md: allow reshape_position for md arrays to be set via sysfs
[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/kernel.h>
37 #include <linux/kthread.h>
38 #include <linux/linkage.h>
39 #include <linux/raid/md.h>
40 #include <linux/raid/bitmap.h>
41 #include <linux/sysctl.h>
42 #include <linux/buffer_head.h> /* for invalidate_bdev */
43 #include <linux/poll.h>
44 #include <linux/mutex.h>
45 #include <linux/ctype.h>
46 #include <linux/freezer.h>
47
48 #include <linux/init.h>
49
50 #include <linux/file.h>
51
52 #ifdef CONFIG_KMOD
53 #include <linux/kmod.h>
54 #endif
55
56 #include <asm/unaligned.h>
57
58 #define MAJOR_NR MD_MAJOR
59 #define MD_DRIVER
60
61 /* 63 partitions with the alternate major number (mdp) */
62 #define MdpMinorShift 6
63
64 #define DEBUG 0
65 #define dprintk(x...) ((void)(DEBUG && printk(x)))
66
67
68 #ifndef MODULE
69 static void autostart_arrays (int part);
70 #endif
71
72 static LIST_HEAD(pers_list);
73 static DEFINE_SPINLOCK(pers_lock);
74
75 static void md_print_devices(void);
76
77 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
78
79 /*
80  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
81  * is 1000 KB/sec, so the extra system load does not show up that much.
82  * Increase it if you want to have more _guaranteed_ speed. Note that
83  * the RAID driver will use the maximum available bandwidth if the IO
84  * subsystem is idle. There is also an 'absolute maximum' reconstruction
85  * speed limit - in case reconstruction slows down your system despite
86  * idle IO detection.
87  *
88  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
89  * or /sys/block/mdX/md/sync_speed_{min,max}
90  */
91
92 static int sysctl_speed_limit_min = 1000;
93 static int sysctl_speed_limit_max = 200000;
94 static inline int speed_min(mddev_t *mddev)
95 {
96         return mddev->sync_speed_min ?
97                 mddev->sync_speed_min : sysctl_speed_limit_min;
98 }
99
100 static inline int speed_max(mddev_t *mddev)
101 {
102         return mddev->sync_speed_max ?
103                 mddev->sync_speed_max : sysctl_speed_limit_max;
104 }
105
106 static struct ctl_table_header *raid_table_header;
107
108 static ctl_table raid_table[] = {
109         {
110                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
111                 .procname       = "speed_limit_min",
112                 .data           = &sysctl_speed_limit_min,
113                 .maxlen         = sizeof(int),
114                 .mode           = S_IRUGO|S_IWUSR,
115                 .proc_handler   = &proc_dointvec,
116         },
117         {
118                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
119                 .procname       = "speed_limit_max",
120                 .data           = &sysctl_speed_limit_max,
121                 .maxlen         = sizeof(int),
122                 .mode           = S_IRUGO|S_IWUSR,
123                 .proc_handler   = &proc_dointvec,
124         },
125         { .ctl_name = 0 }
126 };
127
128 static ctl_table raid_dir_table[] = {
129         {
130                 .ctl_name       = DEV_RAID,
131                 .procname       = "raid",
132                 .maxlen         = 0,
133                 .mode           = S_IRUGO|S_IXUGO,
134                 .child          = raid_table,
135         },
136         { .ctl_name = 0 }
137 };
138
139 static ctl_table raid_root_table[] = {
140         {
141                 .ctl_name       = CTL_DEV,
142                 .procname       = "dev",
143                 .maxlen         = 0,
144                 .mode           = 0555,
145                 .child          = raid_dir_table,
146         },
147         { .ctl_name = 0 }
148 };
149
150 static struct block_device_operations md_fops;
151
152 static int start_readonly;
153
154 /*
155  * We have a system wide 'event count' that is incremented
156  * on any 'interesting' event, and readers of /proc/mdstat
157  * can use 'poll' or 'select' to find out when the event
158  * count increases.
159  *
160  * Events are:
161  *  start array, stop array, error, add device, remove device,
162  *  start build, activate spare
163  */
164 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
165 static atomic_t md_event_count;
166 void md_new_event(mddev_t *mddev)
167 {
168         atomic_inc(&md_event_count);
169         wake_up(&md_event_waiters);
170         sysfs_notify(&mddev->kobj, NULL, "sync_action");
171 }
172 EXPORT_SYMBOL_GPL(md_new_event);
173
174 /* Alternate version that can be called from interrupts
175  * when calling sysfs_notify isn't needed.
176  */
177 static void md_new_event_inintr(mddev_t *mddev)
178 {
179         atomic_inc(&md_event_count);
180         wake_up(&md_event_waiters);
181 }
182
183 /*
184  * Enables to iterate over all existing md arrays
185  * all_mddevs_lock protects this list.
186  */
187 static LIST_HEAD(all_mddevs);
188 static DEFINE_SPINLOCK(all_mddevs_lock);
189
190
191 /*
192  * iterates through all used mddevs in the system.
193  * We take care to grab the all_mddevs_lock whenever navigating
194  * the list, and to always hold a refcount when unlocked.
195  * Any code which breaks out of this loop while own
196  * a reference to the current mddev and must mddev_put it.
197  */
198 #define ITERATE_MDDEV(mddev,tmp)                                        \
199                                                                         \
200         for (({ spin_lock(&all_mddevs_lock);                            \
201                 tmp = all_mddevs.next;                                  \
202                 mddev = NULL;});                                        \
203              ({ if (tmp != &all_mddevs)                                 \
204                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
205                 spin_unlock(&all_mddevs_lock);                          \
206                 if (mddev) mddev_put(mddev);                            \
207                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
208                 tmp != &all_mddevs;});                                  \
209              ({ spin_lock(&all_mddevs_lock);                            \
210                 tmp = tmp->next;})                                      \
211                 )
212
213
214 static int md_fail_request (request_queue_t *q, struct bio *bio)
215 {
216         bio_io_error(bio, bio->bi_size);
217         return 0;
218 }
219
220 static inline mddev_t *mddev_get(mddev_t *mddev)
221 {
222         atomic_inc(&mddev->active);
223         return mddev;
224 }
225
226 static void mddev_put(mddev_t *mddev)
227 {
228         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
229                 return;
230         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
231                 list_del(&mddev->all_mddevs);
232                 spin_unlock(&all_mddevs_lock);
233                 blk_cleanup_queue(mddev->queue);
234                 kobject_unregister(&mddev->kobj);
235         } else
236                 spin_unlock(&all_mddevs_lock);
237 }
238
239 static mddev_t * mddev_find(dev_t unit)
240 {
241         mddev_t *mddev, *new = NULL;
242
243  retry:
244         spin_lock(&all_mddevs_lock);
245         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
246                 if (mddev->unit == unit) {
247                         mddev_get(mddev);
248                         spin_unlock(&all_mddevs_lock);
249                         kfree(new);
250                         return mddev;
251                 }
252
253         if (new) {
254                 list_add(&new->all_mddevs, &all_mddevs);
255                 spin_unlock(&all_mddevs_lock);
256                 return new;
257         }
258         spin_unlock(&all_mddevs_lock);
259
260         new = kzalloc(sizeof(*new), GFP_KERNEL);
261         if (!new)
262                 return NULL;
263
264         new->unit = unit;
265         if (MAJOR(unit) == MD_MAJOR)
266                 new->md_minor = MINOR(unit);
267         else
268                 new->md_minor = MINOR(unit) >> MdpMinorShift;
269
270         mutex_init(&new->reconfig_mutex);
271         INIT_LIST_HEAD(&new->disks);
272         INIT_LIST_HEAD(&new->all_mddevs);
273         init_timer(&new->safemode_timer);
274         atomic_set(&new->active, 1);
275         spin_lock_init(&new->write_lock);
276         init_waitqueue_head(&new->sb_wait);
277         new->reshape_position = MaxSector;
278
279         new->queue = blk_alloc_queue(GFP_KERNEL);
280         if (!new->queue) {
281                 kfree(new);
282                 return NULL;
283         }
284         set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
285
286         blk_queue_make_request(new->queue, md_fail_request);
287
288         goto retry;
289 }
290
291 static inline int mddev_lock(mddev_t * mddev)
292 {
293         return mutex_lock_interruptible(&mddev->reconfig_mutex);
294 }
295
296 static inline int mddev_trylock(mddev_t * mddev)
297 {
298         return mutex_trylock(&mddev->reconfig_mutex);
299 }
300
301 static inline void mddev_unlock(mddev_t * mddev)
302 {
303         mutex_unlock(&mddev->reconfig_mutex);
304
305         md_wakeup_thread(mddev->thread);
306 }
307
308 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
309 {
310         mdk_rdev_t * rdev;
311         struct list_head *tmp;
312
313         ITERATE_RDEV(mddev,rdev,tmp) {
314                 if (rdev->desc_nr == nr)
315                         return rdev;
316         }
317         return NULL;
318 }
319
320 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
321 {
322         struct list_head *tmp;
323         mdk_rdev_t *rdev;
324
325         ITERATE_RDEV(mddev,rdev,tmp) {
326                 if (rdev->bdev->bd_dev == dev)
327                         return rdev;
328         }
329         return NULL;
330 }
331
332 static struct mdk_personality *find_pers(int level, char *clevel)
333 {
334         struct mdk_personality *pers;
335         list_for_each_entry(pers, &pers_list, list) {
336                 if (level != LEVEL_NONE && pers->level == level)
337                         return pers;
338                 if (strcmp(pers->name, clevel)==0)
339                         return pers;
340         }
341         return NULL;
342 }
343
344 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
345 {
346         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
347         return MD_NEW_SIZE_BLOCKS(size);
348 }
349
350 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
351 {
352         sector_t size;
353
354         size = rdev->sb_offset;
355
356         if (chunk_size)
357                 size &= ~((sector_t)chunk_size/1024 - 1);
358         return size;
359 }
360
361 static int alloc_disk_sb(mdk_rdev_t * rdev)
362 {
363         if (rdev->sb_page)
364                 MD_BUG();
365
366         rdev->sb_page = alloc_page(GFP_KERNEL);
367         if (!rdev->sb_page) {
368                 printk(KERN_ALERT "md: out of memory.\n");
369                 return -EINVAL;
370         }
371
372         return 0;
373 }
374
375 static void free_disk_sb(mdk_rdev_t * rdev)
376 {
377         if (rdev->sb_page) {
378                 put_page(rdev->sb_page);
379                 rdev->sb_loaded = 0;
380                 rdev->sb_page = NULL;
381                 rdev->sb_offset = 0;
382                 rdev->size = 0;
383         }
384 }
385
386
387 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
388 {
389         mdk_rdev_t *rdev = bio->bi_private;
390         mddev_t *mddev = rdev->mddev;
391         if (bio->bi_size)
392                 return 1;
393
394         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
395                 printk("md: super_written gets error=%d, uptodate=%d\n",
396                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
397                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
398                 md_error(mddev, rdev);
399         }
400
401         if (atomic_dec_and_test(&mddev->pending_writes))
402                 wake_up(&mddev->sb_wait);
403         bio_put(bio);
404         return 0;
405 }
406
407 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
408 {
409         struct bio *bio2 = bio->bi_private;
410         mdk_rdev_t *rdev = bio2->bi_private;
411         mddev_t *mddev = rdev->mddev;
412         if (bio->bi_size)
413                 return 1;
414
415         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
416             error == -EOPNOTSUPP) {
417                 unsigned long flags;
418                 /* barriers don't appear to be supported :-( */
419                 set_bit(BarriersNotsupp, &rdev->flags);
420                 mddev->barriers_work = 0;
421                 spin_lock_irqsave(&mddev->write_lock, flags);
422                 bio2->bi_next = mddev->biolist;
423                 mddev->biolist = bio2;
424                 spin_unlock_irqrestore(&mddev->write_lock, flags);
425                 wake_up(&mddev->sb_wait);
426                 bio_put(bio);
427                 return 0;
428         }
429         bio_put(bio2);
430         bio->bi_private = rdev;
431         return super_written(bio, bytes_done, error);
432 }
433
434 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
435                    sector_t sector, int size, struct page *page)
436 {
437         /* write first size bytes of page to sector of rdev
438          * Increment mddev->pending_writes before returning
439          * and decrement it on completion, waking up sb_wait
440          * if zero is reached.
441          * If an error occurred, call md_error
442          *
443          * As we might need to resubmit the request if BIO_RW_BARRIER
444          * causes ENOTSUPP, we allocate a spare bio...
445          */
446         struct bio *bio = bio_alloc(GFP_NOIO, 1);
447         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
448
449         bio->bi_bdev = rdev->bdev;
450         bio->bi_sector = sector;
451         bio_add_page(bio, page, size, 0);
452         bio->bi_private = rdev;
453         bio->bi_end_io = super_written;
454         bio->bi_rw = rw;
455
456         atomic_inc(&mddev->pending_writes);
457         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
458                 struct bio *rbio;
459                 rw |= (1<<BIO_RW_BARRIER);
460                 rbio = bio_clone(bio, GFP_NOIO);
461                 rbio->bi_private = bio;
462                 rbio->bi_end_io = super_written_barrier;
463                 submit_bio(rw, rbio);
464         } else
465                 submit_bio(rw, bio);
466 }
467
468 void md_super_wait(mddev_t *mddev)
469 {
470         /* wait for all superblock writes that were scheduled to complete.
471          * if any had to be retried (due to BARRIER problems), retry them
472          */
473         DEFINE_WAIT(wq);
474         for(;;) {
475                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
476                 if (atomic_read(&mddev->pending_writes)==0)
477                         break;
478                 while (mddev->biolist) {
479                         struct bio *bio;
480                         spin_lock_irq(&mddev->write_lock);
481                         bio = mddev->biolist;
482                         mddev->biolist = bio->bi_next ;
483                         bio->bi_next = NULL;
484                         spin_unlock_irq(&mddev->write_lock);
485                         submit_bio(bio->bi_rw, bio);
486                 }
487                 schedule();
488         }
489         finish_wait(&mddev->sb_wait, &wq);
490 }
491
492 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
493 {
494         if (bio->bi_size)
495                 return 1;
496
497         complete((struct completion*)bio->bi_private);
498         return 0;
499 }
500
501 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
502                    struct page *page, int rw)
503 {
504         struct bio *bio = bio_alloc(GFP_NOIO, 1);
505         struct completion event;
506         int ret;
507
508         rw |= (1 << BIO_RW_SYNC);
509
510         bio->bi_bdev = bdev;
511         bio->bi_sector = sector;
512         bio_add_page(bio, page, size, 0);
513         init_completion(&event);
514         bio->bi_private = &event;
515         bio->bi_end_io = bi_complete;
516         submit_bio(rw, bio);
517         wait_for_completion(&event);
518
519         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
520         bio_put(bio);
521         return ret;
522 }
523 EXPORT_SYMBOL_GPL(sync_page_io);
524
525 static int read_disk_sb(mdk_rdev_t * rdev, int size)
526 {
527         char b[BDEVNAME_SIZE];
528         if (!rdev->sb_page) {
529                 MD_BUG();
530                 return -EINVAL;
531         }
532         if (rdev->sb_loaded)
533                 return 0;
534
535
536         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
537                 goto fail;
538         rdev->sb_loaded = 1;
539         return 0;
540
541 fail:
542         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
543                 bdevname(rdev->bdev,b));
544         return -EINVAL;
545 }
546
547 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
548 {
549         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
550                 (sb1->set_uuid1 == sb2->set_uuid1) &&
551                 (sb1->set_uuid2 == sb2->set_uuid2) &&
552                 (sb1->set_uuid3 == sb2->set_uuid3))
553
554                 return 1;
555
556         return 0;
557 }
558
559
560 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
561 {
562         int ret;
563         mdp_super_t *tmp1, *tmp2;
564
565         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
566         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
567
568         if (!tmp1 || !tmp2) {
569                 ret = 0;
570                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
571                 goto abort;
572         }
573
574         *tmp1 = *sb1;
575         *tmp2 = *sb2;
576
577         /*
578          * nr_disks is not constant
579          */
580         tmp1->nr_disks = 0;
581         tmp2->nr_disks = 0;
582
583         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
584                 ret = 0;
585         else
586                 ret = 1;
587
588 abort:
589         kfree(tmp1);
590         kfree(tmp2);
591         return ret;
592 }
593
594
595 static u32 md_csum_fold(u32 csum)
596 {
597         csum = (csum & 0xffff) + (csum >> 16);
598         return (csum & 0xffff) + (csum >> 16);
599 }
600
601 static unsigned int calc_sb_csum(mdp_super_t * sb)
602 {
603         u64 newcsum = 0;
604         u32 *sb32 = (u32*)sb;
605         int i;
606         unsigned int disk_csum, csum;
607
608         disk_csum = sb->sb_csum;
609         sb->sb_csum = 0;
610
611         for (i = 0; i < MD_SB_BYTES/4 ; i++)
612                 newcsum += sb32[i];
613         csum = (newcsum & 0xffffffff) + (newcsum>>32);
614
615
616 #ifdef CONFIG_ALPHA
617         /* This used to use csum_partial, which was wrong for several
618          * reasons including that different results are returned on
619          * different architectures.  It isn't critical that we get exactly
620          * the same return value as before (we always csum_fold before
621          * testing, and that removes any differences).  However as we
622          * know that csum_partial always returned a 16bit value on
623          * alphas, do a fold to maximise conformity to previous behaviour.
624          */
625         sb->sb_csum = md_csum_fold(disk_csum);
626 #else
627         sb->sb_csum = disk_csum;
628 #endif
629         return csum;
630 }
631
632
633 /*
634  * Handle superblock details.
635  * We want to be able to handle multiple superblock formats
636  * so we have a common interface to them all, and an array of
637  * different handlers.
638  * We rely on user-space to write the initial superblock, and support
639  * reading and updating of superblocks.
640  * Interface methods are:
641  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
642  *      loads and validates a superblock on dev.
643  *      if refdev != NULL, compare superblocks on both devices
644  *    Return:
645  *      0 - dev has a superblock that is compatible with refdev
646  *      1 - dev has a superblock that is compatible and newer than refdev
647  *          so dev should be used as the refdev in future
648  *     -EINVAL superblock incompatible or invalid
649  *     -othererror e.g. -EIO
650  *
651  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
652  *      Verify that dev is acceptable into mddev.
653  *       The first time, mddev->raid_disks will be 0, and data from
654  *       dev should be merged in.  Subsequent calls check that dev
655  *       is new enough.  Return 0 or -EINVAL
656  *
657  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
658  *     Update the superblock for rdev with data in mddev
659  *     This does not write to disc.
660  *
661  */
662
663 struct super_type  {
664         char            *name;
665         struct module   *owner;
666         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
667         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
668         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
669 };
670
671 /*
672  * load_super for 0.90.0 
673  */
674 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
675 {
676         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
677         mdp_super_t *sb;
678         int ret;
679         sector_t sb_offset;
680
681         /*
682          * Calculate the position of the superblock,
683          * it's at the end of the disk.
684          *
685          * It also happens to be a multiple of 4Kb.
686          */
687         sb_offset = calc_dev_sboffset(rdev->bdev);
688         rdev->sb_offset = sb_offset;
689
690         ret = read_disk_sb(rdev, MD_SB_BYTES);
691         if (ret) return ret;
692
693         ret = -EINVAL;
694
695         bdevname(rdev->bdev, b);
696         sb = (mdp_super_t*)page_address(rdev->sb_page);
697
698         if (sb->md_magic != MD_SB_MAGIC) {
699                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
700                        b);
701                 goto abort;
702         }
703
704         if (sb->major_version != 0 ||
705             sb->minor_version < 90 ||
706             sb->minor_version > 91) {
707                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
708                         sb->major_version, sb->minor_version,
709                         b);
710                 goto abort;
711         }
712
713         if (sb->raid_disks <= 0)
714                 goto abort;
715
716         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
717                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
718                         b);
719                 goto abort;
720         }
721
722         rdev->preferred_minor = sb->md_minor;
723         rdev->data_offset = 0;
724         rdev->sb_size = MD_SB_BYTES;
725
726         if (sb->state & (1<<MD_SB_BITMAP_PRESENT)) {
727                 if (sb->level != 1 && sb->level != 4
728                     && sb->level != 5 && sb->level != 6
729                     && sb->level != 10) {
730                         /* FIXME use a better test */
731                         printk(KERN_WARNING
732                                "md: bitmaps not supported for this level.\n");
733                         goto abort;
734                 }
735         }
736
737         if (sb->level == LEVEL_MULTIPATH)
738                 rdev->desc_nr = -1;
739         else
740                 rdev->desc_nr = sb->this_disk.number;
741
742         if (refdev == 0)
743                 ret = 1;
744         else {
745                 __u64 ev1, ev2;
746                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
747                 if (!uuid_equal(refsb, sb)) {
748                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
749                                 b, bdevname(refdev->bdev,b2));
750                         goto abort;
751                 }
752                 if (!sb_equal(refsb, sb)) {
753                         printk(KERN_WARNING "md: %s has same UUID"
754                                " but different superblock to %s\n",
755                                b, bdevname(refdev->bdev, b2));
756                         goto abort;
757                 }
758                 ev1 = md_event(sb);
759                 ev2 = md_event(refsb);
760                 if (ev1 > ev2)
761                         ret = 1;
762                 else 
763                         ret = 0;
764         }
765         rdev->size = calc_dev_size(rdev, sb->chunk_size);
766
767         if (rdev->size < sb->size && sb->level > 1)
768                 /* "this cannot possibly happen" ... */
769                 ret = -EINVAL;
770
771  abort:
772         return ret;
773 }
774
775 /*
776  * validate_super for 0.90.0
777  */
778 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
779 {
780         mdp_disk_t *desc;
781         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
782         __u64 ev1 = md_event(sb);
783
784         rdev->raid_disk = -1;
785         rdev->flags = 0;
786         if (mddev->raid_disks == 0) {
787                 mddev->major_version = 0;
788                 mddev->minor_version = sb->minor_version;
789                 mddev->patch_version = sb->patch_version;
790                 mddev->persistent = ! sb->not_persistent;
791                 mddev->chunk_size = sb->chunk_size;
792                 mddev->ctime = sb->ctime;
793                 mddev->utime = sb->utime;
794                 mddev->level = sb->level;
795                 mddev->clevel[0] = 0;
796                 mddev->layout = sb->layout;
797                 mddev->raid_disks = sb->raid_disks;
798                 mddev->size = sb->size;
799                 mddev->events = ev1;
800                 mddev->bitmap_offset = 0;
801                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
802
803                 if (mddev->minor_version >= 91) {
804                         mddev->reshape_position = sb->reshape_position;
805                         mddev->delta_disks = sb->delta_disks;
806                         mddev->new_level = sb->new_level;
807                         mddev->new_layout = sb->new_layout;
808                         mddev->new_chunk = sb->new_chunk;
809                 } else {
810                         mddev->reshape_position = MaxSector;
811                         mddev->delta_disks = 0;
812                         mddev->new_level = mddev->level;
813                         mddev->new_layout = mddev->layout;
814                         mddev->new_chunk = mddev->chunk_size;
815                 }
816
817                 if (sb->state & (1<<MD_SB_CLEAN))
818                         mddev->recovery_cp = MaxSector;
819                 else {
820                         if (sb->events_hi == sb->cp_events_hi && 
821                                 sb->events_lo == sb->cp_events_lo) {
822                                 mddev->recovery_cp = sb->recovery_cp;
823                         } else
824                                 mddev->recovery_cp = 0;
825                 }
826
827                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
828                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
829                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
830                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
831
832                 mddev->max_disks = MD_SB_DISKS;
833
834                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
835                     mddev->bitmap_file == NULL)
836                         mddev->bitmap_offset = mddev->default_bitmap_offset;
837
838         } else if (mddev->pers == NULL) {
839                 /* Insist on good event counter while assembling */
840                 ++ev1;
841                 if (ev1 < mddev->events) 
842                         return -EINVAL;
843         } else if (mddev->bitmap) {
844                 /* if adding to array with a bitmap, then we can accept an
845                  * older device ... but not too old.
846                  */
847                 if (ev1 < mddev->bitmap->events_cleared)
848                         return 0;
849         } else {
850                 if (ev1 < mddev->events)
851                         /* just a hot-add of a new device, leave raid_disk at -1 */
852                         return 0;
853         }
854
855         if (mddev->level != LEVEL_MULTIPATH) {
856                 desc = sb->disks + rdev->desc_nr;
857
858                 if (desc->state & (1<<MD_DISK_FAULTY))
859                         set_bit(Faulty, &rdev->flags);
860                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
861                             desc->raid_disk < mddev->raid_disks */) {
862                         set_bit(In_sync, &rdev->flags);
863                         rdev->raid_disk = desc->raid_disk;
864                 }
865                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
866                         set_bit(WriteMostly, &rdev->flags);
867         } else /* MULTIPATH are always insync */
868                 set_bit(In_sync, &rdev->flags);
869         return 0;
870 }
871
872 /*
873  * sync_super for 0.90.0
874  */
875 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
876 {
877         mdp_super_t *sb;
878         struct list_head *tmp;
879         mdk_rdev_t *rdev2;
880         int next_spare = mddev->raid_disks;
881
882
883         /* make rdev->sb match mddev data..
884          *
885          * 1/ zero out disks
886          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
887          * 3/ any empty disks < next_spare become removed
888          *
889          * disks[0] gets initialised to REMOVED because
890          * we cannot be sure from other fields if it has
891          * been initialised or not.
892          */
893         int i;
894         int active=0, working=0,failed=0,spare=0,nr_disks=0;
895
896         rdev->sb_size = MD_SB_BYTES;
897
898         sb = (mdp_super_t*)page_address(rdev->sb_page);
899
900         memset(sb, 0, sizeof(*sb));
901
902         sb->md_magic = MD_SB_MAGIC;
903         sb->major_version = mddev->major_version;
904         sb->patch_version = mddev->patch_version;
905         sb->gvalid_words  = 0; /* ignored */
906         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
907         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
908         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
909         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
910
911         sb->ctime = mddev->ctime;
912         sb->level = mddev->level;
913         sb->size  = mddev->size;
914         sb->raid_disks = mddev->raid_disks;
915         sb->md_minor = mddev->md_minor;
916         sb->not_persistent = !mddev->persistent;
917         sb->utime = mddev->utime;
918         sb->state = 0;
919         sb->events_hi = (mddev->events>>32);
920         sb->events_lo = (u32)mddev->events;
921
922         if (mddev->reshape_position == MaxSector)
923                 sb->minor_version = 90;
924         else {
925                 sb->minor_version = 91;
926                 sb->reshape_position = mddev->reshape_position;
927                 sb->new_level = mddev->new_level;
928                 sb->delta_disks = mddev->delta_disks;
929                 sb->new_layout = mddev->new_layout;
930                 sb->new_chunk = mddev->new_chunk;
931         }
932         mddev->minor_version = sb->minor_version;
933         if (mddev->in_sync)
934         {
935                 sb->recovery_cp = mddev->recovery_cp;
936                 sb->cp_events_hi = (mddev->events>>32);
937                 sb->cp_events_lo = (u32)mddev->events;
938                 if (mddev->recovery_cp == MaxSector)
939                         sb->state = (1<< MD_SB_CLEAN);
940         } else
941                 sb->recovery_cp = 0;
942
943         sb->layout = mddev->layout;
944         sb->chunk_size = mddev->chunk_size;
945
946         if (mddev->bitmap && mddev->bitmap_file == NULL)
947                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
948
949         sb->disks[0].state = (1<<MD_DISK_REMOVED);
950         ITERATE_RDEV(mddev,rdev2,tmp) {
951                 mdp_disk_t *d;
952                 int desc_nr;
953                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
954                     && !test_bit(Faulty, &rdev2->flags))
955                         desc_nr = rdev2->raid_disk;
956                 else
957                         desc_nr = next_spare++;
958                 rdev2->desc_nr = desc_nr;
959                 d = &sb->disks[rdev2->desc_nr];
960                 nr_disks++;
961                 d->number = rdev2->desc_nr;
962                 d->major = MAJOR(rdev2->bdev->bd_dev);
963                 d->minor = MINOR(rdev2->bdev->bd_dev);
964                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
965                     && !test_bit(Faulty, &rdev2->flags))
966                         d->raid_disk = rdev2->raid_disk;
967                 else
968                         d->raid_disk = rdev2->desc_nr; /* compatibility */
969                 if (test_bit(Faulty, &rdev2->flags))
970                         d->state = (1<<MD_DISK_FAULTY);
971                 else if (test_bit(In_sync, &rdev2->flags)) {
972                         d->state = (1<<MD_DISK_ACTIVE);
973                         d->state |= (1<<MD_DISK_SYNC);
974                         active++;
975                         working++;
976                 } else {
977                         d->state = 0;
978                         spare++;
979                         working++;
980                 }
981                 if (test_bit(WriteMostly, &rdev2->flags))
982                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
983         }
984         /* now set the "removed" and "faulty" bits on any missing devices */
985         for (i=0 ; i < mddev->raid_disks ; i++) {
986                 mdp_disk_t *d = &sb->disks[i];
987                 if (d->state == 0 && d->number == 0) {
988                         d->number = i;
989                         d->raid_disk = i;
990                         d->state = (1<<MD_DISK_REMOVED);
991                         d->state |= (1<<MD_DISK_FAULTY);
992                         failed++;
993                 }
994         }
995         sb->nr_disks = nr_disks;
996         sb->active_disks = active;
997         sb->working_disks = working;
998         sb->failed_disks = failed;
999         sb->spare_disks = spare;
1000
1001         sb->this_disk = sb->disks[rdev->desc_nr];
1002         sb->sb_csum = calc_sb_csum(sb);
1003 }
1004
1005 /*
1006  * version 1 superblock
1007  */
1008
1009 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1010 {
1011         __le32 disk_csum;
1012         u32 csum;
1013         unsigned long long newcsum;
1014         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1015         __le32 *isuper = (__le32*)sb;
1016         int i;
1017
1018         disk_csum = sb->sb_csum;
1019         sb->sb_csum = 0;
1020         newcsum = 0;
1021         for (i=0; size>=4; size -= 4 )
1022                 newcsum += le32_to_cpu(*isuper++);
1023
1024         if (size == 2)
1025                 newcsum += le16_to_cpu(*(__le16*) isuper);
1026
1027         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1028         sb->sb_csum = disk_csum;
1029         return cpu_to_le32(csum);
1030 }
1031
1032 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1033 {
1034         struct mdp_superblock_1 *sb;
1035         int ret;
1036         sector_t sb_offset;
1037         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1038         int bmask;
1039
1040         /*
1041          * Calculate the position of the superblock.
1042          * It is always aligned to a 4K boundary and
1043          * depeding on minor_version, it can be:
1044          * 0: At least 8K, but less than 12K, from end of device
1045          * 1: At start of device
1046          * 2: 4K from start of device.
1047          */
1048         switch(minor_version) {
1049         case 0:
1050                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1051                 sb_offset -= 8*2;
1052                 sb_offset &= ~(sector_t)(4*2-1);
1053                 /* convert from sectors to K */
1054                 sb_offset /= 2;
1055                 break;
1056         case 1:
1057                 sb_offset = 0;
1058                 break;
1059         case 2:
1060                 sb_offset = 4;
1061                 break;
1062         default:
1063                 return -EINVAL;
1064         }
1065         rdev->sb_offset = sb_offset;
1066
1067         /* superblock is rarely larger than 1K, but it can be larger,
1068          * and it is safe to read 4k, so we do that
1069          */
1070         ret = read_disk_sb(rdev, 4096);
1071         if (ret) return ret;
1072
1073
1074         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1075
1076         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1077             sb->major_version != cpu_to_le32(1) ||
1078             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1079             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1080             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1081                 return -EINVAL;
1082
1083         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1084                 printk("md: invalid superblock checksum on %s\n",
1085                         bdevname(rdev->bdev,b));
1086                 return -EINVAL;
1087         }
1088         if (le64_to_cpu(sb->data_size) < 10) {
1089                 printk("md: data_size too small on %s\n",
1090                        bdevname(rdev->bdev,b));
1091                 return -EINVAL;
1092         }
1093         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET)) {
1094                 if (sb->level != cpu_to_le32(1) &&
1095                     sb->level != cpu_to_le32(4) &&
1096                     sb->level != cpu_to_le32(5) &&
1097                     sb->level != cpu_to_le32(6) &&
1098                     sb->level != cpu_to_le32(10)) {
1099                         printk(KERN_WARNING
1100                                "md: bitmaps not supported for this level.\n");
1101                         return -EINVAL;
1102                 }
1103         }
1104
1105         rdev->preferred_minor = 0xffff;
1106         rdev->data_offset = le64_to_cpu(sb->data_offset);
1107         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1108
1109         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1110         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1111         if (rdev->sb_size & bmask)
1112                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1113
1114         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1115                 rdev->desc_nr = -1;
1116         else
1117                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1118
1119         if (refdev == 0)
1120                 ret = 1;
1121         else {
1122                 __u64 ev1, ev2;
1123                 struct mdp_superblock_1 *refsb = 
1124                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1125
1126                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1127                     sb->level != refsb->level ||
1128                     sb->layout != refsb->layout ||
1129                     sb->chunksize != refsb->chunksize) {
1130                         printk(KERN_WARNING "md: %s has strangely different"
1131                                 " superblock to %s\n",
1132                                 bdevname(rdev->bdev,b),
1133                                 bdevname(refdev->bdev,b2));
1134                         return -EINVAL;
1135                 }
1136                 ev1 = le64_to_cpu(sb->events);
1137                 ev2 = le64_to_cpu(refsb->events);
1138
1139                 if (ev1 > ev2)
1140                         ret = 1;
1141                 else
1142                         ret = 0;
1143         }
1144         if (minor_version) 
1145                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1146         else
1147                 rdev->size = rdev->sb_offset;
1148         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1149                 return -EINVAL;
1150         rdev->size = le64_to_cpu(sb->data_size)/2;
1151         if (le32_to_cpu(sb->chunksize))
1152                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1153
1154         if (le64_to_cpu(sb->size) > rdev->size*2)
1155                 return -EINVAL;
1156         return ret;
1157 }
1158
1159 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1160 {
1161         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1162         __u64 ev1 = le64_to_cpu(sb->events);
1163
1164         rdev->raid_disk = -1;
1165         rdev->flags = 0;
1166         if (mddev->raid_disks == 0) {
1167                 mddev->major_version = 1;
1168                 mddev->patch_version = 0;
1169                 mddev->persistent = 1;
1170                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1171                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1172                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1173                 mddev->level = le32_to_cpu(sb->level);
1174                 mddev->clevel[0] = 0;
1175                 mddev->layout = le32_to_cpu(sb->layout);
1176                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1177                 mddev->size = le64_to_cpu(sb->size)/2;
1178                 mddev->events = ev1;
1179                 mddev->bitmap_offset = 0;
1180                 mddev->default_bitmap_offset = 1024 >> 9;
1181                 
1182                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1183                 memcpy(mddev->uuid, sb->set_uuid, 16);
1184
1185                 mddev->max_disks =  (4096-256)/2;
1186
1187                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1188                     mddev->bitmap_file == NULL )
1189                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1190
1191                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1192                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1193                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1194                         mddev->new_level = le32_to_cpu(sb->new_level);
1195                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1196                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1197                 } else {
1198                         mddev->reshape_position = MaxSector;
1199                         mddev->delta_disks = 0;
1200                         mddev->new_level = mddev->level;
1201                         mddev->new_layout = mddev->layout;
1202                         mddev->new_chunk = mddev->chunk_size;
1203                 }
1204
1205         } else if (mddev->pers == NULL) {
1206                 /* Insist of good event counter while assembling */
1207                 ++ev1;
1208                 if (ev1 < mddev->events)
1209                         return -EINVAL;
1210         } else if (mddev->bitmap) {
1211                 /* If adding to array with a bitmap, then we can accept an
1212                  * older device, but not too old.
1213                  */
1214                 if (ev1 < mddev->bitmap->events_cleared)
1215                         return 0;
1216         } else {
1217                 if (ev1 < mddev->events)
1218                         /* just a hot-add of a new device, leave raid_disk at -1 */
1219                         return 0;
1220         }
1221         if (mddev->level != LEVEL_MULTIPATH) {
1222                 int role;
1223                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1224                 switch(role) {
1225                 case 0xffff: /* spare */
1226                         break;
1227                 case 0xfffe: /* faulty */
1228                         set_bit(Faulty, &rdev->flags);
1229                         break;
1230                 default:
1231                         if ((le32_to_cpu(sb->feature_map) &
1232                              MD_FEATURE_RECOVERY_OFFSET))
1233                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1234                         else
1235                                 set_bit(In_sync, &rdev->flags);
1236                         rdev->raid_disk = role;
1237                         break;
1238                 }
1239                 if (sb->devflags & WriteMostly1)
1240                         set_bit(WriteMostly, &rdev->flags);
1241         } else /* MULTIPATH are always insync */
1242                 set_bit(In_sync, &rdev->flags);
1243
1244         return 0;
1245 }
1246
1247 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1248 {
1249         struct mdp_superblock_1 *sb;
1250         struct list_head *tmp;
1251         mdk_rdev_t *rdev2;
1252         int max_dev, i;
1253         /* make rdev->sb match mddev and rdev data. */
1254
1255         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1256
1257         sb->feature_map = 0;
1258         sb->pad0 = 0;
1259         sb->recovery_offset = cpu_to_le64(0);
1260         memset(sb->pad1, 0, sizeof(sb->pad1));
1261         memset(sb->pad2, 0, sizeof(sb->pad2));
1262         memset(sb->pad3, 0, sizeof(sb->pad3));
1263
1264         sb->utime = cpu_to_le64((__u64)mddev->utime);
1265         sb->events = cpu_to_le64(mddev->events);
1266         if (mddev->in_sync)
1267                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1268         else
1269                 sb->resync_offset = cpu_to_le64(0);
1270
1271         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1272
1273         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1274         sb->size = cpu_to_le64(mddev->size<<1);
1275
1276         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1277                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1278                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1279         }
1280
1281         if (rdev->raid_disk >= 0 &&
1282             !test_bit(In_sync, &rdev->flags) &&
1283             rdev->recovery_offset > 0) {
1284                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1285                 sb->recovery_offset = cpu_to_le64(rdev->recovery_offset);
1286         }
1287
1288         if (mddev->reshape_position != MaxSector) {
1289                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1290                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1291                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1292                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1293                 sb->new_level = cpu_to_le32(mddev->new_level);
1294                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1295         }
1296
1297         max_dev = 0;
1298         ITERATE_RDEV(mddev,rdev2,tmp)
1299                 if (rdev2->desc_nr+1 > max_dev)
1300                         max_dev = rdev2->desc_nr+1;
1301         
1302         sb->max_dev = cpu_to_le32(max_dev);
1303         for (i=0; i<max_dev;i++)
1304                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1305         
1306         ITERATE_RDEV(mddev,rdev2,tmp) {
1307                 i = rdev2->desc_nr;
1308                 if (test_bit(Faulty, &rdev2->flags))
1309                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1310                 else if (test_bit(In_sync, &rdev2->flags))
1311                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1312                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1313                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1314                 else
1315                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1316         }
1317
1318         sb->sb_csum = calc_sb_1_csum(sb);
1319 }
1320
1321
1322 static struct super_type super_types[] = {
1323         [0] = {
1324                 .name   = "0.90.0",
1325                 .owner  = THIS_MODULE,
1326                 .load_super     = super_90_load,
1327                 .validate_super = super_90_validate,
1328                 .sync_super     = super_90_sync,
1329         },
1330         [1] = {
1331                 .name   = "md-1",
1332                 .owner  = THIS_MODULE,
1333                 .load_super     = super_1_load,
1334                 .validate_super = super_1_validate,
1335                 .sync_super     = super_1_sync,
1336         },
1337 };
1338
1339 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1340 {
1341         struct list_head *tmp, *tmp2;
1342         mdk_rdev_t *rdev, *rdev2;
1343
1344         ITERATE_RDEV(mddev1,rdev,tmp)
1345                 ITERATE_RDEV(mddev2, rdev2, tmp2)
1346                         if (rdev->bdev->bd_contains ==
1347                             rdev2->bdev->bd_contains)
1348                                 return 1;
1349
1350         return 0;
1351 }
1352
1353 static LIST_HEAD(pending_raid_disks);
1354
1355 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1356 {
1357         char b[BDEVNAME_SIZE];
1358         struct kobject *ko;
1359         char *s;
1360         int err;
1361
1362         if (rdev->mddev) {
1363                 MD_BUG();
1364                 return -EINVAL;
1365         }
1366         /* make sure rdev->size exceeds mddev->size */
1367         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1368                 if (mddev->pers)
1369                         /* Cannot change size, so fail */
1370                         return -ENOSPC;
1371                 else
1372                         mddev->size = rdev->size;
1373         }
1374
1375         /* Verify rdev->desc_nr is unique.
1376          * If it is -1, assign a free number, else
1377          * check number is not in use
1378          */
1379         if (rdev->desc_nr < 0) {
1380                 int choice = 0;
1381                 if (mddev->pers) choice = mddev->raid_disks;
1382                 while (find_rdev_nr(mddev, choice))
1383                         choice++;
1384                 rdev->desc_nr = choice;
1385         } else {
1386                 if (find_rdev_nr(mddev, rdev->desc_nr))
1387                         return -EBUSY;
1388         }
1389         bdevname(rdev->bdev,b);
1390         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1391                 return -ENOMEM;
1392         while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1393                 *s = '!';
1394                         
1395         rdev->mddev = mddev;
1396         printk(KERN_INFO "md: bind<%s>\n", b);
1397
1398         rdev->kobj.parent = &mddev->kobj;
1399         if ((err = kobject_add(&rdev->kobj)))
1400                 goto fail;
1401
1402         if (rdev->bdev->bd_part)
1403                 ko = &rdev->bdev->bd_part->kobj;
1404         else
1405                 ko = &rdev->bdev->bd_disk->kobj;
1406         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1407                 kobject_del(&rdev->kobj);
1408                 goto fail;
1409         }
1410         list_add(&rdev->same_set, &mddev->disks);
1411         bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1412         return 0;
1413
1414  fail:
1415         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1416                b, mdname(mddev));
1417         return err;
1418 }
1419
1420 static void delayed_delete(struct work_struct *ws)
1421 {
1422         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1423         kobject_del(&rdev->kobj);
1424 }
1425
1426 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1427 {
1428         char b[BDEVNAME_SIZE];
1429         if (!rdev->mddev) {
1430                 MD_BUG();
1431                 return;
1432         }
1433         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1434         list_del_init(&rdev->same_set);
1435         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1436         rdev->mddev = NULL;
1437         sysfs_remove_link(&rdev->kobj, "block");
1438
1439         /* We need to delay this, otherwise we can deadlock when
1440          * writing to 'remove' to "dev/state"
1441          */
1442         INIT_WORK(&rdev->del_work, delayed_delete);
1443         schedule_work(&rdev->del_work);
1444 }
1445
1446 /*
1447  * prevent the device from being mounted, repartitioned or
1448  * otherwise reused by a RAID array (or any other kernel
1449  * subsystem), by bd_claiming the device.
1450  */
1451 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1452 {
1453         int err = 0;
1454         struct block_device *bdev;
1455         char b[BDEVNAME_SIZE];
1456
1457         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1458         if (IS_ERR(bdev)) {
1459                 printk(KERN_ERR "md: could not open %s.\n",
1460                         __bdevname(dev, b));
1461                 return PTR_ERR(bdev);
1462         }
1463         err = bd_claim(bdev, rdev);
1464         if (err) {
1465                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1466                         bdevname(bdev, b));
1467                 blkdev_put(bdev);
1468                 return err;
1469         }
1470         rdev->bdev = bdev;
1471         return err;
1472 }
1473
1474 static void unlock_rdev(mdk_rdev_t *rdev)
1475 {
1476         struct block_device *bdev = rdev->bdev;
1477         rdev->bdev = NULL;
1478         if (!bdev)
1479                 MD_BUG();
1480         bd_release(bdev);
1481         blkdev_put(bdev);
1482 }
1483
1484 void md_autodetect_dev(dev_t dev);
1485
1486 static void export_rdev(mdk_rdev_t * rdev)
1487 {
1488         char b[BDEVNAME_SIZE];
1489         printk(KERN_INFO "md: export_rdev(%s)\n",
1490                 bdevname(rdev->bdev,b));
1491         if (rdev->mddev)
1492                 MD_BUG();
1493         free_disk_sb(rdev);
1494         list_del_init(&rdev->same_set);
1495 #ifndef MODULE
1496         md_autodetect_dev(rdev->bdev->bd_dev);
1497 #endif
1498         unlock_rdev(rdev);
1499         kobject_put(&rdev->kobj);
1500 }
1501
1502 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1503 {
1504         unbind_rdev_from_array(rdev);
1505         export_rdev(rdev);
1506 }
1507
1508 static void export_array(mddev_t *mddev)
1509 {
1510         struct list_head *tmp;
1511         mdk_rdev_t *rdev;
1512
1513         ITERATE_RDEV(mddev,rdev,tmp) {
1514                 if (!rdev->mddev) {
1515                         MD_BUG();
1516                         continue;
1517                 }
1518                 kick_rdev_from_array(rdev);
1519         }
1520         if (!list_empty(&mddev->disks))
1521                 MD_BUG();
1522         mddev->raid_disks = 0;
1523         mddev->major_version = 0;
1524 }
1525
1526 static void print_desc(mdp_disk_t *desc)
1527 {
1528         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1529                 desc->major,desc->minor,desc->raid_disk,desc->state);
1530 }
1531
1532 static void print_sb(mdp_super_t *sb)
1533 {
1534         int i;
1535
1536         printk(KERN_INFO 
1537                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1538                 sb->major_version, sb->minor_version, sb->patch_version,
1539                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1540                 sb->ctime);
1541         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1542                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1543                 sb->md_minor, sb->layout, sb->chunk_size);
1544         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1545                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1546                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1547                 sb->failed_disks, sb->spare_disks,
1548                 sb->sb_csum, (unsigned long)sb->events_lo);
1549
1550         printk(KERN_INFO);
1551         for (i = 0; i < MD_SB_DISKS; i++) {
1552                 mdp_disk_t *desc;
1553
1554                 desc = sb->disks + i;
1555                 if (desc->number || desc->major || desc->minor ||
1556                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1557                         printk("     D %2d: ", i);
1558                         print_desc(desc);
1559                 }
1560         }
1561         printk(KERN_INFO "md:     THIS: ");
1562         print_desc(&sb->this_disk);
1563
1564 }
1565
1566 static void print_rdev(mdk_rdev_t *rdev)
1567 {
1568         char b[BDEVNAME_SIZE];
1569         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1570                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1571                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1572                 rdev->desc_nr);
1573         if (rdev->sb_loaded) {
1574                 printk(KERN_INFO "md: rdev superblock:\n");
1575                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1576         } else
1577                 printk(KERN_INFO "md: no rdev superblock!\n");
1578 }
1579
1580 static void md_print_devices(void)
1581 {
1582         struct list_head *tmp, *tmp2;
1583         mdk_rdev_t *rdev;
1584         mddev_t *mddev;
1585         char b[BDEVNAME_SIZE];
1586
1587         printk("\n");
1588         printk("md:     **********************************\n");
1589         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1590         printk("md:     **********************************\n");
1591         ITERATE_MDDEV(mddev,tmp) {
1592
1593                 if (mddev->bitmap)
1594                         bitmap_print_sb(mddev->bitmap);
1595                 else
1596                         printk("%s: ", mdname(mddev));
1597                 ITERATE_RDEV(mddev,rdev,tmp2)
1598                         printk("<%s>", bdevname(rdev->bdev,b));
1599                 printk("\n");
1600
1601                 ITERATE_RDEV(mddev,rdev,tmp2)
1602                         print_rdev(rdev);
1603         }
1604         printk("md:     **********************************\n");
1605         printk("\n");
1606 }
1607
1608
1609 static void sync_sbs(mddev_t * mddev, int nospares)
1610 {
1611         /* Update each superblock (in-memory image), but
1612          * if we are allowed to, skip spares which already
1613          * have the right event counter, or have one earlier
1614          * (which would mean they aren't being marked as dirty
1615          * with the rest of the array)
1616          */
1617         mdk_rdev_t *rdev;
1618         struct list_head *tmp;
1619
1620         ITERATE_RDEV(mddev,rdev,tmp) {
1621                 if (rdev->sb_events == mddev->events ||
1622                     (nospares &&
1623                      rdev->raid_disk < 0 &&
1624                      (rdev->sb_events&1)==0 &&
1625                      rdev->sb_events+1 == mddev->events)) {
1626                         /* Don't update this superblock */
1627                         rdev->sb_loaded = 2;
1628                 } else {
1629                         super_types[mddev->major_version].
1630                                 sync_super(mddev, rdev);
1631                         rdev->sb_loaded = 1;
1632                 }
1633         }
1634 }
1635
1636 static void md_update_sb(mddev_t * mddev, int force_change)
1637 {
1638         int err;
1639         struct list_head *tmp;
1640         mdk_rdev_t *rdev;
1641         int sync_req;
1642         int nospares = 0;
1643
1644 repeat:
1645         spin_lock_irq(&mddev->write_lock);
1646
1647         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1648         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1649                 force_change = 1;
1650         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1651                 /* just a clean<-> dirty transition, possibly leave spares alone,
1652                  * though if events isn't the right even/odd, we will have to do
1653                  * spares after all
1654                  */
1655                 nospares = 1;
1656         if (force_change)
1657                 nospares = 0;
1658         if (mddev->degraded)
1659                 /* If the array is degraded, then skipping spares is both
1660                  * dangerous and fairly pointless.
1661                  * Dangerous because a device that was removed from the array
1662                  * might have a event_count that still looks up-to-date,
1663                  * so it can be re-added without a resync.
1664                  * Pointless because if there are any spares to skip,
1665                  * then a recovery will happen and soon that array won't
1666                  * be degraded any more and the spare can go back to sleep then.
1667                  */
1668                 nospares = 0;
1669
1670         sync_req = mddev->in_sync;
1671         mddev->utime = get_seconds();
1672
1673         /* If this is just a dirty<->clean transition, and the array is clean
1674          * and 'events' is odd, we can roll back to the previous clean state */
1675         if (nospares
1676             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1677             && (mddev->events & 1)
1678             && mddev->events != 1)
1679                 mddev->events--;
1680         else {
1681                 /* otherwise we have to go forward and ... */
1682                 mddev->events ++;
1683                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1684                         /* .. if the array isn't clean, insist on an odd 'events' */
1685                         if ((mddev->events&1)==0) {
1686                                 mddev->events++;
1687                                 nospares = 0;
1688                         }
1689                 } else {
1690                         /* otherwise insist on an even 'events' (for clean states) */
1691                         if ((mddev->events&1)) {
1692                                 mddev->events++;
1693                                 nospares = 0;
1694                         }
1695                 }
1696         }
1697
1698         if (!mddev->events) {
1699                 /*
1700                  * oops, this 64-bit counter should never wrap.
1701                  * Either we are in around ~1 trillion A.C., assuming
1702                  * 1 reboot per second, or we have a bug:
1703                  */
1704                 MD_BUG();
1705                 mddev->events --;
1706         }
1707         sync_sbs(mddev, nospares);
1708
1709         /*
1710          * do not write anything to disk if using
1711          * nonpersistent superblocks
1712          */
1713         if (!mddev->persistent) {
1714                 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1715                 spin_unlock_irq(&mddev->write_lock);
1716                 wake_up(&mddev->sb_wait);
1717                 return;
1718         }
1719         spin_unlock_irq(&mddev->write_lock);
1720
1721         dprintk(KERN_INFO 
1722                 "md: updating %s RAID superblock on device (in sync %d)\n",
1723                 mdname(mddev),mddev->in_sync);
1724
1725         err = bitmap_update_sb(mddev->bitmap);
1726         ITERATE_RDEV(mddev,rdev,tmp) {
1727                 char b[BDEVNAME_SIZE];
1728                 dprintk(KERN_INFO "md: ");
1729                 if (rdev->sb_loaded != 1)
1730                         continue; /* no noise on spare devices */
1731                 if (test_bit(Faulty, &rdev->flags))
1732                         dprintk("(skipping faulty ");
1733
1734                 dprintk("%s ", bdevname(rdev->bdev,b));
1735                 if (!test_bit(Faulty, &rdev->flags)) {
1736                         md_super_write(mddev,rdev,
1737                                        rdev->sb_offset<<1, rdev->sb_size,
1738                                        rdev->sb_page);
1739                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1740                                 bdevname(rdev->bdev,b),
1741                                 (unsigned long long)rdev->sb_offset);
1742                         rdev->sb_events = mddev->events;
1743
1744                 } else
1745                         dprintk(")\n");
1746                 if (mddev->level == LEVEL_MULTIPATH)
1747                         /* only need to write one superblock... */
1748                         break;
1749         }
1750         md_super_wait(mddev);
1751         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
1752
1753         spin_lock_irq(&mddev->write_lock);
1754         if (mddev->in_sync != sync_req ||
1755             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
1756                 /* have to write it out again */
1757                 spin_unlock_irq(&mddev->write_lock);
1758                 goto repeat;
1759         }
1760         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
1761         spin_unlock_irq(&mddev->write_lock);
1762         wake_up(&mddev->sb_wait);
1763
1764 }
1765
1766 /* words written to sysfs files may, or my not, be \n terminated.
1767  * We want to accept with case. For this we use cmd_match.
1768  */
1769 static int cmd_match(const char *cmd, const char *str)
1770 {
1771         /* See if cmd, written into a sysfs file, matches
1772          * str.  They must either be the same, or cmd can
1773          * have a trailing newline
1774          */
1775         while (*cmd && *str && *cmd == *str) {
1776                 cmd++;
1777                 str++;
1778         }
1779         if (*cmd == '\n')
1780                 cmd++;
1781         if (*str || *cmd)
1782                 return 0;
1783         return 1;
1784 }
1785
1786 struct rdev_sysfs_entry {
1787         struct attribute attr;
1788         ssize_t (*show)(mdk_rdev_t *, char *);
1789         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1790 };
1791
1792 static ssize_t
1793 state_show(mdk_rdev_t *rdev, char *page)
1794 {
1795         char *sep = "";
1796         int len=0;
1797
1798         if (test_bit(Faulty, &rdev->flags)) {
1799                 len+= sprintf(page+len, "%sfaulty",sep);
1800                 sep = ",";
1801         }
1802         if (test_bit(In_sync, &rdev->flags)) {
1803                 len += sprintf(page+len, "%sin_sync",sep);
1804                 sep = ",";
1805         }
1806         if (test_bit(WriteMostly, &rdev->flags)) {
1807                 len += sprintf(page+len, "%swrite_mostly",sep);
1808                 sep = ",";
1809         }
1810         if (!test_bit(Faulty, &rdev->flags) &&
1811             !test_bit(In_sync, &rdev->flags)) {
1812                 len += sprintf(page+len, "%sspare", sep);
1813                 sep = ",";
1814         }
1815         return len+sprintf(page+len, "\n");
1816 }
1817
1818 static ssize_t
1819 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1820 {
1821         /* can write
1822          *  faulty  - simulates and error
1823          *  remove  - disconnects the device
1824          *  writemostly - sets write_mostly
1825          *  -writemostly - clears write_mostly
1826          */
1827         int err = -EINVAL;
1828         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
1829                 md_error(rdev->mddev, rdev);
1830                 err = 0;
1831         } else if (cmd_match(buf, "remove")) {
1832                 if (rdev->raid_disk >= 0)
1833                         err = -EBUSY;
1834                 else {
1835                         mddev_t *mddev = rdev->mddev;
1836                         kick_rdev_from_array(rdev);
1837                         if (mddev->pers)
1838                                 md_update_sb(mddev, 1);
1839                         md_new_event(mddev);
1840                         err = 0;
1841                 }
1842         } else if (cmd_match(buf, "writemostly")) {
1843                 set_bit(WriteMostly, &rdev->flags);
1844                 err = 0;
1845         } else if (cmd_match(buf, "-writemostly")) {
1846                 clear_bit(WriteMostly, &rdev->flags);
1847                 err = 0;
1848         }
1849         return err ? err : len;
1850 }
1851 static struct rdev_sysfs_entry rdev_state =
1852 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
1853
1854 static ssize_t
1855 super_show(mdk_rdev_t *rdev, char *page)
1856 {
1857         if (rdev->sb_loaded && rdev->sb_size) {
1858                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1859                 return rdev->sb_size;
1860         } else
1861                 return 0;
1862 }
1863 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1864
1865 static ssize_t
1866 errors_show(mdk_rdev_t *rdev, char *page)
1867 {
1868         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1869 }
1870
1871 static ssize_t
1872 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1873 {
1874         char *e;
1875         unsigned long n = simple_strtoul(buf, &e, 10);
1876         if (*buf && (*e == 0 || *e == '\n')) {
1877                 atomic_set(&rdev->corrected_errors, n);
1878                 return len;
1879         }
1880         return -EINVAL;
1881 }
1882 static struct rdev_sysfs_entry rdev_errors =
1883 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
1884
1885 static ssize_t
1886 slot_show(mdk_rdev_t *rdev, char *page)
1887 {
1888         if (rdev->raid_disk < 0)
1889                 return sprintf(page, "none\n");
1890         else
1891                 return sprintf(page, "%d\n", rdev->raid_disk);
1892 }
1893
1894 static ssize_t
1895 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1896 {
1897         char *e;
1898         int slot = simple_strtoul(buf, &e, 10);
1899         if (strncmp(buf, "none", 4)==0)
1900                 slot = -1;
1901         else if (e==buf || (*e && *e!= '\n'))
1902                 return -EINVAL;
1903         if (rdev->mddev->pers)
1904                 /* Cannot set slot in active array (yet) */
1905                 return -EBUSY;
1906         if (slot >= rdev->mddev->raid_disks)
1907                 return -ENOSPC;
1908         rdev->raid_disk = slot;
1909         /* assume it is working */
1910         rdev->flags = 0;
1911         set_bit(In_sync, &rdev->flags);
1912         return len;
1913 }
1914
1915
1916 static struct rdev_sysfs_entry rdev_slot =
1917 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
1918
1919 static ssize_t
1920 offset_show(mdk_rdev_t *rdev, char *page)
1921 {
1922         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1923 }
1924
1925 static ssize_t
1926 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1927 {
1928         char *e;
1929         unsigned long long offset = simple_strtoull(buf, &e, 10);
1930         if (e==buf || (*e && *e != '\n'))
1931                 return -EINVAL;
1932         if (rdev->mddev->pers)
1933                 return -EBUSY;
1934         rdev->data_offset = offset;
1935         return len;
1936 }
1937
1938 static struct rdev_sysfs_entry rdev_offset =
1939 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
1940
1941 static ssize_t
1942 rdev_size_show(mdk_rdev_t *rdev, char *page)
1943 {
1944         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1945 }
1946
1947 static ssize_t
1948 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1949 {
1950         char *e;
1951         unsigned long long size = simple_strtoull(buf, &e, 10);
1952         if (e==buf || (*e && *e != '\n'))
1953                 return -EINVAL;
1954         if (rdev->mddev->pers)
1955                 return -EBUSY;
1956         rdev->size = size;
1957         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1958                 rdev->mddev->size = size;
1959         return len;
1960 }
1961
1962 static struct rdev_sysfs_entry rdev_size =
1963 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
1964
1965 static struct attribute *rdev_default_attrs[] = {
1966         &rdev_state.attr,
1967         &rdev_super.attr,
1968         &rdev_errors.attr,
1969         &rdev_slot.attr,
1970         &rdev_offset.attr,
1971         &rdev_size.attr,
1972         NULL,
1973 };
1974 static ssize_t
1975 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1976 {
1977         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1978         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1979
1980         if (!entry->show)
1981                 return -EIO;
1982         return entry->show(rdev, page);
1983 }
1984
1985 static ssize_t
1986 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1987               const char *page, size_t length)
1988 {
1989         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1990         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1991
1992         if (!entry->store)
1993                 return -EIO;
1994         if (!capable(CAP_SYS_ADMIN))
1995                 return -EACCES;
1996         return entry->store(rdev, page, length);
1997 }
1998
1999 static void rdev_free(struct kobject *ko)
2000 {
2001         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2002         kfree(rdev);
2003 }
2004 static struct sysfs_ops rdev_sysfs_ops = {
2005         .show           = rdev_attr_show,
2006         .store          = rdev_attr_store,
2007 };
2008 static struct kobj_type rdev_ktype = {
2009         .release        = rdev_free,
2010         .sysfs_ops      = &rdev_sysfs_ops,
2011         .default_attrs  = rdev_default_attrs,
2012 };
2013
2014 /*
2015  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2016  *
2017  * mark the device faulty if:
2018  *
2019  *   - the device is nonexistent (zero size)
2020  *   - the device has no valid superblock
2021  *
2022  * a faulty rdev _never_ has rdev->sb set.
2023  */
2024 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2025 {
2026         char b[BDEVNAME_SIZE];
2027         int err;
2028         mdk_rdev_t *rdev;
2029         sector_t size;
2030
2031         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2032         if (!rdev) {
2033                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2034                 return ERR_PTR(-ENOMEM);
2035         }
2036
2037         if ((err = alloc_disk_sb(rdev)))
2038                 goto abort_free;
2039
2040         err = lock_rdev(rdev, newdev);
2041         if (err)
2042                 goto abort_free;
2043
2044         rdev->kobj.parent = NULL;
2045         rdev->kobj.ktype = &rdev_ktype;
2046         kobject_init(&rdev->kobj);
2047
2048         rdev->desc_nr = -1;
2049         rdev->saved_raid_disk = -1;
2050         rdev->raid_disk = -1;
2051         rdev->flags = 0;
2052         rdev->data_offset = 0;
2053         rdev->sb_events = 0;
2054         atomic_set(&rdev->nr_pending, 0);
2055         atomic_set(&rdev->read_errors, 0);
2056         atomic_set(&rdev->corrected_errors, 0);
2057
2058         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2059         if (!size) {
2060                 printk(KERN_WARNING 
2061                         "md: %s has zero or unknown size, marking faulty!\n",
2062                         bdevname(rdev->bdev,b));
2063                 err = -EINVAL;
2064                 goto abort_free;
2065         }
2066
2067         if (super_format >= 0) {
2068                 err = super_types[super_format].
2069                         load_super(rdev, NULL, super_minor);
2070                 if (err == -EINVAL) {
2071                         printk(KERN_WARNING 
2072                                 "md: %s has invalid sb, not importing!\n",
2073                                 bdevname(rdev->bdev,b));
2074                         goto abort_free;
2075                 }
2076                 if (err < 0) {
2077                         printk(KERN_WARNING 
2078                                 "md: could not read %s's sb, not importing!\n",
2079                                 bdevname(rdev->bdev,b));
2080                         goto abort_free;
2081                 }
2082         }
2083         INIT_LIST_HEAD(&rdev->same_set);
2084
2085         return rdev;
2086
2087 abort_free:
2088         if (rdev->sb_page) {
2089                 if (rdev->bdev)
2090                         unlock_rdev(rdev);
2091                 free_disk_sb(rdev);
2092         }
2093         kfree(rdev);
2094         return ERR_PTR(err);
2095 }
2096
2097 /*
2098  * Check a full RAID array for plausibility
2099  */
2100
2101
2102 static void analyze_sbs(mddev_t * mddev)
2103 {
2104         int i;
2105         struct list_head *tmp;
2106         mdk_rdev_t *rdev, *freshest;
2107         char b[BDEVNAME_SIZE];
2108
2109         freshest = NULL;
2110         ITERATE_RDEV(mddev,rdev,tmp)
2111                 switch (super_types[mddev->major_version].
2112                         load_super(rdev, freshest, mddev->minor_version)) {
2113                 case 1:
2114                         freshest = rdev;
2115                         break;
2116                 case 0:
2117                         break;
2118                 default:
2119                         printk( KERN_ERR \
2120                                 "md: fatal superblock inconsistency in %s"
2121                                 " -- removing from array\n", 
2122                                 bdevname(rdev->bdev,b));
2123                         kick_rdev_from_array(rdev);
2124                 }
2125
2126
2127         super_types[mddev->major_version].
2128                 validate_super(mddev, freshest);
2129
2130         i = 0;
2131         ITERATE_RDEV(mddev,rdev,tmp) {
2132                 if (rdev != freshest)
2133                         if (super_types[mddev->major_version].
2134                             validate_super(mddev, rdev)) {
2135                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2136                                         " from array!\n",
2137                                         bdevname(rdev->bdev,b));
2138                                 kick_rdev_from_array(rdev);
2139                                 continue;
2140                         }
2141                 if (mddev->level == LEVEL_MULTIPATH) {
2142                         rdev->desc_nr = i++;
2143                         rdev->raid_disk = rdev->desc_nr;
2144                         set_bit(In_sync, &rdev->flags);
2145                 }
2146         }
2147
2148
2149
2150         if (mddev->recovery_cp != MaxSector &&
2151             mddev->level >= 1)
2152                 printk(KERN_ERR "md: %s: raid array is not clean"
2153                        " -- starting background reconstruction\n",
2154                        mdname(mddev));
2155
2156 }
2157
2158 static ssize_t
2159 safe_delay_show(mddev_t *mddev, char *page)
2160 {
2161         int msec = (mddev->safemode_delay*1000)/HZ;
2162         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2163 }
2164 static ssize_t
2165 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2166 {
2167         int scale=1;
2168         int dot=0;
2169         int i;
2170         unsigned long msec;
2171         char buf[30];
2172         char *e;
2173         /* remove a period, and count digits after it */
2174         if (len >= sizeof(buf))
2175                 return -EINVAL;
2176         strlcpy(buf, cbuf, len);
2177         buf[len] = 0;
2178         for (i=0; i<len; i++) {
2179                 if (dot) {
2180                         if (isdigit(buf[i])) {
2181                                 buf[i-1] = buf[i];
2182                                 scale *= 10;
2183                         }
2184                         buf[i] = 0;
2185                 } else if (buf[i] == '.') {
2186                         dot=1;
2187                         buf[i] = 0;
2188                 }
2189         }
2190         msec = simple_strtoul(buf, &e, 10);
2191         if (e == buf || (*e && *e != '\n'))
2192                 return -EINVAL;
2193         msec = (msec * 1000) / scale;
2194         if (msec == 0)
2195                 mddev->safemode_delay = 0;
2196         else {
2197                 mddev->safemode_delay = (msec*HZ)/1000;
2198                 if (mddev->safemode_delay == 0)
2199                         mddev->safemode_delay = 1;
2200         }
2201         return len;
2202 }
2203 static struct md_sysfs_entry md_safe_delay =
2204 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2205
2206 static ssize_t
2207 level_show(mddev_t *mddev, char *page)
2208 {
2209         struct mdk_personality *p = mddev->pers;
2210         if (p)
2211                 return sprintf(page, "%s\n", p->name);
2212         else if (mddev->clevel[0])
2213                 return sprintf(page, "%s\n", mddev->clevel);
2214         else if (mddev->level != LEVEL_NONE)
2215                 return sprintf(page, "%d\n", mddev->level);
2216         else
2217                 return 0;
2218 }
2219
2220 static ssize_t
2221 level_store(mddev_t *mddev, const char *buf, size_t len)
2222 {
2223         int rv = len;
2224         if (mddev->pers)
2225                 return -EBUSY;
2226         if (len == 0)
2227                 return 0;
2228         if (len >= sizeof(mddev->clevel))
2229                 return -ENOSPC;
2230         strncpy(mddev->clevel, buf, len);
2231         if (mddev->clevel[len-1] == '\n')
2232                 len--;
2233         mddev->clevel[len] = 0;
2234         mddev->level = LEVEL_NONE;
2235         return rv;
2236 }
2237
2238 static struct md_sysfs_entry md_level =
2239 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2240
2241
2242 static ssize_t
2243 layout_show(mddev_t *mddev, char *page)
2244 {
2245         /* just a number, not meaningful for all levels */
2246         if (mddev->reshape_position != MaxSector &&
2247             mddev->layout != mddev->new_layout)
2248                 return sprintf(page, "%d (%d)\n",
2249                                mddev->new_layout, mddev->layout);
2250         return sprintf(page, "%d\n", mddev->layout);
2251 }
2252
2253 static ssize_t
2254 layout_store(mddev_t *mddev, const char *buf, size_t len)
2255 {
2256         char *e;
2257         unsigned long n = simple_strtoul(buf, &e, 10);
2258
2259         if (!*buf || (*e && *e != '\n'))
2260                 return -EINVAL;
2261
2262         if (mddev->pers)
2263                 return -EBUSY;
2264         if (mddev->reshape_position != MaxSector)
2265                 mddev->new_layout = n;
2266         else
2267                 mddev->layout = n;
2268         return len;
2269 }
2270 static struct md_sysfs_entry md_layout =
2271 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2272
2273
2274 static ssize_t
2275 raid_disks_show(mddev_t *mddev, char *page)
2276 {
2277         if (mddev->raid_disks == 0)
2278                 return 0;
2279         if (mddev->reshape_position != MaxSector &&
2280             mddev->delta_disks != 0)
2281                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2282                                mddev->raid_disks - mddev->delta_disks);
2283         return sprintf(page, "%d\n", mddev->raid_disks);
2284 }
2285
2286 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2287
2288 static ssize_t
2289 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2290 {
2291         char *e;
2292         int rv = 0;
2293         unsigned long n = simple_strtoul(buf, &e, 10);
2294
2295         if (!*buf || (*e && *e != '\n'))
2296                 return -EINVAL;
2297
2298         if (mddev->pers)
2299                 rv = update_raid_disks(mddev, n);
2300         else if (mddev->reshape_position != MaxSector) {
2301                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2302                 mddev->delta_disks = n - olddisks;
2303                 mddev->raid_disks = n;
2304         } else
2305                 mddev->raid_disks = n;
2306         return rv ? rv : len;
2307 }
2308 static struct md_sysfs_entry md_raid_disks =
2309 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2310
2311 static ssize_t
2312 chunk_size_show(mddev_t *mddev, char *page)
2313 {
2314         if (mddev->reshape_position != MaxSector &&
2315             mddev->chunk_size != mddev->new_chunk)
2316                 return sprintf(page, "%d (%d)\n", mddev->new_chunk,
2317                                mddev->chunk_size);
2318         return sprintf(page, "%d\n", mddev->chunk_size);
2319 }
2320
2321 static ssize_t
2322 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2323 {
2324         /* can only set chunk_size if array is not yet active */
2325         char *e;
2326         unsigned long n = simple_strtoul(buf, &e, 10);
2327
2328         if (!*buf || (*e && *e != '\n'))
2329                 return -EINVAL;
2330
2331         if (mddev->pers)
2332                 return -EBUSY;
2333         else if (mddev->reshape_position != MaxSector)
2334                 mddev->new_chunk = n;
2335         else
2336                 mddev->chunk_size = n;
2337         return len;
2338 }
2339 static struct md_sysfs_entry md_chunk_size =
2340 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2341
2342 static ssize_t
2343 resync_start_show(mddev_t *mddev, char *page)
2344 {
2345         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2346 }
2347
2348 static ssize_t
2349 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2350 {
2351         /* can only set chunk_size if array is not yet active */
2352         char *e;
2353         unsigned long long n = simple_strtoull(buf, &e, 10);
2354
2355         if (mddev->pers)
2356                 return -EBUSY;
2357         if (!*buf || (*e && *e != '\n'))
2358                 return -EINVAL;
2359
2360         mddev->recovery_cp = n;
2361         return len;
2362 }
2363 static struct md_sysfs_entry md_resync_start =
2364 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2365
2366 /*
2367  * The array state can be:
2368  *
2369  * clear
2370  *     No devices, no size, no level
2371  *     Equivalent to STOP_ARRAY ioctl
2372  * inactive
2373  *     May have some settings, but array is not active
2374  *        all IO results in error
2375  *     When written, doesn't tear down array, but just stops it
2376  * suspended (not supported yet)
2377  *     All IO requests will block. The array can be reconfigured.
2378  *     Writing this, if accepted, will block until array is quiessent
2379  * readonly
2380  *     no resync can happen.  no superblocks get written.
2381  *     write requests fail
2382  * read-auto
2383  *     like readonly, but behaves like 'clean' on a write request.
2384  *
2385  * clean - no pending writes, but otherwise active.
2386  *     When written to inactive array, starts without resync
2387  *     If a write request arrives then
2388  *       if metadata is known, mark 'dirty' and switch to 'active'.
2389  *       if not known, block and switch to write-pending
2390  *     If written to an active array that has pending writes, then fails.
2391  * active
2392  *     fully active: IO and resync can be happening.
2393  *     When written to inactive array, starts with resync
2394  *
2395  * write-pending
2396  *     clean, but writes are blocked waiting for 'active' to be written.
2397  *
2398  * active-idle
2399  *     like active, but no writes have been seen for a while (100msec).
2400  *
2401  */
2402 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2403                    write_pending, active_idle, bad_word};
2404 static char *array_states[] = {
2405         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2406         "write-pending", "active-idle", NULL };
2407
2408 static int match_word(const char *word, char **list)
2409 {
2410         int n;
2411         for (n=0; list[n]; n++)
2412                 if (cmd_match(word, list[n]))
2413                         break;
2414         return n;
2415 }
2416
2417 static ssize_t
2418 array_state_show(mddev_t *mddev, char *page)
2419 {
2420         enum array_state st = inactive;
2421
2422         if (mddev->pers)
2423                 switch(mddev->ro) {
2424                 case 1:
2425                         st = readonly;
2426                         break;
2427                 case 2:
2428                         st = read_auto;
2429                         break;
2430                 case 0:
2431                         if (mddev->in_sync)
2432                                 st = clean;
2433                         else if (mddev->safemode)
2434                                 st = active_idle;
2435                         else
2436                                 st = active;
2437                 }
2438         else {
2439                 if (list_empty(&mddev->disks) &&
2440                     mddev->raid_disks == 0 &&
2441                     mddev->size == 0)
2442                         st = clear;
2443                 else
2444                         st = inactive;
2445         }
2446         return sprintf(page, "%s\n", array_states[st]);
2447 }
2448
2449 static int do_md_stop(mddev_t * mddev, int ro);
2450 static int do_md_run(mddev_t * mddev);
2451 static int restart_array(mddev_t *mddev);
2452
2453 static ssize_t
2454 array_state_store(mddev_t *mddev, const char *buf, size_t len)
2455 {
2456         int err = -EINVAL;
2457         enum array_state st = match_word(buf, array_states);
2458         switch(st) {
2459         case bad_word:
2460                 break;
2461         case clear:
2462                 /* stopping an active array */
2463                 if (mddev->pers) {
2464                         if (atomic_read(&mddev->active) > 1)
2465                                 return -EBUSY;
2466                         err = do_md_stop(mddev, 0);
2467                 }
2468                 break;
2469         case inactive:
2470                 /* stopping an active array */
2471                 if (mddev->pers) {
2472                         if (atomic_read(&mddev->active) > 1)
2473                                 return -EBUSY;
2474                         err = do_md_stop(mddev, 2);
2475                 }
2476                 break;
2477         case suspended:
2478                 break; /* not supported yet */
2479         case readonly:
2480                 if (mddev->pers)
2481                         err = do_md_stop(mddev, 1);
2482                 else {
2483                         mddev->ro = 1;
2484                         err = do_md_run(mddev);
2485                 }
2486                 break;
2487         case read_auto:
2488                 /* stopping an active array */
2489                 if (mddev->pers) {
2490                         err = do_md_stop(mddev, 1);
2491                         if (err == 0)
2492                                 mddev->ro = 2; /* FIXME mark devices writable */
2493                 } else {
2494                         mddev->ro = 2;
2495                         err = do_md_run(mddev);
2496                 }
2497                 break;
2498         case clean:
2499                 if (mddev->pers) {
2500                         restart_array(mddev);
2501                         spin_lock_irq(&mddev->write_lock);
2502                         if (atomic_read(&mddev->writes_pending) == 0) {
2503                                 mddev->in_sync = 1;
2504                                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
2505                         }
2506                         spin_unlock_irq(&mddev->write_lock);
2507                 } else {
2508                         mddev->ro = 0;
2509                         mddev->recovery_cp = MaxSector;
2510                         err = do_md_run(mddev);
2511                 }
2512                 break;
2513         case active:
2514                 if (mddev->pers) {
2515                         restart_array(mddev);
2516                         clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2517                         wake_up(&mddev->sb_wait);
2518                         err = 0;
2519                 } else {
2520                         mddev->ro = 0;
2521                         err = do_md_run(mddev);
2522                 }
2523                 break;
2524         case write_pending:
2525         case active_idle:
2526                 /* these cannot be set */
2527                 break;
2528         }
2529         if (err)
2530                 return err;
2531         else
2532                 return len;
2533 }
2534 static struct md_sysfs_entry md_array_state =
2535 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
2536
2537 static ssize_t
2538 null_show(mddev_t *mddev, char *page)
2539 {
2540         return -EINVAL;
2541 }
2542
2543 static ssize_t
2544 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2545 {
2546         /* buf must be %d:%d\n? giving major and minor numbers */
2547         /* The new device is added to the array.
2548          * If the array has a persistent superblock, we read the
2549          * superblock to initialise info and check validity.
2550          * Otherwise, only checking done is that in bind_rdev_to_array,
2551          * which mainly checks size.
2552          */
2553         char *e;
2554         int major = simple_strtoul(buf, &e, 10);
2555         int minor;
2556         dev_t dev;
2557         mdk_rdev_t *rdev;
2558         int err;
2559
2560         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2561                 return -EINVAL;
2562         minor = simple_strtoul(e+1, &e, 10);
2563         if (*e && *e != '\n')
2564                 return -EINVAL;
2565         dev = MKDEV(major, minor);
2566         if (major != MAJOR(dev) ||
2567             minor != MINOR(dev))
2568                 return -EOVERFLOW;
2569
2570
2571         if (mddev->persistent) {
2572                 rdev = md_import_device(dev, mddev->major_version,
2573                                         mddev->minor_version);
2574                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2575                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2576                                                        mdk_rdev_t, same_set);
2577                         err = super_types[mddev->major_version]
2578                                 .load_super(rdev, rdev0, mddev->minor_version);
2579                         if (err < 0)
2580                                 goto out;
2581                 }
2582         } else
2583                 rdev = md_import_device(dev, -1, -1);
2584
2585         if (IS_ERR(rdev))
2586                 return PTR_ERR(rdev);
2587         err = bind_rdev_to_array(rdev, mddev);
2588  out:
2589         if (err)
2590                 export_rdev(rdev);
2591         return err ? err : len;
2592 }
2593
2594 static struct md_sysfs_entry md_new_device =
2595 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
2596
2597 static ssize_t
2598 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
2599 {
2600         char *end;
2601         unsigned long chunk, end_chunk;
2602
2603         if (!mddev->bitmap)
2604                 goto out;
2605         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
2606         while (*buf) {
2607                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
2608                 if (buf == end) break;
2609                 if (*end == '-') { /* range */
2610                         buf = end + 1;
2611                         end_chunk = simple_strtoul(buf, &end, 0);
2612                         if (buf == end) break;
2613                 }
2614                 if (*end && !isspace(*end)) break;
2615                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
2616                 buf = end;
2617                 while (isspace(*buf)) buf++;
2618         }
2619         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
2620 out:
2621         return len;
2622 }
2623
2624 static struct md_sysfs_entry md_bitmap =
2625 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
2626
2627 static ssize_t
2628 size_show(mddev_t *mddev, char *page)
2629 {
2630         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2631 }
2632
2633 static int update_size(mddev_t *mddev, unsigned long size);
2634
2635 static ssize_t
2636 size_store(mddev_t *mddev, const char *buf, size_t len)
2637 {
2638         /* If array is inactive, we can reduce the component size, but
2639          * not increase it (except from 0).
2640          * If array is active, we can try an on-line resize
2641          */
2642         char *e;
2643         int err = 0;
2644         unsigned long long size = simple_strtoull(buf, &e, 10);
2645         if (!*buf || *buf == '\n' ||
2646             (*e && *e != '\n'))
2647                 return -EINVAL;
2648
2649         if (mddev->pers) {
2650                 err = update_size(mddev, size);
2651                 md_update_sb(mddev, 1);
2652         } else {
2653                 if (mddev->size == 0 ||
2654                     mddev->size > size)
2655                         mddev->size = size;
2656                 else
2657                         err = -ENOSPC;
2658         }
2659         return err ? err : len;
2660 }
2661
2662 static struct md_sysfs_entry md_size =
2663 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
2664
2665
2666 /* Metdata version.
2667  * This is either 'none' for arrays with externally managed metadata,
2668  * or N.M for internally known formats
2669  */
2670 static ssize_t
2671 metadata_show(mddev_t *mddev, char *page)
2672 {
2673         if (mddev->persistent)
2674                 return sprintf(page, "%d.%d\n",
2675                                mddev->major_version, mddev->minor_version);
2676         else
2677                 return sprintf(page, "none\n");
2678 }
2679
2680 static ssize_t
2681 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2682 {
2683         int major, minor;
2684         char *e;
2685         if (!list_empty(&mddev->disks))
2686                 return -EBUSY;
2687
2688         if (cmd_match(buf, "none")) {
2689                 mddev->persistent = 0;
2690                 mddev->major_version = 0;
2691                 mddev->minor_version = 90;
2692                 return len;
2693         }
2694         major = simple_strtoul(buf, &e, 10);
2695         if (e==buf || *e != '.')
2696                 return -EINVAL;
2697         buf = e+1;
2698         minor = simple_strtoul(buf, &e, 10);
2699         if (e==buf || (*e && *e != '\n') )
2700                 return -EINVAL;
2701         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
2702                 return -ENOENT;
2703         mddev->major_version = major;
2704         mddev->minor_version = minor;
2705         mddev->persistent = 1;
2706         return len;
2707 }
2708
2709 static struct md_sysfs_entry md_metadata =
2710 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2711
2712 static ssize_t
2713 action_show(mddev_t *mddev, char *page)
2714 {
2715         char *type = "idle";
2716         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2717             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2718                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2719                         type = "reshape";
2720                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2721                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2722                                 type = "resync";
2723                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2724                                 type = "check";
2725                         else
2726                                 type = "repair";
2727                 } else
2728                         type = "recover";
2729         }
2730         return sprintf(page, "%s\n", type);
2731 }
2732
2733 static ssize_t
2734 action_store(mddev_t *mddev, const char *page, size_t len)
2735 {
2736         if (!mddev->pers || !mddev->pers->sync_request)
2737                 return -EINVAL;
2738
2739         if (cmd_match(page, "idle")) {
2740                 if (mddev->sync_thread) {
2741                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2742                         md_unregister_thread(mddev->sync_thread);
2743                         mddev->sync_thread = NULL;
2744                         mddev->recovery = 0;
2745                 }
2746         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2747                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2748                 return -EBUSY;
2749         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2750                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2751         else if (cmd_match(page, "reshape")) {
2752                 int err;
2753                 if (mddev->pers->start_reshape == NULL)
2754                         return -EINVAL;
2755                 err = mddev->pers->start_reshape(mddev);
2756                 if (err)
2757                         return err;
2758         } else {
2759                 if (cmd_match(page, "check"))
2760                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2761                 else if (!cmd_match(page, "repair"))
2762                         return -EINVAL;
2763                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2764                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2765         }
2766         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2767         md_wakeup_thread(mddev->thread);
2768         return len;
2769 }
2770
2771 static ssize_t
2772 mismatch_cnt_show(mddev_t *mddev, char *page)
2773 {
2774         return sprintf(page, "%llu\n",
2775                        (unsigned long long) mddev->resync_mismatches);
2776 }
2777
2778 static struct md_sysfs_entry md_scan_mode =
2779 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2780
2781
2782 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
2783
2784 static ssize_t
2785 sync_min_show(mddev_t *mddev, char *page)
2786 {
2787         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2788                        mddev->sync_speed_min ? "local": "system");
2789 }
2790
2791 static ssize_t
2792 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2793 {
2794         int min;
2795         char *e;
2796         if (strncmp(buf, "system", 6)==0) {
2797                 mddev->sync_speed_min = 0;
2798                 return len;
2799         }
2800         min = simple_strtoul(buf, &e, 10);
2801         if (buf == e || (*e && *e != '\n') || min <= 0)
2802                 return -EINVAL;
2803         mddev->sync_speed_min = min;
2804         return len;
2805 }
2806
2807 static struct md_sysfs_entry md_sync_min =
2808 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2809
2810 static ssize_t
2811 sync_max_show(mddev_t *mddev, char *page)
2812 {
2813         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2814                        mddev->sync_speed_max ? "local": "system");
2815 }
2816
2817 static ssize_t
2818 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2819 {
2820         int max;
2821         char *e;
2822         if (strncmp(buf, "system", 6)==0) {
2823                 mddev->sync_speed_max = 0;
2824                 return len;
2825         }
2826         max = simple_strtoul(buf, &e, 10);
2827         if (buf == e || (*e && *e != '\n') || max <= 0)
2828                 return -EINVAL;
2829         mddev->sync_speed_max = max;
2830         return len;
2831 }
2832
2833 static struct md_sysfs_entry md_sync_max =
2834 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2835
2836
2837 static ssize_t
2838 sync_speed_show(mddev_t *mddev, char *page)
2839 {
2840         unsigned long resync, dt, db;
2841         resync = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active));
2842         dt = ((jiffies - mddev->resync_mark) / HZ);
2843         if (!dt) dt++;
2844         db = resync - (mddev->resync_mark_cnt);
2845         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2846 }
2847
2848 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
2849
2850 static ssize_t
2851 sync_completed_show(mddev_t *mddev, char *page)
2852 {
2853         unsigned long max_blocks, resync;
2854
2855         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2856                 max_blocks = mddev->resync_max_sectors;
2857         else
2858                 max_blocks = mddev->size << 1;
2859
2860         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2861         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2862 }
2863
2864 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
2865
2866 static ssize_t
2867 suspend_lo_show(mddev_t *mddev, char *page)
2868 {
2869         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
2870 }
2871
2872 static ssize_t
2873 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
2874 {
2875         char *e;
2876         unsigned long long new = simple_strtoull(buf, &e, 10);
2877
2878         if (mddev->pers->quiesce == NULL)
2879                 return -EINVAL;
2880         if (buf == e || (*e && *e != '\n'))
2881                 return -EINVAL;
2882         if (new >= mddev->suspend_hi ||
2883             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
2884                 mddev->suspend_lo = new;
2885                 mddev->pers->quiesce(mddev, 2);
2886                 return len;
2887         } else
2888                 return -EINVAL;
2889 }
2890 static struct md_sysfs_entry md_suspend_lo =
2891 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
2892
2893
2894 static ssize_t
2895 suspend_hi_show(mddev_t *mddev, char *page)
2896 {
2897         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
2898 }
2899
2900 static ssize_t
2901 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
2902 {
2903         char *e;
2904         unsigned long long new = simple_strtoull(buf, &e, 10);
2905
2906         if (mddev->pers->quiesce == NULL)
2907                 return -EINVAL;
2908         if (buf == e || (*e && *e != '\n'))
2909                 return -EINVAL;
2910         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
2911             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
2912                 mddev->suspend_hi = new;
2913                 mddev->pers->quiesce(mddev, 1);
2914                 mddev->pers->quiesce(mddev, 0);
2915                 return len;
2916         } else
2917                 return -EINVAL;
2918 }
2919 static struct md_sysfs_entry md_suspend_hi =
2920 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
2921
2922 static ssize_t
2923 reshape_position_show(mddev_t *mddev, char *page)
2924 {
2925         if (mddev->reshape_position != MaxSector)
2926                 return sprintf(page, "%llu\n",
2927                                (unsigned long long)mddev->reshape_position);
2928         strcpy(page, "none\n");
2929         return 5;
2930 }
2931
2932 static ssize_t
2933 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
2934 {
2935         char *e;
2936         unsigned long long new = simple_strtoull(buf, &e, 10);
2937         if (mddev->pers)
2938                 return -EBUSY;
2939         if (buf == e || (*e && *e != '\n'))
2940                 return -EINVAL;
2941         mddev->reshape_position = new;
2942         mddev->delta_disks = 0;
2943         mddev->new_level = mddev->level;
2944         mddev->new_layout = mddev->layout;
2945         mddev->new_chunk = mddev->chunk_size;
2946         return len;
2947 }
2948
2949 static struct md_sysfs_entry md_reshape_position =
2950 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
2951        reshape_position_store);
2952
2953
2954 static struct attribute *md_default_attrs[] = {
2955         &md_level.attr,
2956         &md_layout.attr,
2957         &md_raid_disks.attr,
2958         &md_chunk_size.attr,
2959         &md_size.attr,
2960         &md_resync_start.attr,
2961         &md_metadata.attr,
2962         &md_new_device.attr,
2963         &md_safe_delay.attr,
2964         &md_array_state.attr,
2965         &md_reshape_position.attr,
2966         NULL,
2967 };
2968
2969 static struct attribute *md_redundancy_attrs[] = {
2970         &md_scan_mode.attr,
2971         &md_mismatches.attr,
2972         &md_sync_min.attr,
2973         &md_sync_max.attr,
2974         &md_sync_speed.attr,
2975         &md_sync_completed.attr,
2976         &md_suspend_lo.attr,
2977         &md_suspend_hi.attr,
2978         &md_bitmap.attr,
2979         NULL,
2980 };
2981 static struct attribute_group md_redundancy_group = {
2982         .name = NULL,
2983         .attrs = md_redundancy_attrs,
2984 };
2985
2986
2987 static ssize_t
2988 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2989 {
2990         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2991         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2992         ssize_t rv;
2993
2994         if (!entry->show)
2995                 return -EIO;
2996         rv = mddev_lock(mddev);
2997         if (!rv) {
2998                 rv = entry->show(mddev, page);
2999                 mddev_unlock(mddev);
3000         }
3001         return rv;
3002 }
3003
3004 static ssize_t
3005 md_attr_store(struct kobject *kobj, struct attribute *attr,
3006               const char *page, size_t length)
3007 {
3008         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3009         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3010         ssize_t rv;
3011
3012         if (!entry->store)
3013                 return -EIO;
3014         if (!capable(CAP_SYS_ADMIN))
3015                 return -EACCES;
3016         rv = mddev_lock(mddev);
3017         if (!rv) {
3018                 rv = entry->store(mddev, page, length);
3019                 mddev_unlock(mddev);
3020         }
3021         return rv;
3022 }
3023
3024 static void md_free(struct kobject *ko)
3025 {
3026         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3027         kfree(mddev);
3028 }
3029
3030 static struct sysfs_ops md_sysfs_ops = {
3031         .show   = md_attr_show,
3032         .store  = md_attr_store,
3033 };
3034 static struct kobj_type md_ktype = {
3035         .release        = md_free,
3036         .sysfs_ops      = &md_sysfs_ops,
3037         .default_attrs  = md_default_attrs,
3038 };
3039
3040 int mdp_major = 0;
3041
3042 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3043 {
3044         static DEFINE_MUTEX(disks_mutex);
3045         mddev_t *mddev = mddev_find(dev);
3046         struct gendisk *disk;
3047         int partitioned = (MAJOR(dev) != MD_MAJOR);
3048         int shift = partitioned ? MdpMinorShift : 0;
3049         int unit = MINOR(dev) >> shift;
3050
3051         if (!mddev)
3052                 return NULL;
3053
3054         mutex_lock(&disks_mutex);
3055         if (mddev->gendisk) {
3056                 mutex_unlock(&disks_mutex);
3057                 mddev_put(mddev);
3058                 return NULL;
3059         }
3060         disk = alloc_disk(1 << shift);
3061         if (!disk) {
3062                 mutex_unlock(&disks_mutex);
3063                 mddev_put(mddev);
3064                 return NULL;
3065         }
3066         disk->major = MAJOR(dev);
3067         disk->first_minor = unit << shift;
3068         if (partitioned)
3069                 sprintf(disk->disk_name, "md_d%d", unit);
3070         else
3071                 sprintf(disk->disk_name, "md%d", unit);
3072         disk->fops = &md_fops;
3073         disk->private_data = mddev;
3074         disk->queue = mddev->queue;
3075         add_disk(disk);
3076         mddev->gendisk = disk;
3077         mutex_unlock(&disks_mutex);
3078         mddev->kobj.parent = &disk->kobj;
3079         mddev->kobj.k_name = NULL;
3080         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
3081         mddev->kobj.ktype = &md_ktype;
3082         if (kobject_register(&mddev->kobj))
3083                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3084                        disk->disk_name);
3085         return NULL;
3086 }
3087
3088 static void md_safemode_timeout(unsigned long data)
3089 {
3090         mddev_t *mddev = (mddev_t *) data;
3091
3092         mddev->safemode = 1;
3093         md_wakeup_thread(mddev->thread);
3094 }
3095
3096 static int start_dirty_degraded;
3097
3098 static int do_md_run(mddev_t * mddev)
3099 {
3100         int err;
3101         int chunk_size;
3102         struct list_head *tmp;
3103         mdk_rdev_t *rdev;
3104         struct gendisk *disk;
3105         struct mdk_personality *pers;
3106         char b[BDEVNAME_SIZE];
3107
3108         if (list_empty(&mddev->disks))
3109                 /* cannot run an array with no devices.. */
3110                 return -EINVAL;
3111
3112         if (mddev->pers)
3113                 return -EBUSY;
3114
3115         /*
3116          * Analyze all RAID superblock(s)
3117          */
3118         if (!mddev->raid_disks)
3119                 analyze_sbs(mddev);
3120
3121         chunk_size = mddev->chunk_size;
3122
3123         if (chunk_size) {
3124                 if (chunk_size > MAX_CHUNK_SIZE) {
3125                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
3126                                 chunk_size, MAX_CHUNK_SIZE);
3127                         return -EINVAL;
3128                 }
3129                 /*
3130                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
3131                  */
3132                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
3133                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
3134                         return -EINVAL;
3135                 }
3136                 if (chunk_size < PAGE_SIZE) {
3137                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
3138                                 chunk_size, PAGE_SIZE);
3139                         return -EINVAL;
3140                 }
3141
3142                 /* devices must have minimum size of one chunk */
3143                 ITERATE_RDEV(mddev,rdev,tmp) {
3144                         if (test_bit(Faulty, &rdev->flags))
3145                                 continue;
3146                         if (rdev->size < chunk_size / 1024) {
3147                                 printk(KERN_WARNING
3148                                         "md: Dev %s smaller than chunk_size:"
3149                                         " %lluk < %dk\n",
3150                                         bdevname(rdev->bdev,b),
3151                                         (unsigned long long)rdev->size,
3152                                         chunk_size / 1024);
3153                                 return -EINVAL;
3154                         }
3155                 }
3156         }
3157
3158 #ifdef CONFIG_KMOD
3159         if (mddev->level != LEVEL_NONE)
3160                 request_module("md-level-%d", mddev->level);
3161         else if (mddev->clevel[0])
3162                 request_module("md-%s", mddev->clevel);
3163 #endif
3164
3165         /*
3166          * Drop all container device buffers, from now on
3167          * the only valid external interface is through the md
3168          * device.
3169          * Also find largest hardsector size
3170          */
3171         ITERATE_RDEV(mddev,rdev,tmp) {
3172                 if (test_bit(Faulty, &rdev->flags))
3173                         continue;
3174                 sync_blockdev(rdev->bdev);
3175                 invalidate_bdev(rdev->bdev);
3176         }
3177
3178         md_probe(mddev->unit, NULL, NULL);
3179         disk = mddev->gendisk;
3180         if (!disk)
3181                 return -ENOMEM;
3182
3183         spin_lock(&pers_lock);
3184         pers = find_pers(mddev->level, mddev->clevel);
3185         if (!pers || !try_module_get(pers->owner)) {
3186                 spin_unlock(&pers_lock);
3187                 if (mddev->level != LEVEL_NONE)
3188                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
3189                                mddev->level);
3190                 else
3191                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
3192                                mddev->clevel);
3193                 return -EINVAL;
3194         }
3195         mddev->pers = pers;
3196         spin_unlock(&pers_lock);
3197         mddev->level = pers->level;
3198         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3199
3200         if (mddev->reshape_position != MaxSector &&
3201             pers->start_reshape == NULL) {
3202                 /* This personality cannot handle reshaping... */
3203                 mddev->pers = NULL;
3204                 module_put(pers->owner);
3205                 return -EINVAL;
3206         }
3207
3208         if (pers->sync_request) {
3209                 /* Warn if this is a potentially silly
3210                  * configuration.
3211                  */
3212                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3213                 mdk_rdev_t *rdev2;
3214                 struct list_head *tmp2;
3215                 int warned = 0;
3216                 ITERATE_RDEV(mddev, rdev, tmp) {
3217                         ITERATE_RDEV(mddev, rdev2, tmp2) {
3218                                 if (rdev < rdev2 &&
3219                                     rdev->bdev->bd_contains ==
3220                                     rdev2->bdev->bd_contains) {
3221                                         printk(KERN_WARNING
3222                                                "%s: WARNING: %s appears to be"
3223                                                " on the same physical disk as"
3224                                                " %s.\n",
3225                                                mdname(mddev),
3226                                                bdevname(rdev->bdev,b),
3227                                                bdevname(rdev2->bdev,b2));
3228                                         warned = 1;
3229                                 }
3230                         }
3231                 }
3232                 if (warned)
3233                         printk(KERN_WARNING
3234                                "True protection against single-disk"
3235                                " failure might be compromised.\n");
3236         }
3237
3238         mddev->recovery = 0;
3239         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
3240         mddev->barriers_work = 1;
3241         mddev->ok_start_degraded = start_dirty_degraded;
3242
3243         if (start_readonly)
3244                 mddev->ro = 2; /* read-only, but switch on first write */
3245
3246         err = mddev->pers->run(mddev);
3247         if (!err && mddev->pers->sync_request) {
3248                 err = bitmap_create(mddev);
3249                 if (err) {
3250                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
3251                                mdname(mddev), err);
3252                         mddev->pers->stop(mddev);
3253                 }
3254         }
3255         if (err) {
3256                 printk(KERN_ERR "md: pers->run() failed ...\n");
3257                 module_put(mddev->pers->owner);
3258                 mddev->pers = NULL;
3259                 bitmap_destroy(mddev);
3260                 return err;
3261         }
3262         if (mddev->pers->sync_request) {
3263                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3264                         printk(KERN_WARNING
3265                                "md: cannot register extra attributes for %s\n",
3266                                mdname(mddev));
3267         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
3268                 mddev->ro = 0;
3269
3270         atomic_set(&mddev->writes_pending,0);
3271         mddev->safemode = 0;
3272         mddev->safemode_timer.function = md_safemode_timeout;
3273         mddev->safemode_timer.data = (unsigned long) mddev;
3274         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
3275         mddev->in_sync = 1;
3276
3277         ITERATE_RDEV(mddev,rdev,tmp)
3278                 if (rdev->raid_disk >= 0) {
3279                         char nm[20];
3280                         sprintf(nm, "rd%d", rdev->raid_disk);
3281                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
3282                                 printk("md: cannot register %s for %s\n",
3283                                        nm, mdname(mddev));
3284                 }
3285         
3286         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3287         
3288         if (mddev->flags)
3289                 md_update_sb(mddev, 0);
3290
3291         set_capacity(disk, mddev->array_size<<1);
3292
3293         /* If we call blk_queue_make_request here, it will
3294          * re-initialise max_sectors etc which may have been
3295          * refined inside -> run.  So just set the bits we need to set.
3296          * Most initialisation happended when we called
3297          * blk_queue_make_request(..., md_fail_request)
3298          * earlier.
3299          */
3300         mddev->queue->queuedata = mddev;
3301         mddev->queue->make_request_fn = mddev->pers->make_request;
3302
3303         /* If there is a partially-recovered drive we need to
3304          * start recovery here.  If we leave it to md_check_recovery,
3305          * it will remove the drives and not do the right thing
3306          */
3307         if (mddev->degraded && !mddev->sync_thread) {
3308                 struct list_head *rtmp;
3309                 int spares = 0;
3310                 ITERATE_RDEV(mddev,rdev,rtmp)
3311                         if (rdev->raid_disk >= 0 &&
3312                             !test_bit(In_sync, &rdev->flags) &&
3313                             !test_bit(Faulty, &rdev->flags))
3314                                 /* complete an interrupted recovery */
3315                                 spares++;
3316                 if (spares && mddev->pers->sync_request) {
3317                         mddev->recovery = 0;
3318                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
3319                         mddev->sync_thread = md_register_thread(md_do_sync,
3320                                                                 mddev,
3321                                                                 "%s_resync");
3322                         if (!mddev->sync_thread) {
3323                                 printk(KERN_ERR "%s: could not start resync"
3324                                        " thread...\n",
3325                                        mdname(mddev));
3326                                 /* leave the spares where they are, it shouldn't hurt */
3327                                 mddev->recovery = 0;
3328                         }
3329                 }
3330         }
3331         md_wakeup_thread(mddev->thread);
3332         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
3333
3334         mddev->changed = 1;
3335         md_new_event(mddev);
3336         kobject_uevent(&mddev->gendisk->kobj, KOBJ_CHANGE);
3337         return 0;
3338 }
3339
3340 static int restart_array(mddev_t *mddev)
3341 {
3342         struct gendisk *disk = mddev->gendisk;
3343         int err;
3344
3345         /*
3346          * Complain if it has no devices
3347          */
3348         err = -ENXIO;
3349         if (list_empty(&mddev->disks))
3350                 goto out;
3351
3352         if (mddev->pers) {
3353                 err = -EBUSY;
3354                 if (!mddev->ro)
3355                         goto out;
3356
3357                 mddev->safemode = 0;
3358                 mddev->ro = 0;
3359                 set_disk_ro(disk, 0);
3360
3361                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
3362                         mdname(mddev));
3363                 /*
3364                  * Kick recovery or resync if necessary
3365                  */
3366                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3367                 md_wakeup_thread(mddev->thread);
3368                 md_wakeup_thread(mddev->sync_thread);
3369                 err = 0;
3370         } else
3371                 err = -EINVAL;
3372
3373 out:
3374         return err;
3375 }
3376
3377 /* similar to deny_write_access, but accounts for our holding a reference
3378  * to the file ourselves */
3379 static int deny_bitmap_write_access(struct file * file)
3380 {
3381         struct inode *inode = file->f_mapping->host;
3382
3383         spin_lock(&inode->i_lock);
3384         if (atomic_read(&inode->i_writecount) > 1) {
3385                 spin_unlock(&inode->i_lock);
3386                 return -ETXTBSY;
3387         }
3388         atomic_set(&inode->i_writecount, -1);
3389         spin_unlock(&inode->i_lock);
3390
3391         return 0;
3392 }
3393
3394 static void restore_bitmap_write_access(struct file *file)
3395 {
3396         struct inode *inode = file->f_mapping->host;
3397
3398         spin_lock(&inode->i_lock);
3399         atomic_set(&inode->i_writecount, 1);
3400         spin_unlock(&inode->i_lock);
3401 }
3402
3403 /* mode:
3404  *   0 - completely stop and dis-assemble array
3405  *   1 - switch to readonly
3406  *   2 - stop but do not disassemble array
3407  */
3408 static int do_md_stop(mddev_t * mddev, int mode)
3409 {
3410         int err = 0;
3411         struct gendisk *disk = mddev->gendisk;
3412
3413         if (mddev->pers) {
3414                 if (atomic_read(&mddev->active)>2) {
3415                         printk("md: %s still in use.\n",mdname(mddev));
3416                         return -EBUSY;
3417                 }
3418
3419                 if (mddev->sync_thread) {
3420                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3421                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3422                         md_unregister_thread(mddev->sync_thread);
3423                         mddev->sync_thread = NULL;
3424                 }
3425
3426                 del_timer_sync(&mddev->safemode_timer);
3427
3428                 invalidate_partition(disk, 0);
3429
3430                 switch(mode) {
3431                 case 1: /* readonly */
3432                         err  = -ENXIO;
3433                         if (mddev->ro==1)
3434                                 goto out;
3435                         mddev->ro = 1;
3436                         break;
3437                 case 0: /* disassemble */
3438                 case 2: /* stop */
3439                         bitmap_flush(mddev);
3440                         md_super_wait(mddev);
3441                         if (mddev->ro)
3442                                 set_disk_ro(disk, 0);
3443                         blk_queue_make_request(mddev->queue, md_fail_request);
3444                         mddev->pers->stop(mddev);
3445                         mddev->queue->merge_bvec_fn = NULL;
3446                         mddev->queue->unplug_fn = NULL;
3447                         mddev->queue->issue_flush_fn = NULL;
3448                         mddev->queue->backing_dev_info.congested_fn = NULL;
3449                         if (mddev->pers->sync_request)
3450                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3451
3452                         module_put(mddev->pers->owner);
3453                         mddev->pers = NULL;
3454
3455                         set_capacity(disk, 0);
3456                         mddev->changed = 1;
3457
3458                         if (mddev->ro)
3459                                 mddev->ro = 0;
3460                 }
3461                 if (!mddev->in_sync || mddev->flags) {
3462                         /* mark array as shutdown cleanly */
3463                         mddev->in_sync = 1;
3464                         md_update_sb(mddev, 1);
3465                 }
3466                 if (mode == 1)
3467                         set_disk_ro(disk, 1);
3468                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3469         }
3470
3471         /*
3472          * Free resources if final stop
3473          */
3474         if (mode == 0) {
3475                 mdk_rdev_t *rdev;
3476                 struct list_head *tmp;
3477
3478                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
3479
3480                 bitmap_destroy(mddev);
3481                 if (mddev->bitmap_file) {
3482                         restore_bitmap_write_access(mddev->bitmap_file);
3483                         fput(mddev->bitmap_file);
3484                         mddev->bitmap_file = NULL;
3485                 }
3486                 mddev->bitmap_offset = 0;
3487
3488                 ITERATE_RDEV(mddev,rdev,tmp)
3489                         if (rdev->raid_disk >= 0) {
3490                                 char nm[20];
3491                                 sprintf(nm, "rd%d", rdev->raid_disk);
3492                                 sysfs_remove_link(&mddev->kobj, nm);
3493                         }
3494
3495                 /* make sure all delayed_delete calls have finished */
3496                 flush_scheduled_work();
3497
3498                 export_array(mddev);
3499
3500                 mddev->array_size = 0;
3501                 mddev->size = 0;
3502                 mddev->raid_disks = 0;
3503                 mddev->recovery_cp = 0;
3504                 mddev->reshape_position = MaxSector;
3505
3506         } else if (mddev->pers)
3507                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
3508                         mdname(mddev));
3509         err = 0;
3510         md_new_event(mddev);
3511 out:
3512         return err;
3513 }
3514
3515 #ifndef MODULE
3516 static void autorun_array(mddev_t *mddev)
3517 {
3518         mdk_rdev_t *rdev;
3519         struct list_head *tmp;
3520         int err;
3521
3522         if (list_empty(&mddev->disks))
3523                 return;
3524
3525         printk(KERN_INFO "md: running: ");
3526
3527         ITERATE_RDEV(mddev,rdev,tmp) {
3528                 char b[BDEVNAME_SIZE];
3529                 printk("<%s>", bdevname(rdev->bdev,b));
3530         }
3531         printk("\n");
3532
3533         err = do_md_run (mddev);
3534         if (err) {
3535                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
3536                 do_md_stop (mddev, 0);
3537         }
3538 }
3539
3540 /*
3541  * lets try to run arrays based on all disks that have arrived
3542  * until now. (those are in pending_raid_disks)
3543  *
3544  * the method: pick the first pending disk, collect all disks with
3545  * the same UUID, remove all from the pending list and put them into
3546  * the 'same_array' list. Then order this list based on superblock
3547  * update time (freshest comes first), kick out 'old' disks and
3548  * compare superblocks. If everything's fine then run it.
3549  *
3550  * If "unit" is allocated, then bump its reference count
3551  */
3552 static void autorun_devices(int part)
3553 {
3554         struct list_head *tmp;
3555         mdk_rdev_t *rdev0, *rdev;
3556         mddev_t *mddev;
3557         char b[BDEVNAME_SIZE];
3558
3559         printk(KERN_INFO "md: autorun ...\n");
3560         while (!list_empty(&pending_raid_disks)) {
3561                 int unit;
3562                 dev_t dev;
3563                 LIST_HEAD(candidates);
3564                 rdev0 = list_entry(pending_raid_disks.next,
3565                                          mdk_rdev_t, same_set);
3566
3567                 printk(KERN_INFO "md: considering %s ...\n",
3568                         bdevname(rdev0->bdev,b));
3569                 INIT_LIST_HEAD(&candidates);
3570                 ITERATE_RDEV_PENDING(rdev,tmp)
3571                         if (super_90_load(rdev, rdev0, 0) >= 0) {
3572                                 printk(KERN_INFO "md:  adding %s ...\n",
3573                                         bdevname(rdev->bdev,b));
3574                                 list_move(&rdev->same_set, &candidates);
3575                         }
3576                 /*
3577                  * now we have a set of devices, with all of them having
3578                  * mostly sane superblocks. It's time to allocate the
3579                  * mddev.
3580                  */
3581                 if (part) {
3582                         dev = MKDEV(mdp_major,
3583                                     rdev0->preferred_minor << MdpMinorShift);
3584                         unit = MINOR(dev) >> MdpMinorShift;
3585                 } else {
3586                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
3587                         unit = MINOR(dev);
3588                 }
3589                 if (rdev0->preferred_minor != unit) {
3590                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
3591                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
3592                         break;
3593                 }
3594
3595                 md_probe(dev, NULL, NULL);
3596                 mddev = mddev_find(dev);
3597                 if (!mddev) {
3598                         printk(KERN_ERR 
3599                                 "md: cannot allocate memory for md drive.\n");
3600                         break;
3601                 }
3602                 if (mddev_lock(mddev)) 
3603                         printk(KERN_WARNING "md: %s locked, cannot run\n",
3604                                mdname(mddev));
3605                 else if (mddev->raid_disks || mddev->major_version
3606                          || !list_empty(&mddev->disks)) {
3607                         printk(KERN_WARNING 
3608                                 "md: %s already running, cannot run %s\n",
3609                                 mdname(mddev), bdevname(rdev0->bdev,b));
3610                         mddev_unlock(mddev);
3611                 } else {
3612                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
3613                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
3614                                 list_del_init(&rdev->same_set);
3615                                 if (bind_rdev_to_array(rdev, mddev))
3616                                         export_rdev(rdev);
3617                         }
3618                         autorun_array(mddev);
3619                         mddev_unlock(mddev);
3620                 }
3621                 /* on success, candidates will be empty, on error
3622                  * it won't...
3623                  */
3624                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
3625                         export_rdev(rdev);
3626                 mddev_put(mddev);
3627         }
3628         printk(KERN_INFO "md: ... autorun DONE.\n");
3629 }
3630 #endif /* !MODULE */
3631
3632 static int get_version(void __user * arg)
3633 {
3634         mdu_version_t ver;
3635
3636         ver.major = MD_MAJOR_VERSION;
3637         ver.minor = MD_MINOR_VERSION;
3638         ver.patchlevel = MD_PATCHLEVEL_VERSION;
3639
3640         if (copy_to_user(arg, &ver, sizeof(ver)))
3641                 return -EFAULT;
3642
3643         return 0;
3644 }
3645
3646 static int get_array_info(mddev_t * mddev, void __user * arg)
3647 {
3648         mdu_array_info_t info;
3649         int nr,working,active,failed,spare;
3650         mdk_rdev_t *rdev;
3651         struct list_head *tmp;
3652
3653         nr=working=active=failed=spare=0;
3654         ITERATE_RDEV(mddev,rdev,tmp) {
3655                 nr++;
3656                 if (test_bit(Faulty, &rdev->flags))
3657                         failed++;
3658                 else {
3659                         working++;
3660                         if (test_bit(In_sync, &rdev->flags))
3661                                 active++;       
3662                         else
3663                                 spare++;
3664                 }
3665         }
3666
3667         info.major_version = mddev->major_version;
3668         info.minor_version = mddev->minor_version;
3669         info.patch_version = MD_PATCHLEVEL_VERSION;
3670         info.ctime         = mddev->ctime;
3671         info.level         = mddev->level;
3672         info.size          = mddev->size;
3673         if (info.size != mddev->size) /* overflow */
3674                 info.size = -1;
3675         info.nr_disks      = nr;
3676         info.raid_disks    = mddev->raid_disks;
3677         info.md_minor      = mddev->md_minor;
3678         info.not_persistent= !mddev->persistent;
3679
3680         info.utime         = mddev->utime;
3681         info.state         = 0;
3682         if (mddev->in_sync)
3683                 info.state = (1<<MD_SB_CLEAN);
3684         if (mddev->bitmap && mddev->bitmap_offset)
3685                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3686         info.active_disks  = active;
3687         info.working_disks = working;
3688         info.failed_disks  = failed;
3689         info.spare_disks   = spare;
3690
3691         info.layout        = mddev->layout;
3692         info.chunk_size    = mddev->chunk_size;
3693
3694         if (copy_to_user(arg, &info, sizeof(info)))
3695                 return -EFAULT;
3696
3697         return 0;
3698 }
3699
3700 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3701 {
3702         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3703         char *ptr, *buf = NULL;
3704         int err = -ENOMEM;
3705
3706         md_allow_write(mddev);
3707
3708         file = kmalloc(sizeof(*file), GFP_KERNEL);
3709         if (!file)
3710                 goto out;
3711
3712         /* bitmap disabled, zero the first byte and copy out */
3713         if (!mddev->bitmap || !mddev->bitmap->file) {
3714                 file->pathname[0] = '\0';
3715                 goto copy_out;
3716         }
3717
3718         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3719         if (!buf)
3720                 goto out;
3721
3722         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3723         if (!ptr)
3724                 goto out;
3725
3726         strcpy(file->pathname, ptr);
3727
3728 copy_out:
3729         err = 0;
3730         if (copy_to_user(arg, file, sizeof(*file)))
3731                 err = -EFAULT;
3732 out:
3733         kfree(buf);
3734         kfree(file);
3735         return err;
3736 }
3737
3738 static int get_disk_info(mddev_t * mddev, void __user * arg)
3739 {
3740         mdu_disk_info_t info;
3741         unsigned int nr;
3742         mdk_rdev_t *rdev;
3743
3744         if (copy_from_user(&info, arg, sizeof(info)))
3745                 return -EFAULT;
3746
3747         nr = info.number;
3748
3749         rdev = find_rdev_nr(mddev, nr);
3750         if (rdev) {
3751                 info.major = MAJOR(rdev->bdev->bd_dev);
3752                 info.minor = MINOR(rdev->bdev->bd_dev);
3753                 info.raid_disk = rdev->raid_disk;
3754                 info.state = 0;
3755                 if (test_bit(Faulty, &rdev->flags))
3756                         info.state |= (1<<MD_DISK_FAULTY);
3757                 else if (test_bit(In_sync, &rdev->flags)) {
3758                         info.state |= (1<<MD_DISK_ACTIVE);
3759                         info.state |= (1<<MD_DISK_SYNC);
3760                 }
3761                 if (test_bit(WriteMostly, &rdev->flags))
3762                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3763         } else {
3764                 info.major = info.minor = 0;
3765                 info.raid_disk = -1;
3766                 info.state = (1<<MD_DISK_REMOVED);
3767         }
3768
3769         if (copy_to_user(arg, &info, sizeof(info)))
3770                 return -EFAULT;
3771
3772         return 0;
3773 }
3774
3775 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3776 {
3777         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3778         mdk_rdev_t *rdev;
3779         dev_t dev = MKDEV(info->major,info->minor);
3780
3781         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3782                 return -EOVERFLOW;
3783
3784         if (!mddev->raid_disks) {
3785                 int err;
3786                 /* expecting a device which has a superblock */
3787                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3788                 if (IS_ERR(rdev)) {
3789                         printk(KERN_WARNING 
3790                                 "md: md_import_device returned %ld\n",
3791                                 PTR_ERR(rdev));
3792                         return PTR_ERR(rdev);
3793                 }
3794                 if (!list_empty(&mddev->disks)) {
3795                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3796                                                         mdk_rdev_t, same_set);
3797                         int err = super_types[mddev->major_version]
3798                                 .load_super(rdev, rdev0, mddev->minor_version);
3799                         if (err < 0) {
3800                                 printk(KERN_WARNING 
3801                                         "md: %s has different UUID to %s\n",
3802                                         bdevname(rdev->bdev,b), 
3803                                         bdevname(rdev0->bdev,b2));
3804                                 export_rdev(rdev);
3805                                 return -EINVAL;
3806                         }
3807                 }
3808                 err = bind_rdev_to_array(rdev, mddev);
3809                 if (err)
3810                         export_rdev(rdev);
3811                 return err;
3812         }
3813
3814         /*
3815          * add_new_disk can be used once the array is assembled
3816          * to add "hot spares".  They must already have a superblock
3817          * written
3818          */
3819         if (mddev->pers) {
3820                 int err;
3821                 if (!mddev->pers->hot_add_disk) {
3822                         printk(KERN_WARNING 
3823                                 "%s: personality does not support diskops!\n",
3824                                mdname(mddev));
3825                         return -EINVAL;
3826                 }
3827                 if (mddev->persistent)
3828                         rdev = md_import_device(dev, mddev->major_version,
3829                                                 mddev->minor_version);
3830                 else
3831                         rdev = md_import_device(dev, -1, -1);
3832                 if (IS_ERR(rdev)) {
3833                         printk(KERN_WARNING 
3834                                 "md: md_import_device returned %ld\n",
3835                                 PTR_ERR(rdev));
3836                         return PTR_ERR(rdev);
3837                 }
3838                 /* set save_raid_disk if appropriate */
3839                 if (!mddev->persistent) {
3840                         if (info->state & (1<<MD_DISK_SYNC)  &&
3841                             info->raid_disk < mddev->raid_disks)
3842                                 rdev->raid_disk = info->raid_disk;
3843                         else
3844                                 rdev->raid_disk = -1;
3845                 } else
3846                         super_types[mddev->major_version].
3847                                 validate_super(mddev, rdev);
3848                 rdev->saved_raid_disk = rdev->raid_disk;
3849
3850                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3851                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3852                         set_bit(WriteMostly, &rdev->flags);
3853
3854                 rdev->raid_disk = -1;
3855                 err = bind_rdev_to_array(rdev, mddev);
3856                 if (!err && !mddev->pers->hot_remove_disk) {
3857                         /* If there is hot_add_disk but no hot_remove_disk
3858                          * then added disks for geometry changes,
3859                          * and should be added immediately.
3860                          */
3861                         super_types[mddev->major_version].
3862                                 validate_super(mddev, rdev);
3863                         err = mddev->pers->hot_add_disk(mddev, rdev);
3864                         if (err)
3865                                 unbind_rdev_from_array(rdev);
3866                 }
3867                 if (err)
3868                         export_rdev(rdev);
3869
3870                 md_update_sb(mddev, 1);
3871                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3872                 md_wakeup_thread(mddev->thread);
3873                 return err;
3874         }
3875
3876         /* otherwise, add_new_disk is only allowed
3877          * for major_version==0 superblocks
3878          */
3879         if (mddev->major_version != 0) {
3880                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3881                        mdname(mddev));
3882                 return -EINVAL;
3883         }
3884
3885         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3886                 int err;
3887                 rdev = md_import_device (dev, -1, 0);
3888                 if (IS_ERR(rdev)) {
3889                         printk(KERN_WARNING 
3890                                 "md: error, md_import_device() returned %ld\n",
3891                                 PTR_ERR(rdev));
3892                         return PTR_ERR(rdev);
3893                 }
3894                 rdev->desc_nr = info->number;
3895                 if (info->raid_disk < mddev->raid_disks)
3896                         rdev->raid_disk = info->raid_disk;
3897                 else
3898                         rdev->raid_disk = -1;
3899
3900                 rdev->flags = 0;
3901
3902                 if (rdev->raid_disk < mddev->raid_disks)
3903                         if (info->state & (1<<MD_DISK_SYNC))
3904                                 set_bit(In_sync, &rdev->flags);
3905
3906                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3907                         set_bit(WriteMostly, &rdev->flags);
3908
3909                 if (!mddev->persistent) {
3910                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3911                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3912                 } else 
3913                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3914                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3915
3916                 err = bind_rdev_to_array(rdev, mddev);
3917                 if (err) {
3918                         export_rdev(rdev);
3919                         return err;
3920                 }
3921         }
3922
3923         return 0;
3924 }
3925
3926 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3927 {
3928         char b[BDEVNAME_SIZE];
3929         mdk_rdev_t *rdev;
3930
3931         if (!mddev->pers)
3932                 return -ENODEV;
3933
3934         rdev = find_rdev(mddev, dev);
3935         if (!rdev)
3936                 return -ENXIO;
3937
3938         if (rdev->raid_disk >= 0)
3939                 goto busy;
3940
3941         kick_rdev_from_array(rdev);
3942         md_update_sb(mddev, 1);
3943         md_new_event(mddev);
3944
3945         return 0;
3946 busy:
3947         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3948                 bdevname(rdev->bdev,b), mdname(mddev));
3949         return -EBUSY;
3950 }
3951
3952 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3953 {
3954         char b[BDEVNAME_SIZE];
3955         int err;
3956         unsigned int size;
3957         mdk_rdev_t *rdev;
3958
3959         if (!mddev->pers)
3960                 return -ENODEV;
3961
3962         if (mddev->major_version != 0) {
3963                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3964                         " version-0 superblocks.\n",
3965                         mdname(mddev));
3966                 return -EINVAL;
3967         }
3968         if (!mddev->pers->hot_add_disk) {
3969                 printk(KERN_WARNING 
3970                         "%s: personality does not support diskops!\n",
3971                         mdname(mddev));
3972                 return -EINVAL;
3973         }
3974
3975         rdev = md_import_device (dev, -1, 0);
3976         if (IS_ERR(rdev)) {
3977                 printk(KERN_WARNING 
3978                         "md: error, md_import_device() returned %ld\n",
3979                         PTR_ERR(rdev));
3980                 return -EINVAL;
3981         }
3982
3983         if (mddev->persistent)
3984                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3985         else
3986                 rdev->sb_offset =
3987                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3988
3989         size = calc_dev_size(rdev, mddev->chunk_size);
3990         rdev->size = size;
3991
3992         if (test_bit(Faulty, &rdev->flags)) {
3993                 printk(KERN_WARNING 
3994                         "md: can not hot-add faulty %s disk to %s!\n",
3995                         bdevname(rdev->bdev,b), mdname(mddev));
3996                 err = -EINVAL;
3997                 goto abort_export;
3998         }
3999         clear_bit(In_sync, &rdev->flags);
4000         rdev->desc_nr = -1;
4001         rdev->saved_raid_disk = -1;
4002         err = bind_rdev_to_array(rdev, mddev);
4003         if (err)
4004                 goto abort_export;
4005
4006         /*
4007          * The rest should better be atomic, we can have disk failures
4008          * noticed in interrupt contexts ...
4009          */
4010
4011         if (rdev->desc_nr == mddev->max_disks) {
4012                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
4013                         mdname(mddev));
4014                 err = -EBUSY;
4015                 goto abort_unbind_export;
4016         }
4017
4018         rdev->raid_disk = -1;
4019
4020         md_update_sb(mddev, 1);
4021
4022         /*
4023          * Kick recovery, maybe this spare has to be added to the
4024          * array immediately.
4025          */
4026         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4027         md_wakeup_thread(mddev->thread);
4028         md_new_event(mddev);
4029         return 0;
4030
4031 abort_unbind_export:
4032         unbind_rdev_from_array(rdev);
4033
4034 abort_export:
4035         export_rdev(rdev);
4036         return err;
4037 }
4038
4039 static int set_bitmap_file(mddev_t *mddev, int fd)
4040 {
4041         int err;
4042
4043         if (mddev->pers) {
4044                 if (!mddev->pers->quiesce)
4045                         return -EBUSY;
4046                 if (mddev->recovery || mddev->sync_thread)
4047                         return -EBUSY;
4048                 /* we should be able to change the bitmap.. */
4049         }
4050
4051
4052         if (fd >= 0) {
4053                 if (mddev->bitmap)
4054                         return -EEXIST; /* cannot add when bitmap is present */
4055                 mddev->bitmap_file = fget(fd);
4056
4057                 if (mddev->bitmap_file == NULL) {
4058                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4059                                mdname(mddev));
4060                         return -EBADF;
4061                 }
4062
4063                 err = deny_bitmap_write_access(mddev->bitmap_file);
4064                 if (err) {
4065                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4066                                mdname(mddev));
4067                         fput(mddev->bitmap_file);
4068                         mddev->bitmap_file = NULL;
4069                         return err;
4070                 }
4071                 mddev->bitmap_offset = 0; /* file overrides offset */
4072         } else if (mddev->bitmap == NULL)
4073                 return -ENOENT; /* cannot remove what isn't there */
4074         err = 0;
4075         if (mddev->pers) {
4076                 mddev->pers->quiesce(mddev, 1);
4077                 if (fd >= 0)
4078                         err = bitmap_create(mddev);
4079                 if (fd < 0 || err) {
4080                         bitmap_destroy(mddev);
4081                         fd = -1; /* make sure to put the file */
4082                 }
4083                 mddev->pers->quiesce(mddev, 0);
4084         }
4085         if (fd < 0) {
4086                 if (mddev->bitmap_file) {
4087                         restore_bitmap_write_access(mddev->bitmap_file);
4088                         fput(mddev->bitmap_file);
4089                 }
4090                 mddev->bitmap_file = NULL;
4091         }
4092
4093         return err;
4094 }
4095
4096 /*
4097  * set_array_info is used two different ways
4098  * The original usage is when creating a new array.
4099  * In this usage, raid_disks is > 0 and it together with
4100  *  level, size, not_persistent,layout,chunksize determine the
4101  *  shape of the array.
4102  *  This will always create an array with a type-0.90.0 superblock.
4103  * The newer usage is when assembling an array.
4104  *  In this case raid_disks will be 0, and the major_version field is
4105  *  use to determine which style super-blocks are to be found on the devices.
4106  *  The minor and patch _version numbers are also kept incase the
4107  *  super_block handler wishes to interpret them.
4108  */
4109 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
4110 {
4111
4112         if (info->raid_disks == 0) {
4113                 /* just setting version number for superblock loading */
4114                 if (info->major_version < 0 ||
4115                     info->major_version >= ARRAY_SIZE(super_types) ||
4116                     super_types[info->major_version].name == NULL) {
4117                         /* maybe try to auto-load a module? */
4118                         printk(KERN_INFO 
4119                                 "md: superblock version %d not known\n",
4120                                 info->major_version);
4121                         return -EINVAL;
4122                 }
4123                 mddev->major_version = info->major_version;
4124                 mddev->minor_version = info->minor_version;
4125                 mddev->patch_version = info->patch_version;
4126                 mddev->persistent = !info->not_persistent;
4127                 return 0;
4128         }
4129         mddev->major_version = MD_MAJOR_VERSION;
4130         mddev->minor_version = MD_MINOR_VERSION;
4131         mddev->patch_version = MD_PATCHLEVEL_VERSION;
4132         mddev->ctime         = get_seconds();
4133
4134         mddev->level         = info->level;
4135         mddev->clevel[0]     = 0;
4136         mddev->size          = info->size;
4137         mddev->raid_disks    = info->raid_disks;
4138         /* don't set md_minor, it is determined by which /dev/md* was
4139          * openned
4140          */
4141         if (info->state & (1<<MD_SB_CLEAN))
4142                 mddev->recovery_cp = MaxSector;
4143         else
4144                 mddev->recovery_cp = 0;
4145         mddev->persistent    = ! info->not_persistent;
4146
4147         mddev->layout        = info->layout;
4148         mddev->chunk_size    = info->chunk_size;
4149
4150         mddev->max_disks     = MD_SB_DISKS;
4151
4152         mddev->flags         = 0;
4153         set_bit(MD_CHANGE_DEVS, &mddev->flags);
4154
4155         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
4156         mddev->bitmap_offset = 0;
4157
4158         mddev->reshape_position = MaxSector;
4159
4160         /*
4161          * Generate a 128 bit UUID
4162          */
4163         get_random_bytes(mddev->uuid, 16);
4164
4165         mddev->new_level = mddev->level;
4166         mddev->new_chunk = mddev->chunk_size;
4167         mddev->new_layout = mddev->layout;
4168         mddev->delta_disks = 0;
4169
4170         return 0;
4171 }
4172
4173 static int update_size(mddev_t *mddev, unsigned long size)
4174 {
4175         mdk_rdev_t * rdev;
4176         int rv;
4177         struct list_head *tmp;
4178         int fit = (size == 0);
4179
4180         if (mddev->pers->resize == NULL)
4181                 return -EINVAL;
4182         /* The "size" is the amount of each device that is used.
4183          * This can only make sense for arrays with redundancy.
4184          * linear and raid0 always use whatever space is available
4185          * We can only consider changing the size if no resync
4186          * or reconstruction is happening, and if the new size
4187          * is acceptable. It must fit before the sb_offset or,
4188          * if that is <data_offset, it must fit before the
4189          * size of each device.
4190          * If size is zero, we find the largest size that fits.
4191          */
4192         if (mddev->sync_thread)
4193                 return -EBUSY;
4194         ITERATE_RDEV(mddev,rdev,tmp) {
4195                 sector_t avail;
4196                 avail = rdev->size * 2;
4197
4198                 if (fit && (size == 0 || size > avail/2))
4199                         size = avail/2;
4200                 if (avail < ((sector_t)size << 1))
4201                         return -ENOSPC;
4202         }
4203         rv = mddev->pers->resize(mddev, (sector_t)size *2);
4204         if (!rv) {
4205                 struct block_device *bdev;
4206
4207                 bdev = bdget_disk(mddev->gendisk, 0);
4208                 if (bdev) {
4209                         mutex_lock(&bdev->bd_inode->i_mutex);
4210                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
4211                         mutex_unlock(&bdev->bd_inode->i_mutex);
4212                         bdput(bdev);
4213                 }
4214         }
4215         return rv;
4216 }
4217
4218 static int update_raid_disks(mddev_t *mddev, int raid_disks)
4219 {
4220         int rv;
4221         /* change the number of raid disks */
4222         if (mddev->pers->check_reshape == NULL)
4223                 return -EINVAL;
4224         if (raid_disks <= 0 ||
4225             raid_disks >= mddev->max_disks)
4226                 return -EINVAL;
4227         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
4228                 return -EBUSY;
4229         mddev->delta_disks = raid_disks - mddev->raid_disks;
4230
4231         rv = mddev->pers->check_reshape(mddev);
4232         return rv;
4233 }
4234
4235
4236 /*
4237  * update_array_info is used to change the configuration of an
4238  * on-line array.
4239  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
4240  * fields in the info are checked against the array.
4241  * Any differences that cannot be handled will cause an error.
4242  * Normally, only one change can be managed at a time.
4243  */
4244 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
4245 {
4246         int rv = 0;
4247         int cnt = 0;
4248         int state = 0;
4249
4250         /* calculate expected state,ignoring low bits */
4251         if (mddev->bitmap && mddev->bitmap_offset)
4252                 state |= (1 << MD_SB_BITMAP_PRESENT);
4253
4254         if (mddev->major_version != info->major_version ||
4255             mddev->minor_version != info->minor_version ||
4256 /*          mddev->patch_version != info->patch_version || */
4257             mddev->ctime         != info->ctime         ||
4258             mddev->level         != info->level         ||
4259 /*          mddev->layout        != info->layout        || */
4260             !mddev->persistent   != info->not_persistent||
4261             mddev->chunk_size    != info->chunk_size    ||
4262             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
4263             ((state^info->state) & 0xfffffe00)
4264                 )
4265                 return -EINVAL;
4266         /* Check there is only one change */
4267         if (info->size >= 0 && mddev->size != info->size) cnt++;
4268         if (mddev->raid_disks != info->raid_disks) cnt++;
4269         if (mddev->layout != info->layout) cnt++;
4270         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
4271         if (cnt == 0) return 0;
4272         if (cnt > 1) return -EINVAL;
4273
4274         if (mddev->layout != info->layout) {
4275                 /* Change layout
4276                  * we don't need to do anything at the md level, the
4277                  * personality will take care of it all.
4278                  */
4279                 if (mddev->pers->reconfig == NULL)
4280                         return -EINVAL;
4281                 else
4282                         return mddev->pers->reconfig(mddev, info->layout, -1);
4283         }
4284         if (info->size >= 0 && mddev->size != info->size)
4285                 rv = update_size(mddev, info->size);
4286
4287         if (mddev->raid_disks    != info->raid_disks)
4288                 rv = update_raid_disks(mddev, info->raid_disks);
4289
4290         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
4291                 if (mddev->pers->quiesce == NULL)
4292                         return -EINVAL;
4293                 if (mddev->recovery || mddev->sync_thread)
4294                         return -EBUSY;
4295                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
4296                         /* add the bitmap */
4297                         if (mddev->bitmap)
4298                                 return -EEXIST;
4299                         if (mddev->default_bitmap_offset == 0)
4300                                 return -EINVAL;
4301                         mddev->bitmap_offset = mddev->default_bitmap_offset;
4302                         mddev->pers->quiesce(mddev, 1);
4303                         rv = bitmap_create(mddev);
4304                         if (rv)
4305                                 bitmap_destroy(mddev);
4306                         mddev->pers->quiesce(mddev, 0);
4307                 } else {
4308                         /* remove the bitmap */
4309                         if (!mddev->bitmap)
4310                                 return -ENOENT;
4311                         if (mddev->bitmap->file)
4312                                 return -EINVAL;
4313                         mddev->pers->quiesce(mddev, 1);
4314                         bitmap_destroy(mddev);
4315                         mddev->pers->quiesce(mddev, 0);
4316                         mddev->bitmap_offset = 0;
4317                 }
4318         }
4319         md_update_sb(mddev, 1);
4320         return rv;
4321 }
4322
4323 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
4324 {
4325         mdk_rdev_t *rdev;
4326
4327         if (mddev->pers == NULL)
4328                 return -ENODEV;
4329
4330         rdev = find_rdev(mddev, dev);
4331         if (!rdev)
4332                 return -ENODEV;
4333
4334         md_error(mddev, rdev);
4335         return 0;
4336 }
4337
4338 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
4339 {
4340         mddev_t *mddev = bdev->bd_disk->private_data;
4341
4342         geo->heads = 2;
4343         geo->sectors = 4;
4344         geo->cylinders = get_capacity(mddev->gendisk) / 8;
4345         return 0;
4346 }
4347
4348 static int md_ioctl(struct inode *inode, struct file *file,
4349                         unsigned int cmd, unsigned long arg)
4350 {
4351         int err = 0;
4352         void __user *argp = (void __user *)arg;
4353         mddev_t *mddev = NULL;
4354
4355         if (!capable(CAP_SYS_ADMIN))
4356                 return -EACCES;
4357
4358         /*
4359          * Commands dealing with the RAID driver but not any
4360          * particular array:
4361          */
4362         switch (cmd)
4363         {
4364                 case RAID_VERSION:
4365                         err = get_version(argp);
4366                         goto done;
4367
4368                 case PRINT_RAID_DEBUG:
4369                         err = 0;
4370                         md_print_devices();
4371                         goto done;
4372
4373 #ifndef MODULE
4374                 case RAID_AUTORUN:
4375                         err = 0;
4376                         autostart_arrays(arg);
4377                         goto done;
4378 #endif
4379                 default:;
4380         }
4381
4382         /*
4383          * Commands creating/starting a new array:
4384          */
4385
4386         mddev = inode->i_bdev->bd_disk->private_data;
4387
4388         if (!mddev) {
4389                 BUG();
4390                 goto abort;
4391         }
4392
4393         err = mddev_lock(mddev);
4394         if (err) {
4395                 printk(KERN_INFO 
4396                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
4397                         err, cmd);
4398                 goto abort;
4399         }
4400
4401         switch (cmd)
4402         {
4403                 case SET_ARRAY_INFO:
4404                         {
4405                                 mdu_array_info_t info;
4406                                 if (!arg)
4407                                         memset(&info, 0, sizeof(info));
4408                                 else if (copy_from_user(&info, argp, sizeof(info))) {
4409                                         err = -EFAULT;
4410                                         goto abort_unlock;
4411                                 }
4412                                 if (mddev->pers) {
4413                                         err = update_array_info(mddev, &info);
4414                                         if (err) {
4415                                                 printk(KERN_WARNING "md: couldn't update"
4416                                                        " array info. %d\n", err);
4417                                                 goto abort_unlock;
4418                                         }
4419                                         goto done_unlock;
4420                                 }
4421                                 if (!list_empty(&mddev->disks)) {
4422                                         printk(KERN_WARNING
4423                                                "md: array %s already has disks!\n",
4424                                                mdname(mddev));
4425                                         err = -EBUSY;
4426                                         goto abort_unlock;
4427                                 }
4428                                 if (mddev->raid_disks) {
4429                                         printk(KERN_WARNING
4430                                                "md: array %s already initialised!\n",
4431                                                mdname(mddev));
4432                                         err = -EBUSY;
4433                                         goto abort_unlock;
4434                                 }
4435                                 err = set_array_info(mddev, &info);
4436                                 if (err) {
4437                                         printk(KERN_WARNING "md: couldn't set"
4438                                                " array info. %d\n", err);
4439                                         goto abort_unlock;
4440                                 }
4441                         }
4442                         goto done_unlock;
4443
4444                 default:;
4445         }
4446
4447         /*
4448          * Commands querying/configuring an existing array:
4449          */
4450         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
4451          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
4452         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
4453                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
4454                         && cmd != GET_BITMAP_FILE) {
4455                 err = -ENODEV;
4456                 goto abort_unlock;
4457         }
4458
4459         /*
4460          * Commands even a read-only array can execute:
4461          */
4462         switch (cmd)
4463         {
4464                 case GET_ARRAY_INFO:
4465                         err = get_array_info(mddev, argp);
4466                         goto done_unlock;
4467
4468                 case GET_BITMAP_FILE:
4469                         err = get_bitmap_file(mddev, argp);
4470                         goto done_unlock;
4471
4472                 case GET_DISK_INFO:
4473                         err = get_disk_info(mddev, argp);
4474                         goto done_unlock;
4475
4476                 case RESTART_ARRAY_RW:
4477                         err = restart_array(mddev);
4478                         goto done_unlock;
4479
4480                 case STOP_ARRAY:
4481                         err = do_md_stop (mddev, 0);
4482                         goto done_unlock;
4483
4484                 case STOP_ARRAY_RO:
4485                         err = do_md_stop (mddev, 1);
4486                         goto done_unlock;
4487
4488         /*
4489          * We have a problem here : there is no easy way to give a CHS
4490          * virtual geometry. We currently pretend that we have a 2 heads
4491          * 4 sectors (with a BIG number of cylinders...). This drives
4492          * dosfs just mad... ;-)
4493          */
4494         }
4495
4496         /*
4497          * The remaining ioctls are changing the state of the
4498          * superblock, so we do not allow them on read-only arrays.
4499          * However non-MD ioctls (e.g. get-size) will still come through
4500          * here and hit the 'default' below, so only disallow
4501          * 'md' ioctls, and switch to rw mode if started auto-readonly.
4502          */
4503         if (_IOC_TYPE(cmd) == MD_MAJOR &&
4504             mddev->ro && mddev->pers) {
4505                 if (mddev->ro == 2) {
4506                         mddev->ro = 0;
4507                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4508                 md_wakeup_thread(mddev->thread);
4509
4510                 } else {
4511                         err = -EROFS;
4512                         goto abort_unlock;
4513                 }
4514         }
4515
4516         switch (cmd)
4517         {
4518                 case ADD_NEW_DISK:
4519                 {
4520                         mdu_disk_info_t info;
4521                         if (copy_from_user(&info, argp, sizeof(info)))
4522                                 err = -EFAULT;
4523                         else
4524                                 err = add_new_disk(mddev, &info);
4525                         goto done_unlock;
4526                 }
4527
4528                 case HOT_REMOVE_DISK:
4529                         err = hot_remove_disk(mddev, new_decode_dev(arg));
4530                         goto done_unlock;
4531
4532                 case HOT_ADD_DISK:
4533                         err = hot_add_disk(mddev, new_decode_dev(arg));
4534                         goto done_unlock;
4535
4536                 case SET_DISK_FAULTY:
4537                         err = set_disk_faulty(mddev, new_decode_dev(arg));
4538                         goto done_unlock;
4539
4540                 case RUN_ARRAY:
4541                         err = do_md_run (mddev);
4542                         goto done_unlock;
4543
4544                 case SET_BITMAP_FILE:
4545                         err = set_bitmap_file(mddev, (int)arg);
4546                         goto done_unlock;
4547
4548                 default:
4549                         err = -EINVAL;
4550                         goto abort_unlock;
4551         }
4552
4553 done_unlock:
4554 abort_unlock:
4555         mddev_unlock(mddev);
4556
4557         return err;
4558 done:
4559         if (err)
4560                 MD_BUG();
4561 abort:
4562         return err;
4563 }
4564
4565 static int md_open(struct inode *inode, struct file *file)
4566 {
4567         /*
4568          * Succeed if we can lock the mddev, which confirms that
4569          * it isn't being stopped right now.
4570          */
4571         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4572         int err;
4573
4574         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
4575                 goto out;
4576
4577         err = 0;
4578         mddev_get(mddev);
4579         mddev_unlock(mddev);
4580
4581         check_disk_change(inode->i_bdev);
4582  out:
4583         return err;
4584 }
4585
4586 static int md_release(struct inode *inode, struct file * file)
4587 {
4588         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
4589
4590         BUG_ON(!mddev);
4591         mddev_put(mddev);
4592
4593         return 0;
4594 }
4595
4596 static int md_media_changed(struct gendisk *disk)
4597 {
4598         mddev_t *mddev = disk->private_data;
4599
4600         return mddev->changed;
4601 }
4602
4603 static int md_revalidate(struct gendisk *disk)
4604 {
4605         mddev_t *mddev = disk->private_data;
4606
4607         mddev->changed = 0;
4608         return 0;
4609 }
4610 static struct block_device_operations md_fops =
4611 {
4612         .owner          = THIS_MODULE,
4613         .open           = md_open,
4614         .release        = md_release,
4615         .ioctl          = md_ioctl,
4616         .getgeo         = md_getgeo,
4617         .media_changed  = md_media_changed,
4618         .revalidate_disk= md_revalidate,
4619 };
4620
4621 static int md_thread(void * arg)
4622 {
4623         mdk_thread_t *thread = arg;
4624
4625         /*
4626          * md_thread is a 'system-thread', it's priority should be very
4627          * high. We avoid resource deadlocks individually in each
4628          * raid personality. (RAID5 does preallocation) We also use RR and
4629          * the very same RT priority as kswapd, thus we will never get
4630          * into a priority inversion deadlock.
4631          *
4632          * we definitely have to have equal or higher priority than
4633          * bdflush, otherwise bdflush will deadlock if there are too
4634          * many dirty RAID5 blocks.
4635          */
4636
4637         current->flags |= PF_NOFREEZE;
4638         allow_signal(SIGKILL);
4639         while (!kthread_should_stop()) {
4640
4641                 /* We need to wait INTERRUPTIBLE so that
4642                  * we don't add to the load-average.
4643                  * That means we need to be sure no signals are
4644                  * pending
4645                  */
4646                 if (signal_pending(current))
4647                         flush_signals(current);
4648
4649                 wait_event_interruptible_timeout
4650                         (thread->wqueue,
4651                          test_bit(THREAD_WAKEUP, &thread->flags)
4652                          || kthread_should_stop(),
4653                          thread->timeout);
4654
4655                 clear_bit(THREAD_WAKEUP, &thread->flags);
4656
4657                 thread->run(thread->mddev);
4658         }
4659
4660         return 0;
4661 }
4662
4663 void md_wakeup_thread(mdk_thread_t *thread)
4664 {
4665         if (thread) {
4666                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4667                 set_bit(THREAD_WAKEUP, &thread->flags);
4668                 wake_up(&thread->wqueue);
4669         }
4670 }
4671
4672 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4673                                  const char *name)
4674 {
4675         mdk_thread_t *thread;
4676
4677         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4678         if (!thread)
4679                 return NULL;
4680
4681         init_waitqueue_head(&thread->wqueue);
4682
4683         thread->run = run;
4684         thread->mddev = mddev;
4685         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4686         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4687         if (IS_ERR(thread->tsk)) {
4688                 kfree(thread);
4689                 return NULL;
4690         }
4691         return thread;
4692 }
4693
4694 void md_unregister_thread(mdk_thread_t *thread)
4695 {
4696         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4697
4698         kthread_stop(thread->tsk);
4699         kfree(thread);
4700 }
4701
4702 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4703 {
4704         if (!mddev) {
4705                 MD_BUG();
4706                 return;
4707         }
4708
4709         if (!rdev || test_bit(Faulty, &rdev->flags))
4710                 return;
4711 /*
4712         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4713                 mdname(mddev),
4714                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4715                 __builtin_return_address(0),__builtin_return_address(1),
4716                 __builtin_return_address(2),__builtin_return_address(3));
4717 */
4718         if (!mddev->pers)
4719                 return;
4720         if (!mddev->pers->error_handler)
4721                 return;
4722         mddev->pers->error_handler(mddev,rdev);
4723         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4724         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4725         md_wakeup_thread(mddev->thread);
4726         md_new_event_inintr(mddev);
4727 }
4728
4729 /* seq_file implementation /proc/mdstat */
4730
4731 static void status_unused(struct seq_file *seq)
4732 {
4733         int i = 0;
4734         mdk_rdev_t *rdev;
4735         struct list_head *tmp;
4736
4737         seq_printf(seq, "unused devices: ");
4738
4739         ITERATE_RDEV_PENDING(rdev,tmp) {
4740                 char b[BDEVNAME_SIZE];
4741                 i++;
4742                 seq_printf(seq, "%s ",
4743                               bdevname(rdev->bdev,b));
4744         }
4745         if (!i)
4746                 seq_printf(seq, "<none>");
4747
4748         seq_printf(seq, "\n");
4749 }
4750
4751
4752 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4753 {
4754         sector_t max_blocks, resync, res;
4755         unsigned long dt, db, rt;
4756         int scale;
4757         unsigned int per_milli;
4758
4759         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4760
4761         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4762                 max_blocks = mddev->resync_max_sectors >> 1;
4763         else
4764                 max_blocks = mddev->size;
4765
4766         /*
4767          * Should not happen.
4768          */
4769         if (!max_blocks) {
4770                 MD_BUG();
4771                 return;
4772         }
4773         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4774          * in a sector_t, and (max_blocks>>scale) will fit in a
4775          * u32, as those are the requirements for sector_div.
4776          * Thus 'scale' must be at least 10
4777          */
4778         scale = 10;
4779         if (sizeof(sector_t) > sizeof(unsigned long)) {
4780                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4781                         scale++;
4782         }
4783         res = (resync>>scale)*1000;
4784         sector_div(res, (u32)((max_blocks>>scale)+1));
4785
4786         per_milli = res;
4787         {
4788                 int i, x = per_milli/50, y = 20-x;
4789                 seq_printf(seq, "[");
4790                 for (i = 0; i < x; i++)
4791                         seq_printf(seq, "=");
4792                 seq_printf(seq, ">");
4793                 for (i = 0; i < y; i++)
4794                         seq_printf(seq, ".");
4795                 seq_printf(seq, "] ");
4796         }
4797         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4798                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4799                     "reshape" :
4800                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
4801                      "check" :
4802                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4803                       "resync" : "recovery"))),
4804                    per_milli/10, per_milli % 10,
4805                    (unsigned long long) resync,
4806                    (unsigned long long) max_blocks);
4807
4808         /*
4809          * We do not want to overflow, so the order of operands and
4810          * the * 100 / 100 trick are important. We do a +1 to be
4811          * safe against division by zero. We only estimate anyway.
4812          *
4813          * dt: time from mark until now
4814          * db: blocks written from mark until now
4815          * rt: remaining time
4816          */
4817         dt = ((jiffies - mddev->resync_mark) / HZ);
4818         if (!dt) dt++;
4819         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
4820                 - mddev->resync_mark_cnt;
4821         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/2/100+1)))/100;
4822
4823         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4824
4825         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
4826 }
4827
4828 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4829 {
4830         struct list_head *tmp;
4831         loff_t l = *pos;
4832         mddev_t *mddev;
4833
4834         if (l >= 0x10000)
4835                 return NULL;
4836         if (!l--)
4837                 /* header */
4838                 return (void*)1;
4839
4840         spin_lock(&all_mddevs_lock);
4841         list_for_each(tmp,&all_mddevs)
4842                 if (!l--) {
4843                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4844                         mddev_get(mddev);
4845                         spin_unlock(&all_mddevs_lock);
4846                         return mddev;
4847                 }
4848         spin_unlock(&all_mddevs_lock);
4849         if (!l--)
4850                 return (void*)2;/* tail */
4851         return NULL;
4852 }
4853
4854 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4855 {
4856         struct list_head *tmp;
4857         mddev_t *next_mddev, *mddev = v;
4858         
4859         ++*pos;
4860         if (v == (void*)2)
4861                 return NULL;
4862
4863         spin_lock(&all_mddevs_lock);
4864         if (v == (void*)1)
4865                 tmp = all_mddevs.next;
4866         else
4867                 tmp = mddev->all_mddevs.next;
4868         if (tmp != &all_mddevs)
4869                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4870         else {
4871                 next_mddev = (void*)2;
4872                 *pos = 0x10000;
4873         }               
4874         spin_unlock(&all_mddevs_lock);
4875
4876         if (v != (void*)1)
4877                 mddev_put(mddev);
4878         return next_mddev;
4879
4880 }
4881
4882 static void md_seq_stop(struct seq_file *seq, void *v)
4883 {
4884         mddev_t *mddev = v;
4885
4886         if (mddev && v != (void*)1 && v != (void*)2)
4887                 mddev_put(mddev);
4888 }
4889
4890 struct mdstat_info {
4891         int event;
4892 };
4893
4894 static int md_seq_show(struct seq_file *seq, void *v)
4895 {
4896         mddev_t *mddev = v;
4897         sector_t size;
4898         struct list_head *tmp2;
4899         mdk_rdev_t *rdev;
4900         struct mdstat_info *mi = seq->private;
4901         struct bitmap *bitmap;
4902
4903         if (v == (void*)1) {
4904                 struct mdk_personality *pers;
4905                 seq_printf(seq, "Personalities : ");
4906                 spin_lock(&pers_lock);
4907                 list_for_each_entry(pers, &pers_list, list)
4908                         seq_printf(seq, "[%s] ", pers->name);
4909
4910                 spin_unlock(&pers_lock);
4911                 seq_printf(seq, "\n");
4912                 mi->event = atomic_read(&md_event_count);
4913                 return 0;
4914         }
4915         if (v == (void*)2) {
4916                 status_unused(seq);
4917                 return 0;
4918         }
4919
4920         if (mddev_lock(mddev) < 0)
4921                 return -EINTR;
4922
4923         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4924                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4925                                                 mddev->pers ? "" : "in");
4926                 if (mddev->pers) {
4927                         if (mddev->ro==1)
4928                                 seq_printf(seq, " (read-only)");
4929                         if (mddev->ro==2)
4930                                 seq_printf(seq, "(auto-read-only)");
4931                         seq_printf(seq, " %s", mddev->pers->name);
4932                 }
4933
4934                 size = 0;
4935                 ITERATE_RDEV(mddev,rdev,tmp2) {
4936                         char b[BDEVNAME_SIZE];
4937                         seq_printf(seq, " %s[%d]",
4938                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4939                         if (test_bit(WriteMostly, &rdev->flags))
4940                                 seq_printf(seq, "(W)");
4941                         if (test_bit(Faulty, &rdev->flags)) {
4942                                 seq_printf(seq, "(F)");
4943                                 continue;
4944                         } else if (rdev->raid_disk < 0)
4945                                 seq_printf(seq, "(S)"); /* spare */
4946                         size += rdev->size;
4947                 }
4948
4949                 if (!list_empty(&mddev->disks)) {
4950                         if (mddev->pers)
4951                                 seq_printf(seq, "\n      %llu blocks",
4952                                         (unsigned long long)mddev->array_size);
4953                         else
4954                                 seq_printf(seq, "\n      %llu blocks",
4955                                         (unsigned long long)size);
4956                 }
4957                 if (mddev->persistent) {
4958                         if (mddev->major_version != 0 ||
4959                             mddev->minor_version != 90) {
4960                                 seq_printf(seq," super %d.%d",
4961                                            mddev->major_version,
4962                                            mddev->minor_version);
4963                         }
4964                 } else
4965                         seq_printf(seq, " super non-persistent");
4966
4967                 if (mddev->pers) {
4968                         mddev->pers->status (seq, mddev);
4969                         seq_printf(seq, "\n      ");
4970                         if (mddev->pers->sync_request) {
4971                                 if (mddev->curr_resync > 2) {
4972                                         status_resync (seq, mddev);
4973                                         seq_printf(seq, "\n      ");
4974                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4975                                         seq_printf(seq, "\tresync=DELAYED\n      ");
4976                                 else if (mddev->recovery_cp < MaxSector)
4977                                         seq_printf(seq, "\tresync=PENDING\n      ");
4978                         }
4979                 } else
4980                         seq_printf(seq, "\n       ");
4981
4982                 if ((bitmap = mddev->bitmap)) {
4983                         unsigned long chunk_kb;
4984                         unsigned long flags;
4985                         spin_lock_irqsave(&bitmap->lock, flags);
4986                         chunk_kb = bitmap->chunksize >> 10;
4987                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4988                                 "%lu%s chunk",
4989                                 bitmap->pages - bitmap->missing_pages,
4990                                 bitmap->pages,
4991                                 (bitmap->pages - bitmap->missing_pages)
4992                                         << (PAGE_SHIFT - 10),
4993                                 chunk_kb ? chunk_kb : bitmap->chunksize,
4994                                 chunk_kb ? "KB" : "B");
4995                         if (bitmap->file) {
4996                                 seq_printf(seq, ", file: ");
4997                                 seq_path(seq, bitmap->file->f_path.mnt,
4998                                          bitmap->file->f_path.dentry," \t\n");
4999                         }
5000
5001                         seq_printf(seq, "\n");
5002                         spin_unlock_irqrestore(&bitmap->lock, flags);
5003                 }
5004
5005                 seq_printf(seq, "\n");
5006         }
5007         mddev_unlock(mddev);
5008         
5009         return 0;
5010 }
5011
5012 static struct seq_operations md_seq_ops = {
5013         .start  = md_seq_start,
5014         .next   = md_seq_next,
5015         .stop   = md_seq_stop,
5016         .show   = md_seq_show,
5017 };
5018
5019 static int md_seq_open(struct inode *inode, struct file *file)
5020 {
5021         int error;
5022         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5023         if (mi == NULL)
5024                 return -ENOMEM;
5025
5026         error = seq_open(file, &md_seq_ops);
5027         if (error)
5028                 kfree(mi);
5029         else {
5030                 struct seq_file *p = file->private_data;
5031                 p->private = mi;
5032                 mi->event = atomic_read(&md_event_count);
5033         }
5034         return error;
5035 }
5036
5037 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
5038 {
5039         struct seq_file *m = filp->private_data;
5040         struct mdstat_info *mi = m->private;
5041         int mask;
5042
5043         poll_wait(filp, &md_event_waiters, wait);
5044
5045         /* always allow read */
5046         mask = POLLIN | POLLRDNORM;
5047
5048         if (mi->event != atomic_read(&md_event_count))
5049                 mask |= POLLERR | POLLPRI;
5050         return mask;
5051 }
5052
5053 static const struct file_operations md_seq_fops = {
5054         .owner          = THIS_MODULE,
5055         .open           = md_seq_open,
5056         .read           = seq_read,
5057         .llseek         = seq_lseek,
5058         .release        = seq_release_private,
5059         .poll           = mdstat_poll,
5060 };
5061
5062 int register_md_personality(struct mdk_personality *p)
5063 {
5064         spin_lock(&pers_lock);
5065         list_add_tail(&p->list, &pers_list);
5066         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
5067         spin_unlock(&pers_lock);
5068         return 0;
5069 }
5070
5071 int unregister_md_personality(struct mdk_personality *p)
5072 {
5073         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
5074         spin_lock(&pers_lock);
5075         list_del_init(&p->list);
5076         spin_unlock(&pers_lock);
5077         return 0;
5078 }
5079
5080 static int is_mddev_idle(mddev_t *mddev)
5081 {
5082         mdk_rdev_t * rdev;
5083         struct list_head *tmp;
5084         int idle;
5085         unsigned long curr_events;
5086
5087         idle = 1;
5088         ITERATE_RDEV(mddev,rdev,tmp) {
5089                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
5090                 curr_events = disk_stat_read(disk, sectors[0]) + 
5091                                 disk_stat_read(disk, sectors[1]) - 
5092                                 atomic_read(&disk->sync_io);
5093                 /* The difference between curr_events and last_events
5094                  * will be affected by any new non-sync IO (making
5095                  * curr_events bigger) and any difference in the amount of
5096                  * in-flight syncio (making current_events bigger or smaller)
5097                  * The amount in-flight is currently limited to
5098                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
5099                  * which is at most 4096 sectors.
5100                  * These numbers are fairly fragile and should be made
5101                  * more robust, probably by enforcing the
5102                  * 'window size' that md_do_sync sort-of uses.
5103                  *
5104                  * Note: the following is an unsigned comparison.
5105                  */
5106                 if ((curr_events - rdev->last_events + 4096) > 8192) {
5107                         rdev->last_events = curr_events;
5108                         idle = 0;
5109                 }
5110         }
5111         return idle;
5112 }
5113
5114 void md_done_sync(mddev_t *mddev, int blocks, int ok)
5115 {
5116         /* another "blocks" (512byte) blocks have been synced */
5117         atomic_sub(blocks, &mddev->recovery_active);
5118         wake_up(&mddev->recovery_wait);
5119         if (!ok) {
5120                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5121                 md_wakeup_thread(mddev->thread);
5122                 // stop recovery, signal do_sync ....
5123         }
5124 }
5125
5126
5127 /* md_write_start(mddev, bi)
5128  * If we need to update some array metadata (e.g. 'active' flag
5129  * in superblock) before writing, schedule a superblock update
5130  * and wait for it to complete.
5131  */
5132 void md_write_start(mddev_t *mddev, struct bio *bi)
5133 {
5134         if (bio_data_dir(bi) != WRITE)
5135                 return;
5136
5137         BUG_ON(mddev->ro == 1);
5138         if (mddev->ro == 2) {
5139                 /* need to switch to read/write */
5140                 mddev->ro = 0;
5141                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5142                 md_wakeup_thread(mddev->thread);
5143         }
5144         atomic_inc(&mddev->writes_pending);
5145         if (mddev->in_sync) {
5146                 spin_lock_irq(&mddev->write_lock);
5147                 if (mddev->in_sync) {
5148                         mddev->in_sync = 0;
5149                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5150                         md_wakeup_thread(mddev->thread);
5151                 }
5152                 spin_unlock_irq(&mddev->write_lock);
5153         }
5154         wait_event(mddev->sb_wait, mddev->flags==0);
5155 }
5156
5157 void md_write_end(mddev_t *mddev)
5158 {
5159         if (atomic_dec_and_test(&mddev->writes_pending)) {
5160                 if (mddev->safemode == 2)
5161                         md_wakeup_thread(mddev->thread);
5162                 else if (mddev->safemode_delay)
5163                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
5164         }
5165 }
5166
5167 /* md_allow_write(mddev)
5168  * Calling this ensures that the array is marked 'active' so that writes
5169  * may proceed without blocking.  It is important to call this before
5170  * attempting a GFP_KERNEL allocation while holding the mddev lock.
5171  * Must be called with mddev_lock held.
5172  */
5173 void md_allow_write(mddev_t *mddev)
5174 {
5175         if (!mddev->pers)
5176                 return;
5177         if (mddev->ro)
5178                 return;
5179
5180         spin_lock_irq(&mddev->write_lock);
5181         if (mddev->in_sync) {
5182                 mddev->in_sync = 0;
5183                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5184                 if (mddev->safemode_delay &&
5185                     mddev->safemode == 0)
5186                         mddev->safemode = 1;
5187                 spin_unlock_irq(&mddev->write_lock);
5188                 md_update_sb(mddev, 0);
5189         } else
5190                 spin_unlock_irq(&mddev->write_lock);
5191 }
5192 EXPORT_SYMBOL_GPL(md_allow_write);
5193
5194 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
5195
5196 #define SYNC_MARKS      10
5197 #define SYNC_MARK_STEP  (3*HZ)
5198 void md_do_sync(mddev_t *mddev)
5199 {
5200         mddev_t *mddev2;
5201         unsigned int currspeed = 0,
5202                  window;
5203         sector_t max_sectors,j, io_sectors;
5204         unsigned long mark[SYNC_MARKS];
5205         sector_t mark_cnt[SYNC_MARKS];
5206         int last_mark,m;
5207         struct list_head *tmp;
5208         sector_t last_check;
5209         int skipped = 0;
5210         struct list_head *rtmp;
5211         mdk_rdev_t *rdev;
5212         char *desc;
5213
5214         /* just incase thread restarts... */
5215         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
5216                 return;
5217         if (mddev->ro) /* never try to sync a read-only array */
5218                 return;
5219
5220         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5221                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
5222                         desc = "data-check";
5223                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5224                         desc = "requested-resync";
5225                 else
5226                         desc = "resync";
5227         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5228                 desc = "reshape";
5229         else
5230                 desc = "recovery";
5231
5232         /* we overload curr_resync somewhat here.
5233          * 0 == not engaged in resync at all
5234          * 2 == checking that there is no conflict with another sync
5235          * 1 == like 2, but have yielded to allow conflicting resync to
5236          *              commense
5237          * other == active in resync - this many blocks
5238          *
5239          * Before starting a resync we must have set curr_resync to
5240          * 2, and then checked that every "conflicting" array has curr_resync
5241          * less than ours.  When we find one that is the same or higher
5242          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
5243          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
5244          * This will mean we have to start checking from the beginning again.
5245          *
5246          */
5247
5248         do {
5249                 mddev->curr_resync = 2;
5250
5251         try_again:
5252                 if (kthread_should_stop()) {
5253                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5254                         goto skip;
5255                 }
5256                 ITERATE_MDDEV(mddev2,tmp) {
5257                         if (mddev2 == mddev)
5258                                 continue;
5259                         if (mddev2->curr_resync && 
5260                             match_mddev_units(mddev,mddev2)) {
5261                                 DEFINE_WAIT(wq);
5262                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
5263                                         /* arbitrarily yield */
5264                                         mddev->curr_resync = 1;
5265                                         wake_up(&resync_wait);
5266                                 }
5267                                 if (mddev > mddev2 && mddev->curr_resync == 1)
5268                                         /* no need to wait here, we can wait the next
5269                                          * time 'round when curr_resync == 2
5270                                          */
5271                                         continue;
5272                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
5273                                 if (!kthread_should_stop() &&
5274                                     mddev2->curr_resync >= mddev->curr_resync) {
5275                                         printk(KERN_INFO "md: delaying %s of %s"
5276                                                " until %s has finished (they"
5277                                                " share one or more physical units)\n",
5278                                                desc, mdname(mddev), mdname(mddev2));
5279                                         mddev_put(mddev2);
5280                                         schedule();
5281                                         finish_wait(&resync_wait, &wq);
5282                                         goto try_again;
5283                                 }
5284                                 finish_wait(&resync_wait, &wq);
5285                         }
5286                 }
5287         } while (mddev->curr_resync < 2);
5288
5289         j = 0;
5290         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5291                 /* resync follows the size requested by the personality,
5292                  * which defaults to physical size, but can be virtual size
5293                  */
5294                 max_sectors = mddev->resync_max_sectors;
5295                 mddev->resync_mismatches = 0;
5296                 /* we don't use the checkpoint if there's a bitmap */
5297                 if (!mddev->bitmap &&
5298                     !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
5299                         j = mddev->recovery_cp;
5300         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
5301                 max_sectors = mddev->size << 1;
5302         else {
5303                 /* recovery follows the physical size of devices */
5304                 max_sectors = mddev->size << 1;
5305                 j = MaxSector;
5306                 ITERATE_RDEV(mddev,rdev,rtmp)
5307                         if (rdev->raid_disk >= 0 &&
5308                             !test_bit(Faulty, &rdev->flags) &&
5309                             !test_bit(In_sync, &rdev->flags) &&
5310                             rdev->recovery_offset < j)
5311                                 j = rdev->recovery_offset;
5312         }
5313
5314         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
5315         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
5316                 " %d KB/sec/disk.\n", speed_min(mddev));
5317         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
5318                "(but not more than %d KB/sec) for %s.\n",
5319                speed_max(mddev), desc);
5320
5321         is_mddev_idle(mddev); /* this also initializes IO event counters */
5322
5323         io_sectors = 0;
5324         for (m = 0; m < SYNC_MARKS; m++) {
5325                 mark[m] = jiffies;
5326                 mark_cnt[m] = io_sectors;
5327         }
5328         last_mark = 0;
5329         mddev->resync_mark = mark[last_mark];
5330         mddev->resync_mark_cnt = mark_cnt[last_mark];
5331
5332         /*
5333          * Tune reconstruction:
5334          */
5335         window = 32*(PAGE_SIZE/512);
5336         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
5337                 window/2,(unsigned long long) max_sectors/2);
5338
5339         atomic_set(&mddev->recovery_active, 0);
5340         init_waitqueue_head(&mddev->recovery_wait);
5341         last_check = 0;
5342
5343         if (j>2) {
5344                 printk(KERN_INFO 
5345                        "md: resuming %s of %s from checkpoint.\n",
5346                        desc, mdname(mddev));
5347                 mddev->curr_resync = j;
5348         }
5349
5350         while (j < max_sectors) {
5351                 sector_t sectors;
5352
5353                 skipped = 0;
5354                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
5355                                             currspeed < speed_min(mddev));
5356                 if (sectors == 0) {
5357                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
5358                         goto out;
5359                 }
5360
5361                 if (!skipped) { /* actual IO requested */
5362                         io_sectors += sectors;
5363                         atomic_add(sectors, &mddev->recovery_active);
5364                 }
5365
5366                 j += sectors;
5367                 if (j>1) mddev->curr_resync = j;
5368                 mddev->curr_mark_cnt = io_sectors;
5369                 if (last_check == 0)
5370                         /* this is the earliers that rebuilt will be
5371                          * visible in /proc/mdstat
5372                          */
5373                         md_new_event(mddev);
5374
5375                 if (last_check + window > io_sectors || j == max_sectors)
5376                         continue;
5377
5378                 last_check = io_sectors;
5379
5380                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
5381                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
5382                         break;
5383
5384         repeat:
5385                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
5386                         /* step marks */
5387                         int next = (last_mark+1) % SYNC_MARKS;
5388
5389                         mddev->resync_mark = mark[next];
5390                         mddev->resync_mark_cnt = mark_cnt[next];
5391                         mark[next] = jiffies;
5392                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
5393                         last_mark = next;
5394                 }
5395
5396
5397                 if (kthread_should_stop()) {
5398                         /*
5399                          * got a signal, exit.
5400                          */
5401                         printk(KERN_INFO 
5402                                 "md: md_do_sync() got signal ... exiting\n");
5403                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5404                         goto out;
5405                 }
5406
5407                 /*
5408                  * this loop exits only if either when we are slower than
5409                  * the 'hard' speed limit, or the system was IO-idle for
5410                  * a jiffy.
5411                  * the system might be non-idle CPU-wise, but we only care
5412                  * about not overloading the IO subsystem. (things like an
5413                  * e2fsck being done on the RAID array should execute fast)
5414                  */
5415                 mddev->queue->unplug_fn(mddev->queue);
5416                 cond_resched();
5417
5418                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
5419                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
5420
5421                 if (currspeed > speed_min(mddev)) {
5422                         if ((currspeed > speed_max(mddev)) ||
5423                                         !is_mddev_idle(mddev)) {
5424                                 msleep(500);
5425                                 goto repeat;
5426                         }
5427                 }
5428         }
5429         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
5430         /*
5431          * this also signals 'finished resyncing' to md_stop
5432          */
5433  out:
5434         mddev->queue->unplug_fn(mddev->queue);
5435
5436         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
5437
5438         /* tell personality that we are finished */
5439         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
5440
5441         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5442             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
5443             mddev->curr_resync > 2) {
5444                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
5445                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5446                                 if (mddev->curr_resync >= mddev->recovery_cp) {
5447                                         printk(KERN_INFO
5448                                                "md: checkpointing %s of %s.\n",
5449                                                desc, mdname(mddev));
5450                                         mddev->recovery_cp = mddev->curr_resync;
5451                                 }
5452                         } else
5453                                 mddev->recovery_cp = MaxSector;
5454                 } else {
5455                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
5456                                 mddev->curr_resync = MaxSector;
5457                         ITERATE_RDEV(mddev,rdev,rtmp)
5458                                 if (rdev->raid_disk >= 0 &&
5459                                     !test_bit(Faulty, &rdev->flags) &&
5460                                     !test_bit(In_sync, &rdev->flags) &&
5461                                     rdev->recovery_offset < mddev->curr_resync)
5462                                         rdev->recovery_offset = mddev->curr_resync;
5463                 }
5464         }
5465         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5466
5467  skip:
5468         mddev->curr_resync = 0;
5469         wake_up(&resync_wait);
5470         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
5471         md_wakeup_thread(mddev->thread);
5472 }
5473 EXPORT_SYMBOL_GPL(md_do_sync);
5474
5475
5476 static int remove_and_add_spares(mddev_t *mddev)
5477 {
5478         mdk_rdev_t *rdev;
5479         struct list_head *rtmp;
5480         int spares = 0;
5481
5482         ITERATE_RDEV(mddev,rdev,rtmp)
5483                 if (rdev->raid_disk >= 0 &&
5484                     (test_bit(Faulty, &rdev->flags) ||
5485                      ! test_bit(In_sync, &rdev->flags)) &&
5486                     atomic_read(&rdev->nr_pending)==0) {
5487                         if (mddev->pers->hot_remove_disk(
5488                                     mddev, rdev->raid_disk)==0) {
5489                                 char nm[20];
5490                                 sprintf(nm,"rd%d", rdev->raid_disk);
5491                                 sysfs_remove_link(&mddev->kobj, nm);
5492                                 rdev->raid_disk = -1;
5493                         }
5494                 }
5495
5496         if (mddev->degraded) {
5497                 ITERATE_RDEV(mddev,rdev,rtmp)
5498                         if (rdev->raid_disk < 0
5499                             && !test_bit(Faulty, &rdev->flags)) {
5500                                 rdev->recovery_offset = 0;
5501                                 if (mddev->pers->hot_add_disk(mddev,rdev)) {
5502                                         char nm[20];
5503                                         sprintf(nm, "rd%d", rdev->raid_disk);
5504                                         if (sysfs_create_link(&mddev->kobj,
5505                                                               &rdev->kobj, nm))
5506                                                 printk(KERN_WARNING
5507                                                        "md: cannot register "
5508                                                        "%s for %s\n",
5509                                                        nm, mdname(mddev));
5510                                         spares++;
5511                                         md_new_event(mddev);
5512                                 } else
5513                                         break;
5514                         }
5515         }
5516         return spares;
5517 }
5518 /*
5519  * This routine is regularly called by all per-raid-array threads to
5520  * deal with generic issues like resync and super-block update.
5521  * Raid personalities that don't have a thread (linear/raid0) do not
5522  * need this as they never do any recovery or update the superblock.
5523  *
5524  * It does not do any resync itself, but rather "forks" off other threads
5525  * to do that as needed.
5526  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
5527  * "->recovery" and create a thread at ->sync_thread.
5528  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
5529  * and wakeups up this thread which will reap the thread and finish up.
5530  * This thread also removes any faulty devices (with nr_pending == 0).
5531  *
5532  * The overall approach is:
5533  *  1/ if the superblock needs updating, update it.
5534  *  2/ If a recovery thread is running, don't do anything else.
5535  *  3/ If recovery has finished, clean up, possibly marking spares active.
5536  *  4/ If there are any faulty devices, remove them.
5537  *  5/ If array is degraded, try to add spares devices
5538  *  6/ If array has spares or is not in-sync, start a resync thread.
5539  */
5540 void md_check_recovery(mddev_t *mddev)
5541 {
5542         mdk_rdev_t *rdev;
5543         struct list_head *rtmp;
5544
5545
5546         if (mddev->bitmap)
5547                 bitmap_daemon_work(mddev->bitmap);
5548
5549         if (mddev->ro)
5550                 return;
5551
5552         if (signal_pending(current)) {
5553                 if (mddev->pers->sync_request) {
5554                         printk(KERN_INFO "md: %s in immediate safe mode\n",
5555                                mdname(mddev));
5556                         mddev->safemode = 2;
5557                 }
5558                 flush_signals(current);
5559         }
5560
5561         if ( ! (
5562                 mddev->flags ||
5563                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
5564                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
5565                 (mddev->safemode == 1) ||
5566                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
5567                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
5568                 ))
5569                 return;
5570
5571         if (mddev_trylock(mddev)) {
5572                 int spares = 0;
5573
5574                 spin_lock_irq(&mddev->write_lock);
5575                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
5576                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
5577                         mddev->in_sync = 1;
5578                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
5579                 }
5580                 if (mddev->safemode == 1)
5581                         mddev->safemode = 0;
5582                 spin_unlock_irq(&mddev->write_lock);
5583
5584                 if (mddev->flags)
5585                         md_update_sb(mddev, 0);
5586
5587
5588                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
5589                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
5590                         /* resync/recovery still happening */
5591                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5592                         goto unlock;
5593                 }
5594                 if (mddev->sync_thread) {
5595                         /* resync has finished, collect result */
5596                         md_unregister_thread(mddev->sync_thread);
5597                         mddev->sync_thread = NULL;
5598                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
5599                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
5600                                 /* success...*/
5601                                 /* activate any spares */
5602                                 mddev->pers->spare_active(mddev);
5603                         }
5604                         md_update_sb(mddev, 1);
5605
5606                         /* if array is no-longer degraded, then any saved_raid_disk
5607                          * information must be scrapped
5608                          */
5609                         if (!mddev->degraded)
5610                                 ITERATE_RDEV(mddev,rdev,rtmp)
5611                                         rdev->saved_raid_disk = -1;
5612
5613                         mddev->recovery = 0;
5614                         /* flag recovery needed just to double check */
5615                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5616                         md_new_event(mddev);
5617                         goto unlock;
5618                 }
5619                 /* Clear some bits that don't mean anything, but
5620                  * might be left set
5621                  */
5622                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5623                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
5624                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
5625                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
5626
5627                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
5628                         goto unlock;
5629                 /* no recovery is running.
5630                  * remove any failed drives, then
5631                  * add spares if possible.
5632                  * Spare are also removed and re-added, to allow
5633                  * the personality to fail the re-add.
5634                  */
5635
5636                 if (mddev->reshape_position != MaxSector) {
5637                         if (mddev->pers->check_reshape(mddev) != 0)
5638                                 /* Cannot proceed */
5639                                 goto unlock;
5640                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
5641                 } else if ((spares = remove_and_add_spares(mddev))) {
5642                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5643                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
5644                 } else if (mddev->recovery_cp < MaxSector) {
5645                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
5646                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5647                         /* nothing to be done ... */
5648                         goto unlock;
5649
5650                 if (mddev->pers->sync_request) {
5651                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
5652                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
5653                                 /* We are adding a device or devices to an array
5654                                  * which has the bitmap stored on all devices.
5655                                  * So make sure all bitmap pages get written
5656                                  */
5657                                 bitmap_write_all(mddev->bitmap);
5658                         }
5659                         mddev->sync_thread = md_register_thread(md_do_sync,
5660                                                                 mddev,
5661                                                                 "%s_resync");
5662                         if (!mddev->sync_thread) {
5663                                 printk(KERN_ERR "%s: could not start resync"
5664                                         " thread...\n", 
5665                                         mdname(mddev));
5666                                 /* leave the spares where they are, it shouldn't hurt */
5667                                 mddev->recovery = 0;
5668                         } else
5669                                 md_wakeup_thread(mddev->sync_thread);
5670                         md_new_event(mddev);
5671                 }
5672         unlock:
5673                 mddev_unlock(mddev);
5674         }
5675 }
5676
5677 static int md_notify_reboot(struct notifier_block *this,
5678                             unsigned long code, void *x)
5679 {
5680         struct list_head *tmp;
5681         mddev_t *mddev;
5682
5683         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
5684
5685                 printk(KERN_INFO "md: stopping all md devices.\n");
5686
5687                 ITERATE_MDDEV(mddev,tmp)
5688                         if (mddev_trylock(mddev)) {
5689                                 do_md_stop (mddev, 1);
5690                                 mddev_unlock(mddev);
5691                         }
5692                 /*
5693                  * certain more exotic SCSI devices are known to be
5694                  * volatile wrt too early system reboots. While the
5695                  * right place to handle this issue is the given
5696                  * driver, we do want to have a safe RAID driver ...
5697                  */
5698                 mdelay(1000*1);
5699         }
5700         return NOTIFY_DONE;
5701 }
5702
5703 static struct notifier_block md_notifier = {
5704         .notifier_call  = md_notify_reboot,
5705         .next           = NULL,
5706         .priority       = INT_MAX, /* before any real devices */
5707 };
5708
5709 static void md_geninit(void)
5710 {
5711         struct proc_dir_entry *p;
5712
5713         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
5714
5715         p = create_proc_entry("mdstat", S_IRUGO, NULL);
5716         if (p)
5717                 p->proc_fops = &md_seq_fops;
5718 }
5719
5720 static int __init md_init(void)
5721 {
5722         if (register_blkdev(MAJOR_NR, "md"))
5723                 return -1;
5724         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5725                 unregister_blkdev(MAJOR_NR, "md");
5726                 return -1;
5727         }
5728         blk_register_region(MKDEV(MAJOR_NR, 0), 1UL<<MINORBITS, THIS_MODULE,
5729                             md_probe, NULL, NULL);
5730         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
5731                             md_probe, NULL, NULL);
5732
5733         register_reboot_notifier(&md_notifier);
5734         raid_table_header = register_sysctl_table(raid_root_table);
5735
5736         md_geninit();
5737         return (0);
5738 }
5739
5740
5741 #ifndef MODULE
5742
5743 /*
5744  * Searches all registered partitions for autorun RAID arrays
5745  * at boot time.
5746  */
5747 static dev_t detected_devices[128];
5748 static int dev_cnt;
5749
5750 void md_autodetect_dev(dev_t dev)
5751 {
5752         if (dev_cnt >= 0 && dev_cnt < 127)
5753                 detected_devices[dev_cnt++] = dev;
5754 }
5755
5756
5757 static void autostart_arrays(int part)
5758 {
5759         mdk_rdev_t *rdev;
5760         int i;
5761
5762         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5763
5764         for (i = 0; i < dev_cnt; i++) {
5765                 dev_t dev = detected_devices[i];
5766
5767                 rdev = md_import_device(dev,0, 0);
5768                 if (IS_ERR(rdev))
5769                         continue;
5770
5771                 if (test_bit(Faulty, &rdev->flags)) {
5772                         MD_BUG();
5773                         continue;
5774                 }
5775                 list_add(&rdev->same_set, &pending_raid_disks);
5776         }
5777         dev_cnt = 0;
5778
5779         autorun_devices(part);
5780 }
5781
5782 #endif /* !MODULE */
5783
5784 static __exit void md_exit(void)
5785 {
5786         mddev_t *mddev;
5787         struct list_head *tmp;
5788
5789         blk_unregister_region(MKDEV(MAJOR_NR,0), 1U << MINORBITS);
5790         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
5791
5792         unregister_blkdev(MAJOR_NR,"md");
5793         unregister_blkdev(mdp_major, "mdp");
5794         unregister_reboot_notifier(&md_notifier);
5795         unregister_sysctl_table(raid_table_header);
5796         remove_proc_entry("mdstat", NULL);
5797         ITERATE_MDDEV(mddev,tmp) {
5798                 struct gendisk *disk = mddev->gendisk;
5799                 if (!disk)
5800                         continue;
5801                 export_array(mddev);
5802                 del_gendisk(disk);
5803                 put_disk(disk);
5804                 mddev->gendisk = NULL;
5805                 mddev_put(mddev);
5806         }
5807 }
5808
5809 module_init(md_init)
5810 module_exit(md_exit)
5811
5812 static int get_ro(char *buffer, struct kernel_param *kp)
5813 {
5814         return sprintf(buffer, "%d", start_readonly);
5815 }
5816 static int set_ro(const char *val, struct kernel_param *kp)
5817 {
5818         char *e;
5819         int num = simple_strtoul(val, &e, 10);
5820         if (*val && (*e == '\0' || *e == '\n')) {
5821                 start_readonly = num;
5822                 return 0;
5823         }
5824         return -EINVAL;
5825 }
5826
5827 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
5828 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
5829
5830
5831 EXPORT_SYMBOL(register_md_personality);
5832 EXPORT_SYMBOL(unregister_md_personality);
5833 EXPORT_SYMBOL(md_error);
5834 EXPORT_SYMBOL(md_done_sync);
5835 EXPORT_SYMBOL(md_write_start);
5836 EXPORT_SYMBOL(md_write_end);
5837 EXPORT_SYMBOL(md_register_thread);
5838 EXPORT_SYMBOL(md_unregister_thread);
5839 EXPORT_SYMBOL(md_wakeup_thread);
5840 EXPORT_SYMBOL(md_check_recovery);
5841 MODULE_LICENSE("GPL");
5842 MODULE_ALIAS("md");
5843 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);