAULTY);
869                         failed++;
870                 }
871         }
872         sb->nr_disks = nr_disks;
873         sb->active_disks = active;
874         sb->working_disks = working;
875         sb->failed_disks = failed;
876         sb->spare_disks = spare;
877
878         sb->this_disk = sb->disks[rdev->desc_nr];
879         sb->sb_csum = calc_sb_csum(sb);
880 }
881
882 /*
883  * version 1 superblock
884  */
885
886 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
887 {
888         unsigned int disk_csum, csum;
889         unsigned long long newcsum;
890         int size = 256 + le32_to_cpu(sb->max_dev)*2;
891         unsigned int *isuper = (unsigned int*)sb;
892         int i;
893
894         disk_csum = sb->sb_csum;
895         sb->sb_csum = 0;
896         newcsum = 0;
897         for (i=0; size>=4; size -= 4 )
898                 newcsum += le32_to_cpu(*isuper++);
899
900         if (size == 2)
901                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
902
903         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
904         sb->sb_csum = disk_csum;
905         return cpu_to_le32(csum);
906 }
907
908 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
909 {
910         struct mdp_superblock_1 *sb;
911         int ret;
912         sector_t sb_offset;
913         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
914         int bmask;
915
916         /*
917          * Calculate the position of the superblock.
918          * It is always aligned to a 4K boundary and
919          * depeding on minor_version, it can be:
920          * 0: At least 8K, but less than 12K, from end of device
921          * 1: At start of device
922          * 2: 4K from start of device.
923          */
924         switch(minor_version) {
925         case 0:
926                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
927                 sb_offset -= 8*2;
928                 sb_offset &= ~(sector_t)(4*2-1);
929                 /* convert from sectors to K */
930                 sb_offset /= 2;
931                 break;
932         case 1:
933                 sb_offset = 0;
934                 break;
935         case 2:
936                 sb_offset = 4;
937                 break;
938         default:
939                 return -EINVAL;
940         }
941         rdev->sb_offset = sb_offset;
942
943         /* superblock is rarely larger than 1K, but it can be larger,
944          * and it is safe to read 4k, so we do that
945          */
946         ret = read_disk_sb(rdev, 4096);
947         if (ret) return ret;
948
949
950         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
951
952         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
953             sb->major_version != cpu_to_le32(1) ||
954             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
955             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
956             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
957                 return -EINVAL;
958
959         if (calc_sb_1_csum(sb) != sb->sb_csum) {
960                 printk("md: invalid superblock checksum on %s\n",
961                         bdevname(rdev->bdev,b));
962                 return -EINVAL;
963         }
964         if (le64_to_cpu(sb->data_size) < 10) {
965                 printk("md: data_size too small on %s\n",
966                        bdevname(rdev->bdev,b));
967                 return -EINVAL;
968         }
969         rdev->preferred_minor = 0xffff;
970         rdev->data_offset = le64_to_cpu(sb->data_offset);
971
972         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
973         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
974         if (rdev->sb_size & bmask)
975                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
976
977         if (refdev == 0)
978                 return 1;
979         else {
980                 __u64 ev1, ev2;
981                 struct mdp_superblock_1 *refsb = 
982                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
983
984                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
985                     sb->level != refsb->level ||
986                     sb->layout != refsb->layout ||
987                     sb->chunksize != refsb->chunksize) {
988                         printk(KERN_WARNING "md: %s has strangely different"
989                                 " superblock to %s\n",
990                                 bdevname(rdev->bdev,b),
991                                 bdevname(refdev->bdev,b2));
992                         return -EINVAL;
993                 }
994                 ev1 = le64_to_cpu(sb->events);
995                 ev2 = le64_to_cpu(refsb->events);
996
997                 if (ev1 > ev2)
998                         return 1;
999         }
1000         if (minor_version) 
1001                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1002         else
1003                 rdev->size = rdev->sb_offset;
1004         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1005                 return -EINVAL;
1006         rdev->size = le64_to_cpu(sb->data_size)/2;
1007         if (le32_to_cpu(sb->chunksize))
1008                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1009         return 0;
1010 }
1011
1012 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1013 {
1014         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1015
1016         rdev->raid_disk = -1;
1017         rdev->flags = 0;
1018         if (mddev->raid_disks == 0) {
1019                 mddev->major_version = 1;
1020                 mddev->patch_version = 0;
1021                 mddev->persistent = 1;
1022                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1023                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1024                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1025                 mddev->level = le32_to_cpu(sb->level);
1026                 mddev->layout = le32_to_cpu(sb->layout);
1027                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1028                 mddev->size = le64_to_cpu(sb->size)/2;
1029                 mddev->events = le64_to_cpu(sb->events);
1030                 mddev->bitmap_offset = 0;
1031                 mddev->default_bitmap_offset = 0;
1032                 mddev->default_bitmap_offset = 1024;
1033                 
1034                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1035                 memcpy(mddev->uuid, sb->set_uuid, 16);
1036
1037                 mddev->max_disks =  (4096-256)/2;
1038
1039                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1040                     mddev->bitmap_file == NULL ) {
1041                         if (mddev->level != 1) {
1042                                 printk(KERN_WARNING "md: bitmaps only supported for raid1\n");
1043                                 return -EINVAL;
1044                         }
1045                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1046                 }
1047         } else if (mddev->pers == NULL) {
1048                 /* Insist of good event counter while assembling */
1049                 __u64 ev1 = le64_to_cpu(sb->events);
1050                 ++ev1;
1051                 if (ev1 < mddev->events)
1052                         return -EINVAL;
1053         } else if (mddev->bitmap) {
1054                 /* If adding to array with a bitmap, then we can accept an
1055                  * older device, but not too old.
1056                  */
1057                 __u64 ev1 = le64_to_cpu(sb->events);
1058                 if (ev1 < mddev->bitmap->events_cleared)
1059                         return 0;
1060         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1061                 return 0;
1062
1063         if (mddev->level != LEVEL_MULTIPATH) {
1064                 int role;
1065                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1066                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1067                 switch(role) {
1068                 case 0xffff: /* spare */
1069                         break;
1070                 case 0xfffe: /* faulty */
1071                         set_bit(Faulty, &rdev->flags);
1072                         break;
1073                 default:
1074                         set_bit(In_sync, &rdev->flags);
1075                         rdev->raid_disk = role;
1076                         break;
1077                 }
1078                 if (sb->devflags & WriteMostly1)
1079                         set_bit(WriteMostly, &rdev->flags);
1080         } else /* MULTIPATH are always insync */
1081                 set_bit(In_sync, &rdev->flags);
1082
1083         return 0;
1084 }
1085
1086 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1087 {
1088         struct mdp_superblock_1 *sb;
1089         struct list_head *tmp;
1090         mdk_rdev_t *rdev2;
1091         int max_dev, i;
1092         /* make rdev->sb match mddev and rdev data. */
1093
1094         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1095
1096         sb->feature_map = 0;
1097         sb->pad0 = 0;
1098         memset(sb->pad1, 0, sizeof(sb->pad1));
1099         memset(sb->pad2, 0, sizeof(sb->pad2));
1100         memset(sb->pad3, 0, sizeof(sb->pad3));
1101
1102         sb->utime = cpu_to_le64((__u64)mddev->utime);
1103         sb->events = cpu_to_le64(mddev->events);
1104         if (mddev->in_sync)
1105                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1106         else
1107                 sb->resync_offset = cpu_to_le64(0);
1108
1109         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1110                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1111                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1112         }
1113
1114         max_dev = 0;
1115         ITERATE_RDEV(mddev,rdev2,tmp)
1116                 if (rdev2->desc_nr+1 > max_dev)
1117                         max_dev = rdev2->desc_nr+1;
1118         
1119         sb->max_dev = cpu_to_le32(max_dev);
1120         for (i=0; i<max_dev;i++)
1121                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1122         
1123         ITERATE_RDEV(mddev,rdev2,tmp) {
1124                 i = rdev2->desc_nr;
1125                 if (test_bit(Faulty, &rdev2->flags))
1126                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1127                 else if (test_bit(In_sync, &rdev2->flags))
1128                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1129                 else
1130                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1131         }
1132
1133         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1134         sb->sb_csum = calc_sb_1_csum(sb);
1135 }
1136
1137
1138 static struct super_type super_types[] = {
1139         [0] = {
1140                 .name   = "0.90.0",
1141                 .owner  = THIS_MODULE,
1142                 .load_super     = super_90_load,
1143                 .validate_super = super_90_validate,
1144                 .sync_super     = super_90_sync,
1145         },
1146         [1] = {
1147                 .name   = "md-1",
1148                 .owner  = THIS_MODULE,
1149                 .load_super     = super_1_load,
1150                 .validate_super = super_1_validate,
1151                 .sync_super     = super_1_sync,
1152         },
1153 };
1154         
1155 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1156 {
1157         struct list_head *tmp;
1158         mdk_rdev_t *rdev;
1159
1160         ITERATE_RDEV(mddev,rdev,tmp)
1161                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1162                         return rdev;
1163
1164         return NULL;
1165 }
1166
1167 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1168 {
1169         struct list_head *tmp;
1170         mdk_rdev_t *rdev;
1171
1172         ITERATE_RDEV(mddev1,rdev,tmp)
1173                 if (match_dev_unit(mddev2, rdev))
1174                         return 1;
1175
1176         return 0;
1177 }
1178
1179 static LIST_HEAD(pending_raid_disks);
1180
1181 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1182 {
1183         mdk_rdev_t *same_pdev;
1184         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1185         struct kobject *ko;
1186
1187         if (rdev->mddev) {
1188                 MD_BUG();
1189                 return -EINVAL;
1190         }
1191         same_pdev = match_dev_unit(mddev, rdev);
1192         if (same_pdev)
1193                 printk(KERN_WARNING
1194                         "%s: WARNING: %s appears to be on the same physical"
1195                         " disk as %s. True\n     protection against single-disk"
1196                         " failure might be compromised.\n",
1197                         mdname(mddev), bdevname(rdev->bdev,b),
1198                         bdevname(same_pdev->bdev,b2));
1199
1200         /* Verify rdev->desc_nr is unique.
1201          * If it is -1, assign a free number, else
1202          * check number is not in use
1203          */
1204         if (rdev->desc_nr < 0) {
1205                 int choice = 0;
1206                 if (mddev->pers) choice = mddev->raid_disks;
1207                 while (find_rdev_nr(mddev, choice))
1208                         choice++;
1209                 rdev->desc_nr = choice;
1210         } else {
1211                 if (find_rdev_nr(mddev, rdev->desc_nr))
1212                         return -EBUSY;
1213         }
1214         bdevname(rdev->bdev,b);
1215         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1216                 return -ENOMEM;
1217                         
1218         list_add(&rdev->same_set, &mddev->disks);
1219         rdev->mddev = mddev;
1220         printk(KERN_INFO "md: bind<%s>\n", b);
1221
1222         rdev->kobj.parent = &mddev->kobj;
1223         kobject_add(&rdev->kobj);
1224
1225         if (rdev->bdev->bd_part)
1226                 ko = &rdev->bdev->bd_part->kobj;
1227         else
1228                 ko = &rdev->bdev->bd_disk->kobj;
1229         sysfs_create_link(&rdev->kobj, ko, "block");
1230         return 0;
1231 }
1232
1233 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1234 {
1235         char b[BDEVNAME_SIZE];
1236         if (!rdev->mddev) {
1237                 MD_BUG();
1238                 return;
1239         }
1240         list_del_init(&rdev->same_set);
1241         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1242         rdev->mddev = NULL;
1243         sysfs_remove_link(&rdev->kobj, "block");
1244         kobject_del(&rdev->kobj);
1245 }
1246
1247 /*
1248  * prevent the device from being mounted, repartitioned or
1249  * otherwise reused by a RAID array (or any other kernel
1250  * subsystem), by bd_claiming the device.
1251  */
1252 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1253 {
1254         int err = 0;
1255         struct block_device *bdev;
1256         char b[BDEVNAME_SIZE];
1257
1258         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1259         if (IS_ERR(bdev)) {
1260                 printk(KERN_ERR "md: could not open %s.\n",
1261                         __bdevname(dev, b));
1262                 return PTR_ERR(bdev);
1263         }
1264         err = bd_claim(bdev, rdev);
1265         if (err) {
1266                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1267                         bdevname(bdev, b));
1268                 blkdev_put(bdev);
1269                 return err;
1270         }
1271         rdev->bdev = bdev;
1272         return err;
1273 }
1274
1275 static void unlock_rdev(mdk_rdev_t *rdev)
1276 {
1277         struct block_device *bdev = rdev->bdev;
1278         rdev->bdev = NULL;
1279         if (!bdev)
1280                 MD_BUG();
1281         bd_release(bdev);
1282         blkdev_put(bdev);
1283 }
1284
1285 void md_autodetect_dev(dev_t dev);
1286
1287 static void export_rdev(mdk_rdev_t * rdev)
1288 {
1289         char b[BDEVNAME_SIZE];
1290         printk(KERN_INFO "md: export_rdev(%s)\n",
1291                 bdevname(rdev->bdev,b));
1292         if (rdev->mddev)
1293                 MD_BUG();
1294         free_disk_sb(rdev);
1295         list_del_init(&rdev->same_set);
1296 #ifndef MODULE
1297         md_autodetect_dev(rdev->bdev->bd_dev);
1298 #endif
1299         unlock_rdev(rdev);
1300         kobject_put(&rdev->kobj);
1301 }
1302
1303 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1304 {
1305         unbind_rdev_from_array(rdev);
1306         export_rdev(rdev);
1307 }
1308
1309 static void export_array(mddev_t *mddev)
1310 {
1311         struct list_head *tmp;
1312         mdk_rdev_t *rdev;
1313
1314         ITERATE_RDEV(mddev,rdev,tmp) {
1315                 if (!rdev->mddev) {
1316                         MD_BUG();
1317                         continue;
1318                 }
1319                 kick_rdev_from_array(rdev);
1320         }
1321         if (!list_empty(&mddev->disks))
1322                 MD_BUG();
1323         mddev->raid_disks = 0;
1324         mddev->major_version = 0;
1325 }
1326
1327 static void print_desc(mdp_disk_t *desc)
1328 {
1329         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1330                 desc->major,desc->minor,desc->raid_disk,desc->state);
1331 }
1332
1333 static void print_sb(mdp_super_t *sb)
1334 {
1335         int i;
1336
1337         printk(KERN_INFO 
1338                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1339                 sb->major_version, sb->minor_version, sb->patch_version,
1340                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1341                 sb->ctime);
1342         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1343                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1344                 sb->md_minor, sb->layout, sb->chunk_size);
1345         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1346                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1347                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1348                 sb->failed_disks, sb->spare_disks,
1349                 sb->sb_csum, (unsigned long)sb->events_lo);
1350
1351         printk(KERN_INFO);
1352         for (i = 0; i < MD_SB_DISKS; i++) {
1353                 mdp_disk_t *desc;
1354
1355                 desc = sb->disks + i;
1356                 if (desc->number || desc->major || desc->minor ||
1357                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1358                         printk("     D %2d: ", i);
1359                         print_desc(desc);
1360                 }
1361         }
1362         printk(KERN_INFO "md:     THIS: ");
1363         print_desc(&sb->this_disk);
1364
1365 }
1366
1367 static void print_rdev(mdk_rdev_t *rdev)
1368 {
1369         char b[BDEVNAME_SIZE];
1370         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1371                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1372                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1373                 rdev->desc_nr);
1374         if (rdev->sb_loaded) {
1375                 printk(KERN_INFO "md: rdev superblock:\n");
1376                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1377         } else
1378                 printk(KERN_INFO "md: no rdev superblock!\n");
1379 }
1380
1381 void md_print_devices(void)
1382 {
1383         struct list_head *tmp, *tmp2;
1384         mdk_rdev_t *rdev;
1385         mddev_t *mddev;
1386         char b[BDEVNAME_SIZE];
1387
1388         printk("\n");
1389         printk("md:     **********************************\n");
1390         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1391         printk("md:     **********************************\n");
1392         ITERATE_MDDEV(mddev,tmp) {
1393
1394                 if (mddev->bitmap)
1395                         bitmap_print_sb(mddev->bitmap);
1396                 else
1397                         printk("%s: ", mdname(mddev));
1398                 ITERATE_RDEV(mddev,rdev,tmp2)
1399                         printk("<%s>", bdevname(rdev->bdev,b));
1400                 printk("\n");
1401
1402                 ITERATE_RDEV(mddev,rdev,tmp2)
1403                         print_rdev(rdev);
1404         }
1405         printk("md:     **********************************\n");
1406         printk("\n");
1407 }
1408
1409
1410 static void sync_sbs(mddev_t * mddev)
1411 {
1412         mdk_rdev_t *rdev;
1413         struct list_head *tmp;
1414
1415         ITERATE_RDEV(mddev,rdev,tmp) {
1416                 super_types[mddev->major_version].
1417                         sync_super(mddev, rdev);
1418                 rdev->sb_loaded = 1;
1419         }
1420 }
1421
1422 static void md_update_sb(mddev_t * mddev)
1423 {
1424         int err;
1425         struct list_head *tmp;
1426         mdk_rdev_t *rdev;
1427         int sync_req;
1428
1429 repeat:
1430         spin_lock_irq(&mddev->write_lock);
1431         sync_req = mddev->in_sync;
1432         mddev->utime = get_seconds();
1433         mddev->events ++;
1434
1435         if (!mddev->events) {
1436                 /*
1437                  * oops, this 64-bit counter should never wrap.
1438                  * Either we are in around ~1 trillion A.C., assuming
1439                  * 1 reboot per second, or we have a bug:
1440                  */
1441                 MD_BUG();
1442                 mddev->events --;
1443         }
1444         mddev->sb_dirty = 2;
1445         sync_sbs(mddev);
1446
1447         /*
1448          * do not write anything to disk if using
1449          * nonpersistent superblocks
1450          */
1451         if (!mddev->persistent) {
1452                 mddev->sb_dirty = 0;
1453                 spin_unlock_irq(&mddev->write_lock);
1454                 wake_up(&mddev->sb_wait);
1455                 return;
1456         }
1457         spin_unlock_irq(&mddev->write_lock);
1458
1459         dprintk(KERN_INFO 
1460                 "md: updating %s RAID superblock on device (in sync %d)\n",
1461                 mdname(mddev),mddev->in_sync);
1462
1463         err = bitmap_update_sb(mddev->bitmap);
1464         ITERATE_RDEV(mddev,rdev,tmp) {
1465                 char b[BDEVNAME_SIZE];
1466                 dprintk(KERN_INFO "md: ");
1467                 if (test_bit(Faulty, &rdev->flags))
1468                         dprintk("(skipping faulty ");
1469
1470                 dprintk("%s ", bdevname(rdev->bdev,b));
1471                 if (!test_bit(Faulty, &rdev->flags)) {
1472                         md_super_write(mddev,rdev,
1473                                        rdev->sb_offset<<1, rdev->sb_size,
1474                                        rdev->sb_page);
1475                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1476                                 bdevname(rdev->bdev,b),
1477                                 (unsigned long long)rdev->sb_offset);
1478
1479                 } else
1480                         dprintk(")\n");
1481                 if (mddev->level == LEVEL_MULTIPATH)
1482                         /* only need to write one superblock... */
1483                         break;
1484         }
1485         md_super_wait(mddev);
1486         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1487
1488         spin_lock_irq(&mddev->write_lock);
1489         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1490                 /* have to write it out again */
1491                 spin_unlock_irq(&mddev->write_lock);
1492                 goto repeat;
1493         }
1494         mddev->sb_dirty = 0;
1495         spin_unlock_irq(&mddev->write_lock);
1496         wake_up(&mddev->sb_wait);
1497
1498 }
1499
1500 struct rdev_sysfs_entry {
1501         struct attribute attr;
1502         ssize_t (*show)(mdk_rdev_t *, char *);
1503         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1504 };
1505
1506 static ssize_t
1507 state_show(mdk_rdev_t *rdev, char *page)
1508 {
1509         char *sep = "";
1510         int len=0;
1511
1512         if (test_bit(Faulty, &rdev->flags)) {
1513                 len+= sprintf(page+len, "%sfaulty",sep);
1514                 sep = ",";
1515         }
1516         if (test_bit(In_sync, &rdev->flags)) {
1517                 len += sprintf(page+len, "%sin_sync",sep);
1518                 sep = ",";
1519         }
1520         if (!test_bit(Faulty, &rdev->flags) &&
1521             !test_bit(In_sync, &rdev->flags)) {
1522                 len += sprintf(page+len, "%sspare", sep);
1523                 sep = ",";
1524         }
1525         return len+sprintf(page+len, "\n");
1526 }
1527
1528 static struct rdev_sysfs_entry
1529 rdev_state = __ATTR_RO(state);
1530
1531 static ssize_t
1532 super_show(mdk_rdev_t *rdev, char *page)
1533 {
1534         if (rdev->sb_loaded && rdev->sb_size) {
1535                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1536                 return rdev->sb_size;
1537         } else
1538                 return 0;
1539 }
1540 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1541
1542 static struct attribute *rdev_default_attrs[] = {
1543         &rdev_state.attr,
1544         &rdev_super.attr,
1545         NULL,
1546 };
1547 static ssize_t
1548 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1549 {
1550         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1551         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1552
1553         if (!entry->show)
1554                 return -EIO;
1555         return entry->show(rdev, page);
1556 }
1557
1558 static ssize_t
1559 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1560               const char *page, size_t length)
1561 {
1562         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1563         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1564
1565         if (!entry->store)
1566                 return -EIO;
1567         return entry->store(rdev, page, length);
1568 }
1569
1570 static void rdev_free(struct kobject *ko)
1571 {
1572         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1573         kfree(rdev);
1574 }
1575 static struct sysfs_ops rdev_sysfs_ops = {
1576         .show           = rdev_attr_show,
1577         .store          = rdev_attr_store,
1578 };
1579 static struct kobj_type rdev_ktype = {
1580         .release        = rdev_free,
1581         .sysfs_ops      = &rdev_sysfs_ops,
1582         .default_attrs  = rdev_default_attrs,
1583 };
1584
1585 /*
1586  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1587  *
1588  * mark the device faulty if:
1589  *
1590  *   - the device is nonexistent (zero size)
1591  *   - the device has no valid superblock
1592  *
1593  * a faulty rdev _never_ has rdev->sb set.
1594  */
1595 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1596 {
1597         char b[BDEVNAME_SIZE];
1598         int err;
1599         mdk_rdev_t *rdev;
1600         sector_t size;
1601
1602         rdev = (mdk_rdev_t *) kmalloc(sizeof(*rdev), GFP_KERNEL);
1603         if (!rdev) {
1604                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1605                 return ERR_PTR(-ENOMEM);
1606         }
1607         memset(rdev, 0, sizeof(*rdev));
1608
1609         if ((err = alloc_disk_sb(rdev)))
1610                 goto abort_free;
1611
1612         err = lock_rdev(rdev, newdev);
1613         if (err)
1614                 goto abort_free;
1615
1616         rdev->kobj.parent = NULL;
1617         rdev->kobj.ktype = &rdev_ktype;
1618         kobject_init(&rdev->kobj);
1619
1620         rdev->desc_nr = -1;
1621         rdev->flags = 0;
1622         rdev->data_offset = 0;
1623         atomic_set(&rdev->nr_pending, 0);
1624         atomic_set(&rdev->read_errors, 0);
1625
1626         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1627         if (!size) {
1628                 printk(KERN_WARNING 
1629                         "md: %s has zero or unknown size, marking faulty!\n",
1630                         bdevname(rdev->bdev,b));
1631                 err = -EINVAL;
1632                 goto abort_free;
1633         }
1634
1635         if (super_format >= 0) {
1636                 err = super_types[super_format].
1637                         load_super(rdev, NULL, super_minor);
1638                 if (err == -EINVAL) {
1639                         printk(KERN_WARNING 
1640                                 "md: %s has invalid sb, not importing!\n",
1641                                 bdevname(rdev->bdev,b));
1642                         goto abort_free;
1643                 }
1644                 if (err < 0) {
1645                         printk(KERN_WARNING 
1646                                 "md: could not read %s's sb, not importing!\n",
1647                                 bdevname(rdev->bdev,b));
1648                         goto abort_free;
1649                 }
1650         }
1651         INIT_LIST_HEAD(&rdev->same_set);
1652
1653         return rdev;
1654
1655 abort_free:
1656         if (rdev->sb_page) {
1657                 if (rdev->bdev)
1658                         unlock_rdev(rdev);
1659                 free_disk_sb(rdev);
1660         }
1661         kfree(rdev);
1662         return ERR_PTR(err);
1663 }
1664
1665 /*
1666  * Check a full RAID array for plausibility
1667  */
1668
1669
1670 static void analyze_sbs(mddev_t * mddev)
1671 {
1672         int i;
1673         struct list_head *tmp;
1674         mdk_rdev_t *rdev, *freshest;
1675         char b[BDEVNAME_SIZE];
1676
1677         freshest = NULL;
1678         ITERATE_RDEV(mddev,rdev,tmp)
1679                 switch (super_types[mddev->major_version].
1680                         load_super(rdev, freshest, mddev->minor_version)) {
1681                 case 1:
1682                         freshest = rdev;
1683                         break;
1684                 case 0:
1685                         break;
1686                 default:
1687                         printk( KERN_ERR \
1688                                 "md: fatal superblock inconsistency in %s"
1689                                 " -- removing from array\n", 
1690                                 bdevname(rdev->bdev,b));
1691                         kick_rdev_from_array(rdev);
1692                 }
1693
1694
1695         super_types[mddev->major_version].
1696                 validate_super(mddev, freshest);
1697
1698         i = 0;
1699         ITERATE_RDEV(mddev,rdev,tmp) {
1700                 if (rdev != freshest)
1701                         if (super_types[mddev->major_version].
1702                             validate_super(mddev, rdev)) {
1703                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1704                                         " from array!\n",
1705                                         bdevname(rdev->bdev,b));
1706                                 kick_rdev_from_array(rdev);
1707                                 continue;
1708                         }
1709                 if (mddev->level == LEVEL_MULTIPATH) {
1710                         rdev->desc_nr = i++;
1711                         rdev->raid_disk = rdev->desc_nr;
1712                         set_bit(In_sync, &rdev->flags);
1713                 }
1714         }
1715
1716
1717
1718         if (mddev->recovery_cp != MaxSector &&
1719             mddev->level >= 1)
1720                 printk(KERN_ERR "md: %s: raid array is not clean"
1721                        " -- starting background reconstruction\n",
1722                        mdname(mddev));
1723
1724 }
1725
1726 static ssize_t
1727 level_show(mddev_t *mddev, char *page)
1728 {
1729         mdk_personality_t *p = mddev->pers;
1730         if (p == NULL)
1731                 return 0;
1732         if (mddev->level >= 0)
1733                 return sprintf(page, "RAID-%d\n", mddev->level);
1734         else
1735                 return sprintf(page, "%s\n", p->name);
1736 }
1737
1738 static struct md_sysfs_entry md_level = __ATTR_RO(level);
1739
1740 static ssize_t
1741 raid_disks_show(mddev_t *mddev, char *page)
1742 {
1743         return sprintf(page, "%d\n", mddev->raid_disks);
1744 }
1745
1746 static struct md_sysfs_entry md_raid_disks = __ATTR_RO(raid_disks);
1747
1748 static ssize_t
1749 md_show_scan(mddev_t *mddev, char *page)
1750 {
1751         char *type = "none";
1752         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1753             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
1754                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
1755                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1756                                 type = "resync";
1757                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
1758                                 type = "check";
1759                         else
1760                                 type = "repair";
1761                 } else
1762                         type = "recover";
1763         }
1764         return sprintf(page, "%s\n", type);
1765 }
1766
1767 static ssize_t
1768 md_store_scan(mddev_t *mddev, const char *page, size_t len)
1769 {
1770         int canscan=0;
1771
1772         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
1773             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
1774                 return -EBUSY;
1775
1776         if (mddev->pers && mddev->pers->sync_request)
1777                 canscan=1;
1778
1779         if (!canscan)
1780                 return -EINVAL;
1781
1782         if (strcmp(page, "check")==0 || strcmp(page, "check\n")==0)
1783                 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
1784         else if (strcmp(page, "repair")!=0 && strcmp(page, "repair\n")!=0)
1785                 return -EINVAL;
1786         set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
1787         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
1788         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
1789         md_wakeup_thread(mddev->thread);
1790         return len;
1791 }
1792
1793 static ssize_t
1794 mismatch_cnt_show(mddev_t *mddev, char *page)
1795 {
1796         return sprintf(page, "%llu\n",
1797                        (unsigned long long) mddev->resync_mismatches);
1798 }
1799
1800 static struct md_sysfs_entry
1801 md_scan_mode = __ATTR(scan_mode, S_IRUGO|S_IWUSR, md_show_scan, md_store_scan);
1802
1803
1804 static struct md_sysfs_entry
1805 md_mismatches = __ATTR_RO(mismatch_cnt);
1806
1807 static struct attribute *md_default_attrs[] = {
1808         &md_level.attr,
1809         &md_raid_disks.attr,
1810         NULL,
1811 };
1812
1813 static struct attribute *md_redundancy_attrs[] = {
1814         &md_scan_mode.attr,
1815         &md_mismatches.attr,
1816         NULL,
1817 };
1818 static struct attribute_group md_redundancy_group = {
1819         .name = NULL,
1820         .attrs = md_redundancy_attrs,
1821 };
1822
1823
1824 static ssize_t
1825 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1826 {
1827         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1828         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1829         ssize_t rv;
1830
1831         if (!entry->show)
1832                 return -EIO;
1833         mddev_lock(mddev);
1834         rv = entry->show(mddev, page);
1835         mddev_unlock(mddev);
1836         return rv;
1837 }
1838
1839 static ssize_t
1840 md_attr_store(struct kobject *kobj, struct attribute *attr,
1841               const char *page, size_t length)
1842 {
1843         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
1844         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
1845         ssize_t rv;
1846
1847         if (!entry->store)
1848                 return -EIO;
1849         mddev_lock(mddev);
1850         rv = entry->store(mddev, page, length);
1851         mddev_unlock(mddev);
1852         return rv;
1853 }
1854
1855 static void md_free(struct kobject *ko)
1856 {
1857         mddev_t *mddev = container_of(ko, mddev_t, kobj);
1858         kfree(mddev);
1859 }
1860
1861 static struct sysfs_ops md_sysfs_ops = {
1862         .show   = md_attr_show,
1863         .store  = md_attr_store,
1864 };
1865 static struct kobj_type md_ktype = {
1866         .release        = md_free,
1867         .sysfs_ops      = &md_sysfs_ops,
1868         .default_attrs  = md_default_attrs,
1869 };
1870
1871 int mdp_major = 0;
1872
1873 static struct kobject *md_probe(dev_t dev, int *part, void *data)
1874 {
1875         static DECLARE_MUTEX(disks_sem);
1876         mddev_t *mddev = mddev_find(dev);
1877         struct gendisk *disk;
1878         int partitioned = (MAJOR(dev) != MD_MAJOR);
1879         int shift = partitioned ? MdpMinorShift : 0;
1880         int unit = MINOR(dev) >> shift;
1881
1882         if (!mddev)
1883                 return NULL;
1884
1885         down(&disks_sem);
1886         if (mddev->gendisk) {
1887                 up(&disks_sem);
1888                 mddev_put(mddev);
1889                 return NULL;
1890         }
1891         disk = alloc_disk(1 << shift);
1892         if (!disk) {
1893                 up(&disks_sem);
1894                 mddev_put(mddev);
1895                 return NULL;
1896         }
1897         disk->major = MAJOR(dev);
1898         disk->first_minor = unit << shift;
1899         if (partitioned) {
1900                 sprintf(disk->disk_name, "md_d%d", unit);
1901                 sprintf(disk->devfs_name, "md/d%d", unit);
1902         } else {
1903                 sprintf(disk->disk_name, "md%d", unit);
1904                 sprintf(disk->devfs_name, "md/%d", unit);
1905         }
1906         disk->fops = &md_fops;
1907         disk->private_data = mddev;
1908         disk->queue = mddev->queue;
1909         add_disk(disk);
1910         mddev->gendisk = disk;
1911         up(&disks_sem);
1912         mddev->kobj.parent = &disk->kobj;
1913         mddev->kobj.k_name = NULL;
1914         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
1915         mddev->kobj.ktype = &md_ktype;
1916         kobject_register(&mddev->kobj);
1917         return NULL;
1918 }
1919
1920 void md_wakeup_thread(mdk_thread_t *thread);
1921
1922 static void md_safemode_timeout(unsigned long data)
1923 {
1924         mddev_t *mddev = (mddev_t *) data;
1925
1926         mddev->safemode = 1;
1927         md_wakeup_thread(mddev->thread);
1928 }
1929
1930
1931 static int do_md_run(mddev_t * mddev)
1932 {
1933         int pnum, err;
1934         int chunk_size;
1935         struct list_head *tmp;
1936         mdk_rdev_t *rdev;
1937         struct gendisk *disk;
1938         char b[BDEVNAME_SIZE];
1939
1940         if (list_empty(&mddev->disks))
1941                 /* cannot run an array with no devices.. */
1942                 return -EINVAL;
1943
1944         if (mddev->pers)
1945                 return -EBUSY;
1946
1947         /*
1948          * Analyze all RAID superblock(s)
1949          */
1950         if (!mddev->raid_disks)
1951                 analyze_sbs(mddev);
1952
1953         chunk_size = mddev->chunk_size;
1954         pnum = level_to_pers(mddev->level);
1955
1956         if ((pnum != MULTIPATH) && (pnum != RAID1)) {
1957                 if (!chunk_size) {
1958                         /*
1959                          * 'default chunksize' in the old md code used to
1960                          * be PAGE_SIZE, baaad.
1961                          * we abort here to be on the safe side. We don't
1962                          * want to continue the bad practice.
1963                          */
1964                         printk(KERN_ERR 
1965                                 "no chunksize specified, see 'man raidtab'\n");
1966                         return -EINVAL;
1967                 }
1968                 if (chunk_size > MAX_CHUNK_SIZE) {
1969                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
1970                                 chunk_size, MAX_CHUNK_SIZE);
1971                         return -EINVAL;
1972                 }
1973                 /*
1974                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
1975                  */
1976                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
1977                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
1978                         return -EINVAL;
1979                 }
1980                 if (chunk_size < PAGE_SIZE) {
1981                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
1982                                 chunk_size, PAGE_SIZE);
1983                         return -EINVAL;
1984                 }
1985
1986                 /* devices must have minimum size of one chunk */
1987                 ITERATE_RDEV(mddev,rdev,tmp) {
1988                         if (test_bit(Faulty, &rdev->flags))
1989                                 continue;
1990                         if (rdev->size < chunk_size / 1024) {
1991                                 printk(KERN_WARNING
1992                                         "md: Dev %s smaller than chunk_size:"
1993                                         " %lluk < %dk\n",
1994                                         bdevname(rdev->bdev,b),
1995                                         (unsigned long long)rdev->size,
1996                                         chunk_size / 1024);
1997                                 return -EINVAL;
1998                         }
1999                 }
2000         }
2001
2002 #ifdef CONFIG_KMOD
2003         if (!pers[pnum])
2004         {
2005                 request_module("md-personality-%d", pnum);
2006         }
2007 #endif
2008
2009         /*
2010          * Drop all container device buffers, from now on
2011          * the only valid external interface is through the md
2012          * device.
2013          * Also find largest hardsector size
2014          */
2015         ITERATE_RDEV(mddev,rdev,tmp) {
2016                 if (test_bit(Faulty, &rdev->flags))
2017                         continue;
2018                 sync_blockdev(rdev->bdev);
2019                 invalidate_bdev(rdev->bdev, 0);
2020         }
2021
2022         md_probe(mddev->unit, NULL, NULL);
2023         disk = mddev->gendisk;
2024         if (!disk)
2025                 return -ENOMEM;
2026
2027         spin_lock(&pers_lock);
2028         if (!pers[pnum] || !try_module_get(pers[pnum]->owner)) {
2029                 spin_unlock(&pers_lock);
2030                 printk(KERN_WARNING "md: personality %d is not loaded!\n",
2031                        pnum);
2032                 return -EINVAL;
2033         }
2034
2035         mddev->pers = pers[pnum];
2036         spin_unlock(&pers_lock);
2037
2038         mddev->recovery = 0;
2039         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2040         mddev->barriers_work = 1;
2041
2042         if (start_readonly)
2043                 mddev->ro = 2; /* read-only, but switch on first write */
2044
2045         /* before we start the array running, initialise the bitmap */
2046         err = bitmap_create(mddev);
2047         if (err)
2048                 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2049                         mdname(mddev), err);
2050         else
2051                 err = mddev->pers->run(mddev);
2052         if (err) {
2053                 printk(KERN_ERR "md: pers->run() failed ...\n");
2054                 module_put(mddev->pers->owner);
2055                 mddev->pers = NULL;
2056                 bitmap_destroy(mddev);
2057                 return err;
2058         }
2059         if (mddev->pers->sync_request)
2060                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2061         atomic_set(&mddev->writes_pending,0);
2062         mddev->safemode = 0;
2063         mddev->safemode_timer.function = md_safemode_timeout;
2064         mddev->safemode_timer.data = (unsigned long) mddev;
2065         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2066         mddev->in_sync = 1;
2067
2068         ITERATE_RDEV(mddev,rdev,tmp)
2069                 if (rdev->raid_disk >= 0) {
2070                         char nm[20];
2071                         sprintf(nm, "rd%d", rdev->raid_disk);
2072                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2073                 }
2074         
2075         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2076         md_wakeup_thread(mddev->thread);
2077         
2078         if (mddev->sb_dirty)
2079                 md_update_sb(mddev);
2080
2081         set_capacity(disk, mddev->array_size<<1);
2082
2083         /* If we call blk_queue_make_request here, it will
2084          * re-initialise max_sectors etc which may have been
2085          * refined inside -> run.  So just set the bits we need to set.
2086          * Most initialisation happended when we called
2087          * blk_queue_make_request(..., md_fail_request)
2088          * earlier.
2089          */
2090         mddev->queue->queuedata = mddev;
2091         mddev->queue->make_request_fn = mddev->pers->make_request;
2092
2093         mddev->changed = 1;
2094         return 0;
2095 }
2096
2097 static int restart_array(mddev_t *mddev)
2098 {
2099         struct gendisk *disk = mddev->gendisk;
2100         int err;
2101
2102         /*
2103          * Complain if it has no devices
2104          */
2105         err = -ENXIO;
2106         if (list_empty(&mddev->disks))
2107                 goto out;
2108
2109         if (mddev->pers) {
2110                 err = -EBUSY;
2111                 if (!mddev->ro)
2112                         goto out;
2113
2114                 mddev->safemode = 0;
2115                 mddev->ro = 0;
2116                 set_disk_ro(disk, 0);
2117
2118                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2119                         mdname(mddev));
2120                 /*
2121                  * Kick recovery or resync if necessary
2122                  */
2123                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2124                 md_wakeup_thread(mddev->thread);
2125                 err = 0;
2126         } else {
2127                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2128                         mdname(mddev));
2129                 err = -EINVAL;
2130         }
2131
2132 out:
2133         return err;
2134 }
2135
2136 static int do_md_stop(mddev_t * mddev, int ro)
2137 {
2138         int err = 0;
2139         struct gendisk *disk = mddev->gendisk;
2140
2141         if (mddev->pers) {
2142                 if (atomic_read(&mddev->active)>2) {
2143                         printk("md: %s still in use.\n",mdname(mddev));
2144                         return -EBUSY;
2145                 }
2146
2147                 if (mddev->sync_thread) {
2148                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2149                         md_unregister_thread(mddev->sync_thread);
2150                         mddev->sync_thread = NULL;
2151                 }
2152
2153                 del_timer_sync(&mddev->safemode_timer);
2154
2155                 invalidate_partition(disk, 0);
2156
2157                 if (ro) {
2158                         err  = -ENXIO;
2159                         if (mddev->ro==1)
2160                                 goto out;
2161                         mddev->ro = 1;
2162                 } else {
2163                         bitmap_flush(mddev);
2164                         md_super_wait(mddev);
2165                         if (mddev->ro)
2166                                 set_disk_ro(disk, 0);
2167                         blk_queue_make_request(mddev->queue, md_fail_request);
2168                         mddev->pers->stop(mddev);
2169                         if (mddev->pers->sync_request)
2170                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2171
2172                         module_put(mddev->pers->owner);
2173                         mddev->pers = NULL;
2174                         if (mddev->ro)
2175                                 mddev->ro = 0;
2176                 }
2177                 if (!mddev->in_sync) {
2178                         /* mark array as shutdown cleanly */
2179                         mddev->in_sync = 1;
2180                         md_update_sb(mddev);
2181                 }
2182                 if (ro)
2183                         set_disk_ro(disk, 1);
2184         }
2185
2186         bitmap_destroy(mddev);
2187         if (mddev->bitmap_file) {
2188                 atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2189                 fput(mddev->bitmap_file);
2190                 mddev->bitmap_file = NULL;
2191         }
2192         mddev->bitmap_offset = 0;
2193
2194         /*
2195          * Free resources if final stop
2196          */
2197         if (!ro) {
2198                 mdk_rdev_t *rdev;
2199                 struct list_head *tmp;
2200                 struct gendisk *disk;
2201                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2202
2203                 ITERATE_RDEV(mddev,rdev,tmp)
2204                         if (rdev->raid_disk >= 0) {
2205                                 char nm[20];
2206                                 sprintf(nm, "rd%d", rdev->raid_disk);
2207                                 sysfs_remove_link(&mddev->kobj, nm);
2208                         }
2209
2210                 export_array(mddev);
2211
2212                 mddev->array_size = 0;
2213                 disk = mddev->gendisk;
2214                 if (disk)
2215                         set_capacity(disk, 0);
2216                 mddev->changed = 1;
2217         } else
2218                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2219                         mdname(mddev));
2220         err = 0;
2221 out:
2222         return err;
2223 }
2224
2225 static void autorun_array(mddev_t *mddev)
2226 {
2227         mdk_rdev_t *rdev;
2228         struct list_head *tmp;
2229         int err;
2230
2231         if (list_empty(&mddev->disks))
2232                 return;
2233
2234         printk(KERN_INFO "md: running: ");
2235
2236         ITERATE_RDEV(mddev,rdev,tmp) {
2237                 char b[BDEVNAME_SIZE];
2238                 printk("<%s>", bdevname(rdev->bdev,b));
2239         }
2240         printk("\n");
2241
2242         err = do_md_run (mddev);
2243         if (err) {
2244                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2245                 do_md_stop (mddev, 0);
2246         }
2247 }
2248
2249 /*
2250  * lets try to run arrays based on all disks that have arrived
2251  * until now. (those are in pending_raid_disks)
2252  *
2253  * the method: pick the first pending disk, collect all disks with
2254  * the same UUID, remove all from the pending list and put them into
2255  * the 'same_array' list. Then order this list based on superblock
2256  * update time (freshest comes first), kick out 'old' disks and
2257  * compare superblocks. If everything's fine then run it.
2258  *
2259  * If "unit" is allocated, then bump its reference count
2260  */
2261 static void autorun_devices(int part)
2262 {
2263         struct list_head candidates;
2264         struct list_head *tmp;
2265         mdk_rdev_t *rdev0, *rdev;
2266         mddev_t *mddev;
2267         char b[BDEVNAME_SIZE];
2268
2269         printk(KERN_INFO "md: autorun ...\n");
2270         while (!list_empty(&pending_raid_disks)) {
2271                 dev_t dev;
2272                 rdev0 = list_entry(pending_raid_disks.next,
2273                                          mdk_rdev_t, same_set);
2274
2275                 printk(KERN_INFO "md: considering %s ...\n",
2276                         bdevname(rdev0->bdev,b));
2277                 INIT_LIST_HEAD(&candidates);
2278                 ITERATE_RDEV_PENDING(rdev,tmp)
2279                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2280                                 printk(KERN_INFO "md:  adding %s ...\n",
2281                                         bdevname(rdev->bdev,b));
2282                                 list_move(&rdev->same_set, &candidates);
2283                         }
2284                 /*
2285                  * now we have a set of devices, with all of them having
2286                  * mostly sane superblocks. It's time to allocate the
2287                  * mddev.
2288                  */
2289                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2290                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2291                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2292                         break;
2293                 }
2294                 if (part)
2295                         dev = MKDEV(mdp_major,
2296                                     rdev0->preferred_minor << MdpMinorShift);
2297                 else
2298                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2299
2300                 md_probe(dev, NULL, NULL);
2301                 mddev = mddev_find(dev);
2302                 if (!mddev) {
2303                         printk(KERN_ERR 
2304                                 "md: cannot allocate memory for md drive.\n");
2305                         break;
2306                 }
2307                 if (mddev_lock(mddev)) 
2308                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2309                                mdname(mddev));
2310                 else if (mddev->raid_disks || mddev->major_version
2311                          || !list_empty(&mddev->disks)) {
2312                         printk(KERN_WARNING 
2313                                 "md: %s already running, cannot run %s\n",
2314                                 mdname(mddev), bdevname(rdev0->bdev,b));
2315                         mddev_unlock(mddev);
2316                 } else {
2317                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2318                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2319                                 list_del_init(&rdev->same_set);
2320                                 if (bind_rdev_to_array(rdev, mddev))
2321                                         export_rdev(rdev);
2322                         }
2323                         autorun_array(mddev);
2324                         mddev_unlock(mddev);
2325                 }
2326                 /* on success, candidates will be empty, on error
2327                  * it won't...
2328                  */
2329                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2330                         export_rdev(rdev);
2331                 mddev_put(mddev);
2332         }
2333         printk(KERN_INFO "md: ... autorun DONE.\n");
2334 }
2335
2336 /*
2337  * import RAID devices based on one partition
2338  * if possible, the array gets run as well.
2339  */
2340
2341 static int autostart_array(dev_t startdev)
2342 {
2343         char b[BDEVNAME_SIZE];
2344         int err = -EINVAL, i;
2345         mdp_super_t *sb = NULL;
2346         mdk_rdev_t *start_rdev = NULL, *rdev;
2347
2348         start_rdev = md_import_device(startdev, 0, 0);
2349         if (IS_ERR(start_rdev))
2350                 return err;
2351
2352
2353         /* NOTE: this can only work for 0.90.0 superblocks */
2354         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2355         if (sb->major_version != 0 ||
2356             sb->minor_version != 90 ) {
2357                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2358                 export_rdev(start_rdev);
2359                 return err;
2360         }
2361
2362         if (test_bit(Faulty, &start_rdev->flags)) {
2363                 printk(KERN_WARNING 
2364                         "md: can not autostart based on faulty %s!\n",
2365                         bdevname(start_rdev->bdev,b));
2366                 export_rdev(start_rdev);
2367                 return err;
2368         }
2369         list_add(&start_rdev->same_set, &pending_raid_disks);
2370
2371         for (i = 0; i < MD_SB_DISKS; i++) {
2372                 mdp_disk_t *desc = sb->disks + i;
2373                 dev_t dev = MKDEV(desc->major, desc->minor);
2374
2375                 if (!dev)
2376                         continue;
2377                 if (dev == startdev)
2378                         continue;
2379                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2380                         continue;
2381                 rdev = md_import_device(dev, 0, 0);
2382                 if (IS_ERR(rdev))
2383                         continue;
2384
2385                 list_add(&rdev->same_set, &pending_raid_disks);
2386         }
2387
2388         /*
2389          * possibly return codes
2390          */
2391         autorun_devices(0);
2392         return 0;
2393
2394 }
2395
2396
2397 static int get_version(void __user * arg)
2398 {
2399         mdu_version_t ver;
2400
2401         ver.major = MD_MAJOR_VERSION;
2402         ver.minor = MD_MINOR_VERSION;
2403         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2404
2405         if (copy_to_user(arg, &ver, sizeof(ver)))
2406                 return -EFAULT;
2407
2408         return 0;
2409 }
2410
2411 static int get_array_info(mddev_t * mddev, void __user * arg)
2412 {
2413         mdu_array_info_t info;
2414         int nr,working,active,failed,spare;
2415         mdk_rdev_t *rdev;
2416         struct list_head *tmp;
2417
2418         nr=working=active=failed=spare=0;
2419         ITERATE_RDEV(mddev,rdev,tmp) {
2420                 nr++;
2421                 if (test_bit(Faulty, &rdev->flags))
2422                         failed++;
2423                 else {
2424                         working++;
2425                         if (test_bit(In_sync, &rdev->flags))
2426                                 active++;       
2427                         else
2428                                 spare++;
2429                 }
2430         }
2431
2432         info.major_version = mddev->major_version;
2433         info.minor_version = mddev->minor_version;
2434         info.patch_version = MD_PATCHLEVEL_VERSION;
2435         info.ctime         = mddev->ctime;
2436         info.level         = mddev->level;
2437         info.size          = mddev->size;
2438         info.nr_disks      = nr;
2439         info.raid_disks    = mddev->raid_disks;
2440         info.md_minor      = mddev->md_minor;
2441         info.not_persistent= !mddev->persistent;
2442
2443         info.utime         = mddev->utime;
2444         info.state         = 0;
2445         if (mddev->in_sync)
2446                 info.state = (1<<MD_SB_CLEAN);
2447         if (mddev->bitmap && mddev->bitmap_offset)
2448                 info.state = (1<<MD_SB_BITMAP_PRESENT);
2449         info.active_disks  = active;
2450         info.working_disks = working;
2451         info.failed_disks  = failed;
2452         info.spare_disks   = spare;
2453
2454         info.layout        = mddev->layout;
2455         info.chunk_size    = mddev->chunk_size;
2456
2457         if (copy_to_user(arg, &info, sizeof(info)))
2458                 return -EFAULT;
2459
2460         return 0;
2461 }
2462
2463 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
2464 {
2465         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
2466         char *ptr, *buf = NULL;
2467         int err = -ENOMEM;
2468
2469         file = kmalloc(sizeof(*file), GFP_KERNEL);
2470         if (!file)
2471                 goto out;
2472
2473         /* bitmap disabled, zero the first byte and copy out */
2474         if (!mddev->bitmap || !mddev->bitmap->file) {
2475                 file->pathname[0] = '\0';
2476                 goto copy_out;
2477         }
2478
2479         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
2480         if (!buf)
2481                 goto out;
2482
2483         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
2484         if (!ptr)
2485                 goto out;
2486
2487         strcpy(file->pathname, ptr);
2488
2489 copy_out:
2490         err = 0;
2491         if (copy_to_user(arg, file, sizeof(*file)))
2492                 err = -EFAULT;
2493 out:
2494         kfree(buf);
2495         kfree(file);
2496         return err;
2497 }
2498
2499 static int get_disk_info(mddev_t * mddev, void __user * arg)
2500 {
2501         mdu_disk_info_t info;
2502         unsigned int nr;
2503         mdk_rdev_t *rdev;
2504
2505         if (copy_from_user(&info, arg, sizeof(info)))
2506                 return -EFAULT;
2507
2508         nr = info.number;
2509
2510         rdev = find_rdev_nr(mddev, nr);
2511         if (rdev) {
2512                 info.major = MAJOR(rdev->bdev->bd_dev);
2513                 info.minor = MINOR(rdev->bdev->bd_dev);
2514                 info.raid_disk = rdev->raid_disk;
2515                 info.state = 0;
2516                 if (test_bit(Faulty, &rdev->flags))
2517                         info.state |= (1<<MD_DISK_FAULTY);
2518                 else if (test_bit(In_sync, &rdev->flags)) {
2519                         info.state |= (1<<MD_DISK_ACTIVE);
2520                         info.state |= (1<<MD_DISK_SYNC);
2521                 }
2522                 if (test_bit(WriteMostly, &rdev->flags))
2523                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
2524         } else {
2525                 info.major = info.minor = 0;
2526                 info.raid_disk = -1;
2527                 info.state = (1<<MD_DISK_REMOVED);
2528         }
2529
2530         if (copy_to_user(arg, &info, sizeof(info)))
2531                 return -EFAULT;
2532
2533         return 0;
2534 }
2535
2536 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
2537 {
2538         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
2539         mdk_rdev_t *rdev;
2540         dev_t dev = MKDEV(info->major,info->minor);
2541
2542         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
2543                 return -EOVERFLOW;
2544
2545         if (!mddev->raid_disks) {
2546                 int err;
2547                 /* expecting a device which has a superblock */
2548                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
2549                 if (IS_ERR(rdev)) {
2550                         printk(KERN_WARNING 
2551                                 "md: md_import_device returned %ld\n",
2552                                 PTR_ERR(rdev));
2553                         return PTR_ERR(rdev);
2554                 }
2555                 if (!list_empty(&mddev->disks)) {
2556                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2557                                                         mdk_rdev_t, same_set);
2558                         int err = super_types[mddev->major_version]
2559                                 .load_super(rdev, rdev0, mddev->minor_version);
2560                         if (err < 0) {
2561                                 printk(KERN_WARNING 
2562                                         "md: %s has different UUID to %s\n",
2563                                         bdevname(rdev->bdev,b), 
2564                                         bdevname(rdev0->bdev,b2));
2565                                 export_rdev(rdev);
2566                                 return -EINVAL;
2567                         }
2568                 }
2569                 err = bind_rdev_to_array(rdev, mddev);
2570                 if (err)
2571                         export_rdev(rdev);
2572                 return err;
2573         }
2574
2575         /*
2576          * add_new_disk can be used once the array is assembled
2577          * to add "hot spares".  They must already have a superblock
2578          * written
2579          */
2580         if (mddev->pers) {
2581                 int err;
2582                 if (!mddev->pers->hot_add_disk) {
2583                         printk(KERN_WARNING 
2584                                 "%s: personality does not support diskops!\n",
2585                                mdname(mddev));
2586                         return -EINVAL;
2587                 }
2588                 if (mddev->persistent)
2589                         rdev = md_import_device(dev, mddev->major_version,
2590                                                 mddev->minor_version);
2591                 else
2592                         rdev = md_import_device(dev, -1, -1);
2593                 if (IS_ERR(rdev)) {
2594                         printk(KERN_WARNING 
2595                                 "md: md_import_device returned %ld\n",
2596                                 PTR_ERR(rdev));
2597                         return PTR_ERR(rdev);
2598                 }
2599                 /* set save_raid_disk if appropriate */
2600                 if (!mddev->persistent) {
2601                         if (info->state & (1<<MD_DISK_SYNC)  &&
2602                             info->raid_disk < mddev->raid_disks)
2603                                 rdev->raid_disk = info->raid_disk;
2604                         else
2605                                 rdev->raid_disk = -1;
2606                 } else
2607                         super_types[mddev->major_version].
2608                                 validate_super(mddev, rdev);
2609                 rdev->saved_raid_disk = rdev->raid_disk;
2610
2611                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
2612                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2613                         set_bit(WriteMostly, &rdev->flags);
2614
2615                 rdev->raid_disk = -1;
2616                 err = bind_rdev_to_array(rdev, mddev);
2617                 if (err)
2618                         export_rdev(rdev);
2619
2620                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2621                 md_wakeup_thread(mddev->thread);
2622                 return err;
2623         }
2624
2625         /* otherwise, add_new_disk is only allowed
2626          * for major_version==0 superblocks
2627          */
2628         if (mddev->major_version != 0) {
2629                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
2630                        mdname(mddev));
2631                 return -EINVAL;
2632         }
2633
2634         if (!(info->state & (1<<MD_DISK_FAULTY))) {
2635                 int err;
2636                 rdev = md_import_device (dev, -1, 0);
2637                 if (IS_ERR(rdev)) {
2638                         printk(KERN_WARNING 
2639                                 "md: error, md_import_device() returned %ld\n",
2640                                 PTR_ERR(rdev));
2641                         return PTR_ERR(rdev);
2642                 }
2643                 rdev->desc_nr = info->number;
2644                 if (info->raid_disk < mddev->raid_disks)
2645                         rdev->raid_disk = info->raid_disk;
2646                 else
2647                         rdev->raid_disk = -1;
2648
2649                 rdev->flags = 0;
2650
2651                 if (rdev->raid_disk < mddev->raid_disks)
2652                         if (info->state & (1<<MD_DISK_SYNC))
2653                                 set_bit(In_sync, &rdev->flags);
2654
2655                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
2656                         set_bit(WriteMostly, &rdev->flags);
2657
2658                 err = bind_rdev_to_array(rdev, mddev);
2659                 if (err) {
2660                         export_rdev(rdev);
2661                         return err;
2662                 }
2663
2664                 if (!mddev->persistent) {
2665                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
2666                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2667                 } else 
2668                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2669                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
2670
2671                 if (!mddev->size || (mddev->size > rdev->size))
2672                         mddev->size = rdev->size;
2673         }
2674
2675         return 0;
2676 }
2677
2678 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
2679 {
2680         char b[BDEVNAME_SIZE];
2681         mdk_rdev_t *rdev;
2682
2683         if (!mddev->pers)
2684                 return -ENODEV;
2685
2686         rdev = find_rdev(mddev, dev);
2687         if (!rdev)
2688                 return -ENXIO;
2689
2690         if (rdev->raid_disk >= 0)
2691                 goto busy;
2692
2693         kick_rdev_from_array(rdev);
2694         md_update_sb(mddev);
2695
2696         return 0;
2697 busy:
2698         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
2699                 bdevname(rdev->bdev,b), mdname(mddev));
2700         return -EBUSY;
2701 }
2702
2703 static int hot_add_disk(mddev_t * mddev, dev_t dev)
2704 {
2705         char b[BDEVNAME_SIZE];
2706         int err;
2707         unsigned int size;
2708         mdk_rdev_t *rdev;
2709
2710         if (!mddev->pers)
2711                 return -ENODEV;
2712
2713         if (mddev->major_version != 0) {
2714                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
2715                         " version-0 superblocks.\n",
2716                         mdname(mddev));
2717                 return -EINVAL;
2718         }
2719         if (!mddev->pers->hot_add_disk) {
2720                 printk(KERN_WARNING 
2721                         "%s: personality does not support diskops!\n",
2722                         mdname(mddev));
2723                 return -EINVAL;
2724         }
2725
2726         rdev = md_import_device (dev, -1, 0);
2727         if (IS_ERR(rdev)) {
2728                 printk(KERN_WARNING 
2729                         "md: error, md_import_device() returned %ld\n",
2730                         PTR_ERR(rdev));
2731                 return -EINVAL;
2732         }
2733
2734         if (mddev->persistent)
2735                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
2736         else
2737                 rdev->sb_offset =
2738                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2739
2740         size = calc_dev_size(rdev, mddev->chunk_size);
2741         rdev->size = size;
2742
2743         if (size < mddev->size) {
2744                 printk(KERN_WARNING 
2745                         "%s: disk size %llu blocks < array size %llu\n",
2746                         mdname(mddev), (unsigned long long)size,
2747                         (unsigned long long)mddev->size);
2748                 err = -ENOSPC;
2749                 goto abort_export;
2750         }
2751
2752         if (test_bit(Faulty, &rdev->flags)) {
2753                 printk(KERN_WARNING 
2754                         "md: can not hot-add faulty %s disk to %s!\n",
2755                         bdevname(rdev->bdev,b), mdname(mddev));
2756                 err = -EINVAL;
2757                 goto abort_export;
2758         }
2759         clear_bit(In_sync, &rdev->flags);
2760         rdev->desc_nr = -1;
2761         bind_rdev_to_array(rdev, mddev);
2762
2763         /*
2764          * The rest should better be atomic, we can have disk failures
2765          * noticed in interrupt contexts ...
2766          */
2767
2768         if (rdev->desc_nr == mddev->max_disks) {
2769                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
2770                         mdname(mddev));
2771                 err = -EBUSY;
2772                 goto abort_unbind_export;
2773         }
2774
2775         rdev->raid_disk = -1;
2776
2777         md_update_sb(mddev);
2778
2779         /*
2780          * Kick recovery, maybe this spare has to be added to the
2781          * array immediately.
2782          */
2783         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2784         md_wakeup_thread(mddev->thread);
2785
2786         return 0;
2787
2788 abort_unbind_export:
2789         unbind_rdev_from_array(rdev);
2790
2791 abort_export:
2792         export_rdev(rdev);
2793         return err;
2794 }
2795
2796 /* similar to deny_write_access, but accounts for our holding a reference
2797  * to the file ourselves */
2798 static int deny_bitmap_write_access(struct file * file)
2799 {
2800         struct inode *inode = file->f_mapping->host;
2801
2802         spin_lock(&inode->i_lock);
2803         if (atomic_read(&inode->i_writecount) > 1) {
2804                 spin_unlock(&inode->i_lock);
2805                 return -ETXTBSY;
2806         }
2807         atomic_set(&inode->i_writecount, -1);
2808         spin_unlock(&inode->i_lock);
2809
2810         return 0;
2811 }
2812
2813 static int set_bitmap_file(mddev_t *mddev, int fd)
2814 {
2815         int err;
2816
2817         if (mddev->pers) {
2818                 if (!mddev->pers->quiesce)
2819                         return -EBUSY;
2820                 if (mddev->recovery || mddev->sync_thread)
2821                         return -EBUSY;
2822                 /* we should be able to change the bitmap.. */
2823         }
2824
2825
2826         if (fd >= 0) {
2827                 if (mddev->bitmap)
2828                         return -EEXIST; /* cannot add when bitmap is present */
2829                 mddev->bitmap_file = fget(fd);
2830
2831                 if (mddev->bitmap_file == NULL) {
2832                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
2833                                mdname(mddev));
2834                         return -EBADF;
2835                 }
2836
2837                 err = deny_bitmap_write_access(mddev->bitmap_file);
2838                 if (err) {
2839                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
2840                                mdname(mddev));
2841                         fput(mddev->bitmap_file);
2842                         mddev->bitmap_file = NULL;
2843                         return err;
2844                 }
2845                 mddev->bitmap_offset = 0; /* file overrides offset */
2846         } else if (mddev->bitmap == NULL)
2847                 return -ENOENT; /* cannot remove what isn't there */
2848         err = 0;
2849         if (mddev->pers) {
2850                 mddev->pers->quiesce(mddev, 1);
2851                 if (fd >= 0)
2852                         err = bitmap_create(mddev);
2853                 if (fd < 0 || err)
2854                         bitmap_destroy(mddev);
2855                 mddev->pers->quiesce(mddev, 0);
2856         } else if (fd < 0) {
2857                 if (mddev->bitmap_file)
2858                         fput(mddev->bitmap_file);
2859                 mddev->bitmap_file = NULL;
2860         }
2861
2862         return err;
2863 }
2864
2865 /*
2866  * set_array_info is used two different ways
2867  * The original usage is when creating a new array.
2868  * In this usage, raid_disks is > 0 and it together with
2869  *  level, size, not_persistent,layout,chunksize determine the
2870  *  shape of the array.
2871  *  This will always create an array with a type-0.90.0 superblock.
2872  * The newer usage is when assembling an array.
2873  *  In this case raid_disks will be 0, and the major_version field is
2874  *  use to determine which style super-blocks are to be found on the devices.
2875  *  The minor and patch _version numbers are also kept incase the
2876  *  super_block handler wishes to interpret them.
2877  */
2878 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
2879 {
2880
2881         if (info->raid_disks == 0) {
2882                 /* just setting version number for superblock loading */
2883                 if (info->major_version < 0 ||
2884                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
2885                     super_types[info->major_version].name == NULL) {
2886                         /* maybe try to auto-load a module? */
2887                         printk(KERN_INFO 
2888                                 "md: superblock version %d not known\n",
2889                                 info->major_version);
2890                         return -EINVAL;
2891                 }
2892                 mddev->major_version = info->major_version;
2893                 mddev->minor_version = info->minor_version;
2894                 mddev->patch_version = info->patch_version;
2895                 return 0;
2896         }
2897         mddev->major_version = MD_MAJOR_VERSION;
2898         mddev->minor_version = MD_MINOR_VERSION;
2899         mddev->patch_version = MD_PATCHLEVEL_VERSION;
2900         mddev->ctime         = get_seconds();
2901
2902         mddev->level         = info->level;
2903         mddev->size          = info->size;
2904         mddev->raid_disks    = info->raid_disks;
2905         /* don't set md_minor, it is determined by which /dev/md* was
2906          * openned
2907          */
2908         if (info->state & (1<<MD_SB_CLEAN))
2909                 mddev->recovery_cp = MaxSector;
2910         else
2911                 mddev->recovery_cp = 0;
2912         mddev->persistent    = ! info->not_persistent;
2913
2914         mddev->layout        = info->layout;
2915         mddev->chunk_size    = info->chunk_size;
2916
2917         mddev->max_disks     = MD_SB_DISKS;
2918
2919         mddev->sb_dirty      = 1;
2920
2921         /*
2922          * Generate a 128 bit UUID
2923          */
2924         get_random_bytes(mddev->uuid, 16);
2925
2926         return 0;
2927 }
2928
2929 /*
2930  * update_array_info is used to change the configuration of an
2931  * on-line array.
2932  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
2933  * fields in the info are checked against the array.
2934  * Any differences that cannot be handled will cause an error.
2935  * Normally, only one change can be managed at a time.
2936  */
2937 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
2938 {
2939         int rv = 0;
2940         int cnt = 0;
2941         int state = 0;
2942
2943         /* calculate expected state,ignoring low bits */
2944         if (mddev->bitmap && mddev->bitmap_offset)
2945                 state |= (1 << MD_SB_BITMAP_PRESENT);
2946
2947         if (mddev->major_version != info->major_version ||
2948             mddev->minor_version != info->minor_version ||
2949 /*          mddev->patch_version != info->patch_version || */
2950             mddev->ctime         != info->ctime         ||
2951             mddev->level         != info->level         ||
2952 /*          mddev->layout        != info->layout        || */
2953             !mddev->persistent   != info->not_persistent||
2954             mddev->chunk_size    != info->chunk_size    ||
2955             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
2956             ((state^info->state) & 0xfffffe00)
2957                 )
2958                 return -EINVAL;
2959         /* Check there is only one change */
2960         if (mddev->size != info->size) cnt++;
2961         if (mddev->raid_disks != info->raid_disks) cnt++;
2962         if (mddev->layout != info->layout) cnt++;
2963         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
2964         if (cnt == 0) return 0;
2965         if (cnt > 1) return -EINVAL;
2966
2967         if (mddev->layout != info->layout) {
2968                 /* Change layout
2969                  * we don't need to do anything at the md level, the
2970                  * personality will take care of it all.
2971                  */
2972                 if (mddev->pers->reconfig == NULL)
2973                         return -EINVAL;
2974                 else
2975                         return mddev->pers->reconfig(mddev, info->layout, -1);
2976         }
2977         if (mddev->size != info->size) {
2978                 mdk_rdev_t * rdev;
2979                 struct list_head *tmp;
2980                 if (mddev->pers->resize == NULL)
2981                         return -EINVAL;
2982                 /* The "size" is the amount of each device that is used.
2983                  * This can only make sense for arrays with redundancy.
2984                  * linear and raid0 always use whatever space is available
2985                  * We can only consider changing the size if no resync
2986                  * or reconstruction is happening, and if the new size
2987                  * is acceptable. It must fit before the sb_offset or,
2988                  * if that is <data_offset, it must fit before the
2989                  * size of each device.
2990                  * If size is zero, we find the largest size that fits.
2991                  */
2992                 if (mddev->sync_thread)
2993                         return -EBUSY;
2994                 ITERATE_RDEV(mddev,rdev,tmp) {
2995                         sector_t avail;
2996                         int fit = (info->size == 0);
2997                         if (rdev->sb_offset > rdev->data_offset)
2998                                 avail = (rdev->sb_offset*2) - rdev->data_offset;
2999                         else
3000                                 avail = get_capacity(rdev->bdev->bd_disk)
3001                                         - rdev->data_offset;
3002                         if (fit && (info->size == 0 || info->size > avail/2))
3003                                 info->size = avail/2;
3004                         if (avail < ((sector_t)info->size << 1))
3005                                 return -ENOSPC;
3006                 }
3007                 rv = mddev->pers->resize(mddev, (sector_t)info->size *2);
3008                 if (!rv) {
3009                         struct block_device *bdev;
3010
3011                         bdev = bdget_disk(mddev->gendisk, 0);
3012                         if (bdev) {
3013                                 down(&bdev->bd_inode->i_sem);
3014                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3015                                 up(&bdev->bd_inode->i_sem);
3016                                 bdput(bdev);
3017                         }
3018                 }
3019         }
3020         if (mddev->raid_disks    != info->raid_disks) {
3021                 /* change the number of raid disks */
3022                 if (mddev->pers->reshape == NULL)
3023                         return -EINVAL;
3024                 if (info->raid_disks <= 0 ||
3025                     info->raid_disks >= mddev->max_disks)
3026                         return -EINVAL;
3027                 if (mddev->sync_thread)
3028                         return -EBUSY;
3029                 rv = mddev->pers->reshape(mddev, info->raid_disks);
3030                 if (!rv) {
3031                         struct block_device *bdev;
3032
3033                         bdev = bdget_disk(mddev->gendisk, 0);
3034                         if (bdev) {
3035                                 down(&bdev->bd_inode->i_sem);
3036                                 i_size_write(bdev->bd_inode, mddev->array_size << 10);
3037                                 up(&bdev->bd_inode->i_sem);
3038                                 bdput(bdev);
3039                         }
3040                 }
3041         }
3042         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3043                 if (mddev->pers->quiesce == NULL)
3044                         return -EINVAL;
3045                 if (mddev->recovery || mddev->sync_thread)
3046                         return -EBUSY;
3047                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3048                         /* add the bitmap */
3049                         if (mddev->bitmap)
3050                                 return -EEXIST;
3051                         if (mddev->default_bitmap_offset == 0)
3052                                 return -EINVAL;
3053                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3054                         mddev->pers->quiesce(mddev, 1);
3055                         rv = bitmap_create(mddev);
3056                         if (rv)
3057                                 bitmap_destroy(mddev);
3058                         mddev->pers->quiesce(mddev, 0);
3059                 } else {
3060                         /* remove the bitmap */
3061                         if (!mddev->bitmap)
3062                                 return -ENOENT;
3063                         if (mddev->bitmap->file)
3064                                 return -EINVAL;
3065                         mddev->pers->quiesce(mddev, 1);
3066                         bitmap_destroy(mddev);
3067                         mddev->pers->quiesce(mddev, 0);
3068                         mddev->bitmap_offset = 0;
3069                 }
3070         }
3071         md_update_sb(mddev);
3072         return rv;
3073 }
3074
3075 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3076 {
3077         mdk_rdev_t *rdev;
3078
3079         if (mddev->pers == NULL)
3080                 return -ENODEV;
3081
3082         rdev = find_rdev(mddev, dev);
3083         if (!rdev)
3084                 return -ENODEV;
3085
3086         md_error(mddev, rdev);
3087         return 0;
3088 }
3089
3090 static int md_ioctl(struct inode *inode, struct file *file,
3091                         unsigned int cmd, unsigned long arg)
3092 {
3093         int err = 0;
3094         void __user *argp = (void __user *)arg;
3095         struct hd_geometry __user *loc = argp;
3096         mddev_t *mddev = NULL;
3097
3098         if (!capable(CAP_SYS_ADMIN))
3099                 return -EACCES;
3100
3101         /*
3102          * Commands dealing with the RAID driver but not any
3103          * particular array:
3104          */
3105         switch (cmd)
3106         {
3107                 case RAID_VERSION:
3108                         err = get_version(argp);
3109                         goto done;
3110
3111                 case PRINT_RAID_DEBUG:
3112                         err = 0;
3113                         md_print_devices();
3114                         goto done;
3115
3116 #ifndef MODULE
3117                 case RAID_AUTORUN:
3118                         err = 0;
3119                         autostart_arrays(arg);
3120                         goto done;
3121 #endif
3122                 default:;
3123         }
3124
3125         /*
3126          * Commands creating/starting a new array:
3127          */
3128
3129         mddev = inode->i_bdev->bd_disk->private_data;
3130
3131         if (!mddev) {
3132                 BUG();
3133                 goto abort;
3134         }
3135
3136
3137         if (cmd == START_ARRAY) {
3138                 /* START_ARRAY doesn't need to lock the array as autostart_array
3139                  * does the locking, and it could even be a different array
3140                  */
3141                 static int cnt = 3;
3142                 if (cnt > 0 ) {
3143                         printk(KERN_WARNING
3144                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3145                                "This will not be supported beyond 2.6\n",
3146                                current->comm, current->pid);
3147                         cnt--;
3148                 }
3149                 err = autostart_array(new_decode_dev(arg));
3150                 if (err) {
3151                         printk(KERN_WARNING "md: autostart failed!\n");
3152                         goto abort;
3153                 }
3154                 goto done;
3155         }
3156
3157         err = mddev_lock(mddev);
3158         if (err) {
3159                 printk(KERN_INFO 
3160                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3161                         err, cmd);
3162                 goto abort;
3163         }
3164
3165         switch (cmd)
3166         {
3167                 case SET_ARRAY_INFO:
3168                         {
3169                                 mdu_array_info_t info;
3170                                 if (!arg)
3171                                         memset(&info, 0, sizeof(info));
3172                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3173                                         err = -EFAULT;
3174                                         goto abort_unlock;
3175                                 }
3176                                 if (mddev->pers) {
3177                                         err = update_array_info(mddev, &info);
3178                                         if (err) {
3179                                                 printk(KERN_WARNING "md: couldn't update"
3180                                                        " array info. %d\n", err);
3181                                                 goto abort_unlock;
3182                                         }
3183                                         goto done_unlock;
3184                                 }
3185                                 if (!list_empty(&mddev->disks)) {
3186                                         printk(KERN_WARNING
3187                                                "md: array %s already has disks!\n",
3188                                                mdname(mddev));
3189                                         err = -EBUSY;
3190                                         goto abort_unlock;
3191                                 }
3192                                 if (mddev->raid_disks) {
3193                                         printk(KERN_WARNING
3194                                                "md: array %s already initialised!\n",
3195                                                mdname(mddev));
3196                                         err = -EBUSY;
3197                                         goto abort_unlock;
3198                                 }
3199                                 err = set_array_info(mddev, &info);
3200                                 if (err) {
3201                                         printk(KERN_WARNING "md: couldn't set"
3202                                                " array info. %d\n", err);
3203                                         goto abort_unlock;
3204                                 }
3205                         }
3206                         goto done_unlock;
3207
3208                 default:;
3209         }
3210
3211         /*
3212          * Commands querying/configuring an existing array:
3213          */
3214         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3215          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3216         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3217                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3218                 err = -ENODEV;
3219                 goto abort_unlock;
3220         }
3221
3222         /*
3223          * Commands even a read-only array can execute:
3224          */
3225         switch (cmd)
3226         {
3227                 case GET_ARRAY_INFO:
3228                         err = get_array_info(mddev, argp);
3229                         goto done_unlock;
3230
3231                 case GET_BITMAP_FILE:
3232                         err = get_bitmap_file(mddev, argp);
3233                         goto done_unlock;
3234
3235                 case GET_DISK_INFO:
3236                         err = get_disk_info(mddev, argp);
3237                         goto done_unlock;
3238
3239                 case RESTART_ARRAY_RW:
3240                         err = restart_array(mddev);
3241                         goto done_unlock;
3242
3243                 case STOP_ARRAY:
3244                         err = do_md_stop (mddev, 0);
3245                         goto done_unlock;
3246
3247                 case STOP_ARRAY_RO:
3248                         err = do_md_stop (mddev, 1);
3249                         goto done_unlock;
3250
3251         /*
3252          * We have a problem here : there is no easy way to give a CHS
3253          * virtual geometry. We currently pretend that we have a 2 heads
3254          * 4 sectors (with a BIG number of cylinders...). This drives
3255          * dosfs just mad... ;-)
3256          */
3257                 case HDIO_GETGEO:
3258                         if (!loc) {
3259                                 err = -EINVAL;
3260                                 goto abort_unlock;
3261                         }
3262                         err = put_user (2, (char __user *) &loc->heads);
3263                         if (err)
3264                                 goto abort_unlock;
3265                         err = put_user (4, (char __user *) &loc->sectors);
3266                         if (err)
3267                                 goto abort_unlock;
3268                         err = put_user(get_capacity(mddev->gendisk)/8,
3269                                         (short __user *) &loc->cylinders);
3270                         if (err)
3271                                 goto abort_unlock;
3272                         err = put_user (get_start_sect(inode->i_bdev),
3273                                                 (long __user *) &loc->start);
3274                         goto done_unlock;
3275         }
3276
3277         /*
3278          * The remaining ioctls are changing the state of the
3279          * superblock, so we do not allow them on read-only arrays.
3280          * However non-MD ioctls (e.g. get-size) will still come through
3281          * here and hit the 'default' below, so only disallow
3282          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3283          */
3284         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3285             mddev->ro && mddev->pers) {
3286                 if (mddev->ro == 2) {
3287                         mddev->ro = 0;
3288                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3289                 md_wakeup_thread(mddev->thread);
3290
3291                 } else {
3292                         err = -EROFS;
3293                         goto abort_unlock;
3294                 }
3295         }
3296
3297         switch (cmd)
3298         {
3299                 case ADD_NEW_DISK:
3300                 {
3301                         mdu_disk_info_t info;
3302                         if (copy_from_user(&info, argp, sizeof(info)))
3303                                 err = -EFAULT;
3304                         else
3305                                 err = add_new_disk(mddev, &info);
3306                         goto done_unlock;
3307                 }
3308
3309                 case HOT_REMOVE_DISK:
3310                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3311                         goto done_unlock;
3312
3313                 case HOT_ADD_DISK:
3314                         err = hot_add_disk(mddev, new_decode_dev(arg));
3315                         goto done_unlock;
3316
3317                 case SET_DISK_FAULTY:
3318                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3319                         goto done_unlock;
3320
3321                 case RUN_ARRAY:
3322                         err = do_md_run (mddev);
3323                         goto done_unlock;
3324
3325                 case SET_BITMAP_FILE:
3326                         err = set_bitmap_file(mddev, (int)arg);
3327                         goto done_unlock;
3328
3329                 default:
3330                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3331                                 printk(KERN_WARNING "md: %s(pid %d) used"
3332                                         " obsolete MD ioctl, upgrade your"
3333                                         " software to use new ictls.\n",
3334                                         current->comm, current->pid);
3335                         err = -EINVAL;
3336                         goto abort_unlock;
3337         }
3338
3339 done_unlock:
3340 abort_unlock:
3341         mddev_unlock(mddev);
3342
3343         return err;
3344 done:
3345         if (err)
3346                 MD_BUG();
3347 abort:
3348         return err;
3349 }
3350
3351 static int md_open(struct inode *inode, struct file *file)
3352 {
3353         /*
3354          * Succeed if we can lock the mddev, which confirms that
3355          * it isn't being stopped right now.
3356          */
3357         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3358         int err;
3359
3360         if ((err = mddev_lock(mddev)))
3361                 goto out;
3362
3363         err = 0;
3364         mddev_get(mddev);
3365         mddev_unlock(mddev);
3366
3367         check_disk_change(inode->i_bdev);
3368  out:
3369         return err;
3370 }
3371
3372 static int md_release(struct inode *inode, struct file * file)
3373 {
3374         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3375
3376         if (!mddev)
3377                 BUG();
3378         mddev_put(mddev);
3379
3380         return 0;
3381 }
3382
3383 static int md_media_changed(struct gendisk *disk)
3384 {
3385         mddev_t *mddev = disk->private_data;
3386
3387         return mddev->changed;
3388 }
3389
3390 static int md_revalidate(struct gendisk *disk)
3391 {
3392         mddev_t *mddev = disk->private_data;
3393
3394         mddev->changed = 0;
3395         return 0;
3396 }
3397 static struct block_device_operations md_fops =
3398 {
3399         .owner          = THIS_MODULE,
3400         .open           = md_open,
3401         .release        = md_release,
3402         .ioctl          = md_ioctl,
3403         .media_changed  = md_media_changed,
3404         .revalidate_disk= md_revalidate,
3405 };
3406
3407 static int md_thread(void * arg)
3408 {
3409         mdk_thread_t *thread = arg;
3410
3411         /*
3412          * md_thread is a 'system-thread', it's priority should be very
3413          * high. We avoid resource deadlocks individually in each
3414          * raid personality. (RAID5 does preallocation) We also use RR and
3415          * the very same RT priority as kswapd, thus we will never get
3416          * into a priority inversion deadlock.
3417          *
3418          * we definitely have to have equal or higher priority than
3419          * bdflush, otherwise bdflush will deadlock if there are too
3420          * many dirty RAID5 blocks.
3421          */
3422
3423         allow_signal(SIGKILL);
3424         complete(thread->event);
3425         while (!kthread_should_stop()) {
3426                 void (*run)(mddev_t *);
3427
3428                 wait_event_interruptible_timeout(thread->wqueue,
3429                                                  test_bit(THREAD_WAKEUP, &thread->flags)
3430                                                  || kthread_should_stop(),
3431                                                  thread->timeout);
3432                 try_to_freeze();
3433
3434                 clear_bit(THREAD_WAKEUP, &thread->flags);
3435
3436                 run = thread->run;
3437                 if (run)
3438                         run(thread->mddev);
3439         }
3440
3441         return 0;
3442 }
3443
3444 void md_wakeup_thread(mdk_thread_t *thread)
3445 {
3446         if (thread) {
3447                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
3448                 set_bit(THREAD_WAKEUP, &thread->flags);
3449                 wake_up(&thread->wqueue);
3450         }
3451 }
3452
3453 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
3454                                  const char *name)
3455 {
3456         mdk_thread_t *thread;
3457         struct completion event;
3458
3459         thread = kmalloc(sizeof(mdk_thread_t), GFP_KERNEL);
3460         if (!thread)
3461                 return NULL;
3462
3463         memset(thread, 0, sizeof(mdk_thread_t));
3464         init_waitqueue_head(&thread->wqueue);
3465
3466         init_completion(&event);
3467         thread->event = &event;
3468         thread->run = run;
3469         thread->mddev = mddev;
3470         thread->name = name;
3471         thread->timeout = MAX_SCHEDULE_TIMEOUT;
3472         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
3473         if (IS_ERR(thread->tsk)) {
3474                 kfree(thread);
3475                 return NULL;
3476         }
3477         wait_for_completion(&event);
3478         return thread;
3479 }
3480
3481 void md_unregister_thread(mdk_thread_t *thread)
3482 {
3483         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
3484
3485         kthread_stop(thread->tsk);
3486         kfree(thread);
3487 }
3488
3489 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
3490 {
3491         if (!mddev) {
3492                 MD_BUG();
3493                 return;
3494         }
3495
3496         if (!rdev || test_bit(Faulty, &rdev->flags))
3497                 return;
3498 /*
3499         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
3500                 mdname(mddev),
3501                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
3502                 __builtin_return_address(0),__builtin_return_address(1),
3503                 __builtin_return_address(2),__builtin_return_address(3));
3504 */
3505         if (!mddev->pers->error_handler)
3506                 return;
3507         mddev->pers->error_handler(mddev,rdev);
3508         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3509         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3510         md_wakeup_thread(mddev->thread);
3511 }
3512
3513 /* seq_file implementation /proc/mdstat */
3514
3515 static void status_unused(struct seq_file *seq)
3516 {
3517         int i = 0;
3518         mdk_rdev_t *rdev;
3519         struct list_head *tmp;
3520
3521         seq_printf(seq, "unused devices: ");
3522
3523         ITERATE_RDEV_PENDING(rdev,tmp) {
3524                 char b[BDEVNAME_SIZE];
3525                 i++;
3526                 seq_printf(seq, "%s ",
3527                               bdevname(rdev->bdev,b));
3528         }
3529         if (!i)
3530                 seq_printf(seq, "<none>");
3531
3532         seq_printf(seq, "\n");
3533 }
3534
3535
3536 static void status_resync(struct seq_file *seq, mddev_t * mddev)
3537 {
3538         unsigned long max_blocks, resync, res, dt, db, rt;
3539
3540         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
3541
3542         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3543                 max_blocks = mddev->resync_max_sectors >> 1;
3544         else
3545                 max_blocks = mddev->size;
3546
3547         /*
3548          * Should not happen.
3549          */
3550         if (!max_blocks) {
3551                 MD_BUG();
3552                 return;
3553         }
3554         res = (resync/1024)*1000/(max_blocks/1024 + 1);
3555         {
3556                 int i, x = res/50, y = 20-x;
3557                 seq_printf(seq, "[");
3558                 for (i = 0; i < x; i++)
3559                         seq_printf(seq, "=");
3560                 seq_printf(seq, ">");
3561                 for (i = 0; i < y; i++)
3562                         seq_printf(seq, ".");
3563                 seq_printf(seq, "] ");
3564         }
3565         seq_printf(seq, " %s =%3lu.%lu%% (%lu/%lu)",
3566                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
3567                        "resync" : "recovery"),
3568                       res/10, res % 10, resync, max_blocks);
3569
3570         /*
3571          * We do not want to overflow, so the order of operands and
3572          * the * 100 / 100 trick are important. We do a +1 to be
3573          * safe against division by zero. We only estimate anyway.
3574          *
3575          * dt: time from mark until now
3576          * db: blocks written from mark until now
3577          * rt: remaining time
3578          */
3579         dt = ((jiffies - mddev->resync_mark) / HZ);
3580         if (!dt) dt++;
3581         db = resync - (mddev->resync_mark_cnt/2);
3582         rt = (dt * ((max_blocks-resync) / (db/100+1)))/100;
3583
3584         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
3585
3586         seq_printf(seq, " speed=%ldK/sec", db/dt);
3587 }
3588
3589 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
3590 {
3591         struct list_head *tmp;
3592         loff_t l = *pos;
3593         mddev_t *mddev;
3594
3595         if (l >= 0x10000)
3596                 return NULL;
3597         if (!l--)
3598                 /* header */
3599                 return (void*)1;
3600
3601         spin_lock(&all_mddevs_lock);
3602         list_for_each(tmp,&all_mddevs)
3603                 if (!l--) {
3604                         mddev = list_entry(tmp, mddev_t, all_mddevs);
3605                         mddev_get(mddev);
3606                         spin_unlock(&all_mddevs_lock);
3607                         return mddev;
3608                 }
3609         spin_unlock(&all_mddevs_lock);
3610         if (!l--)
3611                 return (void*)2;/* tail */
3612         return NULL;
3613 }
3614
3615 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3616 {
3617         struct list_head *tmp;
3618         mddev_t *next_mddev, *mddev = v;
3619         
3620         ++*pos;
3621         if (v == (void*)2)
3622                 return NULL;
3623
3624         spin_lock(&all_mddevs_lock);
3625         if (v == (void*)1)
3626                 tmp = all_mddevs.next;
3627         else
3628                 tmp = mddev->all_mddevs.next;
3629         if (tmp != &all_mddevs)
3630                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
3631         else {
3632                 next_mddev = (void*)2;
3633                 *pos = 0x10000;
3634         }               
3635         spin_unlock(&all_mddevs_lock);
3636
3637         if (v != (void*)1)
3638                 mddev_put(mddev);
3639         return next_mddev;
3640
3641 }
3642
3643 static void md_seq_stop(struct seq_file *seq, void *v)
3644 {
3645         mddev_t *mddev = v;
3646
3647         if (mddev && v != (void*)1 && v != (void*)2)
3648                 mddev_put(mddev);
3649 }
3650
3651 static int md_seq_show(struct seq_file *seq, void *v)
3652 {
3653         mddev_t *mddev = v;
3654         sector_t size;
3655         struct list_head *tmp2;
3656         mdk_rdev_t *rdev;
3657         int i;
3658         struct bitmap *bitmap;
3659
3660         if (v == (void*)1) {
3661                 seq_printf(seq, "Personalities : ");
3662                 spin_lock(&pers_lock);
3663                 for (i = 0; i < MAX_PERSONALITY; i++)
3664                         if (pers[i])
3665                                 seq_printf(seq, "[%s] ", pers[i]->name);
3666
3667                 spin_unlock(&pers_lock);
3668                 seq_printf(seq, "\n");
3669                 return 0;
3670         }
3671         if (v == (void*)2) {
3672                 status_unused(seq);
3673                 return 0;
3674         }
3675
3676         if (mddev_lock(mddev)!=0) 
3677                 return -EINTR;
3678         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
3679                 seq_printf(seq, "%s : %sactive", mdname(mddev),
3680                                                 mddev->pers ? "" : "in");
3681                 if (mddev->pers) {
3682                         if (mddev->ro==1)
3683                                 seq_printf(seq, " (read-only)");
3684                         if (mddev->ro==2)
3685                                 seq_printf(seq, "(auto-read-only)");
3686                         seq_printf(seq, " %s", mddev->pers->name);
3687                 }
3688
3689                 size = 0;
3690                 ITERATE_RDEV(mddev,rdev,tmp2) {
3691                         char b[BDEVNAME_SIZE];
3692                         seq_printf(seq, " %s[%d]",
3693                                 bdevname(rdev->bdev,b), rdev->desc_nr);
3694                         if (test_bit(WriteMostly, &rdev->flags))
3695                                 seq_printf(seq, "(W)");
3696                         if (test_bit(Faulty, &rdev->flags)) {
3697                                 seq_printf(seq, "(F)");
3698                                 continue;
3699                         } else if (rdev->raid_disk < 0)
3700                                 seq_printf(seq, "(S)"); /* spare */
3701                         size += rdev->size;
3702                 }
3703
3704                 if (!list_empty(&mddev->disks)) {
3705                         if (mddev->pers)
3706                                 seq_printf(seq, "\n      %llu blocks",
3707                                         (unsigned long long)mddev->array_size);
3708                         else
3709                                 seq_printf(seq, "\n      %llu blocks",
3710                                         (unsigned long long)size);
3711                 }
3712                 if (mddev->persistent) {
3713                         if (mddev->major_version != 0 ||
3714                             mddev->minor_version != 90) {
3715                                 seq_printf(seq," super %d.%d",
3716                                            mddev->major_version,
3717                                            mddev->minor_version);
3718                         }
3719                 } else
3720                         seq_printf(seq, " super non-persistent");
3721
3722                 if (mddev->pers) {
3723                         mddev->pers->status (seq, mddev);
3724                         seq_printf(seq, "\n      ");
3725                         if (mddev->curr_resync > 2) {
3726                                 status_resync (seq, mddev);
3727                                 seq_printf(seq, "\n      ");
3728                         } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
3729                                 seq_printf(seq, "\tresync=DELAYED\n      ");
3730                         else if (mddev->recovery_cp < MaxSector)
3731                                 seq_printf(seq, "\tresync=PENDING\n      ");
3732                 } else
3733                         seq_printf(seq, "\n       ");
3734
3735                 if ((bitmap = mddev->bitmap)) {
3736                         unsigned long chunk_kb;
3737                         unsigned long flags;
3738                         spin_lock_irqsave(&bitmap->lock, flags);
3739                         chunk_kb = bitmap->chunksize >> 10;
3740                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
3741                                 "%lu%s chunk",
3742                                 bitmap->pages - bitmap->missing_pages,
3743                                 bitmap->pages,
3744                                 (bitmap->pages - bitmap->missing_pages)
3745                                         << (PAGE_SHIFT - 10),
3746                                 chunk_kb ? chunk_kb : bitmap->chunksize,
3747                                 chunk_kb ? "KB" : "B");
3748                         if (bitmap->file) {
3749                                 seq_printf(seq, ", file: ");
3750                                 seq_path(seq, bitmap->file->f_vfsmnt,
3751                                          bitmap->file->f_dentry," \t\n");
3752                         }
3753
3754                         seq_printf(seq, "\n");
3755                         spin_unlock_irqrestore(&bitmap->lock, flags);
3756                 }
3757
3758                 seq_printf(seq, "\n");
3759         }
3760         mddev_unlock(mddev);
3761         
3762         return 0;
3763 }
3764
3765 static struct seq_operations md_seq_ops = {
3766         .start  = md_seq_start,
3767         .next   = md_seq_next,
3768         .stop   = md_seq_stop,
3769         .show   = md_seq_show,
3770 };
3771
3772 static int md_seq_open(struct inode *inode, struct file *file)
3773 {
3774         int error;
3775
3776         error = seq_open(file, &md_seq_ops);
3777         return error;
3778 }
3779
3780 static struct file_operations md_seq_fops = {
3781         .open           = md_seq_open,
3782         .read           = seq_read,
3783         .llseek         = seq_lseek,
3784         .release        = seq_release,
3785 };
3786
3787 int register_md_personality(int pnum, mdk_personality_t *p)
3788 {
3789         if (pnum >= MAX_PERSONALITY) {
3790                 printk(KERN_ERR
3791                        "md: tried to install personality %s as nr %d, but max is %lu\n",
3792                        p->name, pnum, MAX_PERSONALITY-1);
3793                 return -EINVAL;
3794         }
3795
3796         spin_lock(&pers_lock);
3797         if (pers[pnum]) {
3798                 spin_unlock(&pers_lock);
3799                 return -EBUSY;
3800         }
3801
3802         pers[pnum] = p;
3803         printk(KERN_INFO "md: %s personality registered as nr %d\n", p->name, pnum);
3804         spin_unlock(&pers_lock);
3805         return 0;
3806 }
3807
3808 int unregister_md_personality(int pnum)
3809 {
3810         if (pnum >= MAX_PERSONALITY)
3811                 return -EINVAL;
3812
3813         printk(KERN_INFO "md: %s personality unregistered\n", pers[pnum]->name);
3814         spin_lock(&pers_lock);
3815         pers[pnum] = NULL;
3816         spin_unlock(&pers_lock);
3817         return 0;
3818 }
3819
3820 static int is_mddev_idle(mddev_t *mddev)
3821 {
3822         mdk_rdev_t * rdev;
3823         struct list_head *tmp;
3824         int idle;
3825         unsigned long curr_events;
3826
3827         idle = 1;
3828         ITERATE_RDEV(mddev,rdev,tmp) {
3829                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
3830                 curr_events = disk_stat_read(disk, sectors[0]) + 
3831                                 disk_stat_read(disk, sectors[1]) - 
3832                                 atomic_read(&disk->sync_io);
3833                 /* Allow some slack between valud of curr_events and last_events,
3834                  * as there are some uninteresting races.
3835                  * Note: the following is an unsigned comparison.
3836                  */
3837                 if ((curr_events - rdev->last_events + 32) > 64) {
3838                         rdev->last_events = curr_events;
3839                         idle = 0;
3840                 }
3841         }
3842         return idle;
3843 }
3844
3845 void md_done_sync(mddev_t *mddev, int blocks, int ok)
3846 {
3847         /* another "blocks" (512byte) blocks have been synced */
3848         atomic_sub(blocks, &mddev->recovery_active);
3849         wake_up(&mddev->recovery_wait);
3850         if (!ok) {
3851                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
3852                 md_wakeup_thread(mddev->thread);
3853                 // stop recovery, signal do_sync ....
3854         }
3855 }
3856
3857
3858 /* md_write_start(mddev, bi)
3859  * If we need to update some array metadata (e.g. 'active' flag
3860  * in superblock) before writing, schedule a superblock update
3861  * and wait for it to complete.
3862  */
3863 void md_write_start(mddev_t *mddev, struct bio *bi)
3864 {
3865         if (bio_data_dir(bi) != WRITE)
3866                 return;
3867
3868         BUG_ON(mddev->ro == 1);
3869         if (mddev->ro == 2) {
3870                 /* need to switch to read/write */
3871                 mddev->ro = 0;
3872                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3873                 md_wakeup_thread(mddev->thread);
3874         }
3875         atomic_inc(&mddev->writes_pending);
3876         if (mddev->in_sync) {
3877                 spin_lock_irq(&mddev->write_lock);
3878                 if (mddev->in_sync) {
3879                         mddev->in_sync = 0;
3880                         mddev->sb_dirty = 1;
3881                         md_wakeup_thread(mddev->thread);
3882                 }
3883                 spin_unlock_irq(&mddev->write_lock);
3884         }
3885         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
3886 }
3887
3888 void md_write_end(mddev_t *mddev)
3889 {
3890         if (atomic_dec_and_test(&mddev->writes_pending)) {
3891                 if (mddev->safemode == 2)
3892                         md_wakeup_thread(mddev->thread);
3893                 else
3894                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
3895         }
3896 }
3897
3898 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
3899
3900 #define SYNC_MARKS      10
3901 #define SYNC_MARK_STEP  (3*HZ)
3902 static void md_do_sync(mddev_t *mddev)
3903 {
3904         mddev_t *mddev2;
3905         unsigned int currspeed = 0,
3906                  window;
3907         sector_t max_sectors,j, io_sectors;
3908         unsigned long mark[SYNC_MARKS];
3909         sector_t mark_cnt[SYNC_MARKS];
3910         int last_mark,m;
3911         struct list_head *tmp;
3912         sector_t last_check;
3913         int skipped = 0;
3914
3915         /* just incase thread restarts... */
3916         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
3917                 return;
3918
3919         /* we overload curr_resync somewhat here.
3920          * 0 == not engaged in resync at all
3921          * 2 == checking that there is no conflict with another sync
3922          * 1 == like 2, but have yielded to allow conflicting resync to
3923          *              commense
3924          * other == active in resync - this many blocks
3925          *
3926          * Before starting a resync we must have set curr_resync to
3927          * 2, and then checked that every "conflicting" array has curr_resync
3928          * less than ours.  When we find one that is the same or higher
3929          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
3930          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
3931          * This will mean we have to start checking from the beginning again.
3932          *
3933          */
3934
3935         do {
3936                 mddev->curr_resync = 2;
3937
3938         try_again:
3939                 if (signal_pending(current) ||
3940                     kthread_should_stop()) {
3941                         flush_signals(current);
3942                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3943                         goto skip;
3944                 }
3945                 ITERATE_MDDEV(mddev2,tmp) {
3946                         if (mddev2 == mddev)
3947                                 continue;
3948                         if (mddev2->curr_resync && 
3949                             match_mddev_units(mddev,mddev2)) {
3950                                 DEFINE_WAIT(wq);
3951                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
3952                                         /* arbitrarily yield */
3953                                         mddev->curr_resync = 1;
3954                                         wake_up(&resync_wait);
3955                                 }
3956                                 if (mddev > mddev2 && mddev->curr_resync == 1)
3957                                         /* no need to wait here, we can wait the next
3958                                          * time 'round when curr_resync == 2
3959                                          */
3960                                         continue;
3961                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
3962                                 if (!signal_pending(current) &&
3963                                     !kthread_should_stop() &&
3964                                     mddev2->curr_resync >= mddev->curr_resync) {
3965                                         printk(KERN_INFO "md: delaying resync of %s"
3966                                                " until %s has finished resync (they"
3967                                                " share one or more physical units)\n",
3968                                                mdname(mddev), mdname(mddev2));
3969                                         mddev_put(mddev2);
3970                                         schedule();
3971                                         finish_wait(&resync_wait, &wq);
3972                                         goto try_again;
3973                                 }
3974                                 finish_wait(&resync_wait, &wq);
3975                         }
3976                 }
3977         } while (mddev->curr_resync < 2);
3978
3979         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3980                 /* resync follows the size requested by the personality,
3981                  * which defaults to physical size, but can be virtual size
3982                  */
3983                 max_sectors = mddev->resync_max_sectors;
3984                 mddev->resync_mismatches = 0;
3985         } else
3986                 /* recovery follows the physical size of devices */
3987                 max_sectors = mddev->size << 1;
3988
3989         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
3990         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
3991                 " %d KB/sec/disc.\n", sysctl_speed_limit_min);
3992         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
3993                "(but not more than %d KB/sec) for reconstruction.\n",
3994                sysctl_speed_limit_max);
3995
3996         is_mddev_idle(mddev); /* this also initializes IO event counters */
3997         /* we don't use the checkpoint if there's a bitmap */
3998         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
3999             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4000                 j = mddev->recovery_cp;
4001         else
4002                 j = 0;
4003         io_sectors = 0;
4004         for (m = 0; m < SYNC_MARKS; m++) {
4005                 mark[m] = jiffies;
4006                 mark_cnt[m] = io_sectors;
4007         }
4008         last_mark = 0;
4009         mddev->resync_mark = mark[last_mark];
4010         mddev->resync_mark_cnt = mark_cnt[last_mark];
4011
4012         /*
4013          * Tune reconstruction:
4014          */
4015         window = 32*(PAGE_SIZE/512);
4016         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4017                 window/2,(unsigned long long) max_sectors/2);
4018
4019         atomic_set(&mddev->recovery_active, 0);
4020         init_waitqueue_head(&mddev->recovery_wait);
4021         last_check = 0;
4022
4023         if (j>2) {
4024                 printk(KERN_INFO 
4025                         "md: resuming recovery of %s from checkpoint.\n",
4026                         mdname(mddev));
4027                 mddev->curr_resync = j;
4028         }
4029
4030         while (j < max_sectors) {
4031                 sector_t sectors;
4032
4033                 skipped = 0;
4034                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4035                                             currspeed < sysctl_speed_limit_min);
4036                 if (sectors == 0) {
4037                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4038                         goto out;
4039                 }
4040
4041                 if (!skipped) { /* actual IO requested */
4042                         io_sectors += sectors;
4043                         atomic_add(sectors, &mddev->recovery_active);
4044                 }
4045
4046                 j += sectors;
4047                 if (j>1) mddev->curr_resync = j;
4048
4049
4050                 if (last_check + window > io_sectors || j == max_sectors)
4051                         continue;
4052
4053                 last_check = io_sectors;
4054
4055                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4056                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4057                         break;
4058
4059         repeat:
4060                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4061                         /* step marks */
4062                         int next = (last_mark+1) % SYNC_MARKS;
4063
4064                         mddev->resync_mark = mark[next];
4065                         mddev->resync_mark_cnt = mark_cnt[next];
4066                         mark[next] = jiffies;
4067                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4068                         last_mark = next;
4069                 }
4070
4071
4072                 if (signal_pending(current) || kthread_should_stop()) {
4073                         /*
4074                          * got a signal, exit.
4075                          */
4076                         printk(KERN_INFO 
4077                                 "md: md_do_sync() got signal ... exiting\n");
4078                         flush_signals(current);
4079                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4080                         goto out;
4081                 }
4082
4083                 /*
4084                  * this loop exits only if either when we are slower than
4085                  * the 'hard' speed limit, or the system was IO-idle for
4086                  * a jiffy.
4087                  * the system might be non-idle CPU-wise, but we only care
4088                  * about not overloading the IO subsystem. (things like an
4089                  * e2fsck being done on the RAID array should execute fast)
4090                  */
4091                 mddev->queue->unplug_fn(mddev->queue);
4092                 cond_resched();
4093
4094                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4095                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4096
4097                 if (currspeed > sysctl_speed_limit_min) {
4098                         if ((currspeed > sysctl_speed_limit_max) ||
4099                                         !is_mddev_idle(mddev)) {
4100                                 msleep_interruptible(250);
4101                                 goto repeat;
4102                         }
4103                 }
4104         }
4105         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4106         /*
4107          * this also signals 'finished resyncing' to md_stop
4108          */
4109  out:
4110         mddev->queue->unplug_fn(mddev->queue);
4111
4112         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4113
4114         /* tell personality that we are finished */
4115         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4116
4117         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4118             mddev->curr_resync > 2 &&
4119             mddev->curr_resync >= mddev->recovery_cp) {
4120                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4121                         printk(KERN_INFO 
4122                                 "md: checkpointing recovery of %s.\n",
4123                                 mdname(mddev));
4124                         mddev->recovery_cp = mddev->curr_resync;
4125                 } else
4126                         mddev->recovery_cp = MaxSector;
4127         }
4128
4129  skip:
4130         mddev->curr_resync = 0;
4131         wake_up(&resync_wait);
4132         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4133         md_wakeup_thread(mddev->thread);
4134 }
4135
4136
4137 /*
4138  * This routine is regularly called by all per-raid-array threads to
4139  * deal with generic issues like resync and super-block update.
4140  * Raid personalities that don't have a thread (linear/raid0) do not
4141  * need this as they never do any recovery or update the superblock.
4142  *
4143  * It does not do any resync itself, but rather "forks" off other threads
4144  * to do that as needed.
4145  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4146  * "->recovery" and create a thread at ->sync_thread.
4147  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4148  * and wakeups up this thread which will reap the thread and finish up.
4149  * This thread also removes any faulty devices (with nr_pending == 0).
4150  *
4151  * The overall approach is:
4152  *  1/ if the superblock needs updating, update it.
4153  *  2/ If a recovery thread is running, don't do anything else.
4154  *  3/ If recovery has finished, clean up, possibly marking spares active.
4155  *  4/ If there are any faulty devices, remove them.
4156  *  5/ If array is degraded, try to add spares devices
4157  *  6/ If array has spares or is not in-sync, start a resync thread.
4158  */
4159 void md_check_recovery(mddev_t *mddev)
4160 {
4161         mdk_rdev_t *rdev;
4162         struct list_head *rtmp;
4163
4164
4165         if (mddev->bitmap)
4166                 bitmap_daemon_work(mddev->bitmap);
4167
4168         if (mddev->ro)
4169                 return;
4170
4171         if (signal_pending(current)) {
4172                 if (mddev->pers->sync_request) {
4173                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4174                                mdname(mddev));
4175                         mddev->safemode = 2;
4176                 }
4177                 flush_signals(current);
4178         }
4179
4180         if ( ! (
4181                 mddev->sb_dirty ||
4182                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4183                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4184                 (mddev->safemode == 1) ||
4185                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4186                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4187                 ))
4188                 return;
4189
4190         if (mddev_trylock(mddev)==0) {
4191                 int spares =0;
4192
4193                 spin_lock_irq(&mddev->write_lock);
4194                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4195                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4196                         mddev->in_sync = 1;
4197                         mddev->sb_dirty = 1;
4198                 }
4199                 if (mddev->safemode == 1)
4200                         mddev->safemode = 0;
4201                 spin_unlock_irq(&mddev->write_lock);
4202
4203                 if (mddev->sb_dirty)
4204                         md_update_sb(mddev);
4205
4206
4207                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4208                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4209                         /* resync/recovery still happening */
4210                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4211                         goto unlock;
4212                 }
4213                 if (mddev->sync_thread) {
4214                         /* resync has finished, collect result */
4215                         md_unregister_thread(mddev->sync_thread);
4216                         mddev->sync_thread = NULL;
4217                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4218                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4219                                 /* success...*/
4220                                 /* activate any spares */
4221                                 mddev->pers->spare_active(mddev);
4222                         }
4223                         md_update_sb(mddev);
4224
4225                         /* if array is no-longer degraded, then any saved_raid_disk
4226                          * information must be scrapped
4227                          */
4228                         if (!mddev->degraded)
4229                                 ITERATE_RDEV(mddev,rdev,rtmp)
4230                                         rdev->saved_raid_disk = -1;
4231
4232                         mddev->recovery = 0;
4233                         /* flag recovery needed just to double check */
4234                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4235                         goto unlock;
4236                 }
4237                 /* Clear some bits that don't mean anything, but
4238                  * might be left set
4239                  */
4240                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4241                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4242                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4243                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4244
4245                 /* no recovery is running.
4246                  * remove any failed drives, then
4247                  * add spares if possible.
4248                  * Spare are also removed and re-added, to allow
4249                  * the personality to fail the re-add.
4250                  */
4251                 ITERATE_RDEV(mddev,rdev,rtmp)
4252                         if (rdev->raid_disk >= 0 &&
4253                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4254                             atomic_read(&rdev->nr_pending)==0) {
4255                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4256                                         char nm[20];
4257                                         sprintf(nm,"rd%d", rdev->raid_disk);
4258                                         sysfs_remove_link(&mddev->kobj, nm);
4259                                         rdev->raid_disk = -1;
4260                                 }
4261                         }
4262
4263                 if (mddev->degraded) {
4264                         ITERATE_RDEV(mddev,rdev,rtmp)
4265                                 if (rdev->raid_disk < 0
4266                                     && !test_bit(Faulty, &rdev->flags)) {
4267                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4268                                                 char nm[20];
4269                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4270                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4271                                                 spares++;
4272                                         } else
4273                                                 break;
4274                                 }
4275                 }
4276
4277                 if (spares) {
4278                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4279                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4280                 } else if (mddev->recovery_cp < MaxSector) {
4281                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4282                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4283                         /* nothing to be done ... */
4284                         goto unlock;
4285
4286                 if (mddev->pers->sync_request) {
4287                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4288                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4289                                 /* We are adding a device or devices to an array
4290                                  * which has the bitmap stored on all devices.
4291                                  * So make sure all bitmap pages get written
4292                                  */
4293                                 bitmap_write_all(mddev->bitmap);
4294                         }
4295                         mddev->sync_thread = md_register_thread(md_do_sync,
4296                                                                 mddev,
4297                                                                 "%s_resync");
4298                         if (!mddev->sync_thread) {
4299                                 printk(KERN_ERR "%s: could not start resync"
4300                                         " thread...\n", 
4301                                         mdname(mddev));
4302                                 /* leave the spares where they are, it shouldn't hurt */
4303                                 mddev->recovery = 0;
4304                         } else {
4305                                 md_wakeup_thread(mddev->sync_thread);
4306                         }
4307                 }
4308         unlock:
4309                 mddev_unlock(mddev);
4310         }
4311 }
4312
4313 static int md_notify_reboot(struct notifier_block *this,
4314                             unsigned long code, void *x)
4315 {
4316         struct list_head *tmp;
4317         mddev_t *mddev;
4318
4319         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4320
4321                 printk(KERN_INFO "md: stopping all md devices.\n");
4322
4323                 ITERATE_MDDEV(mddev,tmp)
4324                         if (mddev_trylock(mddev)==0)
4325                                 do_md_stop (mddev, 1);
4326                 /*
4327                  * certain more exotic SCSI devices are known to be
4328                  * volatile wrt too early system reboots. While the
4329                  * right place to handle this issue is the given
4330                  * driver, we do want to have a safe RAID driver ...
4331                  */
4332                 mdelay(1000*1);
4333         }
4334         return NOTIFY_DONE;
4335 }
4336
4337 static struct notifier_block md_notifier = {
4338         .notifier_call  = md_notify_reboot,
4339         .next           = NULL,
4340         .priority       = INT_MAX, /* before any real devices */
4341 };
4342
4343 static void md_geninit(void)
4344 {
4345         struct proc_dir_entry *p;
4346
4347         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4348
4349         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4350         if (p)
4351                 p->proc_fops = &md_seq_fops;
4352 }
4353
4354 static int __init md_init(void)
4355 {
4356         int minor;
4357
4358         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
4359                         " MD_SB_DISKS=%d\n",
4360                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
4361                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
4362         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
4363                         BITMAP_MINOR);
4364
4365         if (register_blkdev(MAJOR_NR, "md"))
4366                 return -1;
4367         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
4368                 unregister_blkdev(MAJOR_NR, "md");
4369                 return -1;
4370         }
4371         devfs_mk_dir("md");
4372         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
4373                                 md_probe, NULL, NULL);
4374         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
4375                             md_probe, NULL, NULL);
4376
4377         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4378                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
4379                                 S_IFBLK|S_IRUSR|S_IWUSR,
4380                                 "md/%d", minor);
4381
4382         for (minor=0; minor < MAX_MD_DEVS; ++minor)
4383                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
4384                               S_IFBLK|S_IRUSR|S_IWUSR,
4385                               "md/mdp%d", minor);
4386
4387
4388         register_reboot_notifier(&md_notifier);
4389         raid_table_header = register_sysctl_table(raid_root_table, 1);
4390
4391         md_geninit();
4392         return (0);
4393 }
4394
4395
4396 #ifndef MODULE
4397
4398 /*
4399  * Searches all registered partitions for autorun RAID arrays
4400  * at boot time.
4401  */
4402 static dev_t detected_devices[128];
4403 static int dev_cnt;
4404
4405 void md_autodetect_dev(dev_t dev)
4406 {
4407         if (dev_cnt >= 0 && dev_cnt < 127)
4408                 detected_devices[dev_cnt++] = dev;
4409 }
4410
4411
4412 static void autostart_arrays(int part)
4413 {
4414         mdk_rdev_t *rdev;
4415         int i;
4416
4417         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
4418
4419         for (i = 0; i < dev_cnt; i++) {
4420                 dev_t dev = detected_devices[i];
4421
4422                 rdev = md_import_device(dev,0, 0);
4423                 if (IS_ERR(rdev))
4424                         continue;
4425
4426                 if (test_bit(Faulty, &rdev->flags)) {
4427                         MD_BUG();
4428                         continue;
4429                 }
4430                 list_add(&rdev->same_set, &pending_raid_disks);
4431         }
4432         dev_cnt = 0;
4433
4434         autorun_devices(part);
4435 }
4436
4437 #endif
4438
4439 static __exit void md_exit(void)
4440 {
4441         mddev_t *mddev;
4442         struct list_head *tmp;
4443         int i;
4444         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
4445         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
4446         for (i=0; i < MAX_MD_DEVS; i++)
4447                 devfs_remove("md/%d", i);
4448         for (i=0; i < MAX_MD_DEVS; i++)
4449                 devfs_remove("md/d%d", i);
4450
4451         devfs_remove("md");
4452
4453         unregister_blkdev(MAJOR_NR,"md");
4454         unregister_blkdev(mdp_major, "mdp");
4455         unregister_reboot_notifier(&md_notifier);
4456         unregister_sysctl_table(raid_table_header);
4457         remove_proc_entry("mdstat", NULL);
4458         ITERATE_MDDEV(mddev,tmp) {
4459                 struct gendisk *disk = mddev->gendisk;
4460                 if (!disk)
4461                         continue;
4462                 export_array(mddev);
4463                 del_gendisk(disk);
4464                 put_disk(disk);
4465                 mddev->gendisk = NULL;
4466                 mddev_put(mddev);
4467         }
4468 }
4469
4470 module_init(md_init)
4471 module_exit(md_exit)
4472
4473 static int get_ro(char *buffer, struct kernel_param *kp)
4474 {
4475         return sprintf(buffer, "%d", start_readonly);
4476 }
4477 static int set_ro(const char *val, struct kernel_param *kp)
4478 {
4479         char *e;
4480         int num = simple_strtoul(val, &e, 10);
4481         if (*val && (*e == '\0' || *e == '\n')) {
4482                 start_readonly = num;
4483                 return 0;;
4484         }
4485         return -EINVAL;
4486 }
4487
4488 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
4489
4490 EXPORT_SYMBOL(register_md_personality);
4491 EXPORT_SYMBOL(unregister_md_personality);
4492 EXPORT_SYMBOL(md_error);
4493 EXPORT_SYMBOL(md_done_sync);
4494 EXPORT_SYMBOL(md_write_start);
4495 EXPORT_SYMBOL(md_write_end);
4496 EXPORT_SYMBOL(md_register_thread);
4497 EXPORT_SYMBOL(md_unregister_thread);
4498 EXPORT_SYMBOL(md_wakeup_thread);
4499 EXPORT_SYMBOL(md_print_devices);
4500 EXPORT_SYMBOL(md_check_recovery);
4501 MODULE_LICENSE("GPL");
4502 MODULE_ALIAS("md");
4503 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);