[PATCH] md: clear the congested_fn when stopping a raid5
[linux-2.6.git] / drivers / md / md.c
1 /*
2    md.c : Multiple Devices driver for Linux
3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5      completely rewritten, based on the MD driver code from Marc Zyngier
6
7    Changes:
8
9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13    - kmod support by: Cyrus Durgin
14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19
20      Neil Brown <neilb@cse.unsw.edu.au>.
21
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25    This program is free software; you can redistribute it and/or modify
26    it under the terms of the GNU General Public License as published by
27    the Free Software Foundation; either version 2, or (at your option)
28    any later version.
29
30    You should have received a copy of the GNU General Public License
31    (for example /usr/src/linux/COPYING); if not, write to the Free
32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/module.h>
36 #include <linux/kthread.h>
37 #include <linux/linkage.h>
38 #include <linux/raid/md.h>
39 #include <linux/raid/bitmap.h>
40 #include <linux/sysctl.h>
41 #include <linux/buffer_head.h> /* for invalidate_bdev */
42 #include <linux/poll.h>
43 #include <linux/mutex.h>
44 #include <linux/ctype.h>
45 #include <linux/freezer.h>
46
47 #include <linux/init.h>
48
49 #include <linux/file.h>
50
51 #ifdef CONFIG_KMOD
52 #include <linux/kmod.h>
53 #endif
54
55 #include <asm/unaligned.h>
56
57 #define MAJOR_NR MD_MAJOR
58 #define MD_DRIVER
59
60 /* 63 partitions with the alternate major number (mdp) */
61 #define MdpMinorShift 6
62
63 #define DEBUG 0
64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
65
66
67 #ifndef MODULE
68 static void autostart_arrays (int part);
69 #endif
70
71 static LIST_HEAD(pers_list);
72 static DEFINE_SPINLOCK(pers_lock);
73
74 static void md_print_devices(void);
75
76 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
77
78 /*
79  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
80  * is 1000 KB/sec, so the extra system load does not show up that much.
81  * Increase it if you want to have more _guaranteed_ speed. Note that
82  * the RAID driver will use the maximum available bandwidth if the IO
83  * subsystem is idle. There is also an 'absolute maximum' reconstruction
84  * speed limit - in case reconstruction slows down your system despite
85  * idle IO detection.
86  *
87  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
88  * or /sys/block/mdX/md/sync_speed_{min,max}
89  */
90
91 static int sysctl_speed_limit_min = 1000;
92 static int sysctl_speed_limit_max = 200000;
93 static inline int speed_min(mddev_t *mddev)
94 {
95         return mddev->sync_speed_min ?
96                 mddev->sync_speed_min : sysctl_speed_limit_min;
97 }
98
99 static inline int speed_max(mddev_t *mddev)
100 {
101         return mddev->sync_speed_max ?
102                 mddev->sync_speed_max : sysctl_speed_limit_max;
103 }
104
105 static struct ctl_table_header *raid_table_header;
106
107 static ctl_table raid_table[] = {
108         {
109                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
110                 .procname       = "speed_limit_min",
111                 .data           = &sysctl_speed_limit_min,
112                 .maxlen         = sizeof(int),
113                 .mode           = S_IRUGO|S_IWUSR,
114                 .proc_handler   = &proc_dointvec,
115         },
116         {
117                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
118                 .procname       = "speed_limit_max",
119                 .data           = &sysctl_speed_limit_max,
120                 .maxlen         = sizeof(int),
121                 .mode           = S_IRUGO|S_IWUSR,
122                 .proc_handler   = &proc_dointvec,
123         },
124         { .ctl_name = 0 }
125 };
126
127 static ctl_table raid_dir_table[] = {
128         {
129                 .ctl_name       = DEV_RAID,
130                 .procname       = "raid",
131                 .maxlen         = 0,
132                 .mode           = S_IRUGO|S_IXUGO,
133                 .child          = raid_table,
134         },
135         { .ctl_name = 0 }
136 };
137
138 static ctl_table raid_root_table[] = {
139         {
140                 .ctl_name       = CTL_DEV,
141                 .procname       = "dev",
142                 .maxlen         = 0,
143                 .mode           = 0555,
144                 .child          = raid_dir_table,
145         },
146         { .ctl_name = 0 }
147 };
148
149 static struct block_device_operations md_fops;
150
151 static int start_readonly;
152
153 /*
154  * We have a system wide 'event count' that is incremented
155  * on any 'interesting' event, and readers of /proc/mdstat
156  * can use 'poll' or 'select' to find out when the event
157  * count increases.
158  *
159  * Events are:
160  *  start array, stop array, error, add device, remove device,
161  *  start build, activate spare
162  */
163 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
164 static atomic_t md_event_count;
165 void md_new_event(mddev_t *mddev)
166 {
167         atomic_inc(&md_event_count);
168         wake_up(&md_event_waiters);
169         sysfs_notify(&mddev->kobj, NULL, "sync_action");
170 }
171 EXPORT_SYMBOL_GPL(md_new_event);
172
173 /* Alternate version that can be called from interrupts
174  * when calling sysfs_notify isn't needed.
175  */
176 static void md_new_event_inintr(mddev_t *mddev)
177 {
178         atomic_inc(&md_event_count);
179         wake_up(&md_event_waiters);
180 }
181
182 /*
183  * Enables to iterate over all existing md arrays
184  * all_mddevs_lock protects this list.
185  */
186 static LIST_HEAD(all_mddevs);
187 static DEFINE_SPINLOCK(all_mddevs_lock);
188
189
190 /*
191  * iterates through all used mddevs in the system.
192  * We take care to grab the all_mddevs_lock whenever navigating
193  * the list, and to always hold a refcount when unlocked.
194  * Any code which breaks out of this loop while own
195  * a reference to the current mddev and must mddev_put it.
196  */
197 #define ITERATE_MDDEV(mddev,tmp)                                        \
198                                                                         \
199         for (({ spin_lock(&all_mddevs_lock);                            \
200                 tmp = all_mddevs.next;                                  \
201                 mddev = NULL;});                                        \
202              ({ if (tmp != &all_mddevs)                                 \
203                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
204                 spin_unlock(&all_mddevs_lock);                          \
205                 if (mddev) mddev_put(mddev);                            \
206                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
207                 tmp != &all_mddevs;});                                  \
208              ({ spin_lock(&all_mddevs_lock);                            \
209                 tmp = tmp->next;})                                      \
210                 )
211
212
213 static int md_fail_request (request_queue_t *q, struct bio *bio)
214 {
215         bio_io_error(bio, bio->bi_size);
216         return 0;
217 }
218
219 static inline mddev_t *mddev_get(mddev_t *mddev)
220 {
221         atomic_inc(&mddev->active);
222         return mddev;
223 }
224
225 static void mddev_put(mddev_t *mddev)
226 {
227         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
228                 return;
229         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
230                 list_del(&mddev->all_mddevs);
231                 spin_unlock(&all_mddevs_lock);
232                 blk_cleanup_queue(mddev->queue);
233                 kobject_unregister(&mddev->kobj);
234         } else
235                 spin_unlock(&all_mddevs_lock);
236 }
237
238 static mddev_t * mddev_find(dev_t unit)
239 {
240         mddev_t *mddev, *new = NULL;
241
242  retry:
243         spin_lock(&all_mddevs_lock);
244         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
245                 if (mddev->unit == unit) {
246                         mddev_get(mddev);
247                         spin_unlock(&all_mddevs_lock);
248                         kfree(new);
249                         return mddev;
250                 }
251
252         if (new) {
253                 list_add(&new->all_mddevs, &all_mddevs);
254                 spin_unlock(&all_mddevs_lock);
255                 return new;
256         }
257         spin_unlock(&all_mddevs_lock);
258
259         new = kzalloc(sizeof(*new), GFP_KERNEL);
260         if (!new)
261                 return NULL;
262
263         new->unit = unit;
264         if (MAJOR(unit) == MD_MAJOR)
265                 new->md_minor = MINOR(unit);
266         else
267                 new->md_minor = MINOR(unit) >> MdpMinorShift;
268
269         mutex_init(&new->reconfig_mutex);
270         INIT_LIST_HEAD(&new->disks);
271         INIT_LIST_HEAD(&new->all_mddevs);
272         init_timer(&new->safemode_timer);
273         atomic_set(&new->active, 1);
274         spin_lock_init(&new->write_lock);
275         init_waitqueue_head(&new->sb_wait);
276
277         new->queue = blk_alloc_queue(GFP_KERNEL);
278         if (!new->queue) {
279                 kfree(new);
280                 return NULL;
281         }
282         set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
283
284         blk_queue_make_request(new->queue, md_fail_request);
285
286         goto retry;
287 }
288
289 static inline int mddev_lock(mddev_t * mddev)
290 {
291         return mutex_lock_interruptible(&mddev->reconfig_mutex);
292 }
293
294 static inline int mddev_trylock(mddev_t * mddev)
295 {
296         return mutex_trylock(&mddev->reconfig_mutex);
297 }
298
299 static inline void mddev_unlock(mddev_t * mddev)
300 {
301         mutex_unlock(&mddev->reconfig_mutex);
302
303         md_wakeup_thread(mddev->thread);
304 }
305
306 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
307 {
308         mdk_rdev_t * rdev;
309         struct list_head *tmp;
310
311         ITERATE_RDEV(mddev,rdev,tmp) {
312                 if (rdev->desc_nr == nr)
313                         return rdev;
314         }
315         return NULL;
316 }
317
318 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
319 {
320         struct list_head *tmp;
321         mdk_rdev_t *rdev;
322
323         ITERATE_RDEV(mddev,rdev,tmp) {
324                 if (rdev->bdev->bd_dev == dev)
325                         return rdev;
326         }
327         return NULL;
328 }
329
330 static struct mdk_personality *find_pers(int level, char *clevel)
331 {
332         struct mdk_personality *pers;
333         list_for_each_entry(pers, &pers_list, list) {
334                 if (level != LEVEL_NONE && pers->level == level)
335                         return pers;
336                 if (strcmp(pers->name, clevel)==0)
337                         return pers;
338         }
339         return NULL;
340 }
341
342 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
343 {
344         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
345         return MD_NEW_SIZE_BLOCKS(size);
346 }
347
348 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
349 {
350         sector_t size;
351
352         size = rdev->sb_offset;
353
354         if (chunk_size)
355                 size &= ~((sector_t)chunk_size/1024 - 1);
356         return size;
357 }
358
359 static int alloc_disk_sb(mdk_rdev_t * rdev)
360 {
361         if (rdev->sb_page)
362                 MD_BUG();
363
364         rdev->sb_page = alloc_page(GFP_KERNEL);
365         if (!rdev->sb_page) {
366                 printk(KERN_ALERT "md: out of memory.\n");
367                 return -EINVAL;
368         }
369
370         return 0;
371 }
372
373 static void free_disk_sb(mdk_rdev_t * rdev)
374 {
375         if (rdev->sb_page) {
376                 put_page(rdev->sb_page);
377                 rdev->sb_loaded = 0;
378                 rdev->sb_page = NULL;
379                 rdev->sb_offset = 0;
380                 rdev->size = 0;
381         }
382 }
383
384
385 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
386 {
387         mdk_rdev_t *rdev = bio->bi_private;
388         mddev_t *mddev = rdev->mddev;
389         if (bio->bi_size)
390                 return 1;
391
392         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
393                 printk("md: super_written gets error=%d, uptodate=%d\n",
394                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
395                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
396                 md_error(mddev, rdev);
397         }
398
399         if (atomic_dec_and_test(&mddev->pending_writes))
400                 wake_up(&mddev->sb_wait);
401         bio_put(bio);
402         return 0;
403 }
404
405 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
406 {
407         struct bio *bio2 = bio->bi_private;
408         mdk_rdev_t *rdev = bio2->bi_private;
409         mddev_t *mddev = rdev->mddev;
410         if (bio->bi_size)
411                 return 1;
412
413         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
414             error == -EOPNOTSUPP) {
415                 unsigned long flags;
416                 /* barriers don't appear to be supported :-( */
417                 set_bit(BarriersNotsupp, &rdev->flags);
418                 mddev->barriers_work = 0;
419                 spin_lock_irqsave(&mddev->write_lock, flags);
420                 bio2->bi_next = mddev->biolist;
421                 mddev->biolist = bio2;
422                 spin_unlock_irqrestore(&mddev->write_lock, flags);
423                 wake_up(&mddev->sb_wait);
424                 bio_put(bio);
425                 return 0;
426         }
427         bio_put(bio2);
428         bio->bi_private = rdev;
429         return super_written(bio, bytes_done, error);
430 }
431
432 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
433                    sector_t sector, int size, struct page *page)
434 {
435         /* write first size bytes of page to sector of rdev
436          * Increment mddev->pending_writes before returning
437          * and decrement it on completion, waking up sb_wait
438          * if zero is reached.
439          * If an error occurred, call md_error
440          *
441          * As we might need to resubmit the request if BIO_RW_BARRIER
442          * causes ENOTSUPP, we allocate a spare bio...
443          */
444         struct bio *bio = bio_alloc(GFP_NOIO, 1);
445         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
446
447         bio->bi_bdev = rdev->bdev;
448         bio->bi_sector = sector;
449         bio_add_page(bio, page, size, 0);
450         bio->bi_private = rdev;
451         bio->bi_end_io = super_written;
452         bio->bi_rw = rw;
453
454         atomic_inc(&mddev->pending_writes);
455         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
456                 struct bio *rbio;
457                 rw |= (1<<BIO_RW_BARRIER);
458                 rbio = bio_clone(bio, GFP_NOIO);
459                 rbio->bi_private = bio;
460                 rbio->bi_end_io = super_written_barrier;
461                 submit_bio(rw, rbio);
462         } else
463                 submit_bio(rw, bio);
464 }
465
466 void md_super_wait(mddev_t *mddev)
467 {
468         /* wait for all superblock writes that were scheduled to complete.
469          * if any had to be retried (due to BARRIER problems), retry them
470          */
471         DEFINE_WAIT(wq);
472         for(;;) {
473                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
474                 if (atomic_read(&mddev->pending_writes)==0)
475                         break;
476                 while (mddev->biolist) {
477                         struct bio *bio;
478                         spin_lock_irq(&mddev->write_lock);
479                         bio = mddev->biolist;
480                         mddev->biolist = bio->bi_next ;
481                         bio->bi_next = NULL;
482                         spin_unlock_irq(&mddev->write_lock);
483                         submit_bio(bio->bi_rw, bio);
484                 }
485                 schedule();
486         }
487         finish_wait(&mddev->sb_wait, &wq);
488 }
489
490 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
491 {
492         if (bio->bi_size)
493                 return 1;
494
495         complete((struct completion*)bio->bi_private);
496         return 0;
497 }
498
499 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
500                    struct page *page, int rw)
501 {
502         struct bio *bio = bio_alloc(GFP_NOIO, 1);
503         struct completion event;
504         int ret;
505
506         rw |= (1 << BIO_RW_SYNC);
507
508         bio->bi_bdev = bdev;
509         bio->bi_sector = sector;
510         bio_add_page(bio, page, size, 0);
511         init_completion(&event);
512         bio->bi_private = &event;
513         bio->bi_end_io = bi_complete;
514         submit_bio(rw, bio);
515         wait_for_completion(&event);
516
517         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
518         bio_put(bio);
519         return ret;
520 }
521 EXPORT_SYMBOL_GPL(sync_page_io);
522
523 static int read_disk_sb(mdk_rdev_t * rdev, int size)
524 {
525         char b[BDEVNAME_SIZE];
526         if (!rdev->sb_page) {
527                 MD_BUG();
528                 return -EINVAL;
529         }
530         if (rdev->sb_loaded)
531                 return 0;
532
533
534         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
535                 goto fail;
536         rdev->sb_loaded = 1;
537         return 0;
538
539 fail:
540         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
541                 bdevname(rdev->bdev,b));
542         return -EINVAL;
543 }
544
545 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
546 {
547         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
548                 (sb1->set_uuid1 == sb2->set_uuid1) &&
549                 (sb1->set_uuid2 == sb2->set_uuid2) &&
550                 (sb1->set_uuid3 == sb2->set_uuid3))
551
552                 return 1;
553
554         return 0;
555 }
556
557
558 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
559 {
560         int ret;
561         mdp_super_t *tmp1, *tmp2;
562
563         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
564         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
565
566         if (!tmp1 || !tmp2) {
567                 ret = 0;
568                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
569                 goto abort;
570         }
571
572         *tmp1 = *sb1;
573         *tmp2 = *sb2;
574
575         /*
576          * nr_disks is not constant
577          */
578         tmp1->nr_disks = 0;
579         tmp2->nr_disks = 0;
580
581         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
582                 ret = 0;
583         else
584                 ret = 1;
585
586 abort:
587         kfree(tmp1);
588         kfree(tmp2);
589         return ret;
590 }
591
592 static unsigned int calc_sb_csum(mdp_super_t * sb)
593 {
594         unsigned int disk_csum, csum;
595
596         disk_csum = sb->sb_csum;
597         sb->sb_csum = 0;
598         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
599         sb->sb_csum = disk_csum;
600         return csum;
601 }
602
603
604 /*
605  * Handle superblock details.
606  * We want to be able to handle multiple superblock formats
607  * so we have a common interface to them all, and an array of
608  * different handlers.
609  * We rely on user-space to write the initial superblock, and support
610  * reading and updating of superblocks.
611  * Interface methods are:
612  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
613  *      loads and validates a superblock on dev.
614  *      if refdev != NULL, compare superblocks on both devices
615  *    Return:
616  *      0 - dev has a superblock that is compatible with refdev
617  *      1 - dev has a superblock that is compatible and newer than refdev
618  *          so dev should be used as the refdev in future
619  *     -EINVAL superblock incompatible or invalid
620  *     -othererror e.g. -EIO
621  *
622  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
623  *      Verify that dev is acceptable into mddev.
624  *       The first time, mddev->raid_disks will be 0, and data from
625  *       dev should be merged in.  Subsequent calls check that dev
626  *       is new enough.  Return 0 or -EINVAL
627  *
628  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
629  *     Update the superblock for rdev with data in mddev
630  *     This does not write to disc.
631  *
632  */
633
634 struct super_type  {
635         char            *name;
636         struct module   *owner;
637         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
638         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
639         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
640 };
641
642 /*
643  * load_super for 0.90.0 
644  */
645 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
646 {
647         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
648         mdp_super_t *sb;
649         int ret;
650         sector_t sb_offset;
651
652         /*
653          * Calculate the position of the superblock,
654          * it's at the end of the disk.
655          *
656          * It also happens to be a multiple of 4Kb.
657          */
658         sb_offset = calc_dev_sboffset(rdev->bdev);
659         rdev->sb_offset = sb_offset;
660
661         ret = read_disk_sb(rdev, MD_SB_BYTES);
662         if (ret) return ret;
663
664         ret = -EINVAL;
665
666         bdevname(rdev->bdev, b);
667         sb = (mdp_super_t*)page_address(rdev->sb_page);
668
669         if (sb->md_magic != MD_SB_MAGIC) {
670                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
671                        b);
672                 goto abort;
673         }
674
675         if (sb->major_version != 0 ||
676             sb->minor_version < 90 ||
677             sb->minor_version > 91) {
678                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
679                         sb->major_version, sb->minor_version,
680                         b);
681                 goto abort;
682         }
683
684         if (sb->raid_disks <= 0)
685                 goto abort;
686
687         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
688                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
689                         b);
690                 goto abort;
691         }
692
693         rdev->preferred_minor = sb->md_minor;
694         rdev->data_offset = 0;
695         rdev->sb_size = MD_SB_BYTES;
696
697         if (sb->level == LEVEL_MULTIPATH)
698                 rdev->desc_nr = -1;
699         else
700                 rdev->desc_nr = sb->this_disk.number;
701
702         if (refdev == 0)
703                 ret = 1;
704         else {
705                 __u64 ev1, ev2;
706                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
707                 if (!uuid_equal(refsb, sb)) {
708                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
709                                 b, bdevname(refdev->bdev,b2));
710                         goto abort;
711                 }
712                 if (!sb_equal(refsb, sb)) {
713                         printk(KERN_WARNING "md: %s has same UUID"
714                                " but different superblock to %s\n",
715                                b, bdevname(refdev->bdev, b2));
716                         goto abort;
717                 }
718                 ev1 = md_event(sb);
719                 ev2 = md_event(refsb);
720                 if (ev1 > ev2)
721                         ret = 1;
722                 else 
723                         ret = 0;
724         }
725         rdev->size = calc_dev_size(rdev, sb->chunk_size);
726
727         if (rdev->size < sb->size && sb->level > 1)
728                 /* "this cannot possibly happen" ... */
729                 ret = -EINVAL;
730
731  abort:
732         return ret;
733 }
734
735 /*
736  * validate_super for 0.90.0
737  */
738 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
739 {
740         mdp_disk_t *desc;
741         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
742         __u64 ev1 = md_event(sb);
743
744         rdev->raid_disk = -1;
745         rdev->flags = 0;
746         if (mddev->raid_disks == 0) {
747                 mddev->major_version = 0;
748                 mddev->minor_version = sb->minor_version;
749                 mddev->patch_version = sb->patch_version;
750                 mddev->persistent = ! sb->not_persistent;
751                 mddev->chunk_size = sb->chunk_size;
752                 mddev->ctime = sb->ctime;
753                 mddev->utime = sb->utime;
754                 mddev->level = sb->level;
755                 mddev->clevel[0] = 0;
756                 mddev->layout = sb->layout;
757                 mddev->raid_disks = sb->raid_disks;
758                 mddev->size = sb->size;
759                 mddev->events = ev1;
760                 mddev->bitmap_offset = 0;
761                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
762
763                 if (mddev->minor_version >= 91) {
764                         mddev->reshape_position = sb->reshape_position;
765                         mddev->delta_disks = sb->delta_disks;
766                         mddev->new_level = sb->new_level;
767                         mddev->new_layout = sb->new_layout;
768                         mddev->new_chunk = sb->new_chunk;
769                 } else {
770                         mddev->reshape_position = MaxSector;
771                         mddev->delta_disks = 0;
772                         mddev->new_level = mddev->level;
773                         mddev->new_layout = mddev->layout;
774                         mddev->new_chunk = mddev->chunk_size;
775                 }
776
777                 if (sb->state & (1<<MD_SB_CLEAN))
778                         mddev->recovery_cp = MaxSector;
779                 else {
780                         if (sb->events_hi == sb->cp_events_hi && 
781                                 sb->events_lo == sb->cp_events_lo) {
782                                 mddev->recovery_cp = sb->recovery_cp;
783                         } else
784                                 mddev->recovery_cp = 0;
785                 }
786
787                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
788                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
789                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
790                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
791
792                 mddev->max_disks = MD_SB_DISKS;
793
794                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
795                     mddev->bitmap_file == NULL) {
796                         if (mddev->level != 1 && mddev->level != 4
797                             && mddev->level != 5 && mddev->level != 6
798                             && mddev->level != 10) {
799                                 /* FIXME use a better test */
800                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
801                                 return -EINVAL;
802                         }
803                         mddev->bitmap_offset = mddev->default_bitmap_offset;
804                 }
805
806         } else if (mddev->pers == NULL) {
807                 /* Insist on good event counter while assembling */
808                 ++ev1;
809                 if (ev1 < mddev->events) 
810                         return -EINVAL;
811         } else if (mddev->bitmap) {
812                 /* if adding to array with a bitmap, then we can accept an
813                  * older device ... but not too old.
814                  */
815                 if (ev1 < mddev->bitmap->events_cleared)
816                         return 0;
817         } else {
818                 if (ev1 < mddev->events)
819                         /* just a hot-add of a new device, leave raid_disk at -1 */
820                         return 0;
821         }
822
823         if (mddev->level != LEVEL_MULTIPATH) {
824                 desc = sb->disks + rdev->desc_nr;
825
826                 if (desc->state & (1<<MD_DISK_FAULTY))
827                         set_bit(Faulty, &rdev->flags);
828                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
829                             desc->raid_disk < mddev->raid_disks */) {
830                         set_bit(In_sync, &rdev->flags);
831                         rdev->raid_disk = desc->raid_disk;
832                 }
833                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
834                         set_bit(WriteMostly, &rdev->flags);
835         } else /* MULTIPATH are always insync */
836                 set_bit(In_sync, &rdev->flags);
837         return 0;
838 }
839
840 /*
841  * sync_super for 0.90.0
842  */
843 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
844 {
845         mdp_super_t *sb;
846         struct list_head *tmp;
847         mdk_rdev_t *rdev2;
848         int next_spare = mddev->raid_disks;
849
850
851         /* make rdev->sb match mddev data..
852          *
853          * 1/ zero out disks
854          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
855          * 3/ any empty disks < next_spare become removed
856          *
857          * disks[0] gets initialised to REMOVED because
858          * we cannot be sure from other fields if it has
859          * been initialised or not.
860          */
861         int i;
862         int active=0, working=0,failed=0,spare=0,nr_disks=0;
863
864         rdev->sb_size = MD_SB_BYTES;
865
866         sb = (mdp_super_t*)page_address(rdev->sb_page);
867
868         memset(sb, 0, sizeof(*sb));
869
870         sb->md_magic = MD_SB_MAGIC;
871         sb->major_version = mddev->major_version;
872         sb->patch_version = mddev->patch_version;
873         sb->gvalid_words  = 0; /* ignored */
874         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
875         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
876         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
877         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
878
879         sb->ctime = mddev->ctime;
880         sb->level = mddev->level;
881         sb->size  = mddev->size;
882         sb->raid_disks = mddev->raid_disks;
883         sb->md_minor = mddev->md_minor;
884         sb->not_persistent = !mddev->persistent;
885         sb->utime = mddev->utime;
886         sb->state = 0;
887         sb->events_hi = (mddev->events>>32);
888         sb->events_lo = (u32)mddev->events;
889
890         if (mddev->reshape_position == MaxSector)
891                 sb->minor_version = 90;
892         else {
893                 sb->minor_version = 91;
894                 sb->reshape_position = mddev->reshape_position;
895                 sb->new_level = mddev->new_level;
896                 sb->delta_disks = mddev->delta_disks;
897                 sb->new_layout = mddev->new_layout;
898                 sb->new_chunk = mddev->new_chunk;
899         }
900         mddev->minor_version = sb->minor_version;
901         if (mddev->in_sync)
902         {
903                 sb->recovery_cp = mddev->recovery_cp;
904                 sb->cp_events_hi = (mddev->events>>32);
905                 sb->cp_events_lo = (u32)mddev->events;
906                 if (mddev->recovery_cp == MaxSector)
907                         sb->state = (1<< MD_SB_CLEAN);
908         } else
909                 sb->recovery_cp = 0;
910
911         sb->layout = mddev->layout;
912         sb->chunk_size = mddev->chunk_size;
913
914         if (mddev->bitmap && mddev->bitmap_file == NULL)
915                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
916
917         sb->disks[0].state = (1<<MD_DISK_REMOVED);
918         ITERATE_RDEV(mddev,rdev2,tmp) {
919                 mdp_disk_t *d;
920                 int desc_nr;
921                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
922                     && !test_bit(Faulty, &rdev2->flags))
923                         desc_nr = rdev2->raid_disk;
924                 else
925                         desc_nr = next_spare++;
926                 rdev2->desc_nr = desc_nr;
927                 d = &sb->disks[rdev2->desc_nr];
928                 nr_disks++;
929                 d->number = rdev2->desc_nr;
930                 d->major = MAJOR(rdev2->bdev->bd_dev);
931                 d->minor = MINOR(rdev2->bdev->bd_dev);
932                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
933                     && !test_bit(Faulty, &rdev2->flags))
934                         d->raid_disk = rdev2->raid_disk;
935                 else
936                         d->raid_disk = rdev2->desc_nr; /* compatibility */
937                 if (test_bit(Faulty, &rdev2->flags))
938                         d->state = (1<<MD_DISK_FAULTY);
939                 else if (test_bit(In_sync, &rdev2->flags)) {
940                         d->state = (1<<MD_DISK_ACTIVE);
941                         d->state |= (1<<MD_DISK_SYNC);
942                         active++;
943                         working++;
944                 } else {
945                         d->state = 0;
946                         spare++;
947                         working++;
948                 }
949                 if (test_bit(WriteMostly, &rdev2->flags))
950                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
951         }
952         /* now set the "removed" and "faulty" bits on any missing devices */
953         for (i=0 ; i < mddev->raid_disks ; i++) {
954                 mdp_disk_t *d = &sb->disks[i];
955                 if (d->state == 0 && d->number == 0) {
956                         d->number = i;
957                         d->raid_disk = i;
958                         d->state = (1<<MD_DISK_REMOVED);
959                         d->state |= (1<<MD_DISK_FAULTY);
960                         failed++;
961                 }
962         }
963         sb->nr_disks = nr_disks;
964         sb->active_disks = active;
965         sb->working_disks = working;
966         sb->failed_disks = failed;
967         sb->spare_disks = spare;
968
969         sb->this_disk = sb->disks[rdev->desc_nr];
970         sb->sb_csum = calc_sb_csum(sb);
971 }
972
973 /*
974  * version 1 superblock
975  */
976
977 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
978 {
979         __le32 disk_csum;
980         u32 csum;
981         unsigned long long newcsum;
982         int size = 256 + le32_to_cpu(sb->max_dev)*2;
983         __le32 *isuper = (__le32*)sb;
984         int i;
985
986         disk_csum = sb->sb_csum;
987         sb->sb_csum = 0;
988         newcsum = 0;
989         for (i=0; size>=4; size -= 4 )
990                 newcsum += le32_to_cpu(*isuper++);
991
992         if (size == 2)
993                 newcsum += le16_to_cpu(*(__le16*) isuper);
994
995         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
996         sb->sb_csum = disk_csum;
997         return cpu_to_le32(csum);
998 }
999
1000 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1001 {
1002         struct mdp_superblock_1 *sb;
1003         int ret;
1004         sector_t sb_offset;
1005         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1006         int bmask;
1007
1008         /*
1009          * Calculate the position of the superblock.
1010          * It is always aligned to a 4K boundary and
1011          * depeding on minor_version, it can be:
1012          * 0: At least 8K, but less than 12K, from end of device
1013          * 1: At start of device
1014          * 2: 4K from start of device.
1015          */
1016         switch(minor_version) {
1017         case 0:
1018                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1019                 sb_offset -= 8*2;
1020                 sb_offset &= ~(sector_t)(4*2-1);
1021                 /* convert from sectors to K */
1022                 sb_offset /= 2;
1023                 break;
1024         case 1:
1025                 sb_offset = 0;
1026                 break;
1027         case 2:
1028                 sb_offset = 4;
1029                 break;
1030         default:
1031                 return -EINVAL;
1032         }
1033         rdev->sb_offset = sb_offset;
1034
1035         /* superblock is rarely larger than 1K, but it can be larger,
1036          * and it is safe to read 4k, so we do that
1037          */
1038         ret = read_disk_sb(rdev, 4096);
1039         if (ret) return ret;
1040
1041
1042         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1043
1044         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1045             sb->major_version != cpu_to_le32(1) ||
1046             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1047             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1048             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1049                 return -EINVAL;
1050
1051         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1052                 printk("md: invalid superblock checksum on %s\n",
1053                         bdevname(rdev->bdev,b));
1054                 return -EINVAL;
1055         }
1056         if (le64_to_cpu(sb->data_size) < 10) {
1057                 printk("md: data_size too small on %s\n",
1058                        bdevname(rdev->bdev,b));
1059                 return -EINVAL;
1060         }
1061         rdev->preferred_minor = 0xffff;
1062         rdev->data_offset = le64_to_cpu(sb->data_offset);
1063         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1064
1065         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1066         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1067         if (rdev->sb_size & bmask)
1068                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1069
1070         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1071                 rdev->desc_nr = -1;
1072         else
1073                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1074
1075         if (refdev == 0)
1076                 ret = 1;
1077         else {
1078                 __u64 ev1, ev2;
1079                 struct mdp_superblock_1 *refsb = 
1080                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1081
1082                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1083                     sb->level != refsb->level ||
1084                     sb->layout != refsb->layout ||
1085                     sb->chunksize != refsb->chunksize) {
1086                         printk(KERN_WARNING "md: %s has strangely different"
1087                                 " superblock to %s\n",
1088                                 bdevname(rdev->bdev,b),
1089                                 bdevname(refdev->bdev,b2));
1090                         return -EINVAL;
1091                 }
1092                 ev1 = le64_to_cpu(sb->events);
1093                 ev2 = le64_to_cpu(refsb->events);
1094
1095                 if (ev1 > ev2)
1096                         ret = 1;
1097                 else
1098                         ret = 0;
1099         }
1100         if (minor_version) 
1101                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1102         else
1103                 rdev->size = rdev->sb_offset;
1104         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1105                 return -EINVAL;
1106         rdev->size = le64_to_cpu(sb->data_size)/2;
1107         if (le32_to_cpu(sb->chunksize))
1108                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1109
1110         if (le64_to_cpu(sb->size) > rdev->size*2)
1111                 return -EINVAL;
1112         return ret;
1113 }
1114
1115 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1116 {
1117         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1118         __u64 ev1 = le64_to_cpu(sb->events);
1119
1120         rdev->raid_disk = -1;
1121         rdev->flags = 0;
1122         if (mddev->raid_disks == 0) {
1123                 mddev->major_version = 1;
1124                 mddev->patch_version = 0;
1125                 mddev->persistent = 1;
1126                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1127                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1128                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1129                 mddev->level = le32_to_cpu(sb->level);
1130                 mddev->clevel[0] = 0;
1131                 mddev->layout = le32_to_cpu(sb->layout);
1132                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1133                 mddev->size = le64_to_cpu(sb->size)/2;
1134                 mddev->events = ev1;
1135                 mddev->bitmap_offset = 0;
1136                 mddev->default_bitmap_offset = 1024 >> 9;
1137                 
1138                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1139                 memcpy(mddev->uuid, sb->set_uuid, 16);
1140
1141                 mddev->max_disks =  (4096-256)/2;
1142
1143                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1144                     mddev->bitmap_file == NULL ) {
1145                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1146                             && mddev->level != 10) {
1147                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1148                                 return -EINVAL;
1149                         }
1150                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1151                 }
1152                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1153                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1154                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1155                         mddev->new_level = le32_to_cpu(sb->new_level);
1156                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1157                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1158                 } else {
1159                         mddev->reshape_position = MaxSector;
1160                         mddev->delta_disks = 0;
1161                         mddev->new_level = mddev->level;
1162                         mddev->new_layout = mddev->layout;
1163                         mddev->new_chunk = mddev->chunk_size;
1164                 }
1165
1166         } else if (mddev->pers == NULL) {
1167                 /* Insist of good event counter while assembling */
1168                 ++ev1;
1169                 if (ev1 < mddev->events)
1170                         return -EINVAL;
1171         } else if (mddev->bitmap) {
1172                 /* If adding to array with a bitmap, then we can accept an
1173                  * older device, but not too old.
1174                  */
1175                 if (ev1 < mddev->bitmap->events_cleared)
1176                         return 0;
1177         } else {
1178                 if (ev1 < mddev->events)
1179                         /* just a hot-add of a new device, leave raid_disk at -1 */
1180                         return 0;
1181         }
1182         if (mddev->level != LEVEL_MULTIPATH) {
1183                 int role;
1184                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1185                 switch(role) {
1186                 case 0xffff: /* spare */
1187                         break;
1188                 case 0xfffe: /* faulty */
1189                         set_bit(Faulty, &rdev->flags);
1190                         break;
1191                 default:
1192                         if ((le32_to_cpu(sb->feature_map) &
1193                              MD_FEATURE_RECOVERY_OFFSET))
1194                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1195                         else
1196                                 set_bit(In_sync, &rdev->flags);
1197                         rdev->raid_disk = role;
1198                         break;
1199                 }
1200                 if (sb->devflags & WriteMostly1)
1201                         set_bit(WriteMostly, &rdev->flags);
1202         } else /* MULTIPATH are always insync */
1203                 set_bit(In_sync, &rdev->flags);
1204
1205         return 0;
1206 }
1207
1208 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1209 {
1210         struct mdp_superblock_1 *sb;
1211         struct list_head *tmp;
1212         mdk_rdev_t *rdev2;
1213         int max_dev, i;
1214         /* make rdev->sb match mddev and rdev data. */
1215
1216         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1217
1218         sb->feature_map = 0;
1219         sb->pad0 = 0;
1220         sb->recovery_offset = cpu_to_le64(0);
1221         memset(sb->pad1, 0, sizeof(sb->pad1));
1222         memset(sb->pad2, 0, sizeof(sb->pad2));
1223         memset(sb->pad3, 0, sizeof(sb->pad3));
1224
1225         sb->utime = cpu_to_le64((__u64)mddev->utime);
1226         sb->events = cpu_to_le64(mddev->events);
1227         if (mddev->in_sync)
1228                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1229         else
1230                 sb->resync_offset = cpu_to_le64(0);
1231
1232         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1233
1234         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1235         sb->size = cpu_to_le64(mddev->size<<1);
1236
1237         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1238                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1239                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1240         }
1241
1242         if (rdev->raid_disk >= 0 &&
1243             !test_bit(In_sync, &rdev->flags) &&
1244             rdev->recovery_offset > 0) {
1245                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1246                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1247         }
1248
1249         if (mddev->reshape_position != MaxSector) {
1250                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1251                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1252                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1253                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1254                 sb->new_level = cpu_to_le32(mddev->new_level);
1255                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1256         }
1257
1258         max_dev = 0;
1259         ITERATE_RDEV(mddev,rdev2,tmp)
1260                 if (rdev2->desc_nr+1 > max_dev)
1261                         max_dev = rdev2->desc_nr+1;
1262         
1263         sb->max_dev = cpu_to_le32(max_dev);
1264         for (i=0; i<max_dev;i++)
1265                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1266         
1267         ITERATE_RDEV(mddev,rdev2,tmp) {
1268                 i = rdev2->desc_nr;
1269                 if (test_bit(Faulty, &rdev2->flags))
1270                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1271                 else if (test_bit(In_sync, &rdev2->flags))
1272                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1273                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1274                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1275                 else
1276                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1277         }
1278
1279         sb->sb_csum = calc_sb_1_csum(sb);
1280 }
1281
1282
1283 static struct super_type super_types[] = {
1284         [0] = {
1285                 .name   = "0.90.0",
1286                 .owner  = THIS_MODULE,
1287                 .load_super     = super_90_load,
1288                 .validate_super = super_90_validate,
1289                 .sync_super     = super_90_sync,
1290         },
1291         [1] = {
1292                 .name   = "md-1",
1293                 .owner  = THIS_MODULE,
1294                 .load_super     = super_1_load,
1295                 .validate_super = super_1_validate,
1296                 .sync_super     = super_1_sync,
1297         },
1298 };
1299
1300 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1301 {
1302         struct list_head *tmp, *tmp2;
1303         mdk_rdev_t *rdev, *rdev2;
1304
1305         ITERATE_RDEV(mddev1,rdev,tmp)
1306                 ITERATE_RDEV(mddev2, rdev2, tmp2)
1307                         if (rdev->bdev->bd_contains ==
1308                             rdev2->bdev->bd_contains)
1309                                 return 1;
1310
1311         return 0;
1312 }
1313
1314 static LIST_HEAD(pending_raid_disks);
1315
1316 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1317 {
1318         char b[BDEVNAME_SIZE];
1319         struct kobject *ko;
1320         char *s;
1321
1322         if (rdev->mddev) {
1323                 MD_BUG();
1324                 return -EINVAL;
1325         }
1326         /* make sure rdev->size exceeds mddev->size */
1327         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1328                 if (mddev->pers)
1329                         /* Cannot change size, so fail */
1330                         return -ENOSPC;
1331                 else
1332                         mddev->size = rdev->size;
1333         }
1334
1335         /* Verify rdev->desc_nr is unique.
1336          * If it is -1, assign a free number, else
1337          * check number is not in use
1338          */
1339         if (rdev->desc_nr < 0) {
1340                 int choice = 0;
1341                 if (mddev->pers) choice = mddev->raid_disks;
1342                 while (find_rdev_nr(mddev, choice))
1343                         choice++;
1344                 rdev->desc_nr = choice;
1345         } else {
1346                 if (find_rdev_nr(mddev, rdev->desc_nr))
1347                         return -EBUSY;
1348         }
1349         bdevname(rdev->bdev,b);
1350         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1351                 return -ENOMEM;
1352         while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1353                 *s = '!';
1354                         
1355         list_add(&rdev->same_set, &mddev->disks);
1356         rdev->mddev = mddev;
1357         printk(KERN_INFO "md: bind<%s>\n", b);
1358
1359         rdev->kobj.parent = &mddev->kobj;
1360         kobject_add(&rdev->kobj);
1361
1362         if (rdev->bdev->bd_part)
1363                 ko = &rdev->bdev->bd_part->kobj;
1364         else
1365                 ko = &rdev->bdev->bd_disk->kobj;
1366         sysfs_create_link(&rdev->kobj, ko, "block");
1367         bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1368         return 0;
1369 }
1370
1371 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1372 {
1373         char b[BDEVNAME_SIZE];
1374         if (!rdev->mddev) {
1375                 MD_BUG();
1376                 return;
1377         }
1378         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1379         list_del_init(&rdev->same_set);
1380         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1381         rdev->mddev = NULL;
1382         sysfs_remove_link(&rdev->kobj, "block");
1383         kobject_del(&rdev->kobj);
1384 }
1385
1386 /*
1387  * prevent the device from being mounted, repartitioned or
1388  * otherwise reused by a RAID array (or any other kernel
1389  * subsystem), by bd_claiming the device.
1390  */
1391 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1392 {
1393         int err = 0;
1394         struct block_device *bdev;
1395         char b[BDEVNAME_SIZE];
1396
1397         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1398         if (IS_ERR(bdev)) {
1399                 printk(KERN_ERR "md: could not open %s.\n",
1400                         __bdevname(dev, b));
1401                 return PTR_ERR(bdev);
1402         }
1403         err = bd_claim(bdev, rdev);
1404         if (err) {
1405                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1406                         bdevname(bdev, b));
1407                 blkdev_put(bdev);
1408                 return err;
1409         }
1410         rdev->bdev = bdev;
1411         return err;
1412 }
1413
1414 static void unlock_rdev(mdk_rdev_t *rdev)
1415 {
1416         struct block_device *bdev = rdev->bdev;
1417         rdev->bdev = NULL;
1418         if (!bdev)
1419                 MD_BUG();
1420         bd_release(bdev);
1421         blkdev_put(bdev);
1422 }
1423
1424 void md_autodetect_dev(dev_t dev);
1425
1426 static void export_rdev(mdk_rdev_t * rdev)
1427 {
1428         char b[BDEVNAME_SIZE];
1429         printk(KERN_INFO "md: export_rdev(%s)\n",
1430                 bdevname(rdev->bdev,b));
1431         if (rdev->mddev)
1432                 MD_BUG();
1433         free_disk_sb(rdev);
1434         list_del_init(&rdev->same_set);
1435 #ifndef MODULE
1436         md_autodetect_dev(rdev->bdev->bd_dev);
1437 #endif
1438         unlock_rdev(rdev);
1439         kobject_put(&rdev->kobj);
1440 }
1441
1442 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1443 {
1444         unbind_rdev_from_array(rdev);
1445         export_rdev(rdev);
1446 }
1447
1448 static void export_array(mddev_t *mddev)
1449 {
1450         struct list_head *tmp;
1451         mdk_rdev_t *rdev;
1452
1453         ITERATE_RDEV(mddev,rdev,tmp) {
1454                 if (!rdev->mddev) {
1455                         MD_BUG();
1456                         continue;
1457                 }
1458                 kick_rdev_from_array(rdev);
1459         }
1460         if (!list_empty(&mddev->disks))
1461                 MD_BUG();
1462         mddev->raid_disks = 0;
1463         mddev->major_version = 0;
1464 }
1465
1466 static void print_desc(mdp_disk_t *desc)
1467 {
1468         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1469                 desc->major,desc->minor,desc->raid_disk,desc->state);
1470 }
1471
1472 static void print_sb(mdp_super_t *sb)
1473 {
1474         int i;
1475
1476         printk(KERN_INFO 
1477                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1478                 sb->major_version, sb->minor_version, sb->patch_version,
1479                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1480                 sb->ctime);
1481         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1482                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1483                 sb->md_minor, sb->layout, sb->chunk_size);
1484         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1485                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1486                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1487                 sb->failed_disks, sb->spare_disks,
1488                 sb->sb_csum, (unsigned long)sb->events_lo);
1489
1490         printk(KERN_INFO);
1491         for (i = 0; i < MD_SB_DISKS; i++) {
1492                 mdp_disk_t *desc;
1493
1494                 desc = sb->disks + i;
1495                 if (desc->number || desc->major || desc->minor ||
1496                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1497                         printk("     D %2d: ", i);
1498                         print_desc(desc);
1499                 }
1500         }
1501         printk(KERN_INFO "md:     THIS: ");
1502         print_desc(&sb->this_disk);
1503
1504 }
1505
1506 static void print_rdev(mdk_rdev_t *rdev)
1507 {
1508         char b[BDEVNAME_SIZE];
1509         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1510                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1511                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1512                 rdev->desc_nr);
1513         if (rdev->sb_loaded) {
1514                 printk(KERN_INFO "md: rdev superblock:\n");
1515                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1516         } else
1517                 printk(KERN_INFO "md: no rdev superblock!\n");
1518 }
1519
1520 static void md_print_devices(void)
1521 {
1522         struct list_head *tmp, *tmp2;
1523         mdk_rdev_t *rdev;
1524         mddev_t *mddev;
1525         char b[BDEVNAME_SIZE];
1526
1527         printk("\n");
1528         printk("md:     **********************************\n");
1529         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1530         printk("md:     **********************************\n");
1531         ITERATE_MDDEV(mddev,tmp) {
1532
1533                 if (mddev->bitmap)
1534                         bitmap_print_sb(mddev->bitmap);
1535                 else
1536                         printk("%s: ", mdname(mddev));
1537                 ITERATE_RDEV(mddev,rdev,tmp2)
1538                         printk("<%s>", bdevname(rdev->bdev,b));
1539                 printk("\n");
1540
1541                 ITERATE_RDEV(mddev,rdev,tmp2)
1542                         print_rdev(rdev);
1543         }
1544         printk("md:     **********************************\n");
1545         printk("\n");
1546 }
1547
1548
1549 static void sync_sbs(mddev_t * mddev, int nospares)
1550 {
1551         /* Update each superblock (in-memory image), but
1552          * if we are allowed to, skip spares which already
1553          * have the right event counter, or have one earlier
1554          * (which would mean they aren't being marked as dirty
1555          * with the rest of the array)
1556          */
1557         mdk_rdev_t *rdev;
1558         struct list_head *tmp;
1559
1560         ITERATE_RDEV(mddev,rdev,tmp) {
1561                 if (rdev->sb_events == mddev->events ||
1562                     (nospares &&
1563                      rdev->raid_disk < 0 &&
1564                      (rdev->sb_events&1)==0 &&
1565                      rdev->sb_events+1 == mddev->events)) {
1566                         /* Don't update this superblock */
1567                         rdev->sb_loaded = 2;
1568                 } else {
1569                         super_types[mddev->major_version].
1570                                 sync_super(mddev, rdev);
1571                         rdev->sb_loaded = 1;
1572                 }
1573         }
1574 }
1575
1576 static void md_update_sb(mddev_t * mddev, int force_change)
1577 {
1578         int err;
1579         struct list_head *tmp;
1580         mdk_rdev_t *rdev;
1581         int sync_req;
1582         int nospares = 0;
1583
1584 repeat:
1585         spin_lock_irq(&mddev->write_lock);
1586
1587         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1588         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1589                 force_change = 1;
1590         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1591                 /* just a clean<-> dirty transition, possibly leave spares alone,
1592                  * though if events isn't the right even/odd, we will have to do
1593                  * spares after all
1594                  */
1595                 nospares = 1;
1596         if (force_change)
1597                 nospares = 0;
1598         if (mddev->degraded)
1599                 /* If the array is degraded, then skipping spares is both
1600                  * dangerous and fairly pointless.
1601                  * Dangerous because a device that was removed from the array
1602                  * might have a event_count that still looks up-to-date,
1603                  * so it can be re-added without a resync.
1604                  * Pointless because if there are any spares to skip,
1605                  * then a recovery will happen and soon that array won't
1606                  * be degraded any more and the spare can go back to sleep then.
1607                  */
1608                 nospares = 0;
1609
1610         sync_req = mddev->in_sync;
1611         mddev->utime = get_seconds();
1612
1613         /* If this is just a dirty<->clean transition, and the array is clean
1614          * and 'events' is odd, we can roll back to the previous clean state */
1615         if (nospares
1616             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1617             && (mddev->events & 1)
1618             && mddev->events != 1)
1619                 mddev->events--;
1620         else {
1621                 /* otherwise we have to go forward and ... */
1622                 mddev->events ++;
1623                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1624                         /* .. if the array isn't clean, insist on an odd 'events' */
1625                         if ((mddev->events&1)==0) {
1626                                 mddev->events++;
1627                                 nospares = 0;
1628                         }
1629                 } else {
1630                         /* otherwise insist on an even 'events' (for clean states) */
1631                         if ((mddev->events&1)) {
1632                                 mddev->events++;
1633                                 nospares = 0;
1634                         }
1635                 }
1636         }
1637
1638         if (!mddev->events) {
1639                 /*
1640                  * oops, this 64-bit counter should never wrap.
1641                  * Either we are in around ~1 trillion A.C., assuming
1642                  * 1 reboot per second, or we have a bug:
1643                  */
1644                 MD_BUG();
1645                 mddev->events --;
1646         }
1647         sync_sbs(mddev, nospares);
1648
1649         /*
1650          * do not write anything to disk if using
1651          * nonpersistent superblocks
1652          */
1653         if (!mddev->persistent) {
1654                 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1655                 spin_unlock_irq(&mddev->write_lock);
1656                 wake_up(&mddev->sb_wait);
1657                 return;
1658         }
1659         spin_unlock_irq(&mddev->write_lock);
1660
1661         dprintk(KERN_INFO 
1662                 "md: updating %s RAID superblock on device (in sync %d)\n",
1663                 mdname(mddev),mddev->in_sync);
1664
1665         err = bitmap_update_sb(mddev->bitmap);
1666         ITERATE_RDEV(mddev,rdev,tmp) {
1667                 char b[BDEVNAME_SIZE];
1668                 dprintk(KERN_INFO "md: ");
1669                 if (rdev->sb_loaded != 1)
1670                         continue; /* no noise on spare devices */
1671                 if (test_bit(Faulty, &rdev->flags))
1672                         dprintk("(skipping faulty ");
1673
1674                 dprintk("%s ", bdevname(rdev->bdev,b));
1675                 if (!test_bit(Faulty, &rdev->flags)) {
1676                         md_super_write(mddev,rdev,
1677                                        rdev->sb_offset<<1, rdev->sb_size,
1678                                        rdev->sb_page);
1679                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1680                                 bdevname(rdev->bdev,b),
1681                                 (unsigned long long)rdev->sb_offset);
1682                         rdev->sb_events = mddev->events;
1683
1684                 } else
1685                         dprintk(")\n");
1686                 if (mddev->level == LEVEL_MULTIPATH)
1687                         /* only need to write one superblock... */
1688                         break;
1689         }
1690         md_super_wait(mddev);
1691         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1692
1693         spin_lock_irq(&mddev->write_lock);
1694         if (mddev->in_sync != sync_req ||
1695             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1696                 /* have to write it out again */
1697                 spin_unlock_irq(&mddev->write_lock);
1698                 goto repeat;
1699         }
1700         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1701         spin_unlock_irq(&mddev->write_lock);
1702         wake_up(&mddev->sb_wait);
1703
1704 }
1705
1706 /* words written to sysfs files may, or my not, be \n terminated.
1707  * We want to accept with case. For this we use cmd_match.
1708  */
1709 static int cmd_match(const char *cmd, const char *str)
1710 {
1711         /* See if cmd, written into a sysfs file, matches
1712          * str.  They must either be the same, or cmd can
1713          * have a trailing newline
1714          */
1715         while (*cmd && *str && *cmd == *str) {
1716                 cmd++;
1717                 str++;
1718         }
1719         if (*cmd == '\n')
1720                 cmd++;
1721         if (*str || *cmd)
1722                 return 0;
1723         return 1;
1724 }
1725
1726 struct rdev_sysfs_entry {
1727         struct attribute attr;
1728         ssize_t (*show)(mdk_rdev_t *, char *);
1729         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1730 };
1731
1732 static ssize_t
1733 state_show(mdk_rdev_t *rdev, char *page)
1734 {
1735         char *sep = "";
1736         int len=0;
1737
1738         if (test_bit(Faulty, &rdev->flags)) {
1739                 len+= sprintf(page+len, "%sfaulty",sep);
1740                 sep = ",";
1741         }
1742         if (test_bit(In_sync, &rdev->flags)) {
1743                 len += sprintf(page+len, "%sin_sync",sep);
1744                 sep = ",";
1745         }
1746         if (test_bit(WriteMostly, &rdev->flags)) {
1747                 len += sprintf(page+len, "%swrite_mostly",sep);
1748                 sep = ",";
1749         }
1750         if (!test_bit(Faulty, &rdev->flags) &&
1751             !test_bit(In_sync, &rdev->flags)) {
1752                 len += sprintf(page+len, "%sspare", sep);
1753                 sep = ",";
1754         }
1755         return len+sprintf(page+len, "\n");
1756 }
1757
1758 static ssize_t
1759 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1760 {
1761         /* can write
1762          *  faulty  - simulates and error
1763          *  remove  - disconnects the device
1764          *  writemostly - sets write_mostly
1765          *  -writemostly - clears write_mostly
1766          */
1767         int err = -EINVAL;
1768         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1769                 md_error(rdev->mddev, rdev);
1770                 err = 0;
1771         } else if (cmd_match(buf, "remove")) {
1772                 if (rdev->raid_disk >= 0)
1773                         err = -EBUSY;
1774                 else {
1775                         mddev_t *mddev = rdev->mddev;
1776                         kick_rdev_from_array(rdev);
1777                         if (mddev->pers)
1778                                 md_update_sb(mddev, 1);
1779                         md_new_event(mddev);
1780                         err = 0;
1781                 }
1782         } else if (cmd_match(buf, "writemostly")) {
1783                 set_bit(WriteMostly, &rdev->flags);
1784                 err = 0;
1785         } else if (cmd_match(buf, "-writemostly")) {
1786                 clear_bit(WriteMostly, &rdev->flags);
1787                 err = 0;
1788         }
1789         return err ? err : len;
1790 }
1791 static struct rdev_sysfs_entry rdev_state =
1792 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1793
1794 static ssize_t
1795 super_show(mdk_rdev_t *rdev, char *page)
1796 {
1797         if (rdev->sb_loaded && rdev->sb_size) {
1798                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1799                 return rdev->sb_size;
1800         } else
1801                 return 0;
1802 }
1803 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1804
1805 static ssize_t
1806 errors_show(mdk_rdev_t *rdev, char *page)
1807 {
1808         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1809 }
1810
1811 static ssize_t
1812 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1813 {
1814         char *e;
1815         unsigned long n = simple_strtoul(buf, &e, 10);
1816         if (*buf && (*e == 0 || *e == '\n')) {
1817                 atomic_set(&rdev->corrected_errors, n);
1818                 return len;
1819         }
1820         return -EINVAL;
1821 }
1822 static struct rdev_sysfs_entry rdev_errors =
1823 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1824
1825 static ssize_t
1826 slot_show(mdk_rdev_t *rdev, char *page)
1827 {
1828         if (rdev->raid_disk < 0)
1829                 return sprintf(page, "none\n");
1830         else
1831                 return sprintf(page, "%d\n", rdev->raid_disk);
1832 }
1833
1834 static ssize_t
1835 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1836 {
1837         char *e;
1838         int slot = simple_strtoul(buf, &e, 10);
1839         if (strncmp(buf, "none", 4)==0)
1840                 slot = -1;
1841         else if (e==buf || (*e && *e!= '\n'))
1842                 return -EINVAL;
1843         if (rdev->mddev->pers)
1844                 /* Cannot set slot in active array (yet) */
1845                 return -EBUSY;
1846         if (slot >= rdev->mddev->raid_disks)
1847                 return -ENOSPC;
1848         rdev->raid_disk = slot;
1849         /* assume it is working */
1850         rdev->flags = 0;
1851         set_bit(In_sync, &rdev->flags);
1852         return len;
1853 }
1854
1855
1856 static struct rdev_sysfs_entry rdev_slot =
1857 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1858
1859 static ssize_t
1860 offset_show(mdk_rdev_t *rdev, char *page)
1861 {
1862         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1863 }
1864
1865 static ssize_t
1866 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1867 {
1868         char *e;
1869         unsigned long long offset = simple_strtoull(buf, &e, 10);
1870         if (e==buf || (*e && *e != '\n'))
1871                 return -EINVAL;
1872         if (rdev->mddev->pers)
1873                 return -EBUSY;
1874         rdev->data_offset = offset;
1875         return len;
1876 }
1877
1878 static struct rdev_sysfs_entry rdev_offset =
1879 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1880
1881 static ssize_t
1882 rdev_size_show(mdk_rdev_t *rdev, char *page)
1883 {
1884         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1885 }
1886
1887 static ssize_t
1888 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1889 {
1890         char *e;
1891         unsigned long long size = simple_strtoull(buf, &e, 10);
1892         if (e==buf || (*e && *e != '\n'))
1893                 return -EINVAL;
1894         if (rdev->mddev->pers)
1895                 return -EBUSY;
1896         rdev->size = size;
1897         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1898                 rdev->mddev->size = size;
1899         return len;
1900 }
1901
1902 static struct rdev_sysfs_entry rdev_size =
1903 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1904
1905 static struct attribute *rdev_default_attrs[] = {
1906         &rdev_state.attr,
1907         &rdev_super.attr,
1908         &rdev_errors.attr,
1909         &rdev_slot.attr,
1910         &rdev_offset.attr,
1911         &rdev_size.attr,
1912         NULL,
1913 };
1914 static ssize_t
1915 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1916 {
1917         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1918         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1919
1920         if (!entry->show)
1921                 return -EIO;
1922         return entry->show(rdev, page);
1923 }
1924
1925 static ssize_t
1926 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1927               const char *page, size_t length)
1928 {
1929         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1930         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1931
1932         if (!entry->store)
1933                 return -EIO;
1934         if (!capable(CAP_SYS_ADMIN))
1935                 return -EACCES;
1936         return entry->store(rdev, page, length);
1937 }
1938
1939 static void rdev_free(struct kobject *ko)
1940 {
1941         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1942         kfree(rdev);
1943 }
1944 static struct sysfs_ops rdev_sysfs_ops = {
1945         .show           = rdev_attr_show,
1946         .store          = rdev_attr_store,
1947 };
1948 static struct kobj_type rdev_ktype = {
1949         .release        = rdev_free,
1950         .sysfs_ops      = &rdev_sysfs_ops,
1951         .default_attrs  = rdev_default_attrs,
1952 };
1953
1954 /*
1955  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1956  *
1957  * mark the device faulty if:
1958  *
1959  *   - the device is nonexistent (zero size)
1960  *   - the device has no valid superblock
1961  *
1962  * a faulty rdev _never_ has rdev->sb set.
1963  */
1964 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1965 {
1966         char b[BDEVNAME_SIZE];
1967         int err;
1968         mdk_rdev_t *rdev;
1969         sector_t size;
1970
1971         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1972         if (!rdev) {
1973                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1974                 return ERR_PTR(-ENOMEM);
1975         }
1976
1977         if ((err = alloc_disk_sb(rdev)))
1978                 goto abort_free;
1979
1980         err = lock_rdev(rdev, newdev);
1981         if (err)
1982                 goto abort_free;
1983
1984         rdev->kobj.parent = NULL;
1985         rdev->kobj.ktype = &rdev_ktype;
1986         kobject_init(&rdev->kobj);
1987
1988         rdev->desc_nr = -1;
1989         rdev->saved_raid_disk = -1;
1990         rdev->raid_disk = -1;
1991         rdev->flags = 0;
1992         rdev->data_offset = 0;
1993         rdev->sb_events = 0;
1994         atomic_set(&rdev->nr_pending, 0);
1995         atomic_set(&rdev->read_errors, 0);
1996         atomic_set(&rdev->corrected_errors, 0);
1997
1998         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1999         if (!size) {
2000                 printk(KERN_WARNING 
2001                         "md: %s has zero or unknown size, marking faulty!\n",
2002                         bdevname(rdev->bdev,b));
2003                 err = -EINVAL;
2004                 goto abort_free;
2005         }
2006
2007         if (super_format >= 0) {
2008                 err = super_types[super_format].
2009                         load_super(rdev, NULL, super_minor);
2010                 if (err == -EINVAL) {
2011                         printk(KERN_WARNING 
2012                                 "md: %s has invalid sb, not importing!\n",
2013                                 bdevname(rdev->bdev,b));
2014                         goto abort_free;
2015                 }
2016                 if (err < 0) {
2017                         printk(KERN_WARNING 
2018                                 "md: could not read %s's sb, not importing!\n",
2019                                 bdevname(rdev->bdev,b));
2020                         goto abort_free;
2021                 }
2022         }
2023         INIT_LIST_HEAD(&rdev->same_set);
2024
2025         return rdev;
2026
2027 abort_free:
2028         if (rdev->sb_page) {
2029                 if (rdev->bdev)
2030                         unlock_rdev(rdev);
2031                 free_disk_sb(rdev);
2032         }
2033         kfree(rdev);
2034         return ERR_PTR(err);
2035 }
2036
2037 /*
2038  * Check a full RAID array for plausibility
2039  */
2040
2041
2042 static void analyze_sbs(mddev_t * mddev)
2043 {
2044         int i;
2045         struct list_head *tmp;
2046         mdk_rdev_t *rdev, *freshest;
2047         char b[BDEVNAME_SIZE];
2048
2049         freshest = NULL;
2050         ITERATE_RDEV(mddev,rdev,tmp)
2051                 switch (super_types[mddev->major_version].
2052                         load_super(rdev, freshest, mddev->minor_version)) {
2053                 case 1:
2054                         freshest = rdev;
2055                         break;
2056                 case 0:
2057                         break;
2058                 default:
2059                         printk( KERN_ERR \
2060                                 "md: fatal superblock inconsistency in %s"
2061                                 " -- removing from array\n", 
2062                                 bdevname(rdev->bdev,b));
2063                         kick_rdev_from_array(rdev);
2064                 }
2065
2066
2067         super_types[mddev->major_version].
2068                 validate_super(mddev, freshest);
2069
2070         i = 0;
2071         ITERATE_RDEV(mddev,rdev,tmp) {
2072                 if (rdev != freshest)
2073                         if (super_types[mddev->major_version].
2074                             validate_super(mddev, rdev)) {
2075                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2076                                         " from array!\n",
2077                                         bdevname(rdev->bdev,b));
2078                                 kick_rdev_from_array(rdev);
2079                                 continue;
2080                         }
2081                 if (mddev->level == LEVEL_MULTIPATH) {
2082                         rdev->desc_nr = i++;
2083                         rdev->raid_disk = rdev->desc_nr;
2084                         set_bit(In_sync, &rdev->flags);
2085                 }
2086         }
2087
2088
2089
2090         if (mddev->recovery_cp != MaxSector &&
2091             mddev->level >= 1)
2092                 printk(KERN_ERR "md: %s: raid array is not clean"
2093                        " -- starting background reconstruction\n",
2094                        mdname(mddev));
2095
2096 }
2097
2098 static ssize_t
2099 safe_delay_show(mddev_t *mddev, char *page)
2100 {
2101         int msec = (mddev->safemode_delay*1000)/HZ;
2102         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2103 }
2104 static ssize_t
2105 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2106 {
2107         int scale=1;
2108         int dot=0;
2109         int i;
2110         unsigned long msec;
2111         char buf[30];
2112         char *e;
2113         /* remove a period, and count digits after it */
2114         if (len >= sizeof(buf))
2115                 return -EINVAL;
2116         strlcpy(buf, cbuf, len);
2117         buf[len] = 0;
2118         for (i=0; i<len; i++) {
2119                 if (dot) {
2120                         if (isdigit(buf[i])) {
2121                                 buf[i-1] = buf[i];
2122                                 scale *= 10;
2123                         }
2124                         buf[i] = 0;
2125                 } else if (buf[i] == '.') {
2126                         dot=1;
2127                         buf[i] = 0;
2128                 }
2129         }
2130         msec = simple_strtoul(buf, &e, 10);
2131         if (e == buf || (*e && *e != '\n'))
2132                 return -EINVAL;
2133         msec = (msec * 1000) / scale;
2134         if (msec == 0)
2135                 mddev->safemode_delay = 0;
2136         else {
2137                 mddev->safemode_delay = (msec*HZ)/1000;
2138                 if (mddev->safemode_delay == 0)
2139                         mddev->safemode_delay = 1;
2140         }
2141         return len;
2142 }
2143 static struct md_sysfs_entry md_safe_delay =
2144 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2145
2146 static ssize_t
2147 level_show(mddev_t *mddev, char *page)
2148 {
2149         struct mdk_personality *p = mddev->pers;
2150         if (p)
2151                 return sprintf(page, "%s\n", p->name);
2152         else if (mddev->clevel[0])
2153                 return sprintf(page, "%s\n", mddev->clevel);
2154         else if (mddev->level != LEVEL_NONE)
2155                 return sprintf(page, "%d\n", mddev->level);
2156         else
2157                 return 0;
2158 }
2159
2160 static ssize_t
2161 level_store(mddev_t *mddev, const char *buf, size_t len)
2162 {
2163         int rv = len;
2164         if (mddev->pers)
2165                 return -EBUSY;
2166         if (len == 0)
2167                 return 0;
2168         if (len >= sizeof(mddev->clevel))
2169                 return -ENOSPC;
2170         strncpy(mddev->clevel, buf, len);
2171         if (mddev->clevel[len-1] == '\n')
2172                 len--;
2173         mddev->clevel[len] = 0;
2174         mddev->level = LEVEL_NONE;
2175         return rv;
2176 }
2177
2178 static struct md_sysfs_entry md_level =
2179 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2180
2181
2182 static ssize_t
2183 layout_show(mddev_t *mddev, char *page)
2184 {
2185         /* just a number, not meaningful for all levels */
2186         return sprintf(page, "%d\n", mddev->layout);
2187 }
2188
2189 static ssize_t
2190 layout_store(mddev_t *mddev, const char *buf, size_t len)
2191 {
2192         char *e;
2193         unsigned long n = simple_strtoul(buf, &e, 10);
2194         if (mddev->pers)
2195                 return -EBUSY;
2196
2197         if (!*buf || (*e && *e != '\n'))
2198                 return -EINVAL;
2199
2200         mddev->layout = n;
2201         return len;
2202 }
2203 static struct md_sysfs_entry md_layout =
2204 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2205
2206
2207 static ssize_t
2208 raid_disks_show(mddev_t *mddev, char *page)
2209 {
2210         if (mddev->raid_disks == 0)
2211                 return 0;
2212         return sprintf(page, "%d\n", mddev->raid_disks);
2213 }
2214
2215 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2216
2217 static ssize_t
2218 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2219 {
2220         char *e;
2221         int rv = 0;
2222         unsigned long n = simple_strtoul(buf, &e, 10);
2223
2224         if (!*buf || (*e && *e != '\n'))
2225                 return -EINVAL;
2226
2227         if (mddev->pers)
2228                 rv = update_raid_disks(mddev, n);
2229         else
2230                 mddev->raid_disks = n;
2231         return rv ? rv : len;
2232 }
2233 static struct md_sysfs_entry md_raid_disks =
2234 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2235
2236 static ssize_t
2237 chunk_size_show(mddev_t *mddev, char *page)
2238 {
2239         return sprintf(page, "%d\n", mddev->chunk_size);
2240 }
2241
2242 static ssize_t
2243 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2244 {
2245         /* can only set chunk_size if array is not yet active */
2246         char *e;
2247         unsigned long n = simple_strtoul(buf, &e, 10);
2248
2249         if (mddev->pers)
2250                 return -EBUSY;
2251         if (!*buf || (*e && *e != '\n'))
2252                 return -EINVAL;
2253
2254         mddev->chunk_size = n;
2255         return len;
2256 }
2257 static struct md_sysfs_entry md_chunk_size =
2258 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2259
2260 static ssize_t
2261 resync_start_show(mddev_t *mddev, char *page)
2262 {
2263         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2264 }
2265
2266 static ssize_t
2267 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2268 {
2269         /* can only set chunk_size if array is not yet active */
2270         char *e;
2271         unsigned long long n = simple_strtoull(buf, &e, 10);
2272
2273         if (mddev->pers)
2274                 return -EBUSY;
2275         if (!*buf || (*e && *e != '\n'))
2276                 return -EINVAL;
2277
2278         mddev->recovery_cp = n;
2279         return len;
2280 }
2281 static struct md_sysfs_entry md_resync_start =
2282 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2283
2284 /*
2285  * The array state can be:
2286  *
2287  * clear
2288  *     No devices, no size, no level
2289  *     Equivalent to STOP_ARRAY ioctl
2290  * inactive
2291  *     May have some settings, but array is not active
2292  *        all IO results in error
2293  *     When written, doesn't tear down array, but just stops it
2294  * suspended (not supported yet)
2295  *     All IO requests will block. The array can be reconfigured.
2296  *     Writing this, if accepted, will block until array is quiessent
2297  * readonly
2298  *     no resync can happen.  no superblocks get written.
2299  *     write requests fail
2300  * read-auto
2301  *     like readonly, but behaves like 'clean' on a write request.
2302  *
2303  * clean - no pending writes, but otherwise active.
2304  *     When written to inactive array, starts without resync
2305  *     If a write request arrives then
2306  *       if metadata is known, mark 'dirty' and switch to 'active'.
2307  *       if not known, block and switch to write-pending
2308  *     If written to an active array that has pending writes, then fails.
2309  * active
2310  *     fully active: IO and resync can be happening.
2311  *     When written to inactive array, starts with resync
2312  *
2313  * write-pending
2314  *     clean, but writes are blocked waiting for 'active' to be written.
2315  *
2316  * active-idle
2317  *     like active, but no writes have been seen for a while (100msec).
2318  *
2319  */
2320 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2321                    write_pending, active_idle, bad_word};
2322 static char *array_states[] = {
2323         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2324         "write-pending", "active-idle", NULL };
2325
2326 static int match_word(const char *word, char **list)
2327 {
2328         int n;
2329         for (n=0; list[n]; n++)
2330                 if (cmd_match(word, list[n]))
2331                         break;
2332         return n;
2333 }
2334
2335 static ssize_t
2336 array_state_show(mddev_t *mddev, char *page)
2337 {
2338         enum array_state st = inactive;
2339
2340         if (mddev->pers)
2341                 switch(mddev->ro) {
2342                 case 1:
2343                         st = readonly;
2344                         break;
2345                 case 2:
2346                         st = read_auto;
2347                         break;
2348                 case 0:
2349                         if (mddev->in_sync)
2350                                 st = clean;
2351                         else if (mddev->safemode)
2352                                 st = active_idle;
2353                         else
2354                                 st = active;
2355                 }
2356         else {
2357                 if (list_empty(&mddev->disks) &&
2358                     mddev->raid_disks == 0 &&
2359                     mddev->size == 0)
2360                         st = clear;
2361                 else
2362                         st = inactive;
2363         }
2364         return sprintf(page, "%s\n", array_states[st]);
2365 }
2366
2367 static int do_md_stop(mddev_t * mddev, int ro);
2368 static int do_md_run(mddev_t * mddev);
2369 static int restart_array(mddev_t *mddev);
2370
2371 static ssize_t
2372 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2373 {
2374         int err = -EINVAL;
2375         enum array_state st = match_word(buf, array_states);
2376         switch(st) {
2377         case bad_word:
2378                 break;
2379         case clear:
2380                 /* stopping an active array */
2381                 if (mddev->pers) {
2382                         if (atomic_read(&mddev->active) > 1)
2383                                 return -EBUSY;
2384                         err = do_md_stop(mddev, 0);
2385                 }
2386                 break;
2387         case inactive:
2388                 /* stopping an active array */
2389                 if (mddev->pers) {
2390                         if (atomic_read(&mddev->active) > 1)
2391                                 return -EBUSY;
2392                         err = do_md_stop(mddev, 2);
2393                 }
2394                 break;
2395         case suspended:
2396                 break; /* not supported yet */
2397         case readonly:
2398                 if (mddev->pers)
2399                         err = do_md_stop(mddev, 1);
2400                 else {
2401                         mddev->ro = 1;
2402                         err = do_md_run(mddev);
2403                 }
2404                 break;
2405         case read_auto:
2406                 /* stopping an active array */
2407                 if (mddev->pers) {
2408                         err = do_md_stop(mddev, 1);
2409                         if (err == 0)
2410                                 mddev->ro = 2; /* FIXME mark devices writable */
2411                 } else {
2412                         mddev->ro = 2;
2413                         err = do_md_run(mddev);
2414                 }
2415                 break;
2416         case clean:
2417                 if (mddev->pers) {
2418                         restart_array(mddev);
2419                         spin_lock_irq(&mddev->write_lock);
2420                         if (atomic_read(&mddev->writes_pending) == 0) {
2421                                 mddev->in_sync = 1;
2422                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
2423                         }
2424                         spin_unlock_irq(&mddev->write_lock);
2425                 } else {
2426                         mddev->ro = 0;
2427                         mddev->recovery_cp = MaxSector;
2428                         err = do_md_run(mddev);
2429                 }
2430                 break;
2431         case active:
2432                 if (mddev->pers) {
2433                         restart_array(mddev);
2434                         clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2435                         wake_up(&mddev->sb_wait);
2436                         err = 0;
2437                 } else {
2438                         mddev->ro = 0;
2439                         err = do_md_run(mddev);
2440                 }
2441                 break;
2442         case write_pending:
2443         case active_idle:
2444                 /* these cannot be set */
2445                 break;
2446         }
2447         if (err)
2448                 return err;
2449         else
2450                 return len;
2451 }
2452 static struct md_sysfs_entry md_array_state =
2453 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2454
2455 static ssize_t
2456 null_show(mddev_t *mddev, char *page)
2457 {
2458         return -EINVAL;
2459 }
2460
2461 static ssize_t
2462 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2463 {
2464         /* buf must be %d:%d\n? giving major and minor numbers */
2465         /* The new device is added to the array.
2466          * If the array has a persistent superblock, we read the
2467          * superblock to initialise info and check validity.
2468          * Otherwise, only checking done is that in bind_rdev_to_array,
2469          * which mainly checks size.
2470          */
2471         char *e;
2472         int major = simple_strtoul(buf, &e, 10);
2473         int minor;
2474         dev_t dev;
2475         mdk_rdev_t *rdev;
2476         int err;
2477
2478         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2479                 return -EINVAL;
2480         minor = simple_strtoul(e+1, &e, 10);
2481         if (*e && *e != '\n')
2482                 return -EINVAL;
2483         dev = MKDEV(major, minor);
2484         if (major != MAJOR(dev) ||
2485             minor != MINOR(dev))
2486                 return -EOVERFLOW;
2487
2488
2489         if (mddev->persistent) {
2490                 rdev = md_import_device(dev, mddev->major_version,
2491                                         mddev->minor_version);
2492                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2493                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2494                                                        mdk_rdev_t, same_set);
2495                         err = super_types[mddev->major_version]
2496                                 .load_super(rdev, rdev0, mddev->minor_version);
2497                         if (err < 0)
2498                                 goto out;
2499                 }
2500         } else
2501                 rdev = md_import_device(dev, -1, -1);
2502
2503         if (IS_ERR(rdev))
2504                 return PTR_ERR(rdev);
2505         err = bind_rdev_to_array(rdev, mddev);
2506  out:
2507         if (err)
2508                 export_rdev(rdev);
2509         return err ? err : len;
2510 }
2511
2512 static struct md_sysfs_entry md_new_device =
2513 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2514
2515 static ssize_t
2516 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2517 {
2518         char *end;
2519         unsigned long chunk, end_chunk;
2520
2521         if (!mddev->bitmap)
2522                 goto out;
2523         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2524         while (*buf) {
2525                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2526                 if (buf == end) break;
2527                 if (*end == '-') { /* range */
2528                         buf = end + 1;
2529                         end_chunk = simple_strtoul(buf, &end, 0);
2530                         if (buf == end) break;
2531                 }
2532                 if (*end && !isspace(*end)) break;
2533                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2534                 buf = end;
2535                 while (isspace(*buf)) buf++;
2536         }
2537         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2538 out:
2539         return len;
2540 }
2541
2542 static struct md_sysfs_entry md_bitmap =
2543 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2544
2545 static ssize_t
2546 size_show(mddev_t *mddev, char *page)
2547 {
2548         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2549 }
2550
2551 static int update_size(mddev_t *mddev, unsigned long size);
2552
2553 static ssize_t
2554 size_store(mddev_t *mddev, const char *buf, size_t len)
2555 {
2556         /* If array is inactive, we can reduce the component size, but
2557          * not increase it (except from 0).
2558          * If array is active, we can try an on-line resize
2559          */
2560         char *e;
2561         int err = 0;
2562         unsigned long long size = simple_strtoull(buf, &e, 10);
2563         if (!*buf || *buf == '\n' ||
2564             (*e && *e != '\n'))
2565                 return -EINVAL;
2566
2567         if (mddev->pers) {
2568                 err = update_size(mddev, size);
2569                 md_update_sb(mddev, 1);
2570         } else {
2571                 if (mddev->size == 0 ||
2572                     mddev->size > size)
2573                         mddev->size = size;
2574                 else
2575                         err = -ENOSPC;
2576         }
2577         return err ? err : len;
2578 }
2579
2580 static struct md_sysfs_entry md_size =
2581 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2582
2583
2584 /* Metdata version.
2585  * This is either 'none' for arrays with externally managed metadata,
2586  * or N.M for internally known formats
2587  */
2588 static ssize_t
2589 metadata_show(mddev_t *mddev, char *page)
2590 {
2591         if (mddev->persistent)
2592                 return sprintf(page, "%d.%d\n",
2593                                mddev->major_version, mddev->minor_version);
2594         else
2595                 return sprintf(page, "none\n");
2596 }
2597
2598 static ssize_t
2599 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2600 {
2601         int major, minor;
2602         char *e;
2603         if (!list_empty(&mddev->disks))
2604                 return -EBUSY;
2605
2606         if (cmd_match(buf, "none")) {
2607                 mddev->persistent = 0;
2608                 mddev->major_version = 0;
2609                 mddev->minor_version = 90;
2610                 return len;
2611         }
2612         major = simple_strtoul(buf, &e, 10);
2613         if (e==buf || *e != '.')
2614                 return -EINVAL;
2615         buf = e+1;
2616         minor = simple_strtoul(buf, &e, 10);
2617         if (e==buf || (*e && *e != '\n') )
2618                 return -EINVAL;
2619         if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2620             super_types[major].name == NULL)
2621                 return -ENOENT;
2622         mddev->major_version = major;
2623         mddev->minor_version = minor;
2624         mddev->persistent = 1;
2625         return len;
2626 }
2627
2628 static struct md_sysfs_entry md_metadata =
2629 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2630
2631 static ssize_t
2632 action_show(mddev_t *mddev, char *page)
2633 {
2634         char *type = "idle";
2635         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2636             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2637                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2638                         type = "reshape";
2639                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2640                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2641                                 type = "resync";
2642                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2643                                 type = "check";
2644                         else
2645                                 type = "repair";
2646                 } else
2647                         type = "recover";
2648         }
2649         return sprintf(page, "%s\n", type);
2650 }
2651
2652 static ssize_t
2653 action_store(mddev_t *mddev, const char *page, size_t len)
2654 {
2655         if (!mddev->pers || !mddev->pers->sync_request)
2656                 return -EINVAL;
2657
2658         if (cmd_match(page, "idle")) {
2659                 if (mddev->sync_thread) {
2660                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2661                         md_unregister_thread(mddev->sync_thread);
2662                         mddev->sync_thread = NULL;
2663                         mddev->recovery = 0;
2664                 }
2665         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2666                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2667                 return -EBUSY;
2668         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2669                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2670         else if (cmd_match(page, "reshape")) {
2671                 int err;
2672                 if (mddev->pers->start_reshape == NULL)
2673                         return -EINVAL;
2674                 err = mddev->pers->start_reshape(mddev);
2675                 if (err)
2676                         return err;
2677         } else {
2678                 if (cmd_match(page, "check"))
2679                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2680                 else if (!cmd_match(page, "repair"))
2681                         return -EINVAL;
2682                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2683                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2684         }
2685         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2686         md_wakeup_thread(mddev->thread);
2687         return len;
2688 }
2689
2690 static ssize_t
2691 mismatch_cnt_show(mddev_t *mddev, char *page)
2692 {
2693         return sprintf(page, "%llu\n",
2694                        (unsigned long long) mddev->resync_mismatches);
2695 }
2696
2697 static struct md_sysfs_entry md_scan_mode =
2698 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2699
2700
2701 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2702
2703 static ssize_t
2704 sync_min_show(mddev_t *mddev, char *page)
2705 {
2706         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2707                        mddev->sync_speed_min ? "local": "system");
2708 }
2709
2710 static ssize_t
2711 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2712 {
2713         int min;
2714         char *e;
2715         if (strncmp(buf, "system", 6)==0) {
2716                 mddev->sync_speed_min = 0;
2717                 return len;
2718         }
2719         min = simple_strtoul(buf, &e, 10);
2720         if (buf == e || (*e && *e != '\n') || min <= 0)
2721                 return -EINVAL;
2722         mddev->sync_speed_min = min;
2723         return len;
2724 }
2725
2726 static struct md_sysfs_entry md_sync_min =
2727 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2728
2729 static ssize_t
2730 sync_max_show(mddev_t *mddev, char *page)
2731 {
2732         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2733                        mddev->sync_speed_max ? "local": "system");
2734 }
2735
2736 static ssize_t
2737 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2738 {
2739         int max;
2740         char *e;
2741         if (strncmp(buf, "system", 6)==0) {
2742                 mddev->sync_speed_max = 0;
2743                 return len;
2744         }
2745         max = simple_strtoul(buf, &e, 10);
2746         if (buf == e || (*e && *e != '\n') || max <= 0)
2747                 return -EINVAL;
2748         mddev->sync_speed_max = max;
2749         return len;
2750 }
2751
2752 static struct md_sysfs_entry md_sync_max =
2753 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2754
2755
2756 static ssize_t
2757 sync_speed_show(mddev_t *mddev, char *page)
2758 {
2759         unsigned long resync, dt, db;
2760         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2761         dt = ((jiffies - mddev->resync_mark) / HZ);
2762         if (!dt) dt++;
2763         db = resync - (mddev->resync_mark_cnt);
2764         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2765 }
2766
2767 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2768
2769 static ssize_t
2770 sync_completed_show(mddev_t *mddev, char *page)
2771 {
2772         unsigned long max_blocks, resync;
2773
2774         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2775                 max_blocks = mddev->resync_max_sectors;
2776         else
2777                 max_blocks = mddev->size << 1;
2778
2779         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2780         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2781 }
2782
2783 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2784
2785 static ssize_t
2786 suspend_lo_show(mddev_t *mddev, char *page)
2787 {
2788         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2789 }
2790
2791 static ssize_t
2792 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2793 {
2794         char *e;
2795         unsigned long long new = simple_strtoull(buf, &e, 10);
2796
2797         if (mddev->pers->quiesce == NULL)
2798                 return -EINVAL;
2799         if (buf == e || (*e && *e != '\n'))
2800                 return -EINVAL;
2801         if (new >= mddev->suspend_hi ||
2802             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2803                 mddev->suspend_lo = new;
2804                 mddev->pers->quiesce(mddev, 2);
2805                 return len;
2806         } else
2807                 return -EINVAL;
2808 }
2809 static struct md_sysfs_entry md_suspend_lo =
2810 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2811
2812
2813 static ssize_t
2814 suspend_hi_show(mddev_t *mddev, char *page)
2815 {
2816         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2817 }
2818
2819 static ssize_t
2820 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2821 {
2822         char *e;
2823         unsigned long long new = simple_strtoull(buf, &e, 10);
2824
2825         if (mddev->pers->quiesce == NULL)
2826                 return -EINVAL;
2827         if (buf == e || (*e && *e != '\n'))
2828                 return -EINVAL;
2829         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2830             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2831                 mddev->suspend_hi = new;
2832                 mddev->pers->quiesce(mddev, 1);
2833                 mddev->pers->quiesce(mddev, 0);
2834                 return len;
2835         } else
2836                 return -EINVAL;
2837 }
2838 static struct md_sysfs_entry md_suspend_hi =
2839 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2840
2841
2842 static struct attribute *md_default_attrs[] = {
2843         &md_level.attr,
2844         &md_layout.attr,
2845         &md_raid_disks.attr,
2846         &md_chunk_size.attr,
2847         &md_size.attr,
2848         &md_resync_start.attr,
2849         &md_metadata.attr,
2850         &md_new_device.attr,
2851         &md_safe_delay.attr,
2852         &md_array_state.attr,
2853         NULL,
2854 };
2855
2856 static struct attribute *md_redundancy_attrs[] = {
2857         &md_scan_mode.attr,
2858         &md_mismatches.attr,
2859         &md_sync_min.attr,
2860         &md_sync_max.attr,
2861         &md_sync_speed.attr,
2862         &md_sync_completed.attr,
2863         &md_suspend_lo.attr,
2864         &md_suspend_hi.attr,
2865         &md_bitmap.attr,
2866         NULL,
2867 };
2868 static struct attribute_group md_redundancy_group = {
2869         .name = NULL,
2870         .attrs = md_redundancy_attrs,
2871 };
2872
2873
2874 static ssize_t
2875 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2876 {
2877         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2878         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2879         ssize_t rv;
2880
2881         if (!entry->show)
2882                 return -EIO;
2883         rv = mddev_lock(mddev);
2884         if (!rv) {
2885                 rv = entry->show(mddev, page);
2886                 mddev_unlock(mddev);
2887         }
2888         return rv;
2889 }
2890
2891 static ssize_t
2892 md_attr_store(struct kobject *kobj, struct attribute *attr,
2893               const char *page, size_t length)
2894 {
2895         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2896         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2897         ssize_t rv;
2898
2899         if (!entry->store)
2900                 return -EIO;
2901         if (!capable(CAP_SYS_ADMIN))
2902                 return -EACCES;
2903         rv = mddev_lock(mddev);
2904         if (!rv) {
2905                 rv = entry->store(mddev, page, length);
2906                 mddev_unlock(mddev);
2907         }
2908         return rv;
2909 }
2910
2911 static void md_free(struct kobject *ko)
2912 {
2913         mddev_t *mddev = container_of(ko, mddev_t, kobj);
2914         kfree(mddev);
2915 }
2916
2917 static struct sysfs_ops md_sysfs_ops = {
2918         .show   = md_attr_show,
2919         .store  = md_attr_store,
2920 };
2921 static struct kobj_type md_ktype = {
2922         .release        = md_free,
2923         .sysfs_ops      = &md_sysfs_ops,
2924         .default_attrs  = md_default_attrs,
2925 };
2926
2927 int mdp_major = 0;
2928
2929 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2930 {
2931         static DEFINE_MUTEX(disks_mutex);
2932         mddev_t *mddev = mddev_find(dev);
2933         struct gendisk *disk;
2934         int partitioned = (MAJOR(dev) != MD_MAJOR);
2935         int shift = partitioned ? MdpMinorShift : 0;
2936         int unit = MINOR(dev) >> shift;
2937
2938         if (!mddev)
2939                 return NULL;
2940
2941         mutex_lock(&disks_mutex);
2942         if (mddev->gendisk) {
2943                 mutex_unlock(&disks_mutex);
2944                 mddev_put(mddev);
2945                 return NULL;
2946         }
2947         disk = alloc_disk(1 << shift);
2948         if (!disk) {
2949                 mutex_unlock(&disks_mutex);
2950                 mddev_put(mddev);
2951                 return NULL;
2952         }
2953         disk->major = MAJOR(dev);
2954         disk->first_minor = unit << shift;
2955         if (partitioned)
2956                 sprintf(disk->disk_name, "md_d%d", unit);
2957         else
2958                 sprintf(disk->disk_name, "md%d", unit);
2959         disk->fops = &md_fops;
2960         disk->private_data = mddev;
2961         disk->queue = mddev->queue;
2962         add_disk(disk);
2963         mddev->gendisk = disk;
2964         mutex_unlock(&disks_mutex);
2965         mddev->kobj.parent = &disk->kobj;
2966         mddev->kobj.k_name = NULL;
2967         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2968         mddev->kobj.ktype = &md_ktype;
2969         kobject_register(&mddev->kobj);
2970         return NULL;
2971 }
2972
2973 static void md_safemode_timeout(unsigned long data)
2974 {
2975         mddev_t *mddev = (mddev_t *) data;
2976
2977         mddev->safemode = 1;
2978         md_wakeup_thread(mddev->thread);
2979 }
2980
2981 static int start_dirty_degraded;
2982
2983 static int do_md_run(mddev_t * mddev)
2984 {
2985         int err;
2986         int chunk_size;
2987         struct list_head *tmp;
2988         mdk_rdev_t *rdev;
2989         struct gendisk *disk;
2990         struct mdk_personality *pers;
2991         char b[BDEVNAME_SIZE];
2992
2993         if (list_empty(&mddev->disks))
2994                 /* cannot run an array with no devices.. */
2995                 return -EINVAL;
2996
2997         if (mddev->pers)
2998                 return -EBUSY;
2999
3000         /*
3001          * Analyze all RAID superblock(s)
3002          */
3003         if (!mddev->raid_disks)
3004                 analyze_sbs(mddev);
3005
3006         chunk_size = mddev->chunk_size;
3007
3008         if (chunk_size) {
3009                 if (chunk_size > MAX_CHUNK_SIZE) {
3010                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3011                                 chunk_size, MAX_CHUNK_SIZE);
3012                         return -EINVAL;
3013                 }
3014                 /*
3015                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3016                  */
3017                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3018                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3019                         return -EINVAL;
3020                 }
3021                 if (chunk_size < PAGE_SIZE) {
3022                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3023                                 chunk_size, PAGE_SIZE);
3024                         return -EINVAL;
3025                 }
3026
3027                 /* devices must have minimum size of one chunk */
3028                 ITERATE_RDEV(mddev,rdev,tmp) {
3029                         if (test_bit(Faulty, &rdev->flags))
3030                                 continue;
3031                         if (rdev->size < chunk_size / 1024) {
3032                                 printk(KERN_WARNING
3033                                         "md: Dev %s smaller than chunk_size:"
3034                                         " %lluk < %dk\n",
3035                                         bdevname(rdev->bdev,b),
3036                                         (unsigned long long)rdev->size,
3037                                         chunk_size / 1024);
3038                                 return -EINVAL;
3039                         }
3040                 }
3041         }
3042
3043 #ifdef CONFIG_KMOD
3044         if (mddev->level != LEVEL_NONE)
3045                 request_module("md-level-%d", mddev->level);
3046         else if (mddev->clevel[0])
3047                 request_module("md-%s", mddev->clevel);
3048 #endif
3049
3050         /*
3051          * Drop all container device buffers, from now on
3052          * the only valid external interface is through the md
3053          * device.
3054          * Also find largest hardsector size
3055          */
3056         ITERATE_RDEV(mddev,rdev,tmp) {
3057                 if (test_bit(Faulty, &rdev->flags))
3058                         continue;
3059                 sync_blockdev(rdev->bdev);
3060                 invalidate_bdev(rdev->bdev, 0);
3061         }
3062
3063         md_probe(mddev->unit, NULL, NULL);
3064         disk = mddev->gendisk;
3065         if (!disk)
3066                 return -ENOMEM;
3067
3068         spin_lock(&pers_lock);
3069         pers = find_pers(mddev->level, mddev->clevel);
3070         if (!pers || !try_module_get(pers->owner)) {
3071                 spin_unlock(&pers_lock);
3072                 if (mddev->level != LEVEL_NONE)
3073                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3074                                mddev->level);
3075                 else
3076                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3077                                mddev->clevel);
3078                 return -EINVAL;
3079         }
3080         mddev->pers = pers;
3081         spin_unlock(&pers_lock);
3082         mddev->level = pers->level;
3083         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3084
3085         if (mddev->reshape_position != MaxSector &&
3086             pers->start_reshape == NULL) {
3087                 /* This personality cannot handle reshaping... */
3088                 mddev->pers = NULL;
3089                 module_put(pers->owner);
3090                 return -EINVAL;
3091         }
3092
3093         if (pers->sync_request) {
3094                 /* Warn if this is a potentially silly
3095                  * configuration.
3096                  */
3097                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3098                 mdk_rdev_t *rdev2;
3099                 struct list_head *tmp2;
3100                 int warned = 0;
3101                 ITERATE_RDEV(mddev, rdev, tmp) {
3102                         ITERATE_RDEV(mddev, rdev2, tmp2) {
3103                                 if (rdev < rdev2 &&
3104                                     rdev->bdev->bd_contains ==
3105                                     rdev2->bdev->bd_contains) {
3106                                         printk(KERN_WARNING
3107                                                "%s: WARNING: %s appears to be"
3108                                                " on the same physical disk as"
3109                                                " %s.\n",
3110                                                mdname(mddev),
3111                                                bdevname(rdev->bdev,b),
3112                                                bdevname(rdev2->bdev,b2));
3113                                         warned = 1;
3114                                 }
3115                         }
3116                 }
3117                 if (warned)
3118                         printk(KERN_WARNING
3119                                "True protection against single-disk"
3120                                " failure might be compromised.\n");
3121         }
3122
3123         mddev->recovery = 0;
3124         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3125         mddev->barriers_work = 1;
3126         mddev->ok_start_degraded = start_dirty_degraded;
3127
3128         if (start_readonly)
3129                 mddev->ro = 2; /* read-only, but switch on first write */
3130
3131         err = mddev->pers->run(mddev);
3132         if (!err && mddev->pers->sync_request) {
3133                 err = bitmap_create(mddev);
3134                 if (err) {
3135                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3136                                mdname(mddev), err);
3137                         mddev->pers->stop(mddev);
3138                 }
3139         }
3140         if (err) {
3141                 printk(KERN_ERR "md: pers->run() failed ...\n");
3142                 module_put(mddev->pers->owner);
3143                 mddev->pers = NULL;
3144                 bitmap_destroy(mddev);
3145                 return err;
3146         }
3147         if (mddev->pers->sync_request)
3148                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
3149         else if (mddev->ro == 2) /* auto-readonly not meaningful */
3150                 mddev->ro = 0;
3151
3152         atomic_set(&mddev->writes_pending,0);
3153         mddev->safemode = 0;
3154         mddev->safemode_timer.function = md_safemode_timeout;
3155         mddev->safemode_timer.data = (unsigned long) mddev;
3156         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3157         mddev->in_sync = 1;
3158
3159         ITERATE_RDEV(mddev,rdev,tmp)
3160                 if (rdev->raid_disk >= 0) {
3161                         char nm[20];
3162                         sprintf(nm, "rd%d", rdev->raid_disk);
3163                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
3164                 }
3165         
3166         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3167         
3168         if (mddev->flags)
3169                 md_update_sb(mddev, 0);
3170
3171         set_capacity(disk, mddev->array_size<<1);
3172
3173         /* If we call blk_queue_make_request here, it will
3174          * re-initialise max_sectors etc which may have been
3175          * refined inside -> run.  So just set the bits we need to set.
3176          * Most initialisation happended when we called
3177          * blk_queue_make_request(..., md_fail_request)
3178          * earlier.
3179          */
3180         mddev->queue->queuedata = mddev;
3181         mddev->queue->make_request_fn = mddev->pers->make_request;
3182
3183         /* If there is a partially-recovered drive we need to
3184          * start recovery here.  If we leave it to md_check_recovery,
3185          * it will remove the drives and not do the right thing
3186          */
3187         if (mddev->degraded && !mddev->sync_thread) {
3188                 struct list_head *rtmp;
3189                 int spares = 0;
3190                 ITERATE_RDEV(mddev,rdev,rtmp)
3191                         if (rdev->raid_disk >= 0 &&
3192                             !test_bit(In_sync, &rdev->flags) &&
3193                             !test_bit(Faulty, &rdev->flags))
3194                                 /* complete an interrupted recovery */
3195                                 spares++;
3196                 if (spares && mddev->pers->sync_request) {
3197                         mddev->recovery = 0;
3198                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3199                         mddev->sync_thread = md_register_thread(md_do_sync,
3200                                                                 mddev,
3201                                                                 "%s_resync");
3202                         if (!mddev->sync_thread) {
3203                                 printk(KERN_ERR "%s: could not start resync"
3204                                        " thread...\n",
3205                                        mdname(mddev));
3206                                 /* leave the spares where they are, it shouldn't hurt */
3207                                 mddev->recovery = 0;
3208                         }
3209                 }
3210         }
3211         md_wakeup_thread(mddev->thread);
3212         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3213
3214         mddev->changed = 1;
3215         md_new_event(mddev);
3216         kobject_uevent(&mddev->gendisk->kobj, KOBJ_CHANGE);
3217         return 0;
3218 }
3219
3220 static int restart_array(mddev_t *mddev)
3221 {
3222         struct gendisk *disk = mddev->gendisk;
3223         int err;
3224
3225         /*
3226          * Complain if it has no devices
3227          */
3228         err = -ENXIO;
3229         if (list_empty(&mddev->disks))
3230                 goto out;
3231
3232         if (mddev->pers) {
3233                 err = -EBUSY;
3234                 if (!mddev->ro)
3235                         goto out;
3236
3237                 mddev->safemode = 0;
3238                 mddev->ro = 0;
3239                 set_disk_ro(disk, 0);
3240
3241                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3242                         mdname(mddev));
3243                 /*
3244                  * Kick recovery or resync if necessary
3245                  */
3246                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3247                 md_wakeup_thread(mddev->thread);
3248                 md_wakeup_thread(mddev->sync_thread);
3249                 err = 0;
3250         } else
3251                 err = -EINVAL;
3252
3253 out:
3254         return err;
3255 }
3256
3257 /* similar to deny_write_access, but accounts for our holding a reference
3258  * to the file ourselves */
3259 static int deny_bitmap_write_access(struct file * file)
3260 {
3261         struct inode *inode = file->f_mapping->host;
3262
3263         spin_lock(&inode->i_lock);
3264         if (atomic_read(&inode->i_writecount) > 1) {
3265                 spin_unlock(&inode->i_lock);
3266                 return -ETXTBSY;
3267         }
3268         atomic_set(&inode->i_writecount, -1);
3269         spin_unlock(&inode->i_lock);
3270
3271         return 0;
3272 }
3273
3274 static void restore_bitmap_write_access(struct file *file)
3275 {
3276         struct inode *inode = file->f_mapping->host;
3277
3278         spin_lock(&inode->i_lock);
3279         atomic_set(&inode->i_writecount, 1);
3280         spin_unlock(&inode->i_lock);
3281 }
3282
3283 /* mode:
3284  *   0 - completely stop and dis-assemble array
3285  *   1 - switch to readonly
3286  *   2 - stop but do not disassemble array
3287  */
3288 static int do_md_stop(mddev_t * mddev, int mode)
3289 {
3290         int err = 0;
3291         struct gendisk *disk = mddev->gendisk;
3292
3293         if (mddev->pers) {
3294                 if (atomic_read(&mddev->active)>2) {
3295                         printk("md: %s still in use.\n",mdname(mddev));
3296                         return -EBUSY;
3297                 }
3298
3299                 if (mddev->sync_thread) {
3300                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3301                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3302                         md_unregister_thread(mddev->sync_thread);
3303                         mddev->sync_thread = NULL;
3304                 }
3305
3306                 del_timer_sync(&mddev->safemode_timer);
3307
3308                 invalidate_partition(disk, 0);
3309
3310                 switch(mode) {
3311                 case 1: /* readonly */
3312                         err  = -ENXIO;
3313                         if (mddev->ro==1)
3314                                 goto out;
3315                         mddev->ro = 1;
3316                         break;
3317                 case 0: /* disassemble */
3318                 case 2: /* stop */
3319                         bitmap_flush(mddev);
3320                         md_super_wait(mddev);
3321                         if (mddev->ro)
3322                                 set_disk_ro(disk, 0);
3323                         blk_queue_make_request(mddev->queue, md_fail_request);
3324                         mddev->pers->stop(mddev);
3325                         mddev->queue->merge_bvec_fn = NULL;
3326                         mddev->queue->unplug_fn = NULL;
3327                         mddev->queue->issue_flush_fn = NULL;
3328                         mddev->queue->backing_dev_info.congested_fn = NULL;
3329                         if (mddev->pers->sync_request)
3330                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3331
3332                         module_put(mddev->pers->owner);
3333                         mddev->pers = NULL;
3334
3335                         set_capacity(disk, 0);
3336                         mddev->changed = 1;
3337
3338                         if (mddev->ro)
3339                                 mddev->ro = 0;
3340                 }
3341                 if (!mddev->in_sync || mddev->flags) {
3342                         /* mark array as shutdown cleanly */
3343                         mddev->in_sync = 1;
3344                         md_update_sb(mddev, 1);
3345                 }
3346                 if (mode == 1)
3347                         set_disk_ro(disk, 1);
3348                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3349         }
3350
3351         /*
3352          * Free resources if final stop
3353          */
3354         if (mode == 0) {
3355                 mdk_rdev_t *rdev;
3356                 struct list_head *tmp;
3357
3358                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3359
3360                 bitmap_destroy(mddev);
3361                 if (mddev->bitmap_file) {
3362                         restore_bitmap_write_access(mddev->bitmap_file);
3363                         fput(mddev->bitmap_file);
3364                         mddev->bitmap_file = NULL;
3365                 }
3366                 mddev->bitmap_offset = 0;
3367
3368                 ITERATE_RDEV(mddev,rdev,tmp)
3369                         if (rdev->raid_disk >= 0) {
3370                                 char nm[20];
3371                                 sprintf(nm, "rd%d", rdev->raid_disk);
3372                                 sysfs_remove_link(&mddev->kobj, nm);
3373                         }
3374
3375                 export_array(mddev);
3376
3377                 mddev->array_size = 0;
3378                 mddev->size = 0;
3379                 mddev->raid_disks = 0;
3380                 mddev->recovery_cp = 0;
3381
3382         } else if (mddev->pers)
3383                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3384                         mdname(mddev));
3385         err = 0;
3386         md_new_event(mddev);
3387 out:
3388         return err;
3389 }
3390
3391 #ifndef MODULE
3392 static void autorun_array(mddev_t *mddev)
3393 {
3394         mdk_rdev_t *rdev;
3395         struct list_head *tmp;
3396         int err;
3397
3398         if (list_empty(&mddev->disks))
3399                 return;
3400
3401         printk(KERN_INFO "md: running: ");
3402
3403         ITERATE_RDEV(mddev,rdev,tmp) {
3404                 char b[BDEVNAME_SIZE];
3405                 printk("<%s>", bdevname(rdev->bdev,b));
3406         }
3407         printk("\n");
3408
3409         err = do_md_run (mddev);
3410         if (err) {
3411                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3412                 do_md_stop (mddev, 0);
3413         }
3414 }
3415
3416 /*
3417  * lets try to run arrays based on all disks that have arrived
3418  * until now. (those are in pending_raid_disks)
3419  *
3420  * the method: pick the first pending disk, collect all disks with
3421  * the same UUID, remove all from the pending list and put them into
3422  * the 'same_array' list. Then order this list based on superblock
3423  * update time (freshest comes first), kick out 'old' disks and
3424  * compare superblocks. If everything's fine then run it.
3425  *
3426  * If "unit" is allocated, then bump its reference count
3427  */
3428 static void autorun_devices(int part)
3429 {
3430         struct list_head *tmp;
3431         mdk_rdev_t *rdev0, *rdev;
3432         mddev_t *mddev;
3433         char b[BDEVNAME_SIZE];
3434
3435         printk(KERN_INFO "md: autorun ...\n");
3436         while (!list_empty(&pending_raid_disks)) {
3437                 int unit;
3438                 dev_t dev;
3439                 LIST_HEAD(candidates);
3440                 rdev0 = list_entry(pending_raid_disks.next,
3441                                          mdk_rdev_t, same_set);
3442
3443                 printk(KERN_INFO "md: considering %s ...\n",
3444                         bdevname(rdev0->bdev,b));
3445                 INIT_LIST_HEAD(&candidates);
3446                 ITERATE_RDEV_PENDING(rdev,tmp)
3447                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3448                                 printk(KERN_INFO "md:  adding %s ...\n",
3449                                         bdevname(rdev->bdev,b));
3450                                 list_move(&rdev->same_set, &candidates);
3451                         }
3452                 /*
3453                  * now we have a set of devices, with all of them having
3454                  * mostly sane superblocks. It's time to allocate the
3455                  * mddev.
3456                  */
3457                 if (part) {
3458                         dev = MKDEV(mdp_major,
3459                                     rdev0->preferred_minor << MdpMinorShift);
3460                         unit = MINOR(dev) >> MdpMinorShift;
3461                 } else {
3462                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3463                         unit = MINOR(dev);
3464                 }
3465                 if (rdev0->preferred_minor != unit) {
3466                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3467                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3468                         break;
3469                 }
3470
3471                 md_probe(dev, NULL, NULL);
3472                 mddev = mddev_find(dev);
3473                 if (!mddev) {
3474                         printk(KERN_ERR 
3475                                 "md: cannot allocate memory for md drive.\n");
3476                         break;
3477                 }
3478                 if (mddev_lock(mddev)) 
3479                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3480                                mdname(mddev));
3481                 else if (mddev->raid_disks || mddev->major_version
3482                          || !list_empty(&mddev->disks)) {
3483                         printk(KERN_WARNING 
3484                                 "md: %s already running, cannot run %s\n",
3485                                 mdname(mddev), bdevname(rdev0->bdev,b));
3486                         mddev_unlock(mddev);
3487                 } else {
3488                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3489                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3490                                 list_del_init(&rdev->same_set);
3491                                 if (bind_rdev_to_array(rdev, mddev))
3492                                         export_rdev(rdev);
3493                         }
3494                         autorun_array(mddev);
3495                         mddev_unlock(mddev);
3496                 }
3497                 /* on success, candidates will be empty, on error
3498                  * it won't...
3499                  */
3500                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3501                         export_rdev(rdev);
3502                 mddev_put(mddev);
3503         }
3504         printk(KERN_INFO "md: ... autorun DONE.\n");
3505 }
3506 #endif /* !MODULE */
3507
3508 static int get_version(void __user * arg)
3509 {
3510         mdu_version_t ver;
3511
3512         ver.major = MD_MAJOR_VERSION;
3513         ver.minor = MD_MINOR_VERSION;
3514         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3515
3516         if (copy_to_user(arg, &ver, sizeof(ver)))
3517                 return -EFAULT;
3518
3519         return 0;
3520 }
3521
3522 static int get_array_info(mddev_t * mddev, void __user * arg)
3523 {
3524         mdu_array_info_t info;
3525         int nr,working,active,failed,spare;
3526         mdk_rdev_t *rdev;
3527         struct list_head *tmp;
3528
3529         nr=working=active=failed=spare=0;
3530         ITERATE_RDEV(mddev,rdev,tmp) {
3531                 nr++;
3532                 if (test_bit(Faulty, &rdev->flags))
3533                         failed++;
3534                 else {
3535                         working++;
3536                         if (test_bit(In_sync, &rdev->flags))
3537                                 active++;       
3538                         else
3539                                 spare++;
3540                 }
3541         }
3542
3543         info.major_version = mddev->major_version;
3544         info.minor_version = mddev->minor_version;
3545         info.patch_version = MD_PATCHLEVEL_VERSION;
3546         info.ctime         = mddev->ctime;
3547         info.level         = mddev->level;
3548         info.size          = mddev->size;
3549         if (info.size != mddev->size) /* overflow */
3550                 info.size = -1;
3551         info.nr_disks      = nr;
3552         info.raid_disks    = mddev->raid_disks;
3553         info.md_minor      = mddev->md_minor;
3554         info.not_persistent= !mddev->persistent;
3555
3556         info.utime         = mddev->utime;
3557         info.state         = 0;
3558         if (mddev->in_sync)
3559                 info.state = (1<<MD_SB_CLEAN);
3560         if (mddev->bitmap && mddev->bitmap_offset)
3561                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3562         info.active_disks  = active;
3563         info.working_disks = working;
3564         info.failed_disks  = failed;
3565         info.spare_disks   = spare;
3566
3567         info.layout        = mddev->layout;
3568         info.chunk_size    = mddev->chunk_size;
3569
3570         if (copy_to_user(arg, &info, sizeof(info)))
3571                 return -EFAULT;
3572
3573         return 0;
3574 }
3575
3576 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3577 {
3578         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3579         char *ptr, *buf = NULL;
3580         int err = -ENOMEM;
3581
3582         md_allow_write(mddev);
3583
3584         file = kmalloc(sizeof(*file), GFP_KERNEL);
3585         if (!file)
3586                 goto out;
3587
3588         /* bitmap disabled, zero the first byte and copy out */
3589         if (!mddev->bitmap || !mddev->bitmap->file) {
3590                 file->pathname[0] = '\0';
3591                 goto copy_out;
3592         }
3593
3594         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3595         if (!buf)
3596                 goto out;
3597
3598         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3599         if (!ptr)
3600                 goto out;
3601
3602         strcpy(file->pathname, ptr);
3603
3604 copy_out:
3605         err = 0;
3606         if (copy_to_user(arg, file, sizeof(*file)))
3607                 err = -EFAULT;
3608 out:
3609         kfree(buf);
3610         kfree(file);
3611         return err;
3612 }
3613
3614 static int get_disk_info(mddev_t * mddev, void __user * arg)
3615 {
3616         mdu_disk_info_t info;
3617         unsigned int nr;
3618         mdk_rdev_t *rdev;
3619
3620         if (copy_from_user(&info, arg, sizeof(info)))
3621                 return -EFAULT;
3622
3623         nr = info.number;
3624
3625         rdev = find_rdev_nr(mddev, nr);
3626         if (rdev) {
3627                 info.major = MAJOR(rdev->bdev->bd_dev);
3628                 info.minor = MINOR(rdev->bdev->bd_dev);
3629                 info.raid_disk = rdev->raid_disk;
3630                 info.state = 0;
3631                 if (test_bit(Faulty, &rdev->flags))
3632                         info.state |= (1<<MD_DISK_FAULTY);
3633                 else if (test_bit(In_sync, &rdev->flags)) {
3634                         info.state |= (1<<MD_DISK_ACTIVE);
3635                         info.state |= (1<<MD_DISK_SYNC);
3636                 }
3637                 if (test_bit(WriteMostly, &rdev->flags))
3638                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3639         } else {
3640                 info.major = info.minor = 0;
3641                 info.raid_disk = -1;
3642                 info.state = (1<<MD_DISK_REMOVED);
3643         }
3644
3645         if (copy_to_user(arg, &info, sizeof(info)))
3646                 return -EFAULT;
3647
3648         return 0;
3649 }
3650
3651 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3652 {
3653         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3654         mdk_rdev_t *rdev;
3655         dev_t dev = MKDEV(info->major,info->minor);
3656
3657         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3658                 return -EOVERFLOW;
3659
3660         if (!mddev->raid_disks) {
3661                 int err;
3662                 /* expecting a device which has a superblock */
3663                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3664                 if (IS_ERR(rdev)) {
3665                         printk(KERN_WARNING 
3666                                 "md: md_import_device returned %ld\n",
3667                                 PTR_ERR(rdev));
3668                         return PTR_ERR(rdev);
3669                 }
3670                 if (!list_empty(&mddev->disks)) {
3671                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3672                                                         mdk_rdev_t, same_set);
3673                         int err = super_types[mddev->major_version]
3674                                 .load_super(rdev, rdev0, mddev->minor_version);
3675                         if (err < 0) {
3676                                 printk(KERN_WARNING 
3677                                         "md: %s has different UUID to %s\n",
3678                                         bdevname(rdev->bdev,b), 
3679                                         bdevname(rdev0->bdev,b2));
3680                                 export_rdev(rdev);
3681                                 return -EINVAL;
3682                         }
3683                 }
3684                 err = bind_rdev_to_array(rdev, mddev);
3685                 if (err)
3686                         export_rdev(rdev);
3687                 return err;
3688         }
3689
3690         /*
3691          * add_new_disk can be used once the array is assembled
3692          * to add "hot spares".  They must already have a superblock
3693          * written
3694          */
3695         if (mddev->pers) {
3696                 int err;
3697                 if (!mddev->pers->hot_add_disk) {
3698                         printk(KERN_WARNING 
3699                                 "%s: personality does not support diskops!\n",
3700                                mdname(mddev));
3701                         return -EINVAL;
3702                 }
3703                 if (mddev->persistent)
3704                         rdev = md_import_device(dev, mddev->major_version,
3705                                                 mddev->minor_version);
3706                 else
3707                         rdev = md_import_device(dev, -1, -1);
3708                 if (IS_ERR(rdev)) {
3709                         printk(KERN_WARNING 
3710                                 "md: md_import_device returned %ld\n",
3711                                 PTR_ERR(rdev));
3712                         return PTR_ERR(rdev);
3713                 }
3714                 /* set save_raid_disk if appropriate */
3715                 if (!mddev->persistent) {
3716                         if (info->state & (1<<MD_DISK_SYNC)  &&
3717                             info->raid_disk < mddev->raid_disks)
3718                                 rdev->raid_disk = info->raid_disk;
3719                         else
3720                                 rdev->raid_disk = -1;
3721                 } else
3722                         super_types[mddev->major_version].
3723                                 validate_super(mddev, rdev);
3724                 rdev->saved_raid_disk = rdev->raid_disk;
3725
3726                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3727                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3728                         set_bit(WriteMostly, &rdev->flags);
3729
3730                 rdev->raid_disk = -1;
3731                 err = bind_rdev_to_array(rdev, mddev);
3732                 if (!err && !mddev->pers->hot_remove_disk) {
3733                         /* If there is hot_add_disk but no hot_remove_disk
3734                          * then added disks for geometry changes,
3735                          * and should be added immediately.
3736                          */
3737                         super_types[mddev->major_version].
3738                                 validate_super(mddev, rdev);
3739                         err = mddev->pers->hot_add_disk(mddev, rdev);
3740                         if (err)
3741                                 unbind_rdev_from_array(rdev);
3742                 }
3743                 if (err)
3744                         export_rdev(rdev);
3745
3746                 md_update_sb(mddev, 1);
3747                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3748                 md_wakeup_thread(mddev->thread);
3749                 return err;
3750         }
3751
3752         /* otherwise, add_new_disk is only allowed
3753          * for major_version==0 superblocks
3754          */
3755         if (mddev->major_version != 0) {
3756                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3757                        mdname(mddev));
3758                 return -EINVAL;
3759         }
3760
3761         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3762                 int err;
3763                 rdev = md_import_device (dev, -1, 0);
3764                 if (IS_ERR(rdev)) {
3765                         printk(KERN_WARNING 
3766                                 "md: error, md_import_device() returned %ld\n",
3767                                 PTR_ERR(rdev));
3768                         return PTR_ERR(rdev);
3769                 }
3770                 rdev->desc_nr = info->number;
3771                 if (info->raid_disk < mddev->raid_disks)
3772                         rdev->raid_disk = info->raid_disk;
3773                 else
3774                         rdev->raid_disk = -1;
3775
3776                 rdev->flags = 0;
3777
3778                 if (rdev->raid_disk < mddev->raid_disks)
3779                         if (info->state & (1<<MD_DISK_SYNC))
3780                                 set_bit(In_sync, &rdev->flags);
3781
3782                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3783                         set_bit(WriteMostly, &rdev->flags);
3784
3785                 if (!mddev->persistent) {
3786                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3787                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3788                 } else 
3789                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3790                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3791
3792                 err = bind_rdev_to_array(rdev, mddev);
3793                 if (err) {
3794                         export_rdev(rdev);
3795                         return err;
3796                 }
3797         }
3798
3799         return 0;
3800 }
3801
3802 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3803 {
3804         char b[BDEVNAME_SIZE];
3805         mdk_rdev_t *rdev;
3806
3807         if (!mddev->pers)
3808                 return -ENODEV;
3809
3810         rdev = find_rdev(mddev, dev);
3811         if (!rdev)
3812                 return -ENXIO;
3813
3814         if (rdev->raid_disk >= 0)
3815                 goto busy;
3816
3817         kick_rdev_from_array(rdev);
3818         md_update_sb(mddev, 1);
3819         md_new_event(mddev);
3820
3821         return 0;
3822 busy:
3823         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3824                 bdevname(rdev->bdev,b), mdname(mddev));
3825         return -EBUSY;
3826 }
3827
3828 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3829 {
3830         char b[BDEVNAME_SIZE];
3831         int err;
3832         unsigned int size;
3833         mdk_rdev_t *rdev;
3834
3835         if (!mddev->pers)
3836                 return -ENODEV;
3837
3838         if (mddev->major_version != 0) {
3839                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3840                         " version-0 superblocks.\n",
3841                         mdname(mddev));
3842                 return -EINVAL;
3843         }
3844         if (!mddev->pers->hot_add_disk) {
3845                 printk(KERN_WARNING 
3846                         "%s: personality does not support diskops!\n",
3847                         mdname(mddev));
3848                 return -EINVAL;
3849         }
3850
3851         rdev = md_import_device (dev, -1, 0);
3852         if (IS_ERR(rdev)) {
3853                 printk(KERN_WARNING 
3854                         "md: error, md_import_device() returned %ld\n",
3855                         PTR_ERR(rdev));
3856                 return -EINVAL;
3857         }
3858
3859         if (mddev->persistent)
3860                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3861         else
3862                 rdev->sb_offset =
3863                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3864
3865         size = calc_dev_size(rdev, mddev->chunk_size);
3866         rdev->size = size;
3867
3868         if (test_bit(Faulty, &rdev->flags)) {
3869                 printk(KERN_WARNING 
3870                         "md: can not hot-add faulty %s disk to %s!\n",
3871                         bdevname(rdev->bdev,b), mdname(mddev));
3872                 err = -EINVAL;
3873                 goto abort_export;
3874         }
3875         clear_bit(In_sync, &rdev->flags);
3876         rdev->desc_nr = -1;
3877         rdev->saved_raid_disk = -1;
3878         err = bind_rdev_to_array(rdev, mddev);
3879         if (err)
3880                 goto abort_export;
3881
3882         /*
3883          * The rest should better be atomic, we can have disk failures
3884          * noticed in interrupt contexts ...
3885          */
3886
3887         if (rdev->desc_nr == mddev->max_disks) {
3888                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3889                         mdname(mddev));
3890                 err = -EBUSY;
3891                 goto abort_unbind_export;
3892         }
3893
3894         rdev->raid_disk = -1;
3895
3896         md_update_sb(mddev, 1);
3897
3898         /*
3899          * Kick recovery, maybe this spare has to be added to the
3900          * array immediately.
3901          */
3902         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3903         md_wakeup_thread(mddev->thread);
3904         md_new_event(mddev);
3905         return 0;
3906
3907 abort_unbind_export:
3908         unbind_rdev_from_array(rdev);
3909
3910 abort_export:
3911         export_rdev(rdev);
3912         return err;
3913 }
3914
3915 static int set_bitmap_file(mddev_t *mddev, int fd)
3916 {
3917         int err;
3918
3919         if (mddev->pers) {
3920                 if (!mddev->pers->quiesce)
3921                         return -EBUSY;
3922                 if (mddev->recovery || mddev->sync_thread)
3923                         return -EBUSY;
3924                 /* we should be able to change the bitmap.. */
3925         }
3926
3927
3928         if (fd >= 0) {
3929                 if (mddev->bitmap)
3930                         return -EEXIST; /* cannot add when bitmap is present */
3931                 mddev->bitmap_file = fget(fd);
3932
3933                 if (mddev->bitmap_file == NULL) {
3934                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3935                                mdname(mddev));
3936                         return -EBADF;
3937                 }
3938
3939                 err = deny_bitmap_write_access(mddev->bitmap_file);
3940                 if (err) {
3941                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3942                                mdname(mddev));
3943                         fput(mddev->bitmap_file);
3944                         mddev->bitmap_file = NULL;
3945                         return err;
3946                 }
3947                 mddev->bitmap_offset = 0; /* file overrides offset */
3948         } else if (mddev->bitmap == NULL)
3949                 return -ENOENT; /* cannot remove what isn't there */
3950         err = 0;
3951         if (mddev->pers) {
3952                 mddev->pers->quiesce(mddev, 1);
3953                 if (fd >= 0)
3954                         err = bitmap_create(mddev);
3955                 if (fd < 0 || err) {
3956                         bitmap_destroy(mddev);
3957                         fd = -1; /* make sure to put the file */
3958                 }
3959                 mddev->pers->quiesce(mddev, 0);
3960         }
3961         if (fd < 0) {
3962                 if (mddev->bitmap_file) {
3963                         restore_bitmap_write_access(mddev->bitmap_file);
3964                         fput(mddev->bitmap_file);
3965                 }
3966                 mddev->bitmap_file = NULL;
3967         }
3968
3969         return err;
3970 }
3971
3972 /*
3973  * set_array_info is used two different ways
3974  * The original usage is when creating a new array.
3975  * In this usage, raid_disks is > 0 and it together with
3976  *  level, size, not_persistent,layout,chunksize determine the
3977  *  shape of the array.
3978  *  This will always create an array with a type-0.90.0 superblock.
3979  * The newer usage is when assembling an array.
3980  *  In this case raid_disks will be 0, and the major_version field is
3981  *  use to determine which style super-blocks are to be found on the devices.
3982  *  The minor and patch _version numbers are also kept incase the
3983  *  super_block handler wishes to interpret them.
3984  */
3985 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3986 {
3987
3988         if (info->raid_disks == 0) {
3989                 /* just setting version number for superblock loading */
3990                 if (info->major_version < 0 ||
3991                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3992                     super_types[info->major_version].name == NULL) {
3993                         /* maybe try to auto-load a module? */
3994                         printk(KERN_INFO 
3995                                 "md: superblock version %d not known\n",
3996                                 info->major_version);
3997                         return -EINVAL;
3998                 }
3999                 mddev->major_version = info->major_version;
4000                 mddev->minor_version = info->minor_version;
4001                 mddev->patch_version = info->patch_version;
4002                 mddev->persistent = !info->not_persistent;
4003                 return 0;
4004         }
4005         mddev->major_version = MD_MAJOR_VERSION;
4006         mddev->minor_version = MD_MINOR_VERSION;
4007         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4008         mddev->ctime         = get_seconds();
4009
4010         mddev->level         = info->level;
4011         mddev->clevel[0]     = 0;
4012         mddev->size          = info->size;
4013         mddev->raid_disks    = info->raid_disks;
4014         /* don't set md_minor, it is determined by which /dev/md* was
4015          * openned
4016          */
4017         if (info->state & (1<<MD_SB_CLEAN))
4018                 mddev->recovery_cp = MaxSector;
4019         else
4020                 mddev->recovery_cp = 0;
4021         mddev->persistent    = ! info->not_persistent;
4022
4023         mddev->layout        = info->layout;
4024         mddev->chunk_size    = info->chunk_size;
4025
4026         mddev->max_disks     = MD_SB_DISKS;
4027
4028         mddev->flags         = 0;
4029         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4030
4031         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4032         mddev->bitmap_offset = 0;
4033
4034         mddev->reshape_position = MaxSector;
4035
4036         /*
4037          * Generate a 128 bit UUID
4038          */
4039         get_random_bytes(mddev->uuid, 16);
4040
4041         mddev->new_level = mddev->level;
4042         mddev->new_chunk = mddev->chunk_size;
4043         mddev->new_layout = mddev->layout;
4044         mddev->delta_disks = 0;
4045
4046         return 0;
4047 }
4048
4049 static int update_size(mddev_t *mddev, unsigned long size)
4050 {
4051         mdk_rdev_t * rdev;
4052         int rv;
4053         struct list_head *tmp;
4054         int fit = (size == 0);
4055
4056         if (mddev->pers->resize == NULL)
4057                 return -EINVAL;
4058         /* The "size" is the amount of each device that is used.
4059          * This can only make sense for arrays with redundancy.
4060          * linear and raid0 always use whatever space is available
4061          * We can only consider changing the size if no resync
4062          * or reconstruction is happening, and if the new size
4063          * is acceptable. It must fit before the sb_offset or,
4064          * if that is <data_offset, it must fit before the
4065          * size of each device.
4066          * If size is zero, we find the largest size that fits.
4067          */
4068         if (mddev->sync_thread)
4069                 return -EBUSY;
4070         ITERATE_RDEV(mddev,rdev,tmp) {
4071                 sector_t avail;
4072                 avail = rdev->size * 2;
4073
4074                 if (fit && (size == 0 || size > avail/2))
4075                         size = avail/2;
4076                 if (avail < ((sector_t)size << 1))
4077                         return -ENOSPC;
4078         }
4079         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4080         if (!rv) {
4081                 struct block_device *bdev;
4082
4083                 bdev = bdget_disk(mddev->gendisk, 0);
4084                 if (bdev) {
4085                         mutex_lock(&bdev->bd_inode->i_mutex);
4086                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4087                         mutex_unlock(&bdev->bd_inode->i_mutex);
4088                         bdput(bdev);
4089                 }
4090         }
4091         return rv;
4092 }
4093
4094 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4095 {
4096         int rv;
4097         /* change the number of raid disks */
4098         if (mddev->pers->check_reshape == NULL)
4099                 return -EINVAL;
4100         if (raid_disks <= 0 ||
4101             raid_disks >= mddev->max_disks)
4102                 return -EINVAL;
4103         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4104                 return -EBUSY;
4105         mddev->delta_disks = raid_disks - mddev->raid_disks;
4106
4107         rv = mddev->pers->check_reshape(mddev);
4108         return rv;
4109 }
4110
4111
4112 /*
4113  * update_array_info is used to change the configuration of an
4114  * on-line array.
4115  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4116  * fields in the info are checked against the array.
4117  * Any differences that cannot be handled will cause an error.
4118  * Normally, only one change can be managed at a time.
4119  */
4120 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4121 {
4122         int rv = 0;
4123         int cnt = 0;
4124         int state = 0;
4125
4126         /* calculate expected state,ignoring low bits */
4127         if (mddev->bitmap && mddev->bitmap_offset)
4128                 state |= (1 << MD_SB_BITMAP_PRESENT);
4129
4130         if (mddev->major_version != info->major_version ||
4131             mddev->minor_version != info->minor_version ||
4132 /*          mddev->patch_version != info->patch_version || */
4133             mddev->ctime         != info->ctime         ||
4134             mddev->level         != info->level         ||
4135 /*          mddev->layout        != info->layout        || */
4136             !mddev->persistent   != info->not_persistent||
4137             mddev->chunk_size    != info->chunk_size    ||
4138             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4139             ((state^info->state) & 0xfffffe00)
4140                 )
4141                 return -EINVAL;
4142         /* Check there is only one change */
4143         if (info->size >= 0 && mddev->size != info->size) cnt++;
4144         if (mddev->raid_disks != info->raid_disks) cnt++;
4145         if (mddev->layout != info->layout) cnt++;
4146         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4147         if (cnt == 0) return 0;
4148         if (cnt > 1) return -EINVAL;
4149
4150         if (mddev->layout != info->layout) {
4151                 /* Change layout
4152                  * we don't need to do anything at the md level, the
4153                  * personality will take care of it all.
4154                  */
4155                 if (mddev->pers->reconfig == NULL)
4156                         return -EINVAL;
4157                 else
4158                         return mddev->pers->reconfig(mddev, info->layout, -1);
4159         }
4160         if (info->size >= 0 && mddev->size != info->size)
4161                 rv = update_size(mddev, info->size);
4162
4163         if (mddev->raid_disks    != info->raid_disks)
4164                 rv = update_raid_disks(mddev, info->raid_disks);
4165
4166         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4167                 if (mddev->pers->quiesce == NULL)
4168                         return -EINVAL;
4169                 if (mddev->recovery || mddev->sync_thread)
4170                         return -EBUSY;
4171                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4172                         /* add the bitmap */
4173                         if (mddev->bitmap)
4174                                 return -EEXIST;
4175                         if (mddev->default_bitmap_offset == 0)
4176                                 return -EINVAL;
4177                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4178                         mddev->pers->quiesce(mddev, 1);
4179                         rv = bitmap_create(mddev);
4180                         if (rv)
4181                                 bitmap_destroy(mddev);
4182                         mddev->pers->quiesce(mddev, 0);
4183                 } else {
4184                         /* remove the bitmap */
4185                         if (!mddev->bitmap)
4186                                 return -ENOENT;
4187                         if (mddev->bitmap->file)
4188                                 return -EINVAL;
4189                         mddev->pers->quiesce(mddev, 1);
4190                         bitmap_destroy(mddev);
4191                         mddev->pers->quiesce(mddev, 0);
4192                         mddev->bitmap_offset = 0;
4193                 }
4194         }
4195         md_update_sb(mddev, 1);
4196         return rv;
4197 }
4198
4199 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4200 {
4201         mdk_rdev_t *rdev;
4202
4203         if (mddev->pers == NULL)
4204                 return -ENODEV;
4205
4206         rdev = find_rdev(mddev, dev);
4207         if (!rdev)
4208                 return -ENODEV;
4209
4210         md_error(mddev, rdev);
4211         return 0;
4212 }
4213
4214 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4215 {
4216         mddev_t *mddev = bdev->bd_disk->private_data;
4217
4218         geo->heads = 2;
4219         geo->sectors = 4;
4220         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4221         return 0;
4222 }
4223
4224 static int md_ioctl(struct inode *inode, struct file *file,
4225                         unsigned int cmd, unsigned long arg)
4226 {
4227         int err = 0;
4228         void __user *argp = (void __user *)arg;
4229         mddev_t *mddev = NULL;
4230
4231         if (!capable(CAP_SYS_ADMIN))
4232                 return -EACCES;
4233
4234         /*
4235          * Commands dealing with the RAID driver but not any
4236          * particular array:
4237          */
4238         switch (cmd)
4239         {
4240                 case RAID_VERSION:
4241                         err = get_version(argp);
4242                         goto done;
4243
4244                 case PRINT_RAID_DEBUG:
4245                         err = 0;
4246                         md_print_devices();
4247                         goto done;
4248
4249 #ifndef MODULE
4250                 case RAID_AUTORUN:
4251                         err = 0;
4252                         autostart_arrays(arg);
4253                         goto done;
4254 #endif
4255                 default:;
4256         }
4257
4258         /*
4259          * Commands creating/starting a new array:
4260          */
4261
4262         mddev = inode->i_bdev->bd_disk->private_data;
4263
4264         if (!mddev) {
4265                 BUG();
4266                 goto abort;
4267         }
4268
4269         err = mddev_lock(mddev);
4270         if (err) {
4271                 printk(KERN_INFO 
4272                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4273                         err, cmd);
4274                 goto abort;
4275         }
4276
4277         switch (cmd)
4278         {
4279                 case SET_ARRAY_INFO:
4280                         {
4281                                 mdu_array_info_t info;
4282                                 if (!arg)
4283                                         memset(&info, 0, sizeof(info));
4284                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4285                                         err = -EFAULT;
4286                                         goto abort_unlock;
4287                                 }
4288                                 if (mddev->pers) {
4289                                         err = update_array_info(mddev, &info);
4290                                         if (err) {
4291                                                 printk(KERN_WARNING "md: couldn't update"
4292                                                        " array info. %d\n", err);
4293                                                 goto abort_unlock;
4294                                         }
4295                                         goto done_unlock;
4296                                 }
4297                                 if (!list_empty(&mddev->disks)) {
4298                                         printk(KERN_WARNING
4299                                                "md: array %s already has disks!\n",
4300                                                mdname(mddev));
4301                                         err = -EBUSY;
4302                                         goto abort_unlock;
4303                                 }
4304                                 if (mddev->raid_disks) {
4305                                         printk(KERN_WARNING
4306                                                "md: array %s already initialised!\n",
4307                                                mdname(mddev));
4308                                         err = -EBUSY;
4309                                         goto abort_unlock;
4310                                 }
4311                                 err = set_array_info(mddev, &info);
4312                                 if (err) {
4313                                         printk(KERN_WARNING "md: couldn't set"
4314                                                " array info. %d\n", err);
4315                                         goto abort_unlock;
4316                                 }
4317                         }
4318                         goto done_unlock;
4319
4320                 default:;
4321         }
4322
4323         /*
4324          * Commands querying/configuring an existing array:
4325          */
4326         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4327          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4328         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4329                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4330                         && cmd != GET_BITMAP_FILE) {
4331                 err = -ENODEV;
4332                 goto abort_unlock;
4333         }
4334
4335         /*
4336          * Commands even a read-only array can execute:
4337          */
4338         switch (cmd)
4339         {
4340                 case GET_ARRAY_INFO:
4341                         err = get_array_info(mddev, argp);
4342                         goto done_unlock;
4343
4344                 case GET_BITMAP_FILE:
4345                         err = get_bitmap_file(mddev, argp);
4346                         goto done_unlock;
4347
4348                 case GET_DISK_INFO:
4349                         err = get_disk_info(mddev, argp);
4350                         goto done_unlock;
4351
4352                 case RESTART_ARRAY_RW:
4353                         err = restart_array(mddev);
4354                         goto done_unlock;
4355
4356                 case STOP_ARRAY:
4357                         err = do_md_stop (mddev, 0);
4358                         goto done_unlock;
4359
4360                 case STOP_ARRAY_RO:
4361                         err = do_md_stop (mddev, 1);
4362                         goto done_unlock;
4363
4364         /*
4365          * We have a problem here : there is no easy way to give a CHS
4366          * virtual geometry. We currently pretend that we have a 2 heads
4367          * 4 sectors (with a BIG number of cylinders...). This drives
4368          * dosfs just mad... ;-)
4369          */
4370         }
4371
4372         /*
4373          * The remaining ioctls are changing the state of the
4374          * superblock, so we do not allow them on read-only arrays.
4375          * However non-MD ioctls (e.g. get-size) will still come through
4376          * here and hit the 'default' below, so only disallow
4377          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4378          */
4379         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4380             mddev->ro && mddev->pers) {
4381                 if (mddev->ro == 2) {
4382                         mddev->ro = 0;
4383                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4384                 md_wakeup_thread(mddev->thread);
4385
4386                 } else {
4387                         err = -EROFS;
4388                         goto abort_unlock;
4389                 }
4390         }
4391
4392         switch (cmd)
4393         {
4394                 case ADD_NEW_DISK:
4395                 {
4396                         mdu_disk_info_t info;
4397                         if (copy_from_user(&info, argp, sizeof(info)))
4398                                 err = -EFAULT;
4399                         else
4400                                 err = add_new_disk(mddev, &info);
4401                         goto done_unlock;
4402                 }
4403
4404                 case HOT_REMOVE_DISK:
4405                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4406                         goto done_unlock;
4407
4408                 case HOT_ADD_DISK:
4409                         err = hot_add_disk(mddev, new_decode_dev(arg));
4410                         goto done_unlock;
4411
4412                 case SET_DISK_FAULTY:
4413                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4414                         goto done_unlock;
4415
4416                 case RUN_ARRAY:
4417                         err = do_md_run (mddev);
4418                         goto done_unlock;
4419
4420                 case SET_BITMAP_FILE:
4421                         err = set_bitmap_file(mddev, (int)arg);
4422                         goto done_unlock;
4423
4424                 default:
4425                         err = -EINVAL;
4426                         goto abort_unlock;
4427         }
4428
4429 done_unlock:
4430 abort_unlock:
4431         mddev_unlock(mddev);
4432
4433         return err;
4434 done:
4435         if (err)
4436                 MD_BUG();
4437 abort:
4438         return err;
4439 }
4440
4441 static int md_open(struct inode *inode, struct file *file)
4442 {
4443         /*
4444          * Succeed if we can lock the mddev, which confirms that
4445          * it isn't being stopped right now.
4446          */
4447         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4448         int err;
4449
4450         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4451                 goto out;
4452
4453         err = 0;
4454         mddev_get(mddev);
4455         mddev_unlock(mddev);
4456
4457         check_disk_change(inode->i_bdev);
4458  out:
4459         return err;
4460 }
4461
4462 static int md_release(struct inode *inode, struct file * file)
4463 {
4464         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4465
4466         BUG_ON(!mddev);
4467         mddev_put(mddev);
4468
4469         return 0;
4470 }
4471
4472 static int md_media_changed(struct gendisk *disk)
4473 {
4474         mddev_t *mddev = disk->private_data;
4475
4476         return mddev->changed;
4477 }
4478
4479 static int md_revalidate(struct gendisk *disk)
4480 {
4481         mddev_t *mddev = disk->private_data;
4482
4483         mddev->changed = 0;
4484         return 0;
4485 }
4486 static struct block_device_operations md_fops =
4487 {
4488         .owner          = THIS_MODULE,
4489         .open           = md_open,
4490         .release        = md_release,
4491         .ioctl          = md_ioctl,
4492         .getgeo         = md_getgeo,
4493         .media_changed  = md_media_changed,
4494         .revalidate_disk= md_revalidate,
4495 };
4496
4497 static int md_thread(void * arg)
4498 {
4499         mdk_thread_t *thread = arg;
4500
4501         /*
4502          * md_thread is a 'system-thread', it's priority should be very
4503          * high. We avoid resource deadlocks individually in each
4504          * raid personality. (RAID5 does preallocation) We also use RR and
4505          * the very same RT priority as kswapd, thus we will never get
4506          * into a priority inversion deadlock.
4507          *
4508          * we definitely have to have equal or higher priority than
4509          * bdflush, otherwise bdflush will deadlock if there are too
4510          * many dirty RAID5 blocks.
4511          */
4512
4513         current->flags |= PF_NOFREEZE;
4514         allow_signal(SIGKILL);
4515         while (!kthread_should_stop()) {
4516
4517                 /* We need to wait INTERRUPTIBLE so that
4518                  * we don't add to the load-average.
4519                  * That means we need to be sure no signals are
4520                  * pending
4521                  */
4522                 if (signal_pending(current))
4523                         flush_signals(current);
4524
4525                 wait_event_interruptible_timeout
4526                         (thread->wqueue,
4527                          test_bit(THREAD_WAKEUP, &thread->flags)
4528                          || kthread_should_stop(),
4529                          thread->timeout);
4530
4531                 clear_bit(THREAD_WAKEUP, &thread->flags);
4532
4533                 thread->run(thread->mddev);
4534         }
4535
4536         return 0;
4537 }
4538
4539 void md_wakeup_thread(mdk_thread_t *thread)
4540 {
4541         if (thread) {
4542                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4543                 set_bit(THREAD_WAKEUP, &thread->flags);
4544                 wake_up(&thread->wqueue);
4545         }
4546 }
4547
4548 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4549                                  const char *name)
4550 {
4551         mdk_thread_t *thread;
4552
4553         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4554         if (!thread)
4555                 return NULL;
4556
4557         init_waitqueue_head(&thread->wqueue);
4558
4559         thread->run = run;
4560         thread->mddev = mddev;
4561         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4562         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4563         if (IS_ERR(thread->tsk)) {
4564                 kfree(thread);
4565                 return NULL;
4566         }
4567         return thread;
4568 }
4569
4570 void md_unregister_thread(mdk_thread_t *thread)
4571 {
4572         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4573
4574         kthread_stop(thread->tsk);
4575         kfree(thread);
4576 }
4577
4578 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4579 {
4580         if (!mddev) {
4581                 MD_BUG();
4582                 return;
4583         }
4584
4585         if (!rdev || test_bit(Faulty, &rdev->flags))
4586                 return;
4587 /*
4588         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4589                 mdname(mddev),
4590                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4591                 __builtin_return_address(0),__builtin_return_address(1),
4592                 __builtin_return_address(2),__builtin_return_address(3));
4593 */
4594         if (!mddev->pers)
4595                 return;
4596         if (!mddev->pers->error_handler)
4597                 return;
4598         mddev->pers->error_handler(mddev,rdev);
4599         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4600         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4601         md_wakeup_thread(mddev->thread);
4602         md_new_event_inintr(mddev);
4603 }
4604
4605 /* seq_file implementation /proc/mdstat */
4606
4607 static void status_unused(struct seq_file *seq)
4608 {
4609         int i = 0;
4610         mdk_rdev_t *rdev;
4611         struct list_head *tmp;
4612
4613         seq_printf(seq, "unused devices: ");
4614
4615         ITERATE_RDEV_PENDING(rdev,tmp) {
4616                 char b[BDEVNAME_SIZE];
4617                 i++;
4618                 seq_printf(seq, "%s ",
4619                               bdevname(rdev->bdev,b));
4620         }
4621         if (!i)
4622                 seq_printf(seq, "<none>");
4623
4624         seq_printf(seq, "\n");
4625 }
4626
4627
4628 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4629 {
4630         sector_t max_blocks, resync, res;
4631         unsigned long dt, db, rt;
4632         int scale;
4633         unsigned int per_milli;
4634
4635         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4636
4637         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4638                 max_blocks = mddev->resync_max_sectors >> 1;
4639         else
4640                 max_blocks = mddev->size;
4641
4642         /*
4643          * Should not happen.
4644          */
4645         if (!max_blocks) {
4646                 MD_BUG();
4647                 return;
4648         }
4649         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4650          * in a sector_t, and (max_blocks>>scale) will fit in a
4651          * u32, as those are the requirements for sector_div.
4652          * Thus 'scale' must be at least 10
4653          */
4654         scale = 10;
4655         if (sizeof(sector_t) > sizeof(unsigned long)) {
4656                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4657                         scale++;
4658         }
4659         res = (resync>>scale)*1000;
4660         sector_div(res, (u32)((max_blocks>>scale)+1));
4661
4662         per_milli = res;
4663         {
4664                 int i, x = per_milli/50, y = 20-x;
4665                 seq_printf(seq, "[");
4666                 for (i = 0; i < x; i++)
4667                         seq_printf(seq, "=");
4668                 seq_printf(seq, ">");
4669                 for (i = 0; i < y; i++)
4670                         seq_printf(seq, ".");
4671                 seq_printf(seq, "] ");
4672         }
4673         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4674                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4675                     "reshape" :
4676                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4677                      "check" :
4678                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4679                       "resync" : "recovery"))),
4680                    per_milli/10, per_milli % 10,
4681                    (unsigned long long) resync,
4682                    (unsigned long long) max_blocks);
4683
4684         /*
4685          * We do not want to overflow, so the order of operands and
4686          * the * 100 / 100 trick are important. We do a +1 to be
4687          * safe against division by zero. We only estimate anyway.
4688          *
4689          * dt: time from mark until now
4690          * db: blocks written from mark until now
4691          * rt: remaining time
4692          */
4693         dt = ((jiffies - mddev->resync_mark) / HZ);
4694         if (!dt) dt++;
4695         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4696                 - mddev->resync_mark_cnt;
4697         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4698
4699         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4700
4701         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4702 }
4703
4704 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4705 {
4706         struct list_head *tmp;
4707         loff_t l = *pos;
4708         mddev_t *mddev;
4709
4710         if (l >= 0x10000)
4711                 return NULL;
4712         if (!l--)
4713                 /* header */
4714                 return (void*)1;
4715
4716         spin_lock(&all_mddevs_lock);
4717         list_for_each(tmp,&all_mddevs)
4718                 if (!l--) {
4719                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4720                         mddev_get(mddev);
4721                         spin_unlock(&all_mddevs_lock);
4722                         return mddev;
4723                 }
4724         spin_unlock(&all_mddevs_lock);
4725         if (!l--)
4726                 return (void*)2;/* tail */
4727         return NULL;
4728 }
4729
4730 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4731 {
4732         struct list_head *tmp;
4733         mddev_t *next_mddev, *mddev = v;
4734         
4735         ++*pos;
4736         if (v == (void*)2)
4737                 return NULL;
4738
4739         spin_lock(&all_mddevs_lock);
4740         if (v == (void*)1)
4741                 tmp = all_mddevs.next;
4742         else
4743                 tmp = mddev->all_mddevs.next;
4744         if (tmp != &all_mddevs)
4745                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4746         else {
4747                 next_mddev = (void*)2;
4748                 *pos = 0x10000;
4749         }               
4750         spin_unlock(&all_mddevs_lock);
4751
4752         if (v != (void*)1)
4753                 mddev_put(mddev);
4754         return next_mddev;
4755
4756 }
4757
4758 static void md_seq_stop(struct seq_file *seq, void *v)
4759 {
4760         mddev_t *mddev = v;
4761
4762         if (mddev && v != (void*)1 && v != (void*)2)
4763                 mddev_put(mddev);
4764 }
4765
4766 struct mdstat_info {
4767         int event;
4768 };
4769
4770 static int md_seq_show(struct seq_file *seq, void *v)
4771 {
4772         mddev_t *mddev = v;
4773         sector_t size;
4774         struct list_head *tmp2;
4775         mdk_rdev_t *rdev;
4776         struct mdstat_info *mi = seq->private;
4777         struct bitmap *bitmap;
4778
4779         if (v == (void*)1) {
4780                 struct mdk_personality *pers;
4781                 seq_printf(seq, "Personalities : ");
4782                 spin_lock(&pers_lock);
4783                 list_for_each_entry(pers, &pers_list, list)
4784                         seq_printf(seq, "[%s] ", pers->name);
4785
4786                 spin_unlock(&pers_lock);
4787                 seq_printf(seq, "\n");
4788                 mi->event = atomic_read(&md_event_count);
4789                 return 0;
4790         }
4791         if (v == (void*)2) {
4792                 status_unused(seq);
4793                 return 0;
4794         }
4795
4796         if (mddev_lock(mddev) < 0)
4797                 return -EINTR;
4798
4799         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4800                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4801                                                 mddev->pers ? "" : "in");
4802                 if (mddev->pers) {
4803                         if (mddev->ro==1)
4804                                 seq_printf(seq, " (read-only)");
4805                         if (mddev->ro==2)
4806                                 seq_printf(seq, "(auto-read-only)");
4807                         seq_printf(seq, " %s", mddev->pers->name);
4808                 }
4809
4810                 size = 0;
4811                 ITERATE_RDEV(mddev,rdev,tmp2) {
4812                         char b[BDEVNAME_SIZE];
4813                         seq_printf(seq, " %s[%d]",
4814                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4815                         if (test_bit(WriteMostly, &rdev->flags))
4816                                 seq_printf(seq, "(W)");
4817                         if (test_bit(Faulty, &rdev->flags)) {
4818                                 seq_printf(seq, "(F)");
4819                                 continue;
4820                         } else if (rdev->raid_disk < 0)
4821                                 seq_printf(seq, "(S)"); /* spare */
4822                         size += rdev->size;
4823                 }
4824
4825                 if (!list_empty(&mddev->disks)) {
4826                         if (mddev->pers)
4827                                 seq_printf(seq, "\n      %llu blocks",
4828                                         (unsigned long long)mddev->array_size);
4829                         else
4830                                 seq_printf(seq, "\n      %llu blocks",
4831                                         (unsigned long long)size);
4832                 }
4833                 if (mddev->persistent) {
4834                         if (mddev->major_version != 0 ||
4835                             mddev->minor_version != 90) {
4836                                 seq_printf(seq," super %d.%d",
4837                                            mddev->major_version,
4838                                            mddev->minor_version);
4839                         }
4840                 } else
4841                         seq_printf(seq, " super non-persistent");
4842
4843                 if (mddev->pers) {
4844                         mddev->pers->status (seq, mddev);
4845                         seq_printf(seq, "\n      ");
4846                         if (mddev->pers->sync_request) {
4847                                 if (mddev->curr_resync > 2) {
4848                                         status_resync (seq, mddev);
4849                                         seq_printf(seq, "\n      ");
4850                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4851                                         seq_printf(seq, "\tresync=DELAYED\n      ");
4852                                 else if (mddev->recovery_cp < MaxSector)
4853                                         seq_printf(seq, "\tresync=PENDING\n      ");
4854                         }
4855                 } else
4856                         seq_printf(seq, "\n       ");
4857
4858                 if ((bitmap = mddev->bitmap)) {
4859                         unsigned long chunk_kb;
4860                         unsigned long flags;
4861                         spin_lock_irqsave(&bitmap->lock, flags);
4862                         chunk_kb = bitmap->chunksize >> 10;
4863                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4864                                 "%lu%s chunk",
4865                                 bitmap->pages - bitmap->missing_pages,
4866                                 bitmap->pages,
4867                                 (bitmap->pages - bitmap->missing_pages)
4868                                         << (PAGE_SHIFT - 10),
4869                                 chunk_kb ? chunk_kb : bitmap->chunksize,
4870                                 chunk_kb ? "KB" : "B");
4871                         if (bitmap->file) {
4872                                 seq_printf(seq, ", file: ");
4873                                 seq_path(seq, bitmap->file->f_path.mnt,
4874                                          bitmap->file->f_path.dentry," \t\n");
4875                         }
4876
4877                         seq_printf(seq, "\n");
4878                         spin_unlock_irqrestore(&bitmap->lock, flags);
4879                 }
4880
4881                 seq_printf(seq, "\n");
4882         }
4883         mddev_unlock(mddev);
4884         
4885         return 0;
4886 }
4887
4888 static struct seq_operations md_seq_ops = {
4889         .start  = md_seq_start,
4890         .next   = md_seq_next,
4891         .stop   = md_seq_stop,
4892         .show   = md_seq_show,
4893 };
4894
4895 static int md_seq_open(struct inode *inode, struct file *file)
4896 {
4897         int error;
4898         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4899         if (mi == NULL)
4900                 return -ENOMEM;
4901
4902         error = seq_open(file, &md_seq_ops);
4903         if (error)
4904                 kfree(mi);
4905         else {
4906                 struct seq_file *p = file->private_data;
4907                 p->private = mi;
4908                 mi->event = atomic_read(&md_event_count);
4909         }
4910         return error;
4911 }
4912
4913 static int md_seq_release(struct inode *inode, struct file *file)
4914 {
4915         struct seq_file *m = file->private_data;
4916         struct mdstat_info *mi = m->private;
4917         m->private = NULL;
4918         kfree(mi);
4919         return seq_release(inode, file);
4920 }
4921
4922 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4923 {
4924         struct seq_file *m = filp->private_data;
4925         struct mdstat_info *mi = m->private;
4926         int mask;
4927
4928         poll_wait(filp, &md_event_waiters, wait);
4929
4930         /* always allow read */
4931         mask = POLLIN | POLLRDNORM;
4932
4933         if (mi->event != atomic_read(&md_event_count))
4934                 mask |= POLLERR | POLLPRI;
4935         return mask;
4936 }
4937
4938 static const struct file_operations md_seq_fops = {
4939         .owner          = THIS_MODULE,
4940         .open           = md_seq_open,
4941         .read           = seq_read,
4942         .llseek         = seq_lseek,
4943         .release        = md_seq_release,
4944         .poll           = mdstat_poll,
4945 };
4946
4947 int register_md_personality(struct mdk_personality *p)
4948 {
4949         spin_lock(&pers_lock);
4950         list_add_tail(&p->list, &pers_list);
4951         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4952         spin_unlock(&pers_lock);
4953         return 0;
4954 }
4955
4956 int unregister_md_personality(struct mdk_personality *p)
4957 {
4958         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4959         spin_lock(&pers_lock);
4960         list_del_init(&p->list);
4961         spin_unlock(&pers_lock);
4962         return 0;
4963 }
4964
4965 static int is_mddev_idle(mddev_t *mddev)
4966 {
4967         mdk_rdev_t * rdev;
4968         struct list_head *tmp;
4969         int idle;
4970         unsigned long curr_events;
4971
4972         idle = 1;
4973         ITERATE_RDEV(mddev,rdev,tmp) {
4974                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4975                 curr_events = disk_stat_read(disk, sectors[0]) + 
4976                                 disk_stat_read(disk, sectors[1]) - 
4977                                 atomic_read(&disk->sync_io);
4978                 /* The difference between curr_events and last_events
4979                  * will be affected by any new non-sync IO (making
4980                  * curr_events bigger) and any difference in the amount of
4981                  * in-flight syncio (making current_events bigger or smaller)
4982                  * The amount in-flight is currently limited to
4983                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4984                  * which is at most 4096 sectors.
4985                  * These numbers are fairly fragile and should be made
4986                  * more robust, probably by enforcing the
4987                  * 'window size' that md_do_sync sort-of uses.
4988                  *
4989                  * Note: the following is an unsigned comparison.
4990                  */
4991                 if ((curr_events - rdev->last_events + 4096) > 8192) {
4992                         rdev->last_events = curr_events;
4993                         idle = 0;
4994                 }
4995         }
4996         return idle;
4997 }
4998
4999 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5000 {
5001         /* another "blocks" (512byte) blocks have been synced */
5002         atomic_sub(blocks, &mddev->recovery_active);
5003         wake_up(&mddev->recovery_wait);
5004         if (!ok) {
5005                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5006                 md_wakeup_thread(mddev->thread);
5007                 // stop recovery, signal do_sync ....
5008         }
5009 }
5010
5011
5012 /* md_write_start(mddev, bi)
5013  * If we need to update some array metadata (e.g. 'active' flag
5014  * in superblock) before writing, schedule a superblock update
5015  * and wait for it to complete.
5016  */
5017 void md_write_start(mddev_t *mddev, struct bio *bi)
5018 {
5019         if (bio_data_dir(bi) != WRITE)
5020                 return;
5021
5022         BUG_ON(mddev->ro == 1);
5023         if (mddev->ro == 2) {
5024                 /* need to switch to read/write */
5025                 mddev->ro = 0;
5026                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5027                 md_wakeup_thread(mddev->thread);
5028         }
5029         atomic_inc(&mddev->writes_pending);
5030         if (mddev->in_sync) {
5031                 spin_lock_irq(&mddev->write_lock);
5032                 if (mddev->in_sync) {
5033                         mddev->in_sync = 0;
5034                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5035                         md_wakeup_thread(mddev->thread);
5036                 }
5037                 spin_unlock_irq(&mddev->write_lock);
5038         }
5039         wait_event(mddev->sb_wait, mddev->flags==0);
5040 }
5041
5042 void md_write_end(mddev_t *mddev)
5043 {
5044         if (atomic_dec_and_test(&mddev->writes_pending)) {
5045                 if (mddev->safemode == 2)
5046                         md_wakeup_thread(mddev->thread);
5047                 else if (mddev->safemode_delay)
5048                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5049         }
5050 }
5051
5052 /* md_allow_write(mddev)
5053  * Calling this ensures that the array is marked 'active' so that writes
5054  * may proceed without blocking.  It is important to call this before
5055  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5056  * Must be called with mddev_lock held.
5057  */
5058 void md_allow_write(mddev_t *mddev)
5059 {
5060         if (!mddev->pers)
5061                 return;
5062         if (mddev->ro)
5063                 return;
5064
5065         spin_lock_irq(&mddev->write_lock);
5066         if (mddev->in_sync) {
5067                 mddev->in_sync = 0;
5068                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5069                 if (mddev->safemode_delay &&
5070                     mddev->safemode == 0)
5071                         mddev->safemode = 1;
5072                 spin_unlock_irq(&mddev->write_lock);
5073                 md_update_sb(mddev, 0);
5074         } else
5075                 spin_unlock_irq(&mddev->write_lock);
5076 }
5077 EXPORT_SYMBOL_GPL(md_allow_write);
5078
5079 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5080
5081 #define SYNC_MARKS      10
5082 #define SYNC_MARK_STEP  (3*HZ)
5083 void md_do_sync(mddev_t *mddev)
5084 {
5085         mddev_t *mddev2;
5086         unsigned int currspeed = 0,
5087                  window;
5088         sector_t max_sectors,j, io_sectors;
5089         unsigned long mark[SYNC_MARKS];
5090         sector_t mark_cnt[SYNC_MARKS];
5091         int last_mark,m;
5092         struct list_head *tmp;
5093         sector_t last_check;
5094         int skipped = 0;
5095         struct list_head *rtmp;
5096         mdk_rdev_t *rdev;
5097         char *desc;
5098
5099         /* just incase thread restarts... */
5100         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5101                 return;
5102         if (mddev->ro) /* never try to sync a read-only array */
5103                 return;
5104
5105         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5106                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5107                         desc = "data-check";
5108                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5109                         desc = "requested-resync";
5110                 else
5111                         desc = "resync";
5112         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5113                 desc = "reshape";
5114         else
5115                 desc = "recovery";
5116
5117         /* we overload curr_resync somewhat here.
5118          * 0 == not engaged in resync at all
5119          * 2 == checking that there is no conflict with another sync
5120          * 1 == like 2, but have yielded to allow conflicting resync to
5121          *              commense
5122          * other == active in resync - this many blocks
5123          *
5124          * Before starting a resync we must have set curr_resync to
5125          * 2, and then checked that every "conflicting" array has curr_resync
5126          * less than ours.  When we find one that is the same or higher
5127          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5128          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5129          * This will mean we have to start checking from the beginning again.
5130          *
5131          */
5132
5133         do {
5134                 mddev->curr_resync = 2;
5135
5136         try_again:
5137                 if (kthread_should_stop()) {
5138                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5139                         goto skip;
5140                 }
5141                 ITERATE_MDDEV(mddev2,tmp) {
5142                         if (mddev2 == mddev)
5143                                 continue;
5144                         if (mddev2->curr_resync && 
5145                             match_mddev_units(mddev,mddev2)) {
5146                                 DEFINE_WAIT(wq);
5147                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5148                                         /* arbitrarily yield */
5149                                         mddev->curr_resync = 1;
5150                                         wake_up(&resync_wait);
5151                                 }
5152                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5153                                         /* no need to wait here, we can wait the next
5154                                          * time 'round when curr_resync == 2
5155                                          */
5156                                         continue;
5157                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5158                                 if (!kthread_should_stop() &&
5159                                     mddev2->curr_resync >= mddev->curr_resync) {
5160                                         printk(KERN_INFO "md: delaying %s of %s"
5161                                                " until %s has finished (they"
5162                                                " share one or more physical units)\n",
5163                                                desc, mdname(mddev), mdname(mddev2));
5164                                         mddev_put(mddev2);
5165                                         schedule();
5166                                         finish_wait(&resync_wait, &wq);
5167                                         goto try_again;
5168                                 }
5169                                 finish_wait(&resync_wait, &wq);
5170                         }
5171                 }
5172         } while (mddev->curr_resync < 2);
5173
5174         j = 0;
5175         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5176                 /* resync follows the size requested by the personality,
5177                  * which defaults to physical size, but can be virtual size
5178                  */
5179                 max_sectors = mddev->resync_max_sectors;
5180                 mddev->resync_mismatches = 0;
5181                 /* we don't use the checkpoint if there's a bitmap */
5182                 if (!mddev->bitmap &&
5183                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5184                         j = mddev->recovery_cp;
5185         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5186                 max_sectors = mddev->size << 1;
5187         else {
5188                 /* recovery follows the physical size of devices */
5189                 max_sectors = mddev->size << 1;
5190                 j = MaxSector;
5191                 ITERATE_RDEV(mddev,rdev,rtmp)
5192                         if (rdev->raid_disk >= 0 &&
5193                             !test_bit(Faulty, &rdev->flags) &&
5194                             !test_bit(In_sync, &rdev->flags) &&
5195                             rdev->recovery_offset < j)
5196                                 j = rdev->recovery_offset;
5197         }
5198
5199         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5200         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5201                 " %d KB/sec/disk.\n", speed_min(mddev));
5202         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5203                "(but not more than %d KB/sec) for %s.\n",
5204                speed_max(mddev), desc);
5205
5206         is_mddev_idle(mddev); /* this also initializes IO event counters */
5207
5208         io_sectors = 0;
5209         for (m = 0; m < SYNC_MARKS; m++) {
5210                 mark[m] = jiffies;
5211                 mark_cnt[m] = io_sectors;
5212         }
5213         last_mark = 0;
5214         mddev->resync_mark = mark[last_mark];
5215         mddev->resync_mark_cnt = mark_cnt[last_mark];
5216
5217         /*
5218          * Tune reconstruction:
5219          */
5220         window = 32*(PAGE_SIZE/512);
5221         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5222                 window/2,(unsigned long long) max_sectors/2);
5223
5224         atomic_set(&mddev->recovery_active, 0);
5225         init_waitqueue_head(&mddev->recovery_wait);
5226         last_check = 0;
5227
5228         if (j>2) {
5229                 printk(KERN_INFO 
5230                        "md: resuming %s of %s from checkpoint.\n",
5231                        desc, mdname(mddev));
5232                 mddev->curr_resync = j;
5233         }
5234
5235         while (j < max_sectors) {
5236                 sector_t sectors;
5237
5238                 skipped = 0;
5239                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5240                                             currspeed < speed_min(mddev));
5241                 if (sectors == 0) {
5242                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5243                         goto out;
5244                 }
5245
5246                 if (!skipped) { /* actual IO requested */
5247                         io_sectors += sectors;
5248                         atomic_add(sectors, &mddev->recovery_active);
5249                 }
5250
5251                 j += sectors;
5252                 if (j>1) mddev->curr_resync = j;
5253                 mddev->curr_mark_cnt = io_sectors;
5254                 if (last_check == 0)
5255                         /* this is the earliers that rebuilt will be
5256                          * visible in /proc/mdstat
5257                          */
5258                         md_new_event(mddev);
5259
5260                 if (last_check + window > io_sectors || j == max_sectors)
5261                         continue;
5262
5263                 last_check = io_sectors;
5264
5265                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5266                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5267                         break;
5268
5269         repeat:
5270                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5271                         /* step marks */
5272                         int next = (last_mark+1) % SYNC_MARKS;
5273
5274                         mddev->resync_mark = mark[next];
5275                         mddev->resync_mark_cnt = mark_cnt[next];
5276                         mark[next] = jiffies;
5277                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5278                         last_mark = next;
5279                 }
5280
5281
5282                 if (kthread_should_stop()) {
5283                         /*
5284                          * got a signal, exit.
5285                          */
5286                         printk(KERN_INFO 
5287                                 "md: md_do_sync() got signal ... exiting\n");
5288                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5289                         goto out;
5290                 }
5291
5292                 /*
5293                  * this loop exits only if either when we are slower than
5294                  * the 'hard' speed limit, or the system was IO-idle for
5295                  * a jiffy.
5296                  * the system might be non-idle CPU-wise, but we only care
5297                  * about not overloading the IO subsystem. (things like an
5298                  * e2fsck being done on the RAID array should execute fast)
5299                  */
5300                 mddev->queue->unplug_fn(mddev->queue);
5301                 cond_resched();
5302
5303                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5304                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5305
5306                 if (currspeed > speed_min(mddev)) {
5307                         if ((currspeed > speed_max(mddev)) ||
5308                                         !is_mddev_idle(mddev)) {
5309                                 msleep(500);
5310                                 goto repeat;
5311                         }
5312                 }
5313         }
5314         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5315         /*
5316          * this also signals 'finished resyncing' to md_stop
5317          */
5318  out:
5319         mddev->queue->unplug_fn(mddev->queue);
5320
5321         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5322
5323         /* tell personality that we are finished */
5324         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5325
5326         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5327             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5328             mddev->curr_resync > 2) {
5329                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5330                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5331                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5332                                         printk(KERN_INFO
5333                                                "md: checkpointing %s of %s.\n",
5334                                                desc, mdname(mddev));
5335                                         mddev->recovery_cp = mddev->curr_resync;
5336                                 }
5337                         } else
5338                                 mddev->recovery_cp = MaxSector;
5339                 } else {
5340                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5341                                 mddev->curr_resync = MaxSector;
5342                         ITERATE_RDEV(mddev,rdev,rtmp)
5343                                 if (rdev->raid_disk >= 0 &&
5344                                     !test_bit(Faulty, &rdev->flags) &&
5345                                     !test_bit(In_sync, &rdev->flags) &&
5346                                     rdev->recovery_offset < mddev->curr_resync)
5347                                         rdev->recovery_offset = mddev->curr_resync;
5348                 }
5349         }
5350         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5351
5352  skip:
5353         mddev->curr_resync = 0;
5354         wake_up(&resync_wait);
5355         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5356         md_wakeup_thread(mddev->thread);
5357 }
5358 EXPORT_SYMBOL_GPL(md_do_sync);
5359
5360
5361 static int remove_and_add_spares(mddev_t *mddev)
5362 {
5363         mdk_rdev_t *rdev;
5364         struct list_head *rtmp;
5365         int spares = 0;
5366
5367         ITERATE_RDEV(mddev,rdev,rtmp)
5368                 if (rdev->raid_disk >= 0 &&
5369                     (test_bit(Faulty, &rdev->flags) ||
5370                      ! test_bit(In_sync, &rdev->flags)) &&
5371                     atomic_read(&rdev->nr_pending)==0) {
5372                         if (mddev->pers->hot_remove_disk(
5373                                     mddev, rdev->raid_disk)==0) {
5374                                 char nm[20];
5375                                 sprintf(nm,"rd%d", rdev->raid_disk);
5376                                 sysfs_remove_link(&mddev->kobj, nm);
5377                                 rdev->raid_disk = -1;
5378                         }
5379                 }
5380
5381         if (mddev->degraded) {
5382                 ITERATE_RDEV(mddev,rdev,rtmp)
5383                         if (rdev->raid_disk < 0
5384                             && !test_bit(Faulty, &rdev->flags)) {
5385                                 rdev->recovery_offset = 0;
5386                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5387                                         char nm[20];
5388                                         sprintf(nm, "rd%d", rdev->raid_disk);
5389                                         sysfs_create_link(&mddev->kobj,
5390                                                           &rdev->kobj, nm);
5391                                         spares++;
5392                                         md_new_event(mddev);
5393                                 } else
5394                                         break;
5395                         }
5396         }
5397         return spares;
5398 }
5399 /*
5400  * This routine is regularly called by all per-raid-array threads to
5401  * deal with generic issues like resync and super-block update.
5402  * Raid personalities that don't have a thread (linear/raid0) do not
5403  * need this as they never do any recovery or update the superblock.
5404  *
5405  * It does not do any resync itself, but rather "forks" off other threads
5406  * to do that as needed.
5407  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5408  * "->recovery" and create a thread at ->sync_thread.
5409  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5410  * and wakeups up this thread which will reap the thread and finish up.
5411  * This thread also removes any faulty devices (with nr_pending == 0).
5412  *
5413  * The overall approach is:
5414  *  1/ if the superblock needs updating, update it.
5415  *  2/ If a recovery thread is running, don't do anything else.
5416  *  3/ If recovery has finished, clean up, possibly marking spares active.
5417  *  4/ If there are any faulty devices, remove them.
5418  *  5/ If array is degraded, try to add spares devices
5419  *  6/ If array has spares or is not in-sync, start a resync thread.
5420  */
5421 void md_check_recovery(mddev_t *mddev)
5422 {
5423         mdk_rdev_t *rdev;
5424         struct list_head *rtmp;
5425
5426
5427         if (mddev->bitmap)
5428                 bitmap_daemon_work(mddev->bitmap);
5429
5430         if (mddev->ro)
5431                 return;
5432
5433         if (signal_pending(current)) {
5434                 if (mddev->pers->sync_request) {
5435                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5436                                mdname(mddev));
5437                         mddev->safemode = 2;
5438                 }
5439                 flush_signals(current);
5440         }
5441
5442         if ( ! (
5443                 mddev->flags ||
5444                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5445                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5446                 (mddev->safemode == 1) ||
5447                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5448                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5449                 ))
5450                 return;
5451
5452         if (mddev_trylock(mddev)) {
5453                 int spares = 0;
5454
5455                 spin_lock_irq(&mddev->write_lock);
5456                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5457                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5458                         mddev->in_sync = 1;
5459                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5460                 }
5461                 if (mddev->safemode == 1)
5462                         mddev->safemode = 0;
5463                 spin_unlock_irq(&mddev->write_lock);
5464
5465                 if (mddev->flags)
5466                         md_update_sb(mddev, 0);
5467
5468
5469                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5470                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5471                         /* resync/recovery still happening */
5472                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5473                         goto unlock;
5474                 }
5475                 if (mddev->sync_thread) {
5476                         /* resync has finished, collect result */
5477                         md_unregister_thread(mddev->sync_thread);
5478                         mddev->sync_thread = NULL;
5479                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5480                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5481                                 /* success...*/
5482                                 /* activate any spares */
5483                                 mddev->pers->spare_active(mddev);
5484                         }
5485                         md_update_sb(mddev, 1);
5486
5487                         /* if array is no-longer degraded, then any saved_raid_disk
5488                          * information must be scrapped
5489                          */
5490                         if (!mddev->degraded)
5491                                 ITERATE_RDEV(mddev,rdev,rtmp)
5492                                         rdev->saved_raid_disk = -1;
5493
5494                         mddev->recovery = 0;
5495                         /* flag recovery needed just to double check */
5496                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5497                         md_new_event(mddev);
5498                         goto unlock;
5499                 }
5500                 /* Clear some bits that don't mean anything, but
5501                  * might be left set
5502                  */
5503                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5504                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5505                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5506                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5507
5508                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5509                         goto unlock;
5510                 /* no recovery is running.
5511                  * remove any failed drives, then
5512                  * add spares if possible.
5513                  * Spare are also removed and re-added, to allow
5514                  * the personality to fail the re-add.
5515                  */
5516
5517                 if (mddev->reshape_position != MaxSector) {
5518                         if (mddev->pers->check_reshape(mddev) != 0)
5519                                 /* Cannot proceed */
5520                                 goto unlock;
5521                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5522                 } else if ((spares = remove_and_add_spares(mddev))) {
5523                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5524                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5525                 } else if (mddev->recovery_cp < MaxSector) {
5526                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5527                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5528                         /* nothing to be done ... */
5529                         goto unlock;
5530
5531                 if (mddev->pers->sync_request) {
5532                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5533                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5534                                 /* We are adding a device or devices to an array
5535                                  * which has the bitmap stored on all devices.
5536                                  * So make sure all bitmap pages get written
5537                                  */
5538                                 bitmap_write_all(mddev->bitmap);
5539                         }
5540                         mddev->sync_thread = md_register_thread(md_do_sync,
5541                                                                 mddev,
5542                                                                 "%s_resync");
5543                         if (!mddev->sync_thread) {
5544                                 printk(KERN_ERR "%s: could not start resync"
5545                                         " thread...\n", 
5546                                         mdname(mddev));
5547                                 /* leave the spares where they are, it shouldn't hurt */
5548                                 mddev->recovery = 0;
5549                         } else
5550                                 md_wakeup_thread(mddev->sync_thread);
5551                         md_new_event(mddev);
5552                 }
5553         unlock:
5554                 mddev_unlock(mddev);
5555         }
5556 }
5557
5558 static int md_notify_reboot(struct notifier_block *this,
5559                             unsigned long code, void *x)
5560 {
5561         struct list_head *tmp;
5562         mddev_t *mddev;
5563
5564         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5565
5566                 printk(KERN_INFO "md: stopping all md devices.\n");
5567
5568                 ITERATE_MDDEV(mddev,tmp)
5569                         if (mddev_trylock(mddev)) {
5570                                 do_md_stop (mddev, 1);
5571                                 mddev_unlock(mddev);
5572                         }
5573                 /*
5574                  * certain more exotic SCSI devices are known to be
5575                  * volatile wrt too early system reboots. While the
5576                  * right place to handle this issue is the given
5577                  * driver, we do want to have a safe RAID driver ...
5578                  */
5579                 mdelay(1000*1);
5580         }
5581         return NOTIFY_DONE;
5582 }
5583
5584 static struct notifier_block md_notifier = {
5585         .notifier_call  = md_notify_reboot,
5586         .next           = NULL,
5587         .priority       = INT_MAX, /* before any real devices */
5588 };
5589
5590 static void md_geninit(void)
5591 {
5592         struct proc_dir_entry *p;
5593
5594         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5595
5596         p = create_proc_entry("mdstat", S_IRUGO, NULL);
5597         if (p)
5598                 p->proc_fops = &md_seq_fops;
5599 }
5600
5601 static int __init md_init(void)
5602 {
5603         if (register_blkdev(MAJOR_NR, "md"))
5604                 return -1;
5605         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5606                 unregister_blkdev(MAJOR_NR, "md");
5607                 return -1;
5608         }
5609         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5610                             md_probe, NULL, NULL);
5611         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5612                             md_probe, NULL, NULL);
5613
5614         register_reboot_notifier(&md_notifier);
5615         raid_table_header = register_sysctl_table(raid_root_table);
5616
5617         md_geninit();
5618         return (0);
5619 }
5620
5621
5622 #ifndef MODULE
5623
5624 /*
5625  * Searches all registered partitions for autorun RAID arrays
5626  * at boot time.
5627  */
5628 static dev_t detected_devices[128];
5629 static int dev_cnt;
5630
5631 void md_autodetect_dev(dev_t dev)
5632 {
5633         if (dev_cnt >= 0 && dev_cnt < 127)
5634                 detected_devices[dev_cnt++] = dev;
5635 }
5636
5637
5638 static void autostart_arrays(int part)
5639 {
5640         mdk_rdev_t *rdev;
5641         int i;
5642
5643         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5644
5645         for (i = 0; i < dev_cnt; i++) {
5646                 dev_t dev = detected_devices[i];
5647
5648                 rdev = md_import_device(dev,0, 0);
5649                 if (IS_ERR(rdev))
5650                         continue;
5651
5652                 if (test_bit(Faulty, &rdev->flags)) {
5653                         MD_BUG();
5654                         continue;
5655                 }
5656                 list_add(&rdev->same_set, &pending_raid_disks);
5657         }
5658         dev_cnt = 0;
5659
5660         autorun_devices(part);
5661 }
5662
5663 #endif /* !MODULE */
5664
5665 static __exit void md_exit(void)
5666 {
5667         mddev_t *mddev;
5668         struct list_head *tmp;
5669
5670         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5671         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5672
5673         unregister_blkdev(MAJOR_NR,"md");
5674         unregister_blkdev(mdp_major, "mdp");
5675         unregister_reboot_notifier(&md_notifier);
5676         unregister_sysctl_table(raid_table_header);
5677         remove_proc_entry("mdstat", NULL);
5678         ITERATE_MDDEV(mddev,tmp) {
5679                 struct gendisk *disk = mddev->gendisk;
5680                 if (!disk)
5681                         continue;
5682                 export_array(mddev);
5683                 del_gendisk(disk);
5684                 put_disk(disk);
5685                 mddev->gendisk = NULL;
5686                 mddev_put(mddev);
5687         }
5688 }
5689
5690 module_init(md_init)
5691 module_exit(md_exit)
5692
5693 static int get_ro(char *buffer, struct kernel_param *kp)
5694 {
5695         return sprintf(buffer, "%d", start_readonly);
5696 }
5697 static int set_ro(const char *val, struct kernel_param *kp)
5698 {
5699         char *e;
5700         int num = simple_strtoul(val, &e, 10);
5701         if (*val && (*e == '\0' || *e == '\n')) {
5702                 start_readonly = num;
5703                 return 0;
5704         }
5705         return -EINVAL;
5706 }
5707
5708 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5709 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5710
5711
5712 EXPORT_SYMBOL(register_md_personality);
5713 EXPORT_SYMBOL(unregister_md_personality);
5714 EXPORT_SYMBOL(md_error);
5715 EXPORT_SYMBOL(md_done_sync);
5716 EXPORT_SYMBOL(md_write_start);
5717 EXPORT_SYMBOL(md_write_end);
5718 EXPORT_SYMBOL(md_register_thread);
5719 EXPORT_SYMBOL(md_unregister_thread);
5720 EXPORT_SYMBOL(md_wakeup_thread);
5721 EXPORT_SYMBOL(md_check_recovery);
5722 MODULE_LICENSE("GPL");
5723 MODULE_ALIAS("md");
5724 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);