md: Use revalidate_disk to effect changes in size of device.
[linux-2.6] / drivers / md / md.c
1 /*
2    md.c : Multiple Devices driver for Linux
3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5      completely rewritten, based on the MD driver code from Marc Zyngier
6
7    Changes:
8
9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13    - kmod support by: Cyrus Durgin
14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19
20      Neil Brown <neilb@cse.unsw.edu.au>.
21
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25    This program is free software; you can redistribute it and/or modify
26    it under the terms of the GNU General Public License as published by
27    the Free Software Foundation; either version 2, or (at your option)
28    any later version.
29
30    You should have received a copy of the GNU General Public License
31    (for example /usr/src/linux/COPYING); if not, write to the Free
32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/buffer_head.h> /* for invalidate_bdev */
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/hdreg.h>
43 #include <linux/proc_fs.h>
44 #include <linux/random.h>
45 #include <linux/reboot.h>
46 #include <linux/file.h>
47 #include <linux/delay.h>
48 #include <linux/raid/md_p.h>
49 #include <linux/raid/md_u.h>
50 #include "md.h"
51 #include "bitmap.h"
52
53 #define DEBUG 0
54 #define dprintk(x...) ((void)(DEBUG && printk(x)))
55
56
57 #ifndef MODULE
58 static void autostart_arrays(int part);
59 #endif
60
61 static LIST_HEAD(pers_list);
62 static DEFINE_SPINLOCK(pers_lock);
63
64 static void md_print_devices(void);
65
66 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
67
68 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
69
70 /*
71  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
72  * is 1000 KB/sec, so the extra system load does not show up that much.
73  * Increase it if you want to have more _guaranteed_ speed. Note that
74  * the RAID driver will use the maximum available bandwidth if the IO
75  * subsystem is idle. There is also an 'absolute maximum' reconstruction
76  * speed limit - in case reconstruction slows down your system despite
77  * idle IO detection.
78  *
79  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
80  * or /sys/block/mdX/md/sync_speed_{min,max}
81  */
82
83 static int sysctl_speed_limit_min = 1000;
84 static int sysctl_speed_limit_max = 200000;
85 static inline int speed_min(mddev_t *mddev)
86 {
87         return mddev->sync_speed_min ?
88                 mddev->sync_speed_min : sysctl_speed_limit_min;
89 }
90
91 static inline int speed_max(mddev_t *mddev)
92 {
93         return mddev->sync_speed_max ?
94                 mddev->sync_speed_max : sysctl_speed_limit_max;
95 }
96
97 static struct ctl_table_header *raid_table_header;
98
99 static ctl_table raid_table[] = {
100         {
101                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
102                 .procname       = "speed_limit_min",
103                 .data           = &sysctl_speed_limit_min,
104                 .maxlen         = sizeof(int),
105                 .mode           = S_IRUGO|S_IWUSR,
106                 .proc_handler   = &proc_dointvec,
107         },
108         {
109                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
110                 .procname       = "speed_limit_max",
111                 .data           = &sysctl_speed_limit_max,
112                 .maxlen         = sizeof(int),
113                 .mode           = S_IRUGO|S_IWUSR,
114                 .proc_handler   = &proc_dointvec,
115         },
116         { .ctl_name = 0 }
117 };
118
119 static ctl_table raid_dir_table[] = {
120         {
121                 .ctl_name       = DEV_RAID,
122                 .procname       = "raid",
123                 .maxlen         = 0,
124                 .mode           = S_IRUGO|S_IXUGO,
125                 .child          = raid_table,
126         },
127         { .ctl_name = 0 }
128 };
129
130 static ctl_table raid_root_table[] = {
131         {
132                 .ctl_name       = CTL_DEV,
133                 .procname       = "dev",
134                 .maxlen         = 0,
135                 .mode           = 0555,
136                 .child          = raid_dir_table,
137         },
138         { .ctl_name = 0 }
139 };
140
141 static struct block_device_operations md_fops;
142
143 static int start_readonly;
144
145 /*
146  * We have a system wide 'event count' that is incremented
147  * on any 'interesting' event, and readers of /proc/mdstat
148  * can use 'poll' or 'select' to find out when the event
149  * count increases.
150  *
151  * Events are:
152  *  start array, stop array, error, add device, remove device,
153  *  start build, activate spare
154  */
155 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
156 static atomic_t md_event_count;
157 void md_new_event(mddev_t *mddev)
158 {
159         atomic_inc(&md_event_count);
160         wake_up(&md_event_waiters);
161 }
162 EXPORT_SYMBOL_GPL(md_new_event);
163
164 /* Alternate version that can be called from interrupts
165  * when calling sysfs_notify isn't needed.
166  */
167 static void md_new_event_inintr(mddev_t *mddev)
168 {
169         atomic_inc(&md_event_count);
170         wake_up(&md_event_waiters);
171 }
172
173 /*
174  * Enables to iterate over all existing md arrays
175  * all_mddevs_lock protects this list.
176  */
177 static LIST_HEAD(all_mddevs);
178 static DEFINE_SPINLOCK(all_mddevs_lock);
179
180
181 /*
182  * iterates through all used mddevs in the system.
183  * We take care to grab the all_mddevs_lock whenever navigating
184  * the list, and to always hold a refcount when unlocked.
185  * Any code which breaks out of this loop while own
186  * a reference to the current mddev and must mddev_put it.
187  */
188 #define for_each_mddev(mddev,tmp)                                       \
189                                                                         \
190         for (({ spin_lock(&all_mddevs_lock);                            \
191                 tmp = all_mddevs.next;                                  \
192                 mddev = NULL;});                                        \
193              ({ if (tmp != &all_mddevs)                                 \
194                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
195                 spin_unlock(&all_mddevs_lock);                          \
196                 if (mddev) mddev_put(mddev);                            \
197                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
198                 tmp != &all_mddevs;});                                  \
199              ({ spin_lock(&all_mddevs_lock);                            \
200                 tmp = tmp->next;})                                      \
201                 )
202
203
204 /* Rather than calling directly into the personality make_request function,
205  * IO requests come here first so that we can check if the device is
206  * being suspended pending a reconfiguration.
207  * We hold a refcount over the call to ->make_request.  By the time that
208  * call has finished, the bio has been linked into some internal structure
209  * and so is visible to ->quiesce(), so we don't need the refcount any more.
210  */
211 static int md_make_request(struct request_queue *q, struct bio *bio)
212 {
213         mddev_t *mddev = q->queuedata;
214         int rv;
215         if (mddev == NULL || mddev->pers == NULL) {
216                 bio_io_error(bio);
217                 return 0;
218         }
219         rcu_read_lock();
220         if (mddev->suspended) {
221                 DEFINE_WAIT(__wait);
222                 for (;;) {
223                         prepare_to_wait(&mddev->sb_wait, &__wait,
224                                         TASK_UNINTERRUPTIBLE);
225                         if (!mddev->suspended)
226                                 break;
227                         rcu_read_unlock();
228                         schedule();
229                         rcu_read_lock();
230                 }
231                 finish_wait(&mddev->sb_wait, &__wait);
232         }
233         atomic_inc(&mddev->active_io);
234         rcu_read_unlock();
235         rv = mddev->pers->make_request(q, bio);
236         if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
237                 wake_up(&mddev->sb_wait);
238
239         return rv;
240 }
241
242 static void mddev_suspend(mddev_t *mddev)
243 {
244         BUG_ON(mddev->suspended);
245         mddev->suspended = 1;
246         synchronize_rcu();
247         wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
248         mddev->pers->quiesce(mddev, 1);
249         md_unregister_thread(mddev->thread);
250         mddev->thread = NULL;
251         /* we now know that no code is executing in the personality module,
252          * except possibly the tail end of a ->bi_end_io function, but that
253          * is certain to complete before the module has a chance to get
254          * unloaded
255          */
256 }
257
258 static void mddev_resume(mddev_t *mddev)
259 {
260         mddev->suspended = 0;
261         wake_up(&mddev->sb_wait);
262         mddev->pers->quiesce(mddev, 0);
263 }
264
265
266 static inline mddev_t *mddev_get(mddev_t *mddev)
267 {
268         atomic_inc(&mddev->active);
269         return mddev;
270 }
271
272 static void mddev_delayed_delete(struct work_struct *ws);
273
274 static void mddev_put(mddev_t *mddev)
275 {
276         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
277                 return;
278         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
279             !mddev->hold_active) {
280                 list_del(&mddev->all_mddevs);
281                 if (mddev->gendisk) {
282                         /* we did a probe so need to clean up.
283                          * Call schedule_work inside the spinlock
284                          * so that flush_scheduled_work() after
285                          * mddev_find will succeed in waiting for the
286                          * work to be done.
287                          */
288                         INIT_WORK(&mddev->del_work, mddev_delayed_delete);
289                         schedule_work(&mddev->del_work);
290                 } else
291                         kfree(mddev);
292         }
293         spin_unlock(&all_mddevs_lock);
294 }
295
296 static mddev_t * mddev_find(dev_t unit)
297 {
298         mddev_t *mddev, *new = NULL;
299
300  retry:
301         spin_lock(&all_mddevs_lock);
302
303         if (unit) {
304                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
305                         if (mddev->unit == unit) {
306                                 mddev_get(mddev);
307                                 spin_unlock(&all_mddevs_lock);
308                                 kfree(new);
309                                 return mddev;
310                         }
311
312                 if (new) {
313                         list_add(&new->all_mddevs, &all_mddevs);
314                         spin_unlock(&all_mddevs_lock);
315                         new->hold_active = UNTIL_IOCTL;
316                         return new;
317                 }
318         } else if (new) {
319                 /* find an unused unit number */
320                 static int next_minor = 512;
321                 int start = next_minor;
322                 int is_free = 0;
323                 int dev = 0;
324                 while (!is_free) {
325                         dev = MKDEV(MD_MAJOR, next_minor);
326                         next_minor++;
327                         if (next_minor > MINORMASK)
328                                 next_minor = 0;
329                         if (next_minor == start) {
330                                 /* Oh dear, all in use. */
331                                 spin_unlock(&all_mddevs_lock);
332                                 kfree(new);
333                                 return NULL;
334                         }
335                                 
336                         is_free = 1;
337                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
338                                 if (mddev->unit == dev) {
339                                         is_free = 0;
340                                         break;
341                                 }
342                 }
343                 new->unit = dev;
344                 new->md_minor = MINOR(dev);
345                 new->hold_active = UNTIL_STOP;
346                 list_add(&new->all_mddevs, &all_mddevs);
347                 spin_unlock(&all_mddevs_lock);
348                 return new;
349         }
350         spin_unlock(&all_mddevs_lock);
351
352         new = kzalloc(sizeof(*new), GFP_KERNEL);
353         if (!new)
354                 return NULL;
355
356         new->unit = unit;
357         if (MAJOR(unit) == MD_MAJOR)
358                 new->md_minor = MINOR(unit);
359         else
360                 new->md_minor = MINOR(unit) >> MdpMinorShift;
361
362         mutex_init(&new->reconfig_mutex);
363         INIT_LIST_HEAD(&new->disks);
364         INIT_LIST_HEAD(&new->all_mddevs);
365         init_timer(&new->safemode_timer);
366         atomic_set(&new->active, 1);
367         atomic_set(&new->openers, 0);
368         atomic_set(&new->active_io, 0);
369         spin_lock_init(&new->write_lock);
370         init_waitqueue_head(&new->sb_wait);
371         init_waitqueue_head(&new->recovery_wait);
372         new->reshape_position = MaxSector;
373         new->resync_min = 0;
374         new->resync_max = MaxSector;
375         new->level = LEVEL_NONE;
376
377         goto retry;
378 }
379
380 static inline int mddev_lock(mddev_t * mddev)
381 {
382         return mutex_lock_interruptible(&mddev->reconfig_mutex);
383 }
384
385 static inline int mddev_is_locked(mddev_t *mddev)
386 {
387         return mutex_is_locked(&mddev->reconfig_mutex);
388 }
389
390 static inline int mddev_trylock(mddev_t * mddev)
391 {
392         return mutex_trylock(&mddev->reconfig_mutex);
393 }
394
395 static inline void mddev_unlock(mddev_t * mddev)
396 {
397         mutex_unlock(&mddev->reconfig_mutex);
398
399         md_wakeup_thread(mddev->thread);
400 }
401
402 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
403 {
404         mdk_rdev_t *rdev;
405
406         list_for_each_entry(rdev, &mddev->disks, same_set)
407                 if (rdev->desc_nr == nr)
408                         return rdev;
409
410         return NULL;
411 }
412
413 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
414 {
415         mdk_rdev_t *rdev;
416
417         list_for_each_entry(rdev, &mddev->disks, same_set)
418                 if (rdev->bdev->bd_dev == dev)
419                         return rdev;
420
421         return NULL;
422 }
423
424 static struct mdk_personality *find_pers(int level, char *clevel)
425 {
426         struct mdk_personality *pers;
427         list_for_each_entry(pers, &pers_list, list) {
428                 if (level != LEVEL_NONE && pers->level == level)
429                         return pers;
430                 if (strcmp(pers->name, clevel)==0)
431                         return pers;
432         }
433         return NULL;
434 }
435
436 /* return the offset of the super block in 512byte sectors */
437 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
438 {
439         sector_t num_sectors = bdev->bd_inode->i_size / 512;
440         return MD_NEW_SIZE_SECTORS(num_sectors);
441 }
442
443 static int alloc_disk_sb(mdk_rdev_t * rdev)
444 {
445         if (rdev->sb_page)
446                 MD_BUG();
447
448         rdev->sb_page = alloc_page(GFP_KERNEL);
449         if (!rdev->sb_page) {
450                 printk(KERN_ALERT "md: out of memory.\n");
451                 return -ENOMEM;
452         }
453
454         return 0;
455 }
456
457 static void free_disk_sb(mdk_rdev_t * rdev)
458 {
459         if (rdev->sb_page) {
460                 put_page(rdev->sb_page);
461                 rdev->sb_loaded = 0;
462                 rdev->sb_page = NULL;
463                 rdev->sb_start = 0;
464                 rdev->sectors = 0;
465         }
466 }
467
468
469 static void super_written(struct bio *bio, int error)
470 {
471         mdk_rdev_t *rdev = bio->bi_private;
472         mddev_t *mddev = rdev->mddev;
473
474         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
475                 printk("md: super_written gets error=%d, uptodate=%d\n",
476                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
477                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
478                 md_error(mddev, rdev);
479         }
480
481         if (atomic_dec_and_test(&mddev->pending_writes))
482                 wake_up(&mddev->sb_wait);
483         bio_put(bio);
484 }
485
486 static void super_written_barrier(struct bio *bio, int error)
487 {
488         struct bio *bio2 = bio->bi_private;
489         mdk_rdev_t *rdev = bio2->bi_private;
490         mddev_t *mddev = rdev->mddev;
491
492         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
493             error == -EOPNOTSUPP) {
494                 unsigned long flags;
495                 /* barriers don't appear to be supported :-( */
496                 set_bit(BarriersNotsupp, &rdev->flags);
497                 mddev->barriers_work = 0;
498                 spin_lock_irqsave(&mddev->write_lock, flags);
499                 bio2->bi_next = mddev->biolist;
500                 mddev->biolist = bio2;
501                 spin_unlock_irqrestore(&mddev->write_lock, flags);
502                 wake_up(&mddev->sb_wait);
503                 bio_put(bio);
504         } else {
505                 bio_put(bio2);
506                 bio->bi_private = rdev;
507                 super_written(bio, error);
508         }
509 }
510
511 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
512                    sector_t sector, int size, struct page *page)
513 {
514         /* write first size bytes of page to sector of rdev
515          * Increment mddev->pending_writes before returning
516          * and decrement it on completion, waking up sb_wait
517          * if zero is reached.
518          * If an error occurred, call md_error
519          *
520          * As we might need to resubmit the request if BIO_RW_BARRIER
521          * causes ENOTSUPP, we allocate a spare bio...
522          */
523         struct bio *bio = bio_alloc(GFP_NOIO, 1);
524         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNCIO) | (1<<BIO_RW_UNPLUG);
525
526         bio->bi_bdev = rdev->bdev;
527         bio->bi_sector = sector;
528         bio_add_page(bio, page, size, 0);
529         bio->bi_private = rdev;
530         bio->bi_end_io = super_written;
531         bio->bi_rw = rw;
532
533         atomic_inc(&mddev->pending_writes);
534         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
535                 struct bio *rbio;
536                 rw |= (1<<BIO_RW_BARRIER);
537                 rbio = bio_clone(bio, GFP_NOIO);
538                 rbio->bi_private = bio;
539                 rbio->bi_end_io = super_written_barrier;
540                 submit_bio(rw, rbio);
541         } else
542                 submit_bio(rw, bio);
543 }
544
545 void md_super_wait(mddev_t *mddev)
546 {
547         /* wait for all superblock writes that were scheduled to complete.
548          * if any had to be retried (due to BARRIER problems), retry them
549          */
550         DEFINE_WAIT(wq);
551         for(;;) {
552                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
553                 if (atomic_read(&mddev->pending_writes)==0)
554                         break;
555                 while (mddev->biolist) {
556                         struct bio *bio;
557                         spin_lock_irq(&mddev->write_lock);
558                         bio = mddev->biolist;
559                         mddev->biolist = bio->bi_next ;
560                         bio->bi_next = NULL;
561                         spin_unlock_irq(&mddev->write_lock);
562                         submit_bio(bio->bi_rw, bio);
563                 }
564                 schedule();
565         }
566         finish_wait(&mddev->sb_wait, &wq);
567 }
568
569 static void bi_complete(struct bio *bio, int error)
570 {
571         complete((struct completion*)bio->bi_private);
572 }
573
574 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
575                    struct page *page, int rw)
576 {
577         struct bio *bio = bio_alloc(GFP_NOIO, 1);
578         struct completion event;
579         int ret;
580
581         rw |= (1 << BIO_RW_SYNCIO) | (1 << BIO_RW_UNPLUG);
582
583         bio->bi_bdev = bdev;
584         bio->bi_sector = sector;
585         bio_add_page(bio, page, size, 0);
586         init_completion(&event);
587         bio->bi_private = &event;
588         bio->bi_end_io = bi_complete;
589         submit_bio(rw, bio);
590         wait_for_completion(&event);
591
592         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
593         bio_put(bio);
594         return ret;
595 }
596 EXPORT_SYMBOL_GPL(sync_page_io);
597
598 static int read_disk_sb(mdk_rdev_t * rdev, int size)
599 {
600         char b[BDEVNAME_SIZE];
601         if (!rdev->sb_page) {
602                 MD_BUG();
603                 return -EINVAL;
604         }
605         if (rdev->sb_loaded)
606                 return 0;
607
608
609         if (!sync_page_io(rdev->bdev, rdev->sb_start, size, rdev->sb_page, READ))
610                 goto fail;
611         rdev->sb_loaded = 1;
612         return 0;
613
614 fail:
615         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
616                 bdevname(rdev->bdev,b));
617         return -EINVAL;
618 }
619
620 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
621 {
622         return  sb1->set_uuid0 == sb2->set_uuid0 &&
623                 sb1->set_uuid1 == sb2->set_uuid1 &&
624                 sb1->set_uuid2 == sb2->set_uuid2 &&
625                 sb1->set_uuid3 == sb2->set_uuid3;
626 }
627
628 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
629 {
630         int ret;
631         mdp_super_t *tmp1, *tmp2;
632
633         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
634         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
635
636         if (!tmp1 || !tmp2) {
637                 ret = 0;
638                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
639                 goto abort;
640         }
641
642         *tmp1 = *sb1;
643         *tmp2 = *sb2;
644
645         /*
646          * nr_disks is not constant
647          */
648         tmp1->nr_disks = 0;
649         tmp2->nr_disks = 0;
650
651         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
652 abort:
653         kfree(tmp1);
654         kfree(tmp2);
655         return ret;
656 }
657
658
659 static u32 md_csum_fold(u32 csum)
660 {
661         csum = (csum & 0xffff) + (csum >> 16);
662         return (csum & 0xffff) + (csum >> 16);
663 }
664
665 static unsigned int calc_sb_csum(mdp_super_t * sb)
666 {
667         u64 newcsum = 0;
668         u32 *sb32 = (u32*)sb;
669         int i;
670         unsigned int disk_csum, csum;
671
672         disk_csum = sb->sb_csum;
673         sb->sb_csum = 0;
674
675         for (i = 0; i < MD_SB_BYTES/4 ; i++)
676                 newcsum += sb32[i];
677         csum = (newcsum & 0xffffffff) + (newcsum>>32);
678
679
680 #ifdef CONFIG_ALPHA
681         /* This used to use csum_partial, which was wrong for several
682          * reasons including that different results are returned on
683          * different architectures.  It isn't critical that we get exactly
684          * the same return value as before (we always csum_fold before
685          * testing, and that removes any differences).  However as we
686          * know that csum_partial always returned a 16bit value on
687          * alphas, do a fold to maximise conformity to previous behaviour.
688          */
689         sb->sb_csum = md_csum_fold(disk_csum);
690 #else
691         sb->sb_csum = disk_csum;
692 #endif
693         return csum;
694 }
695
696
697 /*
698  * Handle superblock details.
699  * We want to be able to handle multiple superblock formats
700  * so we have a common interface to them all, and an array of
701  * different handlers.
702  * We rely on user-space to write the initial superblock, and support
703  * reading and updating of superblocks.
704  * Interface methods are:
705  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
706  *      loads and validates a superblock on dev.
707  *      if refdev != NULL, compare superblocks on both devices
708  *    Return:
709  *      0 - dev has a superblock that is compatible with refdev
710  *      1 - dev has a superblock that is compatible and newer than refdev
711  *          so dev should be used as the refdev in future
712  *     -EINVAL superblock incompatible or invalid
713  *     -othererror e.g. -EIO
714  *
715  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
716  *      Verify that dev is acceptable into mddev.
717  *       The first time, mddev->raid_disks will be 0, and data from
718  *       dev should be merged in.  Subsequent calls check that dev
719  *       is new enough.  Return 0 or -EINVAL
720  *
721  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
722  *     Update the superblock for rdev with data in mddev
723  *     This does not write to disc.
724  *
725  */
726
727 struct super_type  {
728         char                *name;
729         struct module       *owner;
730         int                 (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
731                                           int minor_version);
732         int                 (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
733         void                (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
734         unsigned long long  (*rdev_size_change)(mdk_rdev_t *rdev,
735                                                 sector_t num_sectors);
736 };
737
738 /*
739  * Check that the given mddev has no bitmap.
740  *
741  * This function is called from the run method of all personalities that do not
742  * support bitmaps. It prints an error message and returns non-zero if mddev
743  * has a bitmap. Otherwise, it returns 0.
744  *
745  */
746 int md_check_no_bitmap(mddev_t *mddev)
747 {
748         if (!mddev->bitmap_file && !mddev->bitmap_offset)
749                 return 0;
750         printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
751                 mdname(mddev), mddev->pers->name);
752         return 1;
753 }
754 EXPORT_SYMBOL(md_check_no_bitmap);
755
756 /*
757  * load_super for 0.90.0 
758  */
759 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
760 {
761         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
762         mdp_super_t *sb;
763         int ret;
764
765         /*
766          * Calculate the position of the superblock (512byte sectors),
767          * it's at the end of the disk.
768          *
769          * It also happens to be a multiple of 4Kb.
770          */
771         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
772
773         ret = read_disk_sb(rdev, MD_SB_BYTES);
774         if (ret) return ret;
775
776         ret = -EINVAL;
777
778         bdevname(rdev->bdev, b);
779         sb = (mdp_super_t*)page_address(rdev->sb_page);
780
781         if (sb->md_magic != MD_SB_MAGIC) {
782                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
783                        b);
784                 goto abort;
785         }
786
787         if (sb->major_version != 0 ||
788             sb->minor_version < 90 ||
789             sb->minor_version > 91) {
790                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
791                         sb->major_version, sb->minor_version,
792                         b);
793                 goto abort;
794         }
795
796         if (sb->raid_disks <= 0)
797                 goto abort;
798
799         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
800                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
801                         b);
802                 goto abort;
803         }
804
805         rdev->preferred_minor = sb->md_minor;
806         rdev->data_offset = 0;
807         rdev->sb_size = MD_SB_BYTES;
808
809         if (sb->level == LEVEL_MULTIPATH)
810                 rdev->desc_nr = -1;
811         else
812                 rdev->desc_nr = sb->this_disk.number;
813
814         if (!refdev) {
815                 ret = 1;
816         } else {
817                 __u64 ev1, ev2;
818                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
819                 if (!uuid_equal(refsb, sb)) {
820                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
821                                 b, bdevname(refdev->bdev,b2));
822                         goto abort;
823                 }
824                 if (!sb_equal(refsb, sb)) {
825                         printk(KERN_WARNING "md: %s has same UUID"
826                                " but different superblock to %s\n",
827                                b, bdevname(refdev->bdev, b2));
828                         goto abort;
829                 }
830                 ev1 = md_event(sb);
831                 ev2 = md_event(refsb);
832                 if (ev1 > ev2)
833                         ret = 1;
834                 else 
835                         ret = 0;
836         }
837         rdev->sectors = rdev->sb_start;
838
839         if (rdev->sectors < sb->size * 2 && sb->level > 1)
840                 /* "this cannot possibly happen" ... */
841                 ret = -EINVAL;
842
843  abort:
844         return ret;
845 }
846
847 /*
848  * validate_super for 0.90.0
849  */
850 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
851 {
852         mdp_disk_t *desc;
853         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
854         __u64 ev1 = md_event(sb);
855
856         rdev->raid_disk = -1;
857         clear_bit(Faulty, &rdev->flags);
858         clear_bit(In_sync, &rdev->flags);
859         clear_bit(WriteMostly, &rdev->flags);
860         clear_bit(BarriersNotsupp, &rdev->flags);
861
862         if (mddev->raid_disks == 0) {
863                 mddev->major_version = 0;
864                 mddev->minor_version = sb->minor_version;
865                 mddev->patch_version = sb->patch_version;
866                 mddev->external = 0;
867                 mddev->chunk_sectors = sb->chunk_size >> 9;
868                 mddev->ctime = sb->ctime;
869                 mddev->utime = sb->utime;
870                 mddev->level = sb->level;
871                 mddev->clevel[0] = 0;
872                 mddev->layout = sb->layout;
873                 mddev->raid_disks = sb->raid_disks;
874                 mddev->dev_sectors = sb->size * 2;
875                 mddev->events = ev1;
876                 mddev->bitmap_offset = 0;
877                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
878
879                 if (mddev->minor_version >= 91) {
880                         mddev->reshape_position = sb->reshape_position;
881                         mddev->delta_disks = sb->delta_disks;
882                         mddev->new_level = sb->new_level;
883                         mddev->new_layout = sb->new_layout;
884                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
885                 } else {
886                         mddev->reshape_position = MaxSector;
887                         mddev->delta_disks = 0;
888                         mddev->new_level = mddev->level;
889                         mddev->new_layout = mddev->layout;
890                         mddev->new_chunk_sectors = mddev->chunk_sectors;
891                 }
892
893                 if (sb->state & (1<<MD_SB_CLEAN))
894                         mddev->recovery_cp = MaxSector;
895                 else {
896                         if (sb->events_hi == sb->cp_events_hi && 
897                                 sb->events_lo == sb->cp_events_lo) {
898                                 mddev->recovery_cp = sb->recovery_cp;
899                         } else
900                                 mddev->recovery_cp = 0;
901                 }
902
903                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
904                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
905                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
906                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
907
908                 mddev->max_disks = MD_SB_DISKS;
909
910                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
911                     mddev->bitmap_file == NULL)
912                         mddev->bitmap_offset = mddev->default_bitmap_offset;
913
914         } else if (mddev->pers == NULL) {
915                 /* Insist on good event counter while assembling */
916                 ++ev1;
917                 if (ev1 < mddev->events) 
918                         return -EINVAL;
919         } else if (mddev->bitmap) {
920                 /* if adding to array with a bitmap, then we can accept an
921                  * older device ... but not too old.
922                  */
923                 if (ev1 < mddev->bitmap->events_cleared)
924                         return 0;
925         } else {
926                 if (ev1 < mddev->events)
927                         /* just a hot-add of a new device, leave raid_disk at -1 */
928                         return 0;
929         }
930
931         if (mddev->level != LEVEL_MULTIPATH) {
932                 desc = sb->disks + rdev->desc_nr;
933
934                 if (desc->state & (1<<MD_DISK_FAULTY))
935                         set_bit(Faulty, &rdev->flags);
936                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
937                             desc->raid_disk < mddev->raid_disks */) {
938                         set_bit(In_sync, &rdev->flags);
939                         rdev->raid_disk = desc->raid_disk;
940                 }
941                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
942                         set_bit(WriteMostly, &rdev->flags);
943         } else /* MULTIPATH are always insync */
944                 set_bit(In_sync, &rdev->flags);
945         return 0;
946 }
947
948 /*
949  * sync_super for 0.90.0
950  */
951 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
952 {
953         mdp_super_t *sb;
954         mdk_rdev_t *rdev2;
955         int next_spare = mddev->raid_disks;
956
957
958         /* make rdev->sb match mddev data..
959          *
960          * 1/ zero out disks
961          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
962          * 3/ any empty disks < next_spare become removed
963          *
964          * disks[0] gets initialised to REMOVED because
965          * we cannot be sure from other fields if it has
966          * been initialised or not.
967          */
968         int i;
969         int active=0, working=0,failed=0,spare=0,nr_disks=0;
970
971         rdev->sb_size = MD_SB_BYTES;
972
973         sb = (mdp_super_t*)page_address(rdev->sb_page);
974
975         memset(sb, 0, sizeof(*sb));
976
977         sb->md_magic = MD_SB_MAGIC;
978         sb->major_version = mddev->major_version;
979         sb->patch_version = mddev->patch_version;
980         sb->gvalid_words  = 0; /* ignored */
981         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
982         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
983         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
984         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
985
986         sb->ctime = mddev->ctime;
987         sb->level = mddev->level;
988         sb->size = mddev->dev_sectors / 2;
989         sb->raid_disks = mddev->raid_disks;
990         sb->md_minor = mddev->md_minor;
991         sb->not_persistent = 0;
992         sb->utime = mddev->utime;
993         sb->state = 0;
994         sb->events_hi = (mddev->events>>32);
995         sb->events_lo = (u32)mddev->events;
996
997         if (mddev->reshape_position == MaxSector)
998                 sb->minor_version = 90;
999         else {
1000                 sb->minor_version = 91;
1001                 sb->reshape_position = mddev->reshape_position;
1002                 sb->new_level = mddev->new_level;
1003                 sb->delta_disks = mddev->delta_disks;
1004                 sb->new_layout = mddev->new_layout;
1005                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1006         }
1007         mddev->minor_version = sb->minor_version;
1008         if (mddev->in_sync)
1009         {
1010                 sb->recovery_cp = mddev->recovery_cp;
1011                 sb->cp_events_hi = (mddev->events>>32);
1012                 sb->cp_events_lo = (u32)mddev->events;
1013                 if (mddev->recovery_cp == MaxSector)
1014                         sb->state = (1<< MD_SB_CLEAN);
1015         } else
1016                 sb->recovery_cp = 0;
1017
1018         sb->layout = mddev->layout;
1019         sb->chunk_size = mddev->chunk_sectors << 9;
1020
1021         if (mddev->bitmap && mddev->bitmap_file == NULL)
1022                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1023
1024         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1025         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1026                 mdp_disk_t *d;
1027                 int desc_nr;
1028                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1029                     && !test_bit(Faulty, &rdev2->flags))
1030                         desc_nr = rdev2->raid_disk;
1031                 else
1032                         desc_nr = next_spare++;
1033                 rdev2->desc_nr = desc_nr;
1034                 d = &sb->disks[rdev2->desc_nr];
1035                 nr_disks++;
1036                 d->number = rdev2->desc_nr;
1037                 d->major = MAJOR(rdev2->bdev->bd_dev);
1038                 d->minor = MINOR(rdev2->bdev->bd_dev);
1039                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
1040                     && !test_bit(Faulty, &rdev2->flags))
1041                         d->raid_disk = rdev2->raid_disk;
1042                 else
1043                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1044                 if (test_bit(Faulty, &rdev2->flags))
1045                         d->state = (1<<MD_DISK_FAULTY);
1046                 else if (test_bit(In_sync, &rdev2->flags)) {
1047                         d->state = (1<<MD_DISK_ACTIVE);
1048                         d->state |= (1<<MD_DISK_SYNC);
1049                         active++;
1050                         working++;
1051                 } else {
1052                         d->state = 0;
1053                         spare++;
1054                         working++;
1055                 }
1056                 if (test_bit(WriteMostly, &rdev2->flags))
1057                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1058         }
1059         /* now set the "removed" and "faulty" bits on any missing devices */
1060         for (i=0 ; i < mddev->raid_disks ; i++) {
1061                 mdp_disk_t *d = &sb->disks[i];
1062                 if (d->state == 0 && d->number == 0) {
1063                         d->number = i;
1064                         d->raid_disk = i;
1065                         d->state = (1<<MD_DISK_REMOVED);
1066                         d->state |= (1<<MD_DISK_FAULTY);
1067                         failed++;
1068                 }
1069         }
1070         sb->nr_disks = nr_disks;
1071         sb->active_disks = active;
1072         sb->working_disks = working;
1073         sb->failed_disks = failed;
1074         sb->spare_disks = spare;
1075
1076         sb->this_disk = sb->disks[rdev->desc_nr];
1077         sb->sb_csum = calc_sb_csum(sb);
1078 }
1079
1080 /*
1081  * rdev_size_change for 0.90.0
1082  */
1083 static unsigned long long
1084 super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1085 {
1086         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1087                 return 0; /* component must fit device */
1088         if (rdev->mddev->bitmap_offset)
1089                 return 0; /* can't move bitmap */
1090         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
1091         if (!num_sectors || num_sectors > rdev->sb_start)
1092                 num_sectors = rdev->sb_start;
1093         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1094                        rdev->sb_page);
1095         md_super_wait(rdev->mddev);
1096         return num_sectors / 2; /* kB for sysfs */
1097 }
1098
1099
1100 /*
1101  * version 1 superblock
1102  */
1103
1104 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1105 {
1106         __le32 disk_csum;
1107         u32 csum;
1108         unsigned long long newcsum;
1109         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1110         __le32 *isuper = (__le32*)sb;
1111         int i;
1112
1113         disk_csum = sb->sb_csum;
1114         sb->sb_csum = 0;
1115         newcsum = 0;
1116         for (i=0; size>=4; size -= 4 )
1117                 newcsum += le32_to_cpu(*isuper++);
1118
1119         if (size == 2)
1120                 newcsum += le16_to_cpu(*(__le16*) isuper);
1121
1122         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1123         sb->sb_csum = disk_csum;
1124         return cpu_to_le32(csum);
1125 }
1126
1127 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1128 {
1129         struct mdp_superblock_1 *sb;
1130         int ret;
1131         sector_t sb_start;
1132         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1133         int bmask;
1134
1135         /*
1136          * Calculate the position of the superblock in 512byte sectors.
1137          * It is always aligned to a 4K boundary and
1138          * depeding on minor_version, it can be:
1139          * 0: At least 8K, but less than 12K, from end of device
1140          * 1: At start of device
1141          * 2: 4K from start of device.
1142          */
1143         switch(minor_version) {
1144         case 0:
1145                 sb_start = rdev->bdev->bd_inode->i_size >> 9;
1146                 sb_start -= 8*2;
1147                 sb_start &= ~(sector_t)(4*2-1);
1148                 break;
1149         case 1:
1150                 sb_start = 0;
1151                 break;
1152         case 2:
1153                 sb_start = 8;
1154                 break;
1155         default:
1156                 return -EINVAL;
1157         }
1158         rdev->sb_start = sb_start;
1159
1160         /* superblock is rarely larger than 1K, but it can be larger,
1161          * and it is safe to read 4k, so we do that
1162          */
1163         ret = read_disk_sb(rdev, 4096);
1164         if (ret) return ret;
1165
1166
1167         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1168
1169         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1170             sb->major_version != cpu_to_le32(1) ||
1171             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1172             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1173             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1174                 return -EINVAL;
1175
1176         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1177                 printk("md: invalid superblock checksum on %s\n",
1178                         bdevname(rdev->bdev,b));
1179                 return -EINVAL;
1180         }
1181         if (le64_to_cpu(sb->data_size) < 10) {
1182                 printk("md: data_size too small on %s\n",
1183                        bdevname(rdev->bdev,b));
1184                 return -EINVAL;
1185         }
1186
1187         rdev->preferred_minor = 0xffff;
1188         rdev->data_offset = le64_to_cpu(sb->data_offset);
1189         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1190
1191         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1192         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1193         if (rdev->sb_size & bmask)
1194                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1195
1196         if (minor_version
1197             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1198                 return -EINVAL;
1199
1200         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1201                 rdev->desc_nr = -1;
1202         else
1203                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1204
1205         if (!refdev) {
1206                 ret = 1;
1207         } else {
1208                 __u64 ev1, ev2;
1209                 struct mdp_superblock_1 *refsb = 
1210                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1211
1212                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1213                     sb->level != refsb->level ||
1214                     sb->layout != refsb->layout ||
1215                     sb->chunksize != refsb->chunksize) {
1216                         printk(KERN_WARNING "md: %s has strangely different"
1217                                 " superblock to %s\n",
1218                                 bdevname(rdev->bdev,b),
1219                                 bdevname(refdev->bdev,b2));
1220                         return -EINVAL;
1221                 }
1222                 ev1 = le64_to_cpu(sb->events);
1223                 ev2 = le64_to_cpu(refsb->events);
1224
1225                 if (ev1 > ev2)
1226                         ret = 1;
1227                 else
1228                         ret = 0;
1229         }
1230         if (minor_version)
1231                 rdev->sectors = (rdev->bdev->bd_inode->i_size >> 9) -
1232                         le64_to_cpu(sb->data_offset);
1233         else
1234                 rdev->sectors = rdev->sb_start;
1235         if (rdev->sectors < le64_to_cpu(sb->data_size))
1236                 return -EINVAL;
1237         rdev->sectors = le64_to_cpu(sb->data_size);
1238         if (le64_to_cpu(sb->size) > rdev->sectors)
1239                 return -EINVAL;
1240         return ret;
1241 }
1242
1243 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1244 {
1245         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1246         __u64 ev1 = le64_to_cpu(sb->events);
1247
1248         rdev->raid_disk = -1;
1249         clear_bit(Faulty, &rdev->flags);
1250         clear_bit(In_sync, &rdev->flags);
1251         clear_bit(WriteMostly, &rdev->flags);
1252         clear_bit(BarriersNotsupp, &rdev->flags);
1253
1254         if (mddev->raid_disks == 0) {
1255                 mddev->major_version = 1;
1256                 mddev->patch_version = 0;
1257                 mddev->external = 0;
1258                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1259                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1260                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1261                 mddev->level = le32_to_cpu(sb->level);
1262                 mddev->clevel[0] = 0;
1263                 mddev->layout = le32_to_cpu(sb->layout);
1264                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1265                 mddev->dev_sectors = le64_to_cpu(sb->size);
1266                 mddev->events = ev1;
1267                 mddev->bitmap_offset = 0;
1268                 mddev->default_bitmap_offset = 1024 >> 9;
1269                 
1270                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1271                 memcpy(mddev->uuid, sb->set_uuid, 16);
1272
1273                 mddev->max_disks =  (4096-256)/2;
1274
1275                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1276                     mddev->bitmap_file == NULL )
1277                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1278
1279                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1280                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1281                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1282                         mddev->new_level = le32_to_cpu(sb->new_level);
1283                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1284                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1285                 } else {
1286                         mddev->reshape_position = MaxSector;
1287                         mddev->delta_disks = 0;
1288                         mddev->new_level = mddev->level;
1289                         mddev->new_layout = mddev->layout;
1290                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1291                 }
1292
1293         } else if (mddev->pers == NULL) {
1294                 /* Insist of good event counter while assembling */
1295                 ++ev1;
1296                 if (ev1 < mddev->events)
1297                         return -EINVAL;
1298         } else if (mddev->bitmap) {
1299                 /* If adding to array with a bitmap, then we can accept an
1300                  * older device, but not too old.
1301                  */
1302                 if (ev1 < mddev->bitmap->events_cleared)
1303                         return 0;
1304         } else {
1305                 if (ev1 < mddev->events)
1306                         /* just a hot-add of a new device, leave raid_disk at -1 */
1307                         return 0;
1308         }
1309         if (mddev->level != LEVEL_MULTIPATH) {
1310                 int role;
1311                 if (rdev->desc_nr < 0 ||
1312                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1313                         role = 0xffff;
1314                         rdev->desc_nr = -1;
1315                 } else
1316                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1317                 switch(role) {
1318                 case 0xffff: /* spare */
1319                         break;
1320                 case 0xfffe: /* faulty */
1321                         set_bit(Faulty, &rdev->flags);
1322                         break;
1323                 default:
1324                         if ((le32_to_cpu(sb->feature_map) &
1325                              MD_FEATURE_RECOVERY_OFFSET))
1326                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1327                         else
1328                                 set_bit(In_sync, &rdev->flags);
1329                         rdev->raid_disk = role;
1330                         break;
1331                 }
1332                 if (sb->devflags & WriteMostly1)
1333                         set_bit(WriteMostly, &rdev->flags);
1334         } else /* MULTIPATH are always insync */
1335                 set_bit(In_sync, &rdev->flags);
1336
1337         return 0;
1338 }
1339
1340 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1341 {
1342         struct mdp_superblock_1 *sb;
1343         mdk_rdev_t *rdev2;
1344         int max_dev, i;
1345         /* make rdev->sb match mddev and rdev data. */
1346
1347         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1348
1349         sb->feature_map = 0;
1350         sb->pad0 = 0;
1351         sb->recovery_offset = cpu_to_le64(0);
1352         memset(sb->pad1, 0, sizeof(sb->pad1));
1353         memset(sb->pad2, 0, sizeof(sb->pad2));
1354         memset(sb->pad3, 0, sizeof(sb->pad3));
1355
1356         sb->utime = cpu_to_le64((__u64)mddev->utime);
1357         sb->events = cpu_to_le64(mddev->events);
1358         if (mddev->in_sync)
1359                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1360         else
1361                 sb->resync_offset = cpu_to_le64(0);
1362
1363         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1364
1365         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1366         sb->size = cpu_to_le64(mddev->dev_sectors);
1367         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1368         sb->level = cpu_to_le32(mddev->level);
1369         sb->layout = cpu_to_le32(mddev->layout);
1370
1371         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1372                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1373                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1374         }
1375
1376         if (rdev->raid_disk >= 0 &&
1377             !test_bit(In_sync, &rdev->flags)) {
1378                 if (mddev->curr_resync_completed > rdev->recovery_offset)
1379                         rdev->recovery_offset = mddev->curr_resync_completed;
1380                 if (rdev->recovery_offset > 0) {
1381                         sb->feature_map |=
1382                                 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1383                         sb->recovery_offset =
1384                                 cpu_to_le64(rdev->recovery_offset);
1385                 }
1386         }
1387
1388         if (mddev->reshape_position != MaxSector) {
1389                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1390                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1391                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1392                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1393                 sb->new_level = cpu_to_le32(mddev->new_level);
1394                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1395         }
1396
1397         max_dev = 0;
1398         list_for_each_entry(rdev2, &mddev->disks, same_set)
1399                 if (rdev2->desc_nr+1 > max_dev)
1400                         max_dev = rdev2->desc_nr+1;
1401
1402         if (max_dev > le32_to_cpu(sb->max_dev)) {
1403                 int bmask;
1404                 sb->max_dev = cpu_to_le32(max_dev);
1405                 rdev->sb_size = max_dev * 2 + 256;
1406                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1407                 if (rdev->sb_size & bmask)
1408                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
1409         }
1410         for (i=0; i<max_dev;i++)
1411                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1412         
1413         list_for_each_entry(rdev2, &mddev->disks, same_set) {
1414                 i = rdev2->desc_nr;
1415                 if (test_bit(Faulty, &rdev2->flags))
1416                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1417                 else if (test_bit(In_sync, &rdev2->flags))
1418                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1419                 else if (rdev2->raid_disk >= 0 && rdev2->recovery_offset > 0)
1420                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1421                 else
1422                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1423         }
1424
1425         sb->sb_csum = calc_sb_1_csum(sb);
1426 }
1427
1428 static unsigned long long
1429 super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1430 {
1431         struct mdp_superblock_1 *sb;
1432         sector_t max_sectors;
1433         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1434                 return 0; /* component must fit device */
1435         if (rdev->sb_start < rdev->data_offset) {
1436                 /* minor versions 1 and 2; superblock before data */
1437                 max_sectors = rdev->bdev->bd_inode->i_size >> 9;
1438                 max_sectors -= rdev->data_offset;
1439                 if (!num_sectors || num_sectors > max_sectors)
1440                         num_sectors = max_sectors;
1441         } else if (rdev->mddev->bitmap_offset) {
1442                 /* minor version 0 with bitmap we can't move */
1443                 return 0;
1444         } else {
1445                 /* minor version 0; superblock after data */
1446                 sector_t sb_start;
1447                 sb_start = (rdev->bdev->bd_inode->i_size >> 9) - 8*2;
1448                 sb_start &= ~(sector_t)(4*2 - 1);
1449                 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1450                 if (!num_sectors || num_sectors > max_sectors)
1451                         num_sectors = max_sectors;
1452                 rdev->sb_start = sb_start;
1453         }
1454         sb = (struct mdp_superblock_1 *) page_address(rdev->sb_page);
1455         sb->data_size = cpu_to_le64(num_sectors);
1456         sb->super_offset = rdev->sb_start;
1457         sb->sb_csum = calc_sb_1_csum(sb);
1458         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1459                        rdev->sb_page);
1460         md_super_wait(rdev->mddev);
1461         return num_sectors / 2; /* kB for sysfs */
1462 }
1463
1464 static struct super_type super_types[] = {
1465         [0] = {
1466                 .name   = "0.90.0",
1467                 .owner  = THIS_MODULE,
1468                 .load_super         = super_90_load,
1469                 .validate_super     = super_90_validate,
1470                 .sync_super         = super_90_sync,
1471                 .rdev_size_change   = super_90_rdev_size_change,
1472         },
1473         [1] = {
1474                 .name   = "md-1",
1475                 .owner  = THIS_MODULE,
1476                 .load_super         = super_1_load,
1477                 .validate_super     = super_1_validate,
1478                 .sync_super         = super_1_sync,
1479                 .rdev_size_change   = super_1_rdev_size_change,
1480         },
1481 };
1482
1483 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1484 {
1485         mdk_rdev_t *rdev, *rdev2;
1486
1487         rcu_read_lock();
1488         rdev_for_each_rcu(rdev, mddev1)
1489                 rdev_for_each_rcu(rdev2, mddev2)
1490                         if (rdev->bdev->bd_contains ==
1491                             rdev2->bdev->bd_contains) {
1492                                 rcu_read_unlock();
1493                                 return 1;
1494                         }
1495         rcu_read_unlock();
1496         return 0;
1497 }
1498
1499 static LIST_HEAD(pending_raid_disks);
1500
1501 /*
1502  * Try to register data integrity profile for an mddev
1503  *
1504  * This is called when an array is started and after a disk has been kicked
1505  * from the array. It only succeeds if all working and active component devices
1506  * are integrity capable with matching profiles.
1507  */
1508 int md_integrity_register(mddev_t *mddev)
1509 {
1510         mdk_rdev_t *rdev, *reference = NULL;
1511
1512         if (list_empty(&mddev->disks))
1513                 return 0; /* nothing to do */
1514         if (blk_get_integrity(mddev->gendisk))
1515                 return 0; /* already registered */
1516         list_for_each_entry(rdev, &mddev->disks, same_set) {
1517                 /* skip spares and non-functional disks */
1518                 if (test_bit(Faulty, &rdev->flags))
1519                         continue;
1520                 if (rdev->raid_disk < 0)
1521                         continue;
1522                 /*
1523                  * If at least one rdev is not integrity capable, we can not
1524                  * enable data integrity for the md device.
1525                  */
1526                 if (!bdev_get_integrity(rdev->bdev))
1527                         return -EINVAL;
1528                 if (!reference) {
1529                         /* Use the first rdev as the reference */
1530                         reference = rdev;
1531                         continue;
1532                 }
1533                 /* does this rdev's profile match the reference profile? */
1534                 if (blk_integrity_compare(reference->bdev->bd_disk,
1535                                 rdev->bdev->bd_disk) < 0)
1536                         return -EINVAL;
1537         }
1538         /*
1539          * All component devices are integrity capable and have matching
1540          * profiles, register the common profile for the md device.
1541          */
1542         if (blk_integrity_register(mddev->gendisk,
1543                         bdev_get_integrity(reference->bdev)) != 0) {
1544                 printk(KERN_ERR "md: failed to register integrity for %s\n",
1545                         mdname(mddev));
1546                 return -EINVAL;
1547         }
1548         printk(KERN_NOTICE "md: data integrity on %s enabled\n",
1549                 mdname(mddev));
1550         return 0;
1551 }
1552 EXPORT_SYMBOL(md_integrity_register);
1553
1554 /* Disable data integrity if non-capable/non-matching disk is being added */
1555 void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1556 {
1557         struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1558         struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1559
1560         if (!bi_mddev) /* nothing to do */
1561                 return;
1562         if (rdev->raid_disk < 0) /* skip spares */
1563                 return;
1564         if (bi_rdev && blk_integrity_compare(mddev->gendisk,
1565                                              rdev->bdev->bd_disk) >= 0)
1566                 return;
1567         printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
1568         blk_integrity_unregister(mddev->gendisk);
1569 }
1570 EXPORT_SYMBOL(md_integrity_add_rdev);
1571
1572 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1573 {
1574         char b[BDEVNAME_SIZE];
1575         struct kobject *ko;
1576         char *s;
1577         int err;
1578
1579         if (rdev->mddev) {
1580                 MD_BUG();
1581                 return -EINVAL;
1582         }
1583
1584         /* prevent duplicates */
1585         if (find_rdev(mddev, rdev->bdev->bd_dev))
1586                 return -EEXIST;
1587
1588         /* make sure rdev->sectors exceeds mddev->dev_sectors */
1589         if (rdev->sectors && (mddev->dev_sectors == 0 ||
1590                         rdev->sectors < mddev->dev_sectors)) {
1591                 if (mddev->pers) {
1592                         /* Cannot change size, so fail
1593                          * If mddev->level <= 0, then we don't care
1594                          * about aligning sizes (e.g. linear)
1595                          */
1596                         if (mddev->level > 0)
1597                                 return -ENOSPC;
1598                 } else
1599                         mddev->dev_sectors = rdev->sectors;
1600         }
1601
1602         /* Verify rdev->desc_nr is unique.
1603          * If it is -1, assign a free number, else
1604          * check number is not in use
1605          */
1606         if (rdev->desc_nr < 0) {
1607                 int choice = 0;
1608                 if (mddev->pers) choice = mddev->raid_disks;
1609                 while (find_rdev_nr(mddev, choice))
1610                         choice++;
1611                 rdev->desc_nr = choice;
1612         } else {
1613                 if (find_rdev_nr(mddev, rdev->desc_nr))
1614                         return -EBUSY;
1615         }
1616         if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
1617                 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
1618                        mdname(mddev), mddev->max_disks);
1619                 return -EBUSY;
1620         }
1621         bdevname(rdev->bdev,b);
1622         while ( (s=strchr(b, '/')) != NULL)
1623                 *s = '!';
1624
1625         rdev->mddev = mddev;
1626         printk(KERN_INFO "md: bind<%s>\n", b);
1627
1628         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
1629                 goto fail;
1630
1631         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
1632         if ((err = sysfs_create_link(&rdev->kobj, ko, "block"))) {
1633                 kobject_del(&rdev->kobj);
1634                 goto fail;
1635         }
1636         rdev->sysfs_state = sysfs_get_dirent(rdev->kobj.sd, "state");
1637
1638         list_add_rcu(&rdev->same_set, &mddev->disks);
1639         bd_claim_by_disk(rdev->bdev, rdev->bdev->bd_holder, mddev->gendisk);
1640
1641         /* May as well allow recovery to be retried once */
1642         mddev->recovery_disabled = 0;
1643
1644         return 0;
1645
1646  fail:
1647         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
1648                b, mdname(mddev));
1649         return err;
1650 }
1651
1652 static void md_delayed_delete(struct work_struct *ws)
1653 {
1654         mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
1655         kobject_del(&rdev->kobj);
1656         kobject_put(&rdev->kobj);
1657 }
1658
1659 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1660 {
1661         char b[BDEVNAME_SIZE];
1662         if (!rdev->mddev) {
1663                 MD_BUG();
1664                 return;
1665         }
1666         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1667         list_del_rcu(&rdev->same_set);
1668         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1669         rdev->mddev = NULL;
1670         sysfs_remove_link(&rdev->kobj, "block");
1671         sysfs_put(rdev->sysfs_state);
1672         rdev->sysfs_state = NULL;
1673         /* We need to delay this, otherwise we can deadlock when
1674          * writing to 'remove' to "dev/state".  We also need
1675          * to delay it due to rcu usage.
1676          */
1677         synchronize_rcu();
1678         INIT_WORK(&rdev->del_work, md_delayed_delete);
1679         kobject_get(&rdev->kobj);
1680         schedule_work(&rdev->del_work);
1681 }
1682
1683 /*
1684  * prevent the device from being mounted, repartitioned or
1685  * otherwise reused by a RAID array (or any other kernel
1686  * subsystem), by bd_claiming the device.
1687  */
1688 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
1689 {
1690         int err = 0;
1691         struct block_device *bdev;
1692         char b[BDEVNAME_SIZE];
1693
1694         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1695         if (IS_ERR(bdev)) {
1696                 printk(KERN_ERR "md: could not open %s.\n",
1697                         __bdevname(dev, b));
1698                 return PTR_ERR(bdev);
1699         }
1700         err = bd_claim(bdev, shared ? (mdk_rdev_t *)lock_rdev : rdev);
1701         if (err) {
1702                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1703                         bdevname(bdev, b));
1704                 blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1705                 return err;
1706         }
1707         if (!shared)
1708                 set_bit(AllReserved, &rdev->flags);
1709         rdev->bdev = bdev;
1710         return err;
1711 }
1712
1713 static void unlock_rdev(mdk_rdev_t *rdev)
1714 {
1715         struct block_device *bdev = rdev->bdev;
1716         rdev->bdev = NULL;
1717         if (!bdev)
1718                 MD_BUG();
1719         bd_release(bdev);
1720         blkdev_put(bdev, FMODE_READ|FMODE_WRITE);
1721 }
1722
1723 void md_autodetect_dev(dev_t dev);
1724
1725 static void export_rdev(mdk_rdev_t * rdev)
1726 {
1727         char b[BDEVNAME_SIZE];
1728         printk(KERN_INFO "md: export_rdev(%s)\n",
1729                 bdevname(rdev->bdev,b));
1730         if (rdev->mddev)
1731                 MD_BUG();
1732         free_disk_sb(rdev);
1733 #ifndef MODULE
1734         if (test_bit(AutoDetected, &rdev->flags))
1735                 md_autodetect_dev(rdev->bdev->bd_dev);
1736 #endif
1737         unlock_rdev(rdev);
1738         kobject_put(&rdev->kobj);
1739 }
1740
1741 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1742 {
1743         unbind_rdev_from_array(rdev);
1744         export_rdev(rdev);
1745 }
1746
1747 static void export_array(mddev_t *mddev)
1748 {
1749         mdk_rdev_t *rdev, *tmp;
1750
1751         rdev_for_each(rdev, tmp, mddev) {
1752                 if (!rdev->mddev) {
1753                         MD_BUG();
1754                         continue;
1755                 }
1756                 kick_rdev_from_array(rdev);
1757         }
1758         if (!list_empty(&mddev->disks))
1759                 MD_BUG();
1760         mddev->raid_disks = 0;
1761         mddev->major_version = 0;
1762 }
1763
1764 static void print_desc(mdp_disk_t *desc)
1765 {
1766         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1767                 desc->major,desc->minor,desc->raid_disk,desc->state);
1768 }
1769
1770 static void print_sb_90(mdp_super_t *sb)
1771 {
1772         int i;
1773
1774         printk(KERN_INFO 
1775                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1776                 sb->major_version, sb->minor_version, sb->patch_version,
1777                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1778                 sb->ctime);
1779         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1780                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1781                 sb->md_minor, sb->layout, sb->chunk_size);
1782         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1783                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1784                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1785                 sb->failed_disks, sb->spare_disks,
1786                 sb->sb_csum, (unsigned long)sb->events_lo);
1787
1788         printk(KERN_INFO);
1789         for (i = 0; i < MD_SB_DISKS; i++) {
1790                 mdp_disk_t *desc;
1791
1792                 desc = sb->disks + i;
1793                 if (desc->number || desc->major || desc->minor ||
1794                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1795                         printk("     D %2d: ", i);
1796                         print_desc(desc);
1797                 }
1798         }
1799         printk(KERN_INFO "md:     THIS: ");
1800         print_desc(&sb->this_disk);
1801 }
1802
1803 static void print_sb_1(struct mdp_superblock_1 *sb)
1804 {
1805         __u8 *uuid;
1806
1807         uuid = sb->set_uuid;
1808         printk(KERN_INFO
1809                "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%02x%02x%02x%02x"
1810                ":%02x%02x:%02x%02x:%02x%02x:%02x%02x%02x%02x%02x%02x>\n"
1811                "md:    Name: \"%s\" CT:%llu\n",
1812                 le32_to_cpu(sb->major_version),
1813                 le32_to_cpu(sb->feature_map),
1814                 uuid[0], uuid[1], uuid[2], uuid[3],
1815                 uuid[4], uuid[5], uuid[6], uuid[7],
1816                 uuid[8], uuid[9], uuid[10], uuid[11],
1817                 uuid[12], uuid[13], uuid[14], uuid[15],
1818                 sb->set_name,
1819                 (unsigned long long)le64_to_cpu(sb->ctime)
1820                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
1821
1822         uuid = sb->device_uuid;
1823         printk(KERN_INFO
1824                "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
1825                         " RO:%llu\n"
1826                "md:     Dev:%08x UUID: %02x%02x%02x%02x:%02x%02x:%02x%02x:%02x%02x"
1827                         ":%02x%02x%02x%02x%02x%02x\n"
1828                "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
1829                "md:         (MaxDev:%u) \n",
1830                 le32_to_cpu(sb->level),
1831                 (unsigned long long)le64_to_cpu(sb->size),
1832                 le32_to_cpu(sb->raid_disks),
1833                 le32_to_cpu(sb->layout),
1834                 le32_to_cpu(sb->chunksize),
1835                 (unsigned long long)le64_to_cpu(sb->data_offset),
1836                 (unsigned long long)le64_to_cpu(sb->data_size),
1837                 (unsigned long long)le64_to_cpu(sb->super_offset),
1838                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
1839                 le32_to_cpu(sb->dev_number),
1840                 uuid[0], uuid[1], uuid[2], uuid[3],
1841                 uuid[4], uuid[5], uuid[6], uuid[7],
1842                 uuid[8], uuid[9], uuid[10], uuid[11],
1843                 uuid[12], uuid[13], uuid[14], uuid[15],
1844                 sb->devflags,
1845                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
1846                 (unsigned long long)le64_to_cpu(sb->events),
1847                 (unsigned long long)le64_to_cpu(sb->resync_offset),
1848                 le32_to_cpu(sb->sb_csum),
1849                 le32_to_cpu(sb->max_dev)
1850                 );
1851 }
1852
1853 static void print_rdev(mdk_rdev_t *rdev, int major_version)
1854 {
1855         char b[BDEVNAME_SIZE];
1856         printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
1857                 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
1858                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1859                 rdev->desc_nr);
1860         if (rdev->sb_loaded) {
1861                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
1862                 switch (major_version) {
1863                 case 0:
1864                         print_sb_90((mdp_super_t*)page_address(rdev->sb_page));
1865                         break;
1866                 case 1:
1867                         print_sb_1((struct mdp_superblock_1 *)page_address(rdev->sb_page));
1868                         break;
1869                 }
1870         } else
1871                 printk(KERN_INFO "md: no rdev superblock!\n");
1872 }
1873
1874 static void md_print_devices(void)
1875 {
1876         struct list_head *tmp;
1877         mdk_rdev_t *rdev;
1878         mddev_t *mddev;
1879         char b[BDEVNAME_SIZE];
1880
1881         printk("\n");
1882         printk("md:     **********************************\n");
1883         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1884         printk("md:     **********************************\n");
1885         for_each_mddev(mddev, tmp) {
1886
1887                 if (mddev->bitmap)
1888                         bitmap_print_sb(mddev->bitmap);
1889                 else
1890                         printk("%s: ", mdname(mddev));
1891                 list_for_each_entry(rdev, &mddev->disks, same_set)
1892                         printk("<%s>", bdevname(rdev->bdev,b));
1893                 printk("\n");
1894
1895                 list_for_each_entry(rdev, &mddev->disks, same_set)
1896                         print_rdev(rdev, mddev->major_version);
1897         }
1898         printk("md:     **********************************\n");
1899         printk("\n");
1900 }
1901
1902
1903 static void sync_sbs(mddev_t * mddev, int nospares)
1904 {
1905         /* Update each superblock (in-memory image), but
1906          * if we are allowed to, skip spares which already
1907          * have the right event counter, or have one earlier
1908          * (which would mean they aren't being marked as dirty
1909          * with the rest of the array)
1910          */
1911         mdk_rdev_t *rdev;
1912
1913         list_for_each_entry(rdev, &mddev->disks, same_set) {
1914                 if (rdev->sb_events == mddev->events ||
1915                     (nospares &&
1916                      rdev->raid_disk < 0 &&
1917                      (rdev->sb_events&1)==0 &&
1918                      rdev->sb_events+1 == mddev->events)) {
1919                         /* Don't update this superblock */
1920                         rdev->sb_loaded = 2;
1921                 } else {
1922                         super_types[mddev->major_version].
1923                                 sync_super(mddev, rdev);
1924                         rdev->sb_loaded = 1;
1925                 }
1926         }
1927 }
1928
1929 static void md_update_sb(mddev_t * mddev, int force_change)
1930 {
1931         mdk_rdev_t *rdev;
1932         int sync_req;
1933         int nospares = 0;
1934
1935         mddev->utime = get_seconds();
1936         if (mddev->external)
1937                 return;
1938 repeat:
1939         spin_lock_irq(&mddev->write_lock);
1940
1941         set_bit(MD_CHANGE_PENDING, &mddev->flags);
1942         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
1943                 force_change = 1;
1944         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
1945                 /* just a clean<-> dirty transition, possibly leave spares alone,
1946                  * though if events isn't the right even/odd, we will have to do
1947                  * spares after all
1948                  */
1949                 nospares = 1;
1950         if (force_change)
1951                 nospares = 0;
1952         if (mddev->degraded)
1953                 /* If the array is degraded, then skipping spares is both
1954                  * dangerous and fairly pointless.
1955                  * Dangerous because a device that was removed from the array
1956                  * might have a event_count that still looks up-to-date,
1957                  * so it can be re-added without a resync.
1958                  * Pointless because if there are any spares to skip,
1959                  * then a recovery will happen and soon that array won't
1960                  * be degraded any more and the spare can go back to sleep then.
1961                  */
1962                 nospares = 0;
1963
1964         sync_req = mddev->in_sync;
1965
1966         /* If this is just a dirty<->clean transition, and the array is clean
1967          * and 'events' is odd, we can roll back to the previous clean state */
1968         if (nospares
1969             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
1970             && (mddev->events & 1)
1971             && mddev->events != 1)
1972                 mddev->events--;
1973         else {
1974                 /* otherwise we have to go forward and ... */
1975                 mddev->events ++;
1976                 if (!mddev->in_sync || mddev->recovery_cp != MaxSector) { /* not clean */
1977                         /* .. if the array isn't clean, insist on an odd 'events' */
1978                         if ((mddev->events&1)==0) {
1979                                 mddev->events++;
1980                                 nospares = 0;
1981                         }
1982                 } else {
1983                         /* otherwise insist on an even 'events' (for clean states) */
1984                         if ((mddev->events&1)) {
1985                                 mddev->events++;
1986                                 nospares = 0;
1987                         }
1988                 }
1989         }
1990
1991         if (!mddev->events) {
1992                 /*
1993                  * oops, this 64-bit counter should never wrap.
1994                  * Either we are in around ~1 trillion A.C., assuming
1995                  * 1 reboot per second, or we have a bug:
1996                  */
1997                 MD_BUG();
1998                 mddev->events --;
1999         }
2000
2001         /*
2002          * do not write anything to disk if using
2003          * nonpersistent superblocks
2004          */
2005         if (!mddev->persistent) {
2006                 if (!mddev->external)
2007                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2008
2009                 spin_unlock_irq(&mddev->write_lock);
2010                 wake_up(&mddev->sb_wait);
2011                 return;
2012         }
2013         sync_sbs(mddev, nospares);
2014         spin_unlock_irq(&mddev->write_lock);
2015
2016         dprintk(KERN_INFO 
2017                 "md: updating %s RAID superblock on device (in sync %d)\n",
2018                 mdname(mddev),mddev->in_sync);
2019
2020         bitmap_update_sb(mddev->bitmap);
2021         list_for_each_entry(rdev, &mddev->disks, same_set) {
2022                 char b[BDEVNAME_SIZE];
2023                 dprintk(KERN_INFO "md: ");
2024                 if (rdev->sb_loaded != 1)
2025                         continue; /* no noise on spare devices */
2026                 if (test_bit(Faulty, &rdev->flags))
2027                         dprintk("(skipping faulty ");
2028
2029                 dprintk("%s ", bdevname(rdev->bdev,b));
2030                 if (!test_bit(Faulty, &rdev->flags)) {
2031                         md_super_write(mddev,rdev,
2032                                        rdev->sb_start, rdev->sb_size,
2033                                        rdev->sb_page);
2034                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2035                                 bdevname(rdev->bdev,b),
2036                                 (unsigned long long)rdev->sb_start);
2037                         rdev->sb_events = mddev->events;
2038
2039                 } else
2040                         dprintk(")\n");
2041                 if (mddev->level == LEVEL_MULTIPATH)
2042                         /* only need to write one superblock... */
2043                         break;
2044         }
2045         md_super_wait(mddev);
2046         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2047
2048         spin_lock_irq(&mddev->write_lock);
2049         if (mddev->in_sync != sync_req ||
2050             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2051                 /* have to write it out again */
2052                 spin_unlock_irq(&mddev->write_lock);
2053                 goto repeat;
2054         }
2055         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2056         spin_unlock_irq(&mddev->write_lock);
2057         wake_up(&mddev->sb_wait);
2058         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2059                 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2060
2061 }
2062
2063 /* words written to sysfs files may, or may not, be \n terminated.
2064  * We want to accept with case. For this we use cmd_match.
2065  */
2066 static int cmd_match(const char *cmd, const char *str)
2067 {
2068         /* See if cmd, written into a sysfs file, matches
2069          * str.  They must either be the same, or cmd can
2070          * have a trailing newline
2071          */
2072         while (*cmd && *str && *cmd == *str) {
2073                 cmd++;
2074                 str++;
2075         }
2076         if (*cmd == '\n')
2077                 cmd++;
2078         if (*str || *cmd)
2079                 return 0;
2080         return 1;
2081 }
2082
2083 struct rdev_sysfs_entry {
2084         struct attribute attr;
2085         ssize_t (*show)(mdk_rdev_t *, char *);
2086         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2087 };
2088
2089 static ssize_t
2090 state_show(mdk_rdev_t *rdev, char *page)
2091 {
2092         char *sep = "";
2093         size_t len = 0;
2094
2095         if (test_bit(Faulty, &rdev->flags)) {
2096                 len+= sprintf(page+len, "%sfaulty",sep);
2097                 sep = ",";
2098         }
2099         if (test_bit(In_sync, &rdev->flags)) {
2100                 len += sprintf(page+len, "%sin_sync",sep);
2101                 sep = ",";
2102         }
2103         if (test_bit(WriteMostly, &rdev->flags)) {
2104                 len += sprintf(page+len, "%swrite_mostly",sep);
2105                 sep = ",";
2106         }
2107         if (test_bit(Blocked, &rdev->flags)) {
2108                 len += sprintf(page+len, "%sblocked", sep);
2109                 sep = ",";
2110         }
2111         if (!test_bit(Faulty, &rdev->flags) &&
2112             !test_bit(In_sync, &rdev->flags)) {
2113                 len += sprintf(page+len, "%sspare", sep);
2114                 sep = ",";
2115         }
2116         return len+sprintf(page+len, "\n");
2117 }
2118
2119 static ssize_t
2120 state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2121 {
2122         /* can write
2123          *  faulty  - simulates and error
2124          *  remove  - disconnects the device
2125          *  writemostly - sets write_mostly
2126          *  -writemostly - clears write_mostly
2127          *  blocked - sets the Blocked flag
2128          *  -blocked - clears the Blocked flag
2129          *  insync - sets Insync providing device isn't active
2130          */
2131         int err = -EINVAL;
2132         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2133                 md_error(rdev->mddev, rdev);
2134                 err = 0;
2135         } else if (cmd_match(buf, "remove")) {
2136                 if (rdev->raid_disk >= 0)
2137                         err = -EBUSY;
2138                 else {
2139                         mddev_t *mddev = rdev->mddev;
2140                         kick_rdev_from_array(rdev);
2141                         if (mddev->pers)
2142                                 md_update_sb(mddev, 1);
2143                         md_new_event(mddev);
2144                         err = 0;
2145                 }
2146         } else if (cmd_match(buf, "writemostly")) {
2147                 set_bit(WriteMostly, &rdev->flags);
2148                 err = 0;
2149         } else if (cmd_match(buf, "-writemostly")) {
2150                 clear_bit(WriteMostly, &rdev->flags);
2151                 err = 0;
2152         } else if (cmd_match(buf, "blocked")) {
2153                 set_bit(Blocked, &rdev->flags);
2154                 err = 0;
2155         } else if (cmd_match(buf, "-blocked")) {
2156                 clear_bit(Blocked, &rdev->flags);
2157                 wake_up(&rdev->blocked_wait);
2158                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2159                 md_wakeup_thread(rdev->mddev->thread);
2160
2161                 err = 0;
2162         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2163                 set_bit(In_sync, &rdev->flags);
2164                 err = 0;
2165         }
2166         if (!err && rdev->sysfs_state)
2167                 sysfs_notify_dirent(rdev->sysfs_state);
2168         return err ? err : len;
2169 }
2170 static struct rdev_sysfs_entry rdev_state =
2171 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2172
2173 static ssize_t
2174 errors_show(mdk_rdev_t *rdev, char *page)
2175 {
2176         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2177 }
2178
2179 static ssize_t
2180 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2181 {
2182         char *e;
2183         unsigned long n = simple_strtoul(buf, &e, 10);
2184         if (*buf && (*e == 0 || *e == '\n')) {
2185                 atomic_set(&rdev->corrected_errors, n);
2186                 return len;
2187         }
2188         return -EINVAL;
2189 }
2190 static struct rdev_sysfs_entry rdev_errors =
2191 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2192
2193 static ssize_t
2194 slot_show(mdk_rdev_t *rdev, char *page)
2195 {
2196         if (rdev->raid_disk < 0)
2197                 return sprintf(page, "none\n");
2198         else
2199                 return sprintf(page, "%d\n", rdev->raid_disk);
2200 }
2201
2202 static ssize_t
2203 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2204 {
2205         char *e;
2206         int err;
2207         char nm[20];
2208         int slot = simple_strtoul(buf, &e, 10);
2209         if (strncmp(buf, "none", 4)==0)
2210                 slot = -1;
2211         else if (e==buf || (*e && *e!= '\n'))
2212                 return -EINVAL;
2213         if (rdev->mddev->pers && slot == -1) {
2214                 /* Setting 'slot' on an active array requires also
2215                  * updating the 'rd%d' link, and communicating
2216                  * with the personality with ->hot_*_disk.
2217                  * For now we only support removing
2218                  * failed/spare devices.  This normally happens automatically,
2219                  * but not when the metadata is externally managed.
2220                  */
2221                 if (rdev->raid_disk == -1)
2222                         return -EEXIST;
2223                 /* personality does all needed checks */
2224                 if (rdev->mddev->pers->hot_add_disk == NULL)
2225                         return -EINVAL;
2226                 err = rdev->mddev->pers->
2227                         hot_remove_disk(rdev->mddev, rdev->raid_disk);
2228                 if (err)
2229                         return err;
2230                 sprintf(nm, "rd%d", rdev->raid_disk);
2231                 sysfs_remove_link(&rdev->mddev->kobj, nm);
2232                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2233                 md_wakeup_thread(rdev->mddev->thread);
2234         } else if (rdev->mddev->pers) {
2235                 mdk_rdev_t *rdev2;
2236                 /* Activating a spare .. or possibly reactivating
2237                  * if we ever get bitmaps working here.
2238                  */
2239
2240                 if (rdev->raid_disk != -1)
2241                         return -EBUSY;
2242
2243                 if (rdev->mddev->pers->hot_add_disk == NULL)
2244                         return -EINVAL;
2245
2246                 list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2247                         if (rdev2->raid_disk == slot)
2248                                 return -EEXIST;
2249
2250                 rdev->raid_disk = slot;
2251                 if (test_bit(In_sync, &rdev->flags))
2252                         rdev->saved_raid_disk = slot;
2253                 else
2254                         rdev->saved_raid_disk = -1;
2255                 err = rdev->mddev->pers->
2256                         hot_add_disk(rdev->mddev, rdev);
2257                 if (err) {
2258                         rdev->raid_disk = -1;
2259                         return err;
2260                 } else
2261                         sysfs_notify_dirent(rdev->sysfs_state);
2262                 sprintf(nm, "rd%d", rdev->raid_disk);
2263                 if (sysfs_create_link(&rdev->mddev->kobj, &rdev->kobj, nm))
2264                         printk(KERN_WARNING
2265                                "md: cannot register "
2266                                "%s for %s\n",
2267                                nm, mdname(rdev->mddev));
2268
2269                 /* don't wakeup anyone, leave that to userspace. */
2270         } else {
2271                 if (slot >= rdev->mddev->raid_disks)
2272                         return -ENOSPC;
2273                 rdev->raid_disk = slot;
2274                 /* assume it is working */
2275                 clear_bit(Faulty, &rdev->flags);
2276                 clear_bit(WriteMostly, &rdev->flags);
2277                 set_bit(In_sync, &rdev->flags);
2278                 sysfs_notify_dirent(rdev->sysfs_state);
2279         }
2280         return len;
2281 }
2282
2283
2284 static struct rdev_sysfs_entry rdev_slot =
2285 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2286
2287 static ssize_t
2288 offset_show(mdk_rdev_t *rdev, char *page)
2289 {
2290         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2291 }
2292
2293 static ssize_t
2294 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2295 {
2296         char *e;
2297         unsigned long long offset = simple_strtoull(buf, &e, 10);
2298         if (e==buf || (*e && *e != '\n'))
2299                 return -EINVAL;
2300         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2301                 return -EBUSY;
2302         if (rdev->sectors && rdev->mddev->external)
2303                 /* Must set offset before size, so overlap checks
2304                  * can be sane */
2305                 return -EBUSY;
2306         rdev->data_offset = offset;
2307         return len;
2308 }
2309
2310 static struct rdev_sysfs_entry rdev_offset =
2311 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2312
2313 static ssize_t
2314 rdev_size_show(mdk_rdev_t *rdev, char *page)
2315 {
2316         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2317 }
2318
2319 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2320 {
2321         /* check if two start/length pairs overlap */
2322         if (s1+l1 <= s2)
2323                 return 0;
2324         if (s2+l2 <= s1)
2325                 return 0;
2326         return 1;
2327 }
2328
2329 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2330 {
2331         unsigned long long blocks;
2332         sector_t new;
2333
2334         if (strict_strtoull(buf, 10, &blocks) < 0)
2335                 return -EINVAL;
2336
2337         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2338                 return -EINVAL; /* sector conversion overflow */
2339
2340         new = blocks * 2;
2341         if (new != blocks * 2)
2342                 return -EINVAL; /* unsigned long long to sector_t overflow */
2343
2344         *sectors = new;
2345         return 0;
2346 }
2347
2348 static ssize_t
2349 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2350 {
2351         mddev_t *my_mddev = rdev->mddev;
2352         sector_t oldsectors = rdev->sectors;
2353         sector_t sectors;
2354
2355         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2356                 return -EINVAL;
2357         if (my_mddev->pers && rdev->raid_disk >= 0) {
2358                 if (my_mddev->persistent) {
2359                         sectors = super_types[my_mddev->major_version].
2360                                 rdev_size_change(rdev, sectors);
2361                         if (!sectors)
2362                                 return -EBUSY;
2363                 } else if (!sectors)
2364                         sectors = (rdev->bdev->bd_inode->i_size >> 9) -
2365                                 rdev->data_offset;
2366         }
2367         if (sectors < my_mddev->dev_sectors)
2368                 return -EINVAL; /* component must fit device */
2369
2370         rdev->sectors = sectors;
2371         if (sectors > oldsectors && my_mddev->external) {
2372                 /* need to check that all other rdevs with the same ->bdev
2373                  * do not overlap.  We need to unlock the mddev to avoid
2374                  * a deadlock.  We have already changed rdev->sectors, and if
2375                  * we have to change it back, we will have the lock again.
2376                  */
2377                 mddev_t *mddev;
2378                 int overlap = 0;
2379                 struct list_head *tmp;
2380
2381                 mddev_unlock(my_mddev);
2382                 for_each_mddev(mddev, tmp) {
2383                         mdk_rdev_t *rdev2;
2384
2385                         mddev_lock(mddev);
2386                         list_for_each_entry(rdev2, &mddev->disks, same_set)
2387                                 if (test_bit(AllReserved, &rdev2->flags) ||
2388                                     (rdev->bdev == rdev2->bdev &&
2389                                      rdev != rdev2 &&
2390                                      overlaps(rdev->data_offset, rdev->sectors,
2391                                               rdev2->data_offset,
2392                                               rdev2->sectors))) {
2393                                         overlap = 1;
2394                                         break;
2395                                 }
2396                         mddev_unlock(mddev);
2397                         if (overlap) {
2398                                 mddev_put(mddev);
2399                                 break;
2400                         }
2401                 }
2402                 mddev_lock(my_mddev);
2403                 if (overlap) {
2404                         /* Someone else could have slipped in a size
2405                          * change here, but doing so is just silly.
2406                          * We put oldsectors back because we *know* it is
2407                          * safe, and trust userspace not to race with
2408                          * itself
2409                          */
2410                         rdev->sectors = oldsectors;
2411                         return -EBUSY;
2412                 }
2413         }
2414         return len;
2415 }
2416
2417 static struct rdev_sysfs_entry rdev_size =
2418 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2419
2420 static struct attribute *rdev_default_attrs[] = {
2421         &rdev_state.attr,
2422         &rdev_errors.attr,
2423         &rdev_slot.attr,
2424         &rdev_offset.attr,
2425         &rdev_size.attr,
2426         NULL,
2427 };
2428 static ssize_t
2429 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2430 {
2431         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2432         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2433         mddev_t *mddev = rdev->mddev;
2434         ssize_t rv;
2435
2436         if (!entry->show)
2437                 return -EIO;
2438
2439         rv = mddev ? mddev_lock(mddev) : -EBUSY;
2440         if (!rv) {
2441                 if (rdev->mddev == NULL)
2442                         rv = -EBUSY;
2443                 else
2444                         rv = entry->show(rdev, page);
2445                 mddev_unlock(mddev);
2446         }
2447         return rv;
2448 }
2449
2450 static ssize_t
2451 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2452               const char *page, size_t length)
2453 {
2454         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2455         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2456         ssize_t rv;
2457         mddev_t *mddev = rdev->mddev;
2458
2459         if (!entry->store)
2460                 return -EIO;
2461         if (!capable(CAP_SYS_ADMIN))
2462                 return -EACCES;
2463         rv = mddev ? mddev_lock(mddev): -EBUSY;
2464         if (!rv) {
2465                 if (rdev->mddev == NULL)
2466                         rv = -EBUSY;
2467                 else
2468                         rv = entry->store(rdev, page, length);
2469                 mddev_unlock(mddev);
2470         }
2471         return rv;
2472 }
2473
2474 static void rdev_free(struct kobject *ko)
2475 {
2476         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
2477         kfree(rdev);
2478 }
2479 static struct sysfs_ops rdev_sysfs_ops = {
2480         .show           = rdev_attr_show,
2481         .store          = rdev_attr_store,
2482 };
2483 static struct kobj_type rdev_ktype = {
2484         .release        = rdev_free,
2485         .sysfs_ops      = &rdev_sysfs_ops,
2486         .default_attrs  = rdev_default_attrs,
2487 };
2488
2489 /*
2490  * Import a device. If 'super_format' >= 0, then sanity check the superblock
2491  *
2492  * mark the device faulty if:
2493  *
2494  *   - the device is nonexistent (zero size)
2495  *   - the device has no valid superblock
2496  *
2497  * a faulty rdev _never_ has rdev->sb set.
2498  */
2499 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
2500 {
2501         char b[BDEVNAME_SIZE];
2502         int err;
2503         mdk_rdev_t *rdev;
2504         sector_t size;
2505
2506         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
2507         if (!rdev) {
2508                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
2509                 return ERR_PTR(-ENOMEM);
2510         }
2511
2512         if ((err = alloc_disk_sb(rdev)))
2513                 goto abort_free;
2514
2515         err = lock_rdev(rdev, newdev, super_format == -2);
2516         if (err)
2517                 goto abort_free;
2518
2519         kobject_init(&rdev->kobj, &rdev_ktype);
2520
2521         rdev->desc_nr = -1;
2522         rdev->saved_raid_disk = -1;
2523         rdev->raid_disk = -1;
2524         rdev->flags = 0;
2525         rdev->data_offset = 0;
2526         rdev->sb_events = 0;
2527         atomic_set(&rdev->nr_pending, 0);
2528         atomic_set(&rdev->read_errors, 0);
2529         atomic_set(&rdev->corrected_errors, 0);
2530
2531         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
2532         if (!size) {
2533                 printk(KERN_WARNING 
2534                         "md: %s has zero or unknown size, marking faulty!\n",
2535                         bdevname(rdev->bdev,b));
2536                 err = -EINVAL;
2537                 goto abort_free;
2538         }
2539
2540         if (super_format >= 0) {
2541                 err = super_types[super_format].
2542                         load_super(rdev, NULL, super_minor);
2543                 if (err == -EINVAL) {
2544                         printk(KERN_WARNING
2545                                 "md: %s does not have a valid v%d.%d "
2546                                "superblock, not importing!\n",
2547                                 bdevname(rdev->bdev,b),
2548                                super_format, super_minor);
2549                         goto abort_free;
2550                 }
2551                 if (err < 0) {
2552                         printk(KERN_WARNING 
2553                                 "md: could not read %s's sb, not importing!\n",
2554                                 bdevname(rdev->bdev,b));
2555                         goto abort_free;
2556                 }
2557         }
2558
2559         INIT_LIST_HEAD(&rdev->same_set);
2560         init_waitqueue_head(&rdev->blocked_wait);
2561
2562         return rdev;
2563
2564 abort_free:
2565         if (rdev->sb_page) {
2566                 if (rdev->bdev)
2567                         unlock_rdev(rdev);
2568                 free_disk_sb(rdev);
2569         }
2570         kfree(rdev);
2571         return ERR_PTR(err);
2572 }
2573
2574 /*
2575  * Check a full RAID array for plausibility
2576  */
2577
2578
2579 static void analyze_sbs(mddev_t * mddev)
2580 {
2581         int i;
2582         mdk_rdev_t *rdev, *freshest, *tmp;
2583         char b[BDEVNAME_SIZE];
2584
2585         freshest = NULL;
2586         rdev_for_each(rdev, tmp, mddev)
2587                 switch (super_types[mddev->major_version].
2588                         load_super(rdev, freshest, mddev->minor_version)) {
2589                 case 1:
2590                         freshest = rdev;
2591                         break;
2592                 case 0:
2593                         break;
2594                 default:
2595                         printk( KERN_ERR \
2596                                 "md: fatal superblock inconsistency in %s"
2597                                 " -- removing from array\n", 
2598                                 bdevname(rdev->bdev,b));
2599                         kick_rdev_from_array(rdev);
2600                 }
2601
2602
2603         super_types[mddev->major_version].
2604                 validate_super(mddev, freshest);
2605
2606         i = 0;
2607         rdev_for_each(rdev, tmp, mddev) {
2608                 if (rdev->desc_nr >= mddev->max_disks ||
2609                     i > mddev->max_disks) {
2610                         printk(KERN_WARNING
2611                                "md: %s: %s: only %d devices permitted\n",
2612                                mdname(mddev), bdevname(rdev->bdev, b),
2613                                mddev->max_disks);
2614                         kick_rdev_from_array(rdev);
2615                         continue;
2616                 }
2617                 if (rdev != freshest)
2618                         if (super_types[mddev->major_version].
2619                             validate_super(mddev, rdev)) {
2620                                 printk(KERN_WARNING "md: kicking non-fresh %s"
2621                                         " from array!\n",
2622                                         bdevname(rdev->bdev,b));
2623                                 kick_rdev_from_array(rdev);
2624                                 continue;
2625                         }
2626                 if (mddev->level == LEVEL_MULTIPATH) {
2627                         rdev->desc_nr = i++;
2628                         rdev->raid_disk = rdev->desc_nr;
2629                         set_bit(In_sync, &rdev->flags);
2630                 } else if (rdev->raid_disk >= mddev->raid_disks) {
2631                         rdev->raid_disk = -1;
2632                         clear_bit(In_sync, &rdev->flags);
2633                 }
2634         }
2635 }
2636
2637 static void md_safemode_timeout(unsigned long data);
2638
2639 static ssize_t
2640 safe_delay_show(mddev_t *mddev, char *page)
2641 {
2642         int msec = (mddev->safemode_delay*1000)/HZ;
2643         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
2644 }
2645 static ssize_t
2646 safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
2647 {
2648         int scale=1;
2649         int dot=0;
2650         int i;
2651         unsigned long msec;
2652         char buf[30];
2653
2654         /* remove a period, and count digits after it */
2655         if (len >= sizeof(buf))
2656                 return -EINVAL;
2657         strlcpy(buf, cbuf, sizeof(buf));
2658         for (i=0; i<len; i++) {
2659                 if (dot) {
2660                         if (isdigit(buf[i])) {
2661                                 buf[i-1] = buf[i];
2662                                 scale *= 10;
2663                         }
2664                         buf[i] = 0;
2665                 } else if (buf[i] == '.') {
2666                         dot=1;
2667                         buf[i] = 0;
2668                 }
2669         }
2670         if (strict_strtoul(buf, 10, &msec) < 0)
2671                 return -EINVAL;
2672         msec = (msec * 1000) / scale;
2673         if (msec == 0)
2674                 mddev->safemode_delay = 0;
2675         else {
2676                 unsigned long old_delay = mddev->safemode_delay;
2677                 mddev->safemode_delay = (msec*HZ)/1000;
2678                 if (mddev->safemode_delay == 0)
2679                         mddev->safemode_delay = 1;
2680                 if (mddev->safemode_delay < old_delay)
2681                         md_safemode_timeout((unsigned long)mddev);
2682         }
2683         return len;
2684 }
2685 static struct md_sysfs_entry md_safe_delay =
2686 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
2687
2688 static ssize_t
2689 level_show(mddev_t *mddev, char *page)
2690 {
2691         struct mdk_personality *p = mddev->pers;
2692         if (p)
2693                 return sprintf(page, "%s\n", p->name);
2694         else if (mddev->clevel[0])
2695                 return sprintf(page, "%s\n", mddev->clevel);
2696         else if (mddev->level != LEVEL_NONE)
2697                 return sprintf(page, "%d\n", mddev->level);
2698         else
2699                 return 0;
2700 }
2701
2702 static ssize_t
2703 level_store(mddev_t *mddev, const char *buf, size_t len)
2704 {
2705         char level[16];
2706         ssize_t rv = len;
2707         struct mdk_personality *pers;
2708         void *priv;
2709         mdk_rdev_t *rdev;
2710
2711         if (mddev->pers == NULL) {
2712                 if (len == 0)
2713                         return 0;
2714                 if (len >= sizeof(mddev->clevel))
2715                         return -ENOSPC;
2716                 strncpy(mddev->clevel, buf, len);
2717                 if (mddev->clevel[len-1] == '\n')
2718                         len--;
2719                 mddev->clevel[len] = 0;
2720                 mddev->level = LEVEL_NONE;
2721                 return rv;
2722         }
2723
2724         /* request to change the personality.  Need to ensure:
2725          *  - array is not engaged in resync/recovery/reshape
2726          *  - old personality can be suspended
2727          *  - new personality will access other array.
2728          */
2729
2730         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
2731                 return -EBUSY;
2732
2733         if (!mddev->pers->quiesce) {
2734                 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
2735                        mdname(mddev), mddev->pers->name);
2736                 return -EINVAL;
2737         }
2738
2739         /* Now find the new personality */
2740         if (len == 0 || len >= sizeof(level))
2741                 return -EINVAL;
2742         strncpy(level, buf, len);
2743         if (level[len-1] == '\n')
2744                 len--;
2745         level[len] = 0;
2746
2747         request_module("md-%s", level);
2748         spin_lock(&pers_lock);
2749         pers = find_pers(LEVEL_NONE, level);
2750         if (!pers || !try_module_get(pers->owner)) {
2751                 spin_unlock(&pers_lock);
2752                 printk(KERN_WARNING "md: personality %s not loaded\n", level);
2753                 return -EINVAL;
2754         }
2755         spin_unlock(&pers_lock);
2756
2757         if (pers == mddev->pers) {
2758                 /* Nothing to do! */
2759                 module_put(pers->owner);
2760                 return rv;
2761         }
2762         if (!pers->takeover) {
2763                 module_put(pers->owner);
2764                 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
2765                        mdname(mddev), level);
2766                 return -EINVAL;
2767         }
2768
2769         /* ->takeover must set new_* and/or delta_disks
2770          * if it succeeds, and may set them when it fails.
2771          */
2772         priv = pers->takeover(mddev);
2773         if (IS_ERR(priv)) {
2774                 mddev->new_level = mddev->level;
2775                 mddev->new_layout = mddev->layout;
2776                 mddev->new_chunk_sectors = mddev->chunk_sectors;
2777                 mddev->raid_disks -= mddev->delta_disks;
2778                 mddev->delta_disks = 0;
2779                 module_put(pers->owner);
2780                 printk(KERN_WARNING "md: %s: %s would not accept array\n",
2781                        mdname(mddev), level);
2782                 return PTR_ERR(priv);
2783         }
2784
2785         /* Looks like we have a winner */
2786         mddev_suspend(mddev);
2787         mddev->pers->stop(mddev);
2788         module_put(mddev->pers->owner);
2789         /* Invalidate devices that are now superfluous */
2790         list_for_each_entry(rdev, &mddev->disks, same_set)
2791                 if (rdev->raid_disk >= mddev->raid_disks) {
2792                         rdev->raid_disk = -1;
2793                         clear_bit(In_sync, &rdev->flags);
2794                 }
2795         mddev->pers = pers;
2796         mddev->private = priv;
2797         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2798         mddev->level = mddev->new_level;
2799         mddev->layout = mddev->new_layout;
2800         mddev->chunk_sectors = mddev->new_chunk_sectors;
2801         mddev->delta_disks = 0;
2802         pers->run(mddev);
2803         mddev_resume(mddev);
2804         set_bit(MD_CHANGE_DEVS, &mddev->flags);
2805         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2806         md_wakeup_thread(mddev->thread);
2807         return rv;
2808 }
2809
2810 static struct md_sysfs_entry md_level =
2811 __ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
2812
2813
2814 static ssize_t
2815 layout_show(mddev_t *mddev, char *page)
2816 {
2817         /* just a number, not meaningful for all levels */
2818         if (mddev->reshape_position != MaxSector &&
2819             mddev->layout != mddev->new_layout)
2820                 return sprintf(page, "%d (%d)\n",
2821                                mddev->new_layout, mddev->layout);
2822         return sprintf(page, "%d\n", mddev->layout);
2823 }
2824
2825 static ssize_t
2826 layout_store(mddev_t *mddev, const char *buf, size_t len)
2827 {
2828         char *e;
2829         unsigned long n = simple_strtoul(buf, &e, 10);
2830
2831         if (!*buf || (*e && *e != '\n'))
2832                 return -EINVAL;
2833
2834         if (mddev->pers) {
2835                 int err;
2836                 if (mddev->pers->check_reshape == NULL)
2837                         return -EBUSY;
2838                 mddev->new_layout = n;
2839                 err = mddev->pers->check_reshape(mddev);
2840                 if (err) {
2841                         mddev->new_layout = mddev->layout;
2842                         return err;
2843                 }
2844         } else {
2845                 mddev->new_layout = n;
2846                 if (mddev->reshape_position == MaxSector)
2847                         mddev->layout = n;
2848         }
2849         return len;
2850 }
2851 static struct md_sysfs_entry md_layout =
2852 __ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
2853
2854
2855 static ssize_t
2856 raid_disks_show(mddev_t *mddev, char *page)
2857 {
2858         if (mddev->raid_disks == 0)
2859                 return 0;
2860         if (mddev->reshape_position != MaxSector &&
2861             mddev->delta_disks != 0)
2862                 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
2863                                mddev->raid_disks - mddev->delta_disks);
2864         return sprintf(page, "%d\n", mddev->raid_disks);
2865 }
2866
2867 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2868
2869 static ssize_t
2870 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2871 {
2872         char *e;
2873         int rv = 0;
2874         unsigned long n = simple_strtoul(buf, &e, 10);
2875
2876         if (!*buf || (*e && *e != '\n'))
2877                 return -EINVAL;
2878
2879         if (mddev->pers)
2880                 rv = update_raid_disks(mddev, n);
2881         else if (mddev->reshape_position != MaxSector) {
2882                 int olddisks = mddev->raid_disks - mddev->delta_disks;
2883                 mddev->delta_disks = n - olddisks;
2884                 mddev->raid_disks = n;
2885         } else
2886                 mddev->raid_disks = n;
2887         return rv ? rv : len;
2888 }
2889 static struct md_sysfs_entry md_raid_disks =
2890 __ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
2891
2892 static ssize_t
2893 chunk_size_show(mddev_t *mddev, char *page)
2894 {
2895         if (mddev->reshape_position != MaxSector &&
2896             mddev->chunk_sectors != mddev->new_chunk_sectors)
2897                 return sprintf(page, "%d (%d)\n",
2898                                mddev->new_chunk_sectors << 9,
2899                                mddev->chunk_sectors << 9);
2900         return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
2901 }
2902
2903 static ssize_t
2904 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2905 {
2906         char *e;
2907         unsigned long n = simple_strtoul(buf, &e, 10);
2908
2909         if (!*buf || (*e && *e != '\n'))
2910                 return -EINVAL;
2911
2912         if (mddev->pers) {
2913                 int err;
2914                 if (mddev->pers->check_reshape == NULL)
2915                         return -EBUSY;
2916                 mddev->new_chunk_sectors = n >> 9;
2917                 err = mddev->pers->check_reshape(mddev);
2918                 if (err) {
2919                         mddev->new_chunk_sectors = mddev->chunk_sectors;
2920                         return err;
2921                 }
2922         } else {
2923                 mddev->new_chunk_sectors = n >> 9;
2924                 if (mddev->reshape_position == MaxSector)
2925                         mddev->chunk_sectors = n >> 9;
2926         }
2927         return len;
2928 }
2929 static struct md_sysfs_entry md_chunk_size =
2930 __ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
2931
2932 static ssize_t
2933 resync_start_show(mddev_t *mddev, char *page)
2934 {
2935         if (mddev->recovery_cp == MaxSector)
2936                 return sprintf(page, "none\n");
2937         return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
2938 }
2939
2940 static ssize_t
2941 resync_start_store(mddev_t *mddev, const char *buf, size_t len)
2942 {
2943         char *e;
2944         unsigned long long n = simple_strtoull(buf, &e, 10);
2945
2946         if (mddev->pers)
2947                 return -EBUSY;
2948         if (!*buf || (*e && *e != '\n'))
2949                 return -EINVAL;
2950
2951         mddev->recovery_cp = n;
2952         return len;
2953 }
2954 static struct md_sysfs_entry md_resync_start =
2955 __ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
2956
2957 /*
2958  * The array state can be:
2959  *
2960  * clear
2961  *     No devices, no size, no level
2962  *     Equivalent to STOP_ARRAY ioctl
2963  * inactive
2964  *     May have some settings, but array is not active
2965  *        all IO results in error
2966  *     When written, doesn't tear down array, but just stops it
2967  * suspended (not supported yet)
2968  *     All IO requests will block. The array can be reconfigured.
2969  *     Writing this, if accepted, will block until array is quiescent
2970  * readonly
2971  *     no resync can happen.  no superblocks get written.
2972  *     write requests fail
2973  * read-auto
2974  *     like readonly, but behaves like 'clean' on a write request.
2975  *
2976  * clean - no pending writes, but otherwise active.
2977  *     When written to inactive array, starts without resync
2978  *     If a write request arrives then
2979  *       if metadata is known, mark 'dirty' and switch to 'active'.
2980  *       if not known, block and switch to write-pending
2981  *     If written to an active array that has pending writes, then fails.
2982  * active
2983  *     fully active: IO and resync can be happening.
2984  *     When written to inactive array, starts with resync
2985  *
2986  * write-pending
2987  *     clean, but writes are blocked waiting for 'active' to be written.
2988  *
2989  * active-idle
2990  *     like active, but no writes have been seen for a while (100msec).
2991  *
2992  */
2993 enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
2994                    write_pending, active_idle, bad_word};
2995 static char *array_states[] = {
2996         "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
2997         "write-pending", "active-idle", NULL };
2998
2999 static int match_word(const char *word, char **list)
3000 {
3001         int n;
3002         for (n=0; list[n]; n++)
3003                 if (cmd_match(word, list[n]))
3004                         break;
3005         return n;
3006 }
3007
3008 static ssize_t
3009 array_state_show(mddev_t *mddev, char *page)
3010 {
3011         enum array_state st = inactive;
3012
3013         if (mddev->pers)
3014                 switch(mddev->ro) {
3015                 case 1:
3016                         st = readonly;
3017                         break;
3018                 case 2:
3019                         st = read_auto;
3020                         break;
3021                 case 0:
3022                         if (mddev->in_sync)
3023                                 st = clean;
3024                         else if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
3025                                 st = write_pending;
3026                         else if (mddev->safemode)
3027                                 st = active_idle;
3028                         else
3029                                 st = active;
3030                 }
3031         else {
3032                 if (list_empty(&mddev->disks) &&
3033                     mddev->raid_disks == 0 &&
3034                     mddev->dev_sectors == 0)
3035                         st = clear;
3036                 else
3037                         st = inactive;
3038         }
3039         return sprintf(page, "%s\n", array_states[st]);
3040 }
3041
3042 static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3043 static int do_md_run(mddev_t * mddev);
3044 static int restart_array(mddev_t *mddev);
3045
3046 static ssize_t
3047 array_state_store(mddev_t *mddev, const char *buf, size_t len)
3048 {
3049         int err = -EINVAL;
3050         enum array_state st = match_word(buf, array_states);
3051         switch(st) {
3052         case bad_word:
3053                 break;
3054         case clear:
3055                 /* stopping an active array */
3056                 if (atomic_read(&mddev->openers) > 0)
3057                         return -EBUSY;
3058                 err = do_md_stop(mddev, 0, 0);
3059                 break;
3060         case inactive:
3061                 /* stopping an active array */
3062                 if (mddev->pers) {
3063                         if (atomic_read(&mddev->openers) > 0)
3064                                 return -EBUSY;
3065                         err = do_md_stop(mddev, 2, 0);
3066                 } else
3067                         err = 0; /* already inactive */
3068                 break;
3069         case suspended:
3070                 break; /* not supported yet */
3071         case readonly:
3072                 if (mddev->pers)
3073                         err = do_md_stop(mddev, 1, 0);
3074                 else {
3075                         mddev->ro = 1;
3076                         set_disk_ro(mddev->gendisk, 1);
3077                         err = do_md_run(mddev);
3078                 }
3079                 break;
3080         case read_auto:
3081                 if (mddev->pers) {
3082                         if (mddev->ro == 0)
3083                                 err = do_md_stop(mddev, 1, 0);
3084                         else if (mddev->ro == 1)
3085                                 err = restart_array(mddev);
3086                         if (err == 0) {
3087                                 mddev->ro = 2;
3088                                 set_disk_ro(mddev->gendisk, 0);
3089                         }
3090                 } else {
3091                         mddev->ro = 2;
3092                         err = do_md_run(mddev);
3093                 }
3094                 break;
3095         case clean:
3096                 if (mddev->pers) {
3097                         restart_array(mddev);
3098                         spin_lock_irq(&mddev->write_lock);
3099                         if (atomic_read(&mddev->writes_pending) == 0) {
3100                                 if (mddev->in_sync == 0) {
3101                                         mddev->in_sync = 1;
3102                                         if (mddev->safemode == 1)
3103                                                 mddev->safemode = 0;
3104                                         if (mddev->persistent)
3105                                                 set_bit(MD_CHANGE_CLEAN,
3106                                                         &mddev->flags);
3107                                 }
3108                                 err = 0;
3109                         } else
3110                                 err = -EBUSY;
3111                         spin_unlock_irq(&mddev->write_lock);
3112                 } else
3113                         err = -EINVAL;
3114                 break;
3115         case active:
3116                 if (mddev->pers) {
3117                         restart_array(mddev);
3118                         if (mddev->external)
3119                                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
3120                         wake_up(&mddev->sb_wait);
3121                         err = 0;
3122                 } else {
3123                         mddev->ro = 0;
3124                         set_disk_ro(mddev->gendisk, 0);
3125                         err = do_md_run(mddev);
3126                 }
3127                 break;
3128         case write_pending:
3129         case active_idle:
3130                 /* these cannot be set */
3131                 break;
3132         }
3133         if (err)
3134                 return err;
3135         else {
3136                 sysfs_notify_dirent(mddev->sysfs_state);
3137                 return len;
3138         }
3139 }
3140 static struct md_sysfs_entry md_array_state =
3141 __ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3142
3143 static ssize_t
3144 null_show(mddev_t *mddev, char *page)
3145 {
3146         return -EINVAL;
3147 }
3148
3149 static ssize_t
3150 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3151 {
3152         /* buf must be %d:%d\n? giving major and minor numbers */
3153         /* The new device is added to the array.
3154          * If the array has a persistent superblock, we read the
3155          * superblock to initialise info and check validity.
3156          * Otherwise, only checking done is that in bind_rdev_to_array,
3157          * which mainly checks size.
3158          */
3159         char *e;
3160         int major = simple_strtoul(buf, &e, 10);
3161         int minor;
3162         dev_t dev;
3163         mdk_rdev_t *rdev;
3164         int err;
3165
3166         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3167                 return -EINVAL;
3168         minor = simple_strtoul(e+1, &e, 10);
3169         if (*e && *e != '\n')
3170                 return -EINVAL;
3171         dev = MKDEV(major, minor);
3172         if (major != MAJOR(dev) ||
3173             minor != MINOR(dev))
3174                 return -EOVERFLOW;
3175
3176
3177         if (mddev->persistent) {
3178                 rdev = md_import_device(dev, mddev->major_version,
3179                                         mddev->minor_version);
3180                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3181                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3182                                                        mdk_rdev_t, same_set);
3183                         err = super_types[mddev->major_version]
3184                                 .load_super(rdev, rdev0, mddev->minor_version);
3185                         if (err < 0)
3186                                 goto out;
3187                 }
3188         } else if (mddev->external)
3189                 rdev = md_import_device(dev, -2, -1);
3190         else
3191                 rdev = md_import_device(dev, -1, -1);
3192
3193         if (IS_ERR(rdev))
3194                 return PTR_ERR(rdev);
3195         err = bind_rdev_to_array(rdev, mddev);
3196  out:
3197         if (err)
3198                 export_rdev(rdev);
3199         return err ? err : len;
3200 }
3201
3202 static struct md_sysfs_entry md_new_device =
3203 __ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3204
3205 static ssize_t
3206 bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3207 {
3208         char *end;
3209         unsigned long chunk, end_chunk;
3210
3211         if (!mddev->bitmap)
3212                 goto out;
3213         /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3214         while (*buf) {
3215                 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3216                 if (buf == end) break;
3217                 if (*end == '-') { /* range */
3218                         buf = end + 1;
3219                         end_chunk = simple_strtoul(buf, &end, 0);
3220                         if (buf == end) break;
3221                 }
3222                 if (*end && !isspace(*end)) break;
3223                 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3224                 buf = end;
3225                 while (isspace(*buf)) buf++;
3226         }
3227         bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3228 out:
3229         return len;
3230 }
3231
3232 static struct md_sysfs_entry md_bitmap =
3233 __ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3234
3235 static ssize_t
3236 size_show(mddev_t *mddev, char *page)
3237 {
3238         return sprintf(page, "%llu\n",
3239                 (unsigned long long)mddev->dev_sectors / 2);
3240 }
3241
3242 static int update_size(mddev_t *mddev, sector_t num_sectors);
3243
3244 static ssize_t
3245 size_store(mddev_t *mddev, const char *buf, size_t len)
3246 {
3247         /* If array is inactive, we can reduce the component size, but
3248          * not increase it (except from 0).
3249          * If array is active, we can try an on-line resize
3250          */
3251         sector_t sectors;
3252         int err = strict_blocks_to_sectors(buf, &sectors);
3253
3254         if (err < 0)
3255                 return err;
3256         if (mddev->pers) {
3257                 err = update_size(mddev, sectors);
3258                 md_update_sb(mddev, 1);
3259         } else {
3260                 if (mddev->dev_sectors == 0 ||
3261                     mddev->dev_sectors > sectors)
3262                         mddev->dev_sectors = sectors;
3263                 else
3264                         err = -ENOSPC;
3265         }
3266         return err ? err : len;
3267 }
3268
3269 static struct md_sysfs_entry md_size =
3270 __ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3271
3272
3273 /* Metdata version.
3274  * This is one of
3275  *   'none' for arrays with no metadata (good luck...)
3276  *   'external' for arrays with externally managed metadata,
3277  * or N.M for internally known formats
3278  */
3279 static ssize_t
3280 metadata_show(mddev_t *mddev, char *page)
3281 {
3282         if (mddev->persistent)
3283                 return sprintf(page, "%d.%d\n",
3284                                mddev->major_version, mddev->minor_version);
3285         else if (mddev->external)
3286                 return sprintf(page, "external:%s\n", mddev->metadata_type);
3287         else
3288                 return sprintf(page, "none\n");
3289 }
3290
3291 static ssize_t
3292 metadata_store(mddev_t *mddev, const char *buf, size_t len)
3293 {
3294         int major, minor;
3295         char *e;
3296         /* Changing the details of 'external' metadata is
3297          * always permitted.  Otherwise there must be
3298          * no devices attached to the array.
3299          */
3300         if (mddev->external && strncmp(buf, "external:", 9) == 0)
3301                 ;
3302         else if (!list_empty(&mddev->disks))
3303                 return -EBUSY;
3304
3305         if (cmd_match(buf, "none")) {
3306                 mddev->persistent = 0;
3307                 mddev->external = 0;
3308                 mddev->major_version = 0;
3309                 mddev->minor_version = 90;
3310                 return len;
3311         }
3312         if (strncmp(buf, "external:", 9) == 0) {
3313                 size_t namelen = len-9;
3314                 if (namelen >= sizeof(mddev->metadata_type))
3315                         namelen = sizeof(mddev->metadata_type)-1;
3316                 strncpy(mddev->metadata_type, buf+9, namelen);
3317                 mddev->metadata_type[namelen] = 0;
3318                 if (namelen && mddev->metadata_type[namelen-1] == '\n')
3319                         mddev->metadata_type[--namelen] = 0;
3320                 mddev->persistent = 0;
3321                 mddev->external = 1;
3322                 mddev->major_version = 0;
3323                 mddev->minor_version = 90;
3324                 return len;
3325         }
3326         major = simple_strtoul(buf, &e, 10);
3327         if (e==buf || *e != '.')
3328                 return -EINVAL;
3329         buf = e+1;
3330         minor = simple_strtoul(buf, &e, 10);
3331         if (e==buf || (*e && *e != '\n') )
3332                 return -EINVAL;
3333         if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
3334                 return -ENOENT;
3335         mddev->major_version = major;
3336         mddev->minor_version = minor;
3337         mddev->persistent = 1;
3338         mddev->external = 0;
3339         return len;
3340 }
3341
3342 static struct md_sysfs_entry md_metadata =
3343 __ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
3344
3345 static ssize_t
3346 action_show(mddev_t *mddev, char *page)
3347 {
3348         char *type = "idle";
3349         if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3350                 type = "frozen";
3351         else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3352             (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
3353                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
3354                         type = "reshape";
3355                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
3356                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
3357                                 type = "resync";
3358                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
3359                                 type = "check";
3360                         else
3361                                 type = "repair";
3362                 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
3363                         type = "recover";
3364         }
3365         return sprintf(page, "%s\n", type);
3366 }
3367
3368 static ssize_t
3369 action_store(mddev_t *mddev, const char *page, size_t len)
3370 {
3371         if (!mddev->pers || !mddev->pers->sync_request)
3372                 return -EINVAL;
3373
3374         if (cmd_match(page, "frozen"))
3375                 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3376         else
3377                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
3378
3379         if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
3380                 if (mddev->sync_thread) {
3381                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
3382                         md_unregister_thread(mddev->sync_thread);
3383                         mddev->sync_thread = NULL;
3384                         mddev->recovery = 0;
3385                 }
3386         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
3387                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
3388                 return -EBUSY;
3389         else if (cmd_match(page, "resync"))
3390                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3391         else if (cmd_match(page, "recover")) {
3392                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
3393                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3394         } else if (cmd_match(page, "reshape")) {
3395                 int err;
3396                 if (mddev->pers->start_reshape == NULL)
3397                         return -EINVAL;
3398                 err = mddev->pers->start_reshape(mddev);
3399                 if (err)
3400                         return err;
3401                 sysfs_notify(&mddev->kobj, NULL, "degraded");
3402         } else {
3403                 if (cmd_match(page, "check"))
3404                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
3405                 else if (!cmd_match(page, "repair"))
3406                         return -EINVAL;
3407                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
3408                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
3409         }
3410         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3411         md_wakeup_thread(mddev->thread);
3412         sysfs_notify_dirent(mddev->sysfs_action);
3413         return len;
3414 }
3415
3416 static ssize_t
3417 mismatch_cnt_show(mddev_t *mddev, char *page)
3418 {
3419         return sprintf(page, "%llu\n",
3420                        (unsigned long long) mddev->resync_mismatches);
3421 }
3422
3423 static struct md_sysfs_entry md_scan_mode =
3424 __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
3425
3426
3427 static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
3428
3429 static ssize_t
3430 sync_min_show(mddev_t *mddev, char *page)
3431 {
3432         return sprintf(page, "%d (%s)\n", speed_min(mddev),
3433                        mddev->sync_speed_min ? "local": "system");
3434 }
3435
3436 static ssize_t
3437 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
3438 {
3439         int min;
3440         char *e;
3441         if (strncmp(buf, "system", 6)==0) {
3442                 mddev->sync_speed_min = 0;
3443                 return len;
3444         }
3445         min = simple_strtoul(buf, &e, 10);
3446         if (buf == e || (*e && *e != '\n') || min <= 0)
3447                 return -EINVAL;
3448         mddev->sync_speed_min = min;
3449         return len;
3450 }
3451
3452 static struct md_sysfs_entry md_sync_min =
3453 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
3454
3455 static ssize_t
3456 sync_max_show(mddev_t *mddev, char *page)
3457 {
3458         return sprintf(page, "%d (%s)\n", speed_max(mddev),
3459                        mddev->sync_speed_max ? "local": "system");
3460 }
3461
3462 static ssize_t
3463 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
3464 {
3465         int max;
3466         char *e;
3467         if (strncmp(buf, "system", 6)==0) {
3468                 mddev->sync_speed_max = 0;
3469                 return len;
3470         }
3471         max = simple_strtoul(buf, &e, 10);
3472         if (buf == e || (*e && *e != '\n') || max <= 0)
3473                 return -EINVAL;
3474         mddev->sync_speed_max = max;
3475         return len;
3476 }
3477
3478 static struct md_sysfs_entry md_sync_max =
3479 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
3480
3481 static ssize_t
3482 degraded_show(mddev_t *mddev, char *page)
3483 {
3484         return sprintf(page, "%d\n", mddev->degraded);
3485 }
3486 static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
3487
3488 static ssize_t
3489 sync_force_parallel_show(mddev_t *mddev, char *page)
3490 {
3491         return sprintf(page, "%d\n", mddev->parallel_resync);
3492 }
3493
3494 static ssize_t
3495 sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
3496 {
3497         long n;
3498
3499         if (strict_strtol(buf, 10, &n))
3500                 return -EINVAL;
3501
3502         if (n != 0 && n != 1)
3503                 return -EINVAL;
3504
3505         mddev->parallel_resync = n;
3506
3507         if (mddev->sync_thread)
3508                 wake_up(&resync_wait);
3509
3510         return len;
3511 }
3512
3513 /* force parallel resync, even with shared block devices */
3514 static struct md_sysfs_entry md_sync_force_parallel =
3515 __ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
3516        sync_force_parallel_show, sync_force_parallel_store);
3517
3518 static ssize_t
3519 sync_speed_show(mddev_t *mddev, char *page)
3520 {
3521         unsigned long resync, dt, db;
3522         if (mddev->curr_resync == 0)
3523                 return sprintf(page, "none\n");
3524         resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
3525         dt = (jiffies - mddev->resync_mark) / HZ;
3526         if (!dt) dt++;
3527         db = resync - mddev->resync_mark_cnt;
3528         return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
3529 }
3530
3531 static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
3532
3533 static ssize_t
3534 sync_completed_show(mddev_t *mddev, char *page)
3535 {
3536         unsigned long max_sectors, resync;
3537
3538         if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3539                 return sprintf(page, "none\n");
3540
3541         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
3542                 max_sectors = mddev->resync_max_sectors;
3543         else
3544                 max_sectors = mddev->dev_sectors;
3545
3546         resync = mddev->curr_resync_completed;
3547         return sprintf(page, "%lu / %lu\n", resync, max_sectors);
3548 }
3549
3550 static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
3551
3552 static ssize_t
3553 min_sync_show(mddev_t *mddev, char *page)
3554 {
3555         return sprintf(page, "%llu\n",
3556                        (unsigned long long)mddev->resync_min);
3557 }
3558 static ssize_t
3559 min_sync_store(mddev_t *mddev, const char *buf, size_t len)
3560 {
3561         unsigned long long min;
3562         if (strict_strtoull(buf, 10, &min))
3563                 return -EINVAL;
3564         if (min > mddev->resync_max)
3565                 return -EINVAL;
3566         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3567                 return -EBUSY;
3568
3569         /* Must be a multiple of chunk_size */
3570         if (mddev->chunk_sectors) {
3571                 sector_t temp = min;
3572                 if (sector_div(temp, mddev->chunk_sectors))
3573                         return -EINVAL;
3574         }
3575         mddev->resync_min = min;
3576
3577         return len;
3578 }
3579
3580 static struct md_sysfs_entry md_min_sync =
3581 __ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
3582
3583 static ssize_t
3584 max_sync_show(mddev_t *mddev, char *page)
3585 {
3586         if (mddev->resync_max == MaxSector)
3587                 return sprintf(page, "max\n");
3588         else
3589                 return sprintf(page, "%llu\n",
3590                                (unsigned long long)mddev->resync_max);
3591 }
3592 static ssize_t
3593 max_sync_store(mddev_t *mddev, const char *buf, size_t len)
3594 {
3595         if (strncmp(buf, "max", 3) == 0)
3596                 mddev->resync_max = MaxSector;
3597         else {
3598                 unsigned long long max;
3599                 if (strict_strtoull(buf, 10, &max))
3600                         return -EINVAL;
3601                 if (max < mddev->resync_min)
3602                         return -EINVAL;
3603                 if (max < mddev->resync_max &&
3604                     test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
3605                         return -EBUSY;
3606
3607                 /* Must be a multiple of chunk_size */
3608                 if (mddev->chunk_sectors) {
3609                         sector_t temp = max;
3610                         if (sector_div(temp, mddev->chunk_sectors))
3611                                 return -EINVAL;
3612                 }
3613                 mddev->resync_max = max;
3614         }
3615         wake_up(&mddev->recovery_wait);
3616         return len;
3617 }
3618
3619 static struct md_sysfs_entry md_max_sync =
3620 __ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
3621
3622 static ssize_t
3623 suspend_lo_show(mddev_t *mddev, char *page)
3624 {
3625         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
3626 }
3627
3628 static ssize_t
3629 suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
3630 {
3631         char *e;
3632         unsigned long long new = simple_strtoull(buf, &e, 10);
3633
3634         if (mddev->pers == NULL || 
3635             mddev->pers->quiesce == NULL)
3636                 return -EINVAL;
3637         if (buf == e || (*e && *e != '\n'))
3638                 return -EINVAL;
3639         if (new >= mddev->suspend_hi ||
3640             (new > mddev->suspend_lo && new < mddev->suspend_hi)) {
3641                 mddev->suspend_lo = new;
3642                 mddev->pers->quiesce(mddev, 2);
3643                 return len;
3644         } else
3645                 return -EINVAL;
3646 }
3647 static struct md_sysfs_entry md_suspend_lo =
3648 __ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
3649
3650
3651 static ssize_t
3652 suspend_hi_show(mddev_t *mddev, char *page)
3653 {
3654         return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
3655 }
3656
3657 static ssize_t
3658 suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
3659 {
3660         char *e;
3661         unsigned long long new = simple_strtoull(buf, &e, 10);
3662
3663         if (mddev->pers == NULL ||
3664             mddev->pers->quiesce == NULL)
3665                 return -EINVAL;
3666         if (buf == e || (*e && *e != '\n'))
3667                 return -EINVAL;
3668         if ((new <= mddev->suspend_lo && mddev->suspend_lo >= mddev->suspend_hi) ||
3669             (new > mddev->suspend_lo && new > mddev->suspend_hi)) {
3670                 mddev->suspend_hi = new;
3671                 mddev->pers->quiesce(mddev, 1);
3672                 mddev->pers->quiesce(mddev, 0);
3673                 return len;
3674         } else
3675                 return -EINVAL;
3676 }
3677 static struct md_sysfs_entry md_suspend_hi =
3678 __ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
3679
3680 static ssize_t
3681 reshape_position_show(mddev_t *mddev, char *page)
3682 {
3683         if (mddev->reshape_position != MaxSector)
3684                 return sprintf(page, "%llu\n",
3685                                (unsigned long long)mddev->reshape_position);
3686         strcpy(page, "none\n");
3687         return 5;
3688 }
3689
3690 static ssize_t
3691 reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
3692 {
3693         char *e;
3694         unsigned long long new = simple_strtoull(buf, &e, 10);
3695         if (mddev->pers)
3696                 return -EBUSY;
3697         if (buf == e || (*e && *e != '\n'))
3698                 return -EINVAL;
3699         mddev->reshape_position = new;
3700         mddev->delta_disks = 0;
3701         mddev->new_level = mddev->level;
3702         mddev->new_layout = mddev->layout;
3703         mddev->new_chunk_sectors = mddev->chunk_sectors;
3704         return len;
3705 }
3706
3707 static struct md_sysfs_entry md_reshape_position =
3708 __ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
3709        reshape_position_store);
3710
3711 static ssize_t
3712 array_size_show(mddev_t *mddev, char *page)
3713 {
3714         if (mddev->external_size)
3715                 return sprintf(page, "%llu\n",
3716                                (unsigned long long)mddev->array_sectors/2);
3717         else
3718                 return sprintf(page, "default\n");
3719 }
3720
3721 static ssize_t
3722 array_size_store(mddev_t *mddev, const char *buf, size_t len)
3723 {
3724         sector_t sectors;
3725
3726         if (strncmp(buf, "default", 7) == 0) {
3727                 if (mddev->pers)
3728                         sectors = mddev->pers->size(mddev, 0, 0);
3729                 else
3730                         sectors = mddev->array_sectors;
3731
3732                 mddev->external_size = 0;
3733         } else {
3734                 if (strict_blocks_to_sectors(buf, &sectors) < 0)
3735                         return -EINVAL;
3736                 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
3737                         return -E2BIG;
3738
3739                 mddev->external_size = 1;
3740         }
3741
3742         mddev->array_sectors = sectors;
3743         set_capacity(mddev->gendisk, mddev->array_sectors);
3744         if (mddev->pers)
3745                 revalidate_disk(mddev->gendisk);
3746
3747         return len;
3748 }
3749
3750 static struct md_sysfs_entry md_array_size =
3751 __ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
3752        array_size_store);
3753
3754 static struct attribute *md_default_attrs[] = {
3755         &md_level.attr,
3756         &md_layout.attr,
3757         &md_raid_disks.attr,
3758         &md_chunk_size.attr,
3759         &md_size.attr,
3760         &md_resync_start.attr,
3761         &md_metadata.attr,
3762         &md_new_device.attr,
3763         &md_safe_delay.attr,
3764         &md_array_state.attr,
3765         &md_reshape_position.attr,
3766         &md_array_size.attr,
3767         NULL,
3768 };
3769
3770 static struct attribute *md_redundancy_attrs[] = {
3771         &md_scan_mode.attr,
3772         &md_mismatches.attr,
3773         &md_sync_min.attr,
3774         &md_sync_max.attr,
3775         &md_sync_speed.attr,
3776         &md_sync_force_parallel.attr,
3777         &md_sync_completed.attr,
3778         &md_min_sync.attr,
3779         &md_max_sync.attr,
3780         &md_suspend_lo.attr,
3781         &md_suspend_hi.attr,
3782         &md_bitmap.attr,
3783         &md_degraded.attr,
3784         NULL,
3785 };
3786 static struct attribute_group md_redundancy_group = {
3787         .name = NULL,
3788         .attrs = md_redundancy_attrs,
3789 };
3790
3791
3792 static ssize_t
3793 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3794 {
3795         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3796         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3797         ssize_t rv;
3798
3799         if (!entry->show)
3800                 return -EIO;
3801         rv = mddev_lock(mddev);
3802         if (!rv) {
3803                 rv = entry->show(mddev, page);
3804                 mddev_unlock(mddev);
3805         }
3806         return rv;
3807 }
3808
3809 static ssize_t
3810 md_attr_store(struct kobject *kobj, struct attribute *attr,
3811               const char *page, size_t length)
3812 {
3813         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
3814         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
3815         ssize_t rv;
3816
3817         if (!entry->store)
3818                 return -EIO;
3819         if (!capable(CAP_SYS_ADMIN))
3820                 return -EACCES;
3821         rv = mddev_lock(mddev);
3822         if (mddev->hold_active == UNTIL_IOCTL)
3823                 mddev->hold_active = 0;
3824         if (!rv) {
3825                 rv = entry->store(mddev, page, length);
3826                 mddev_unlock(mddev);
3827         }
3828         return rv;
3829 }
3830
3831 static void md_free(struct kobject *ko)
3832 {
3833         mddev_t *mddev = container_of(ko, mddev_t, kobj);
3834
3835         if (mddev->sysfs_state)
3836                 sysfs_put(mddev->sysfs_state);
3837
3838         if (mddev->gendisk) {
3839                 del_gendisk(mddev->gendisk);
3840                 put_disk(mddev->gendisk);
3841         }
3842         if (mddev->queue)
3843                 blk_cleanup_queue(mddev->queue);
3844
3845         kfree(mddev);
3846 }
3847
3848 static struct sysfs_ops md_sysfs_ops = {
3849         .show   = md_attr_show,
3850         .store  = md_attr_store,
3851 };
3852 static struct kobj_type md_ktype = {
3853         .release        = md_free,
3854         .sysfs_ops      = &md_sysfs_ops,
3855         .default_attrs  = md_default_attrs,
3856 };
3857
3858 int mdp_major = 0;
3859
3860 static void mddev_delayed_delete(struct work_struct *ws)
3861 {
3862         mddev_t *mddev = container_of(ws, mddev_t, del_work);
3863
3864         if (mddev->private == &md_redundancy_group) {
3865                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
3866                 if (mddev->sysfs_action)
3867                         sysfs_put(mddev->sysfs_action);
3868                 mddev->sysfs_action = NULL;
3869                 mddev->private = NULL;
3870         }
3871         kobject_del(&mddev->kobj);
3872         kobject_put(&mddev->kobj);
3873 }
3874
3875 static int md_alloc(dev_t dev, char *name)
3876 {
3877         static DEFINE_MUTEX(disks_mutex);
3878         mddev_t *mddev = mddev_find(dev);
3879         struct gendisk *disk;
3880         int partitioned;
3881         int shift;
3882         int unit;
3883         int error;
3884
3885         if (!mddev)
3886                 return -ENODEV;
3887
3888         partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
3889         shift = partitioned ? MdpMinorShift : 0;
3890         unit = MINOR(mddev->unit) >> shift;
3891
3892         /* wait for any previous instance if this device
3893          * to be completed removed (mddev_delayed_delete).
3894          */
3895         flush_scheduled_work();
3896
3897         mutex_lock(&disks_mutex);
3898         error = -EEXIST;
3899         if (mddev->gendisk)
3900                 goto abort;
3901
3902         if (name) {
3903                 /* Need to ensure that 'name' is not a duplicate.
3904                  */
3905                 mddev_t *mddev2;
3906                 spin_lock(&all_mddevs_lock);
3907
3908                 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
3909                         if (mddev2->gendisk &&
3910                             strcmp(mddev2->gendisk->disk_name, name) == 0) {
3911                                 spin_unlock(&all_mddevs_lock);
3912                                 goto abort;
3913                         }
3914                 spin_unlock(&all_mddevs_lock);
3915         }
3916
3917         error = -ENOMEM;
3918         mddev->queue = blk_alloc_queue(GFP_KERNEL);
3919         if (!mddev->queue)
3920                 goto abort;
3921         mddev->queue->queuedata = mddev;
3922
3923         /* Can be unlocked because the queue is new: no concurrency */
3924         queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue);
3925
3926         blk_queue_make_request(mddev->queue, md_make_request);
3927
3928         disk = alloc_disk(1 << shift);
3929         if (!disk) {
3930                 blk_cleanup_queue(mddev->queue);
3931                 mddev->queue = NULL;
3932                 goto abort;
3933         }
3934         disk->major = MAJOR(mddev->unit);
3935         disk->first_minor = unit << shift;
3936         if (name)
3937                 strcpy(disk->disk_name, name);
3938         else if (partitioned)
3939                 sprintf(disk->disk_name, "md_d%d", unit);
3940         else
3941                 sprintf(disk->disk_name, "md%d", unit);
3942         disk->fops = &md_fops;
3943         disk->private_data = mddev;
3944         disk->queue = mddev->queue;
3945         /* Allow extended partitions.  This makes the
3946          * 'mdp' device redundant, but we can't really
3947          * remove it now.
3948          */
3949         disk->flags |= GENHD_FL_EXT_DEVT;
3950         add_disk(disk);
3951         mddev->gendisk = disk;
3952         error = kobject_init_and_add(&mddev->kobj, &md_ktype,
3953                                      &disk_to_dev(disk)->kobj, "%s", "md");
3954         if (error) {
3955                 /* This isn't possible, but as kobject_init_and_add is marked
3956                  * __must_check, we must do something with the result
3957                  */
3958                 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
3959                        disk->disk_name);
3960                 error = 0;
3961         }
3962  abort:
3963         mutex_unlock(&disks_mutex);
3964         if (!error) {
3965                 kobject_uevent(&mddev->kobj, KOBJ_ADD);
3966                 mddev->sysfs_state = sysfs_get_dirent(mddev->kobj.sd, "array_state");
3967         }
3968         mddev_put(mddev);
3969         return error;
3970 }
3971
3972 static struct kobject *md_probe(dev_t dev, int *part, void *data)
3973 {
3974         md_alloc(dev, NULL);
3975         return NULL;
3976 }
3977
3978 static int add_named_array(const char *val, struct kernel_param *kp)
3979 {
3980         /* val must be "md_*" where * is not all digits.
3981          * We allocate an array with a large free minor number, and
3982          * set the name to val.  val must not already be an active name.
3983          */
3984         int len = strlen(val);
3985         char buf[DISK_NAME_LEN];
3986
3987         while (len && val[len-1] == '\n')
3988                 len--;
3989         if (len >= DISK_NAME_LEN)
3990                 return -E2BIG;
3991         strlcpy(buf, val, len+1);
3992         if (strncmp(buf, "md_", 3) != 0)
3993                 return -EINVAL;
3994         return md_alloc(0, buf);
3995 }
3996
3997 static void md_safemode_timeout(unsigned long data)
3998 {
3999         mddev_t *mddev = (mddev_t *) data;
4000
4001         if (!atomic_read(&mddev->writes_pending)) {
4002                 mddev->safemode = 1;
4003                 if (mddev->external)
4004                         sysfs_notify_dirent(mddev->sysfs_state);
4005         }
4006         md_wakeup_thread(mddev->thread);
4007 }
4008
4009 static int start_dirty_degraded;
4010
4011 static int do_md_run(mddev_t * mddev)
4012 {
4013         int err;
4014         mdk_rdev_t *rdev;
4015         struct gendisk *disk;
4016         struct mdk_personality *pers;
4017
4018         if (list_empty(&mddev->disks))
4019                 /* cannot run an array with no devices.. */
4020                 return -EINVAL;
4021
4022         if (mddev->pers)
4023                 return -EBUSY;
4024
4025         /*
4026          * Analyze all RAID superblock(s)
4027          */
4028         if (!mddev->raid_disks) {
4029                 if (!mddev->persistent)
4030                         return -EINVAL;
4031                 analyze_sbs(mddev);
4032         }
4033
4034         if (mddev->level != LEVEL_NONE)
4035                 request_module("md-level-%d", mddev->level);
4036         else if (mddev->clevel[0])
4037                 request_module("md-%s", mddev->clevel);
4038
4039         /*
4040          * Drop all container device buffers, from now on
4041          * the only valid external interface is through the md
4042          * device.
4043          */
4044         list_for_each_entry(rdev, &mddev->disks, same_set) {
4045                 if (test_bit(Faulty, &rdev->flags))
4046                         continue;
4047                 sync_blockdev(rdev->bdev);
4048                 invalidate_bdev(rdev->bdev);
4049
4050                 /* perform some consistency tests on the device.
4051                  * We don't want the data to overlap the metadata,
4052                  * Internal Bitmap issues have been handled elsewhere.
4053                  */
4054                 if (rdev->data_offset < rdev->sb_start) {
4055                         if (mddev->dev_sectors &&
4056                             rdev->data_offset + mddev->dev_sectors
4057                             > rdev->sb_start) {
4058                                 printk("md: %s: data overlaps metadata\n",
4059                                        mdname(mddev));
4060                                 return -EINVAL;
4061                         }
4062                 } else {
4063                         if (rdev->sb_start + rdev->sb_size/512
4064                             > rdev->data_offset) {
4065                                 printk("md: %s: metadata overlaps data\n",
4066                                        mdname(mddev));
4067                                 return -EINVAL;
4068                         }
4069                 }
4070                 sysfs_notify_dirent(rdev->sysfs_state);
4071         }
4072
4073         md_probe(mddev->unit, NULL, NULL);
4074         disk = mddev->gendisk;
4075         if (!disk)
4076                 return -ENOMEM;
4077
4078         spin_lock(&pers_lock);
4079         pers = find_pers(mddev->level, mddev->clevel);
4080         if (!pers || !try_module_get(pers->owner)) {
4081                 spin_unlock(&pers_lock);
4082                 if (mddev->level != LEVEL_NONE)
4083                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4084                                mddev->level);
4085                 else
4086                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4087                                mddev->clevel);
4088                 return -EINVAL;
4089         }
4090         mddev->pers = pers;
4091         spin_unlock(&pers_lock);
4092         if (mddev->level != pers->level) {
4093                 mddev->level = pers->level;
4094                 mddev->new_level = pers->level;
4095         }
4096         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4097
4098         if (mddev->reshape_position != MaxSector &&
4099             pers->start_reshape == NULL) {
4100                 /* This personality cannot handle reshaping... */
4101                 mddev->pers = NULL;
4102                 module_put(pers->owner);
4103                 return -EINVAL;
4104         }
4105
4106         if (pers->sync_request) {
4107                 /* Warn if this is a potentially silly
4108                  * configuration.
4109                  */
4110                 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4111                 mdk_rdev_t *rdev2;
4112                 int warned = 0;
4113
4114                 list_for_each_entry(rdev, &mddev->disks, same_set)
4115                         list_for_each_entry(rdev2, &mddev->disks, same_set) {
4116                                 if (rdev < rdev2 &&
4117                                     rdev->bdev->bd_contains ==
4118                                     rdev2->bdev->bd_contains) {
4119                                         printk(KERN_WARNING
4120                                                "%s: WARNING: %s appears to be"
4121                                                " on the same physical disk as"
4122                                                " %s.\n",
4123                                                mdname(mddev),
4124                                                bdevname(rdev->bdev,b),
4125                                                bdevname(rdev2->bdev,b2));
4126                                         warned = 1;
4127                                 }
4128                         }
4129
4130                 if (warned)
4131                         printk(KERN_WARNING
4132                                "True protection against single-disk"
4133                                " failure might be compromised.\n");
4134         }
4135
4136         mddev->recovery = 0;
4137         /* may be over-ridden by personality */
4138         mddev->resync_max_sectors = mddev->dev_sectors;
4139
4140         mddev->barriers_work = 1;
4141         mddev->ok_start_degraded = start_dirty_degraded;
4142
4143         if (start_readonly)
4144                 mddev->ro = 2; /* read-only, but switch on first write */
4145
4146         err = mddev->pers->run(mddev);
4147         if (err)
4148                 printk(KERN_ERR "md: pers->run() failed ...\n");
4149         else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4150                 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4151                           " but 'external_size' not in effect?\n", __func__);
4152                 printk(KERN_ERR
4153                        "md: invalid array_size %llu > default size %llu\n",
4154                        (unsigned long long)mddev->array_sectors / 2,
4155                        (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4156                 err = -EINVAL;
4157                 mddev->pers->stop(mddev);
4158         }
4159         if (err == 0 && mddev->pers->sync_request) {
4160                 err = bitmap_create(mddev);
4161                 if (err) {
4162                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4163                                mdname(mddev), err);
4164                         mddev->pers->stop(mddev);
4165                 }
4166         }
4167         if (err) {
4168                 module_put(mddev->pers->owner);
4169                 mddev->pers = NULL;
4170                 bitmap_destroy(mddev);
4171                 return err;
4172         }
4173         if (mddev->pers->sync_request) {
4174                 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4175                         printk(KERN_WARNING
4176                                "md: cannot register extra attributes for %s\n",
4177                                mdname(mddev));
4178                 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
4179         } else if (mddev->ro == 2) /* auto-readonly not meaningful */
4180                 mddev->ro = 0;
4181
4182         atomic_set(&mddev->writes_pending,0);
4183         mddev->safemode = 0;
4184         mddev->safemode_timer.function = md_safemode_timeout;
4185         mddev->safemode_timer.data = (unsigned long) mddev;
4186         mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4187         mddev->in_sync = 1;
4188
4189         list_for_each_entry(rdev, &mddev->disks, same_set)
4190                 if (rdev->raid_disk >= 0) {
4191                         char nm[20];
4192                         sprintf(nm, "rd%d", rdev->raid_disk);
4193                         if (sysfs_create_link(&mddev->kobj, &rdev->kobj, nm))
4194                                 printk("md: cannot register %s for %s\n",
4195                                        nm, mdname(mddev));
4196                 }
4197         
4198         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4199         
4200         if (mddev->flags)
4201                 md_update_sb(mddev, 0);
4202
4203         set_capacity(disk, mddev->array_sectors);
4204
4205         /* If there is a partially-recovered drive we need to
4206          * start recovery here.  If we leave it to md_check_recovery,
4207          * it will remove the drives and not do the right thing
4208          */
4209         if (mddev->degraded && !mddev->sync_thread) {
4210                 int spares = 0;
4211                 list_for_each_entry(rdev, &mddev->disks, same_set)
4212                         if (rdev->raid_disk >= 0 &&
4213                             !test_bit(In_sync, &rdev->flags) &&
4214                             !test_bit(Faulty, &rdev->flags))
4215                                 /* complete an interrupted recovery */
4216                                 spares++;
4217                 if (spares && mddev->pers->sync_request) {
4218                         mddev->recovery = 0;
4219                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4220                         mddev->sync_thread = md_register_thread(md_do_sync,
4221                                                                 mddev,
4222                                                                 "%s_resync");
4223                         if (!mddev->sync_thread) {
4224                                 printk(KERN_ERR "%s: could not start resync"
4225                                        " thread...\n",
4226                                        mdname(mddev));
4227                                 /* leave the spares where they are, it shouldn't hurt */
4228                                 mddev->recovery = 0;
4229                         }
4230                 }
4231         }
4232         md_wakeup_thread(mddev->thread);
4233         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4234
4235         revalidate_disk(mddev->gendisk);
4236         mddev->changed = 1;
4237         md_new_event(mddev);
4238         sysfs_notify_dirent(mddev->sysfs_state);
4239         if (mddev->sysfs_action)
4240                 sysfs_notify_dirent(mddev->sysfs_action);
4241         sysfs_notify(&mddev->kobj, NULL, "degraded");
4242         kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4243         return 0;
4244 }
4245
4246 static int restart_array(mddev_t *mddev)
4247 {
4248         struct gendisk *disk = mddev->gendisk;
4249
4250         /* Complain if it has no devices */
4251         if (list_empty(&mddev->disks))
4252                 return -ENXIO;
4253         if (!mddev->pers)
4254                 return -EINVAL;
4255         if (!mddev->ro)
4256                 return -EBUSY;
4257         mddev->safemode = 0;
4258         mddev->ro = 0;
4259         set_disk_ro(disk, 0);
4260         printk(KERN_INFO "md: %s switched to read-write mode.\n",
4261                 mdname(mddev));
4262         /* Kick recovery or resync if necessary */
4263         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4264         md_wakeup_thread(mddev->thread);
4265         md_wakeup_thread(mddev->sync_thread);
4266         sysfs_notify_dirent(mddev->sysfs_state);
4267         return 0;
4268 }
4269
4270 /* similar to deny_write_access, but accounts for our holding a reference
4271  * to the file ourselves */
4272 static int deny_bitmap_write_access(struct file * file)
4273 {
4274         struct inode *inode = file->f_mapping->host;
4275
4276         spin_lock(&inode->i_lock);
4277         if (atomic_read(&inode->i_writecount) > 1) {
4278                 spin_unlock(&inode->i_lock);
4279                 return -ETXTBSY;
4280         }
4281         atomic_set(&inode->i_writecount, -1);
4282         spin_unlock(&inode->i_lock);
4283
4284         return 0;
4285 }
4286
4287 static void restore_bitmap_write_access(struct file *file)
4288 {
4289         struct inode *inode = file->f_mapping->host;
4290
4291         spin_lock(&inode->i_lock);
4292         atomic_set(&inode->i_writecount, 1);
4293         spin_unlock(&inode->i_lock);
4294 }
4295
4296 /* mode:
4297  *   0 - completely stop and dis-assemble array
4298  *   1 - switch to readonly
4299  *   2 - stop but do not disassemble array
4300  */
4301 static int do_md_stop(mddev_t * mddev, int mode, int is_open)
4302 {
4303         int err = 0;
4304         struct gendisk *disk = mddev->gendisk;
4305         mdk_rdev_t *rdev;
4306
4307         if (atomic_read(&mddev->openers) > is_open) {
4308                 printk("md: %s still in use.\n",mdname(mddev));
4309                 return -EBUSY;
4310         }
4311
4312         if (mddev->pers) {
4313
4314                 if (mddev->sync_thread) {
4315                         set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4316                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4317                         md_unregister_thread(mddev->sync_thread);
4318                         mddev->sync_thread = NULL;
4319                 }
4320
4321                 del_timer_sync(&mddev->safemode_timer);
4322
4323                 switch(mode) {
4324                 case 1: /* readonly */
4325                         err  = -ENXIO;
4326                         if (mddev->ro==1)
4327                                 goto out;
4328                         mddev->ro = 1;
4329                         break;
4330                 case 0: /* disassemble */
4331                 case 2: /* stop */
4332                         bitmap_flush(mddev);
4333                         md_super_wait(mddev);
4334                         if (mddev->ro)
4335                                 set_disk_ro(disk, 0);
4336
4337                         mddev->pers->stop(mddev);
4338                         mddev->queue->merge_bvec_fn = NULL;
4339                         mddev->queue->unplug_fn = NULL;
4340                         mddev->queue->backing_dev_info.congested_fn = NULL;
4341                         module_put(mddev->pers->owner);
4342                         if (mddev->pers->sync_request)
4343                                 mddev->private = &md_redundancy_group;
4344                         mddev->pers = NULL;
4345                         /* tell userspace to handle 'inactive' */
4346                         sysfs_notify_dirent(mddev->sysfs_state);
4347
4348                         list_for_each_entry(rdev, &mddev->disks, same_set)
4349                                 if (rdev->raid_disk >= 0) {
4350                                         char nm[20];
4351                                         sprintf(nm, "rd%d", rdev->raid_disk);
4352                                         sysfs_remove_link(&mddev->kobj, nm);
4353                                 }
4354
4355                         set_capacity(disk, 0);
4356                         mddev->changed = 1;
4357
4358                         if (mddev->ro)
4359                                 mddev->ro = 0;
4360                 }
4361                 if (!mddev->in_sync || mddev->flags) {
4362                         /* mark array as shutdown cleanly */
4363                         mddev->in_sync = 1;
4364                         md_update_sb(mddev, 1);
4365                 }
4366                 if (mode == 1)
4367                         set_disk_ro(disk, 1);
4368                 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4369         }
4370
4371         /*
4372          * Free resources if final stop
4373          */
4374         if (mode == 0) {
4375
4376                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
4377
4378                 bitmap_destroy(mddev);
4379                 if (mddev->bitmap_file) {
4380                         restore_bitmap_write_access(mddev->bitmap_file);
4381                         fput(mddev->bitmap_file);
4382                         mddev->bitmap_file = NULL;
4383                 }
4384                 mddev->bitmap_offset = 0;
4385
4386                 /* make sure all md_delayed_delete calls have finished */
4387                 flush_scheduled_work();
4388
4389                 export_array(mddev);
4390
4391                 mddev->array_sectors = 0;
4392                 mddev->external_size = 0;
4393                 mddev->dev_sectors = 0;
4394                 mddev->raid_disks = 0;
4395                 mddev->recovery_cp = 0;
4396                 mddev->resync_min = 0;
4397                 mddev->resync_max = MaxSector;
4398                 mddev->reshape_position = MaxSector;
4399                 mddev->external = 0;
4400                 mddev->persistent = 0;
4401                 mddev->level = LEVEL_NONE;
4402                 mddev->clevel[0] = 0;
4403                 mddev->flags = 0;
4404                 mddev->ro = 0;
4405                 mddev->metadata_type[0] = 0;
4406                 mddev->chunk_sectors = 0;
4407                 mddev->ctime = mddev->utime = 0;
4408                 mddev->layout = 0;
4409                 mddev->max_disks = 0;
4410                 mddev->events = 0;
4411                 mddev->delta_disks = 0;
4412                 mddev->new_level = LEVEL_NONE;
4413                 mddev->new_layout = 0;
4414                 mddev->new_chunk_sectors = 0;
4415                 mddev->curr_resync = 0;
4416                 mddev->resync_mismatches = 0;
4417                 mddev->suspend_lo = mddev->suspend_hi = 0;
4418                 mddev->sync_speed_min = mddev->sync_speed_max = 0;
4419                 mddev->recovery = 0;
4420                 mddev->in_sync = 0;
4421                 mddev->changed = 0;
4422                 mddev->degraded = 0;
4423                 mddev->barriers_work = 0;
4424                 mddev->safemode = 0;
4425                 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4426                 if (mddev->hold_active == UNTIL_STOP)
4427                         mddev->hold_active = 0;
4428
4429         } else if (mddev->pers)
4430                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
4431                         mdname(mddev));
4432         err = 0;
4433         blk_integrity_unregister(disk);
4434         md_new_event(mddev);
4435         sysfs_notify_dirent(mddev->sysfs_state);
4436 out:
4437         return err;
4438 }
4439
4440 #ifndef MODULE
4441 static void autorun_array(mddev_t *mddev)
4442 {
4443         mdk_rdev_t *rdev;
4444         int err;
4445
4446         if (list_empty(&mddev->disks))
4447                 return;
4448
4449         printk(KERN_INFO "md: running: ");
4450
4451         list_for_each_entry(rdev, &mddev->disks, same_set) {
4452                 char b[BDEVNAME_SIZE];
4453                 printk("<%s>", bdevname(rdev->bdev,b));
4454         }
4455         printk("\n");
4456
4457         err = do_md_run(mddev);
4458         if (err) {
4459                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
4460                 do_md_stop(mddev, 0, 0);
4461         }
4462 }
4463
4464 /*
4465  * lets try to run arrays based on all disks that have arrived
4466  * until now. (those are in pending_raid_disks)
4467  *
4468  * the method: pick the first pending disk, collect all disks with
4469  * the same UUID, remove all from the pending list and put them into
4470  * the 'same_array' list. Then order this list based on superblock
4471  * update time (freshest comes first), kick out 'old' disks and
4472  * compare superblocks. If everything's fine then run it.
4473  *
4474  * If "unit" is allocated, then bump its reference count
4475  */
4476 static void autorun_devices(int part)
4477 {
4478         mdk_rdev_t *rdev0, *rdev, *tmp;
4479         mddev_t *mddev;
4480         char b[BDEVNAME_SIZE];
4481
4482         printk(KERN_INFO "md: autorun ...\n");
4483         while (!list_empty(&pending_raid_disks)) {
4484                 int unit;
4485                 dev_t dev;
4486                 LIST_HEAD(candidates);
4487                 rdev0 = list_entry(pending_raid_disks.next,
4488                                          mdk_rdev_t, same_set);
4489
4490                 printk(KERN_INFO "md: considering %s ...\n",
4491                         bdevname(rdev0->bdev,b));
4492                 INIT_LIST_HEAD(&candidates);
4493                 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
4494                         if (super_90_load(rdev, rdev0, 0) >= 0) {
4495                                 printk(KERN_INFO "md:  adding %s ...\n",
4496                                         bdevname(rdev->bdev,b));
4497                                 list_move(&rdev->same_set, &candidates);
4498                         }
4499                 /*
4500                  * now we have a set of devices, with all of them having
4501                  * mostly sane superblocks. It's time to allocate the
4502                  * mddev.
4503                  */
4504                 if (part) {
4505                         dev = MKDEV(mdp_major,
4506                                     rdev0->preferred_minor << MdpMinorShift);
4507                         unit = MINOR(dev) >> MdpMinorShift;
4508                 } else {
4509                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
4510                         unit = MINOR(dev);
4511                 }
4512                 if (rdev0->preferred_minor != unit) {
4513                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
4514                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
4515                         break;
4516                 }
4517
4518                 md_probe(dev, NULL, NULL);
4519                 mddev = mddev_find(dev);
4520                 if (!mddev || !mddev->gendisk) {
4521                         if (mddev)
4522                                 mddev_put(mddev);
4523                         printk(KERN_ERR
4524                                 "md: cannot allocate memory for md drive.\n");
4525                         break;
4526                 }
4527                 if (mddev_lock(mddev)) 
4528                         printk(KERN_WARNING "md: %s locked, cannot run\n",
4529                                mdname(mddev));
4530                 else if (mddev->raid_disks || mddev->major_version
4531                          || !list_empty(&mddev->disks)) {
4532                         printk(KERN_WARNING 
4533                                 "md: %s already running, cannot run %s\n",
4534                                 mdname(mddev), bdevname(rdev0->bdev,b));
4535                         mddev_unlock(mddev);
4536                 } else {
4537                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
4538                         mddev->persistent = 1;
4539                         rdev_for_each_list(rdev, tmp, &candidates) {
4540                                 list_del_init(&rdev->same_set);
4541                                 if (bind_rdev_to_array(rdev, mddev))
4542                                         export_rdev(rdev);
4543                         }
4544                         autorun_array(mddev);
4545                         mddev_unlock(mddev);
4546                 }
4547                 /* on success, candidates will be empty, on error
4548                  * it won't...
4549                  */
4550                 rdev_for_each_list(rdev, tmp, &candidates) {
4551                         list_del_init(&rdev->same_set);
4552                         export_rdev(rdev);
4553                 }
4554                 mddev_put(mddev);
4555         }
4556         printk(KERN_INFO "md: ... autorun DONE.\n");
4557 }
4558 #endif /* !MODULE */
4559
4560 static int get_version(void __user * arg)
4561 {
4562         mdu_version_t ver;
4563
4564         ver.major = MD_MAJOR_VERSION;
4565         ver.minor = MD_MINOR_VERSION;
4566         ver.patchlevel = MD_PATCHLEVEL_VERSION;
4567
4568         if (copy_to_user(arg, &ver, sizeof(ver)))
4569                 return -EFAULT;
4570
4571         return 0;
4572 }
4573
4574 static int get_array_info(mddev_t * mddev, void __user * arg)
4575 {
4576         mdu_array_info_t info;
4577         int nr,working,active,failed,spare;
4578         mdk_rdev_t *rdev;
4579
4580         nr=working=active=failed=spare=0;
4581         list_for_each_entry(rdev, &mddev->disks, same_set) {
4582                 nr++;
4583                 if (test_bit(Faulty, &rdev->flags))
4584                         failed++;
4585                 else {
4586                         working++;
4587                         if (test_bit(In_sync, &rdev->flags))
4588                                 active++;       
4589                         else
4590                                 spare++;
4591                 }
4592         }
4593
4594         info.major_version = mddev->major_version;
4595         info.minor_version = mddev->minor_version;
4596         info.patch_version = MD_PATCHLEVEL_VERSION;
4597         info.ctime         = mddev->ctime;
4598         info.level         = mddev->level;
4599         info.size          = mddev->dev_sectors / 2;
4600         if (info.size != mddev->dev_sectors / 2) /* overflow */
4601                 info.size = -1;
4602         info.nr_disks      = nr;
4603         info.raid_disks    = mddev->raid_disks;
4604         info.md_minor      = mddev->md_minor;
4605         info.not_persistent= !mddev->persistent;
4606
4607         info.utime         = mddev->utime;
4608         info.state         = 0;
4609         if (mddev->in_sync)
4610                 info.state = (1<<MD_SB_CLEAN);
4611         if (mddev->bitmap && mddev->bitmap_offset)
4612                 info.state = (1<<MD_SB_BITMAP_PRESENT);
4613         info.active_disks  = active;
4614         info.working_disks = working;
4615         info.failed_disks  = failed;
4616         info.spare_disks   = spare;
4617
4618         info.layout        = mddev->layout;
4619         info.chunk_size    = mddev->chunk_sectors << 9;
4620
4621         if (copy_to_user(arg, &info, sizeof(info)))
4622                 return -EFAULT;
4623
4624         return 0;
4625 }
4626
4627 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
4628 {
4629         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
4630         char *ptr, *buf = NULL;
4631         int err = -ENOMEM;
4632
4633         if (md_allow_write(mddev))
4634                 file = kmalloc(sizeof(*file), GFP_NOIO);
4635         else
4636                 file = kmalloc(sizeof(*file), GFP_KERNEL);
4637
4638         if (!file)
4639                 goto out;
4640
4641         /* bitmap disabled, zero the first byte and copy out */
4642         if (!mddev->bitmap || !mddev->bitmap->file) {
4643                 file->pathname[0] = '\0';
4644                 goto copy_out;
4645         }
4646
4647         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
4648         if (!buf)
4649                 goto out;
4650
4651         ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
4652         if (IS_ERR(ptr))
4653                 goto out;
4654
4655         strcpy(file->pathname, ptr);
4656
4657 copy_out:
4658         err = 0;
4659         if (copy_to_user(arg, file, sizeof(*file)))
4660                 err = -EFAULT;
4661 out:
4662         kfree(buf);
4663         kfree(file);
4664         return err;
4665 }
4666
4667 static int get_disk_info(mddev_t * mddev, void __user * arg)
4668 {
4669         mdu_disk_info_t info;
4670         mdk_rdev_t *rdev;
4671
4672         if (copy_from_user(&info, arg, sizeof(info)))
4673                 return -EFAULT;
4674
4675         rdev = find_rdev_nr(mddev, info.number);
4676         if (rdev) {
4677                 info.major = MAJOR(rdev->bdev->bd_dev);
4678                 info.minor = MINOR(rdev->bdev->bd_dev);
4679                 info.raid_disk = rdev->raid_disk;
4680                 info.state = 0;
4681                 if (test_bit(Faulty, &rdev->flags))
4682                         info.state |= (1<<MD_DISK_FAULTY);
4683                 else if (test_bit(In_sync, &rdev->flags)) {
4684                         info.state |= (1<<MD_DISK_ACTIVE);
4685                         info.state |= (1<<MD_DISK_SYNC);
4686                 }
4687                 if (test_bit(WriteMostly, &rdev->flags))
4688                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
4689         } else {
4690                 info.major = info.minor = 0;
4691                 info.raid_disk = -1;
4692                 info.state = (1<<MD_DISK_REMOVED);
4693         }
4694
4695         if (copy_to_user(arg, &info, sizeof(info)))
4696                 return -EFAULT;
4697
4698         return 0;
4699 }
4700
4701 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
4702 {
4703         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4704         mdk_rdev_t *rdev;
4705         dev_t dev = MKDEV(info->major,info->minor);
4706
4707         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
4708                 return -EOVERFLOW;
4709
4710         if (!mddev->raid_disks) {
4711                 int err;
4712                 /* expecting a device which has a superblock */
4713                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
4714                 if (IS_ERR(rdev)) {
4715                         printk(KERN_WARNING 
4716                                 "md: md_import_device returned %ld\n",
4717                                 PTR_ERR(rdev));
4718                         return PTR_ERR(rdev);
4719                 }
4720                 if (!list_empty(&mddev->disks)) {
4721                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
4722                                                         mdk_rdev_t, same_set);
4723                         int err = super_types[mddev->major_version]
4724                                 .load_super(rdev, rdev0, mddev->minor_version);
4725                         if (err < 0) {
4726                                 printk(KERN_WARNING 
4727                                         "md: %s has different UUID to %s\n",
4728                                         bdevname(rdev->bdev,b), 
4729                                         bdevname(rdev0->bdev,b2));
4730                                 export_rdev(rdev);
4731                                 return -EINVAL;
4732                         }
4733                 }
4734                 err = bind_rdev_to_array(rdev, mddev);
4735                 if (err)
4736                         export_rdev(rdev);
4737                 return err;
4738         }
4739
4740         /*
4741          * add_new_disk can be used once the array is assembled
4742          * to add "hot spares".  They must already have a superblock
4743          * written
4744          */
4745         if (mddev->pers) {
4746                 int err;
4747                 if (!mddev->pers->hot_add_disk) {
4748                         printk(KERN_WARNING 
4749                                 "%s: personality does not support diskops!\n",
4750                                mdname(mddev));
4751                         return -EINVAL;
4752                 }
4753                 if (mddev->persistent)
4754                         rdev = md_import_device(dev, mddev->major_version,
4755                                                 mddev->minor_version);
4756                 else
4757                         rdev = md_import_device(dev, -1, -1);
4758                 if (IS_ERR(rdev)) {
4759                         printk(KERN_WARNING 
4760                                 "md: md_import_device returned %ld\n",
4761                                 PTR_ERR(rdev));
4762                         return PTR_ERR(rdev);
4763                 }
4764                 /* set save_raid_disk if appropriate */
4765                 if (!mddev->persistent) {
4766                         if (info->state & (1<<MD_DISK_SYNC)  &&
4767                             info->raid_disk < mddev->raid_disks)
4768                                 rdev->raid_disk = info->raid_disk;
4769                         else
4770                                 rdev->raid_disk = -1;
4771                 } else
4772                         super_types[mddev->major_version].
4773                                 validate_super(mddev, rdev);
4774                 rdev->saved_raid_disk = rdev->raid_disk;
4775
4776                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
4777                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4778                         set_bit(WriteMostly, &rdev->flags);
4779                 else
4780                         clear_bit(WriteMostly, &rdev->flags);
4781
4782                 rdev->raid_disk = -1;
4783                 err = bind_rdev_to_array(rdev, mddev);
4784                 if (!err && !mddev->pers->hot_remove_disk) {
4785                         /* If there is hot_add_disk but no hot_remove_disk
4786                          * then added disks for geometry changes,
4787                          * and should be added immediately.
4788                          */
4789                         super_types[mddev->major_version].
4790                                 validate_super(mddev, rdev);
4791                         err = mddev->pers->hot_add_disk(mddev, rdev);
4792                         if (err)
4793                                 unbind_rdev_from_array(rdev);
4794                 }
4795                 if (err)
4796                         export_rdev(rdev);
4797                 else
4798                         sysfs_notify_dirent(rdev->sysfs_state);
4799
4800                 md_update_sb(mddev, 1);
4801                 if (mddev->degraded)
4802                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4803                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4804                 md_wakeup_thread(mddev->thread);
4805                 return err;
4806         }
4807
4808         /* otherwise, add_new_disk is only allowed
4809          * for major_version==0 superblocks
4810          */
4811         if (mddev->major_version != 0) {
4812                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
4813                        mdname(mddev));
4814                 return -EINVAL;
4815         }
4816
4817         if (!(info->state & (1<<MD_DISK_FAULTY))) {
4818                 int err;
4819                 rdev = md_import_device(dev, -1, 0);
4820                 if (IS_ERR(rdev)) {
4821                         printk(KERN_WARNING 
4822                                 "md: error, md_import_device() returned %ld\n",
4823                                 PTR_ERR(rdev));
4824                         return PTR_ERR(rdev);
4825                 }
4826                 rdev->desc_nr = info->number;
4827                 if (info->raid_disk < mddev->raid_disks)
4828                         rdev->raid_disk = info->raid_disk;
4829                 else
4830                         rdev->raid_disk = -1;
4831
4832                 if (rdev->raid_disk < mddev->raid_disks)
4833                         if (info->state & (1<<MD_DISK_SYNC))
4834                                 set_bit(In_sync, &rdev->flags);
4835
4836                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
4837                         set_bit(WriteMostly, &rdev->flags);
4838
4839                 if (!mddev->persistent) {
4840                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
4841                         rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4842                 } else 
4843                         rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4844                 rdev->sectors = rdev->sb_start;
4845
4846                 err = bind_rdev_to_array(rdev, mddev);
4847                 if (err) {
4848                         export_rdev(rdev);
4849                         return err;
4850                 }
4851         }
4852
4853         return 0;
4854 }
4855
4856 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
4857 {
4858         char b[BDEVNAME_SIZE];
4859         mdk_rdev_t *rdev;
4860
4861         rdev = find_rdev(mddev, dev);
4862         if (!rdev)
4863                 return -ENXIO;
4864
4865         if (rdev->raid_disk >= 0)
4866                 goto busy;
4867
4868         kick_rdev_from_array(rdev);
4869         md_update_sb(mddev, 1);
4870         md_new_event(mddev);
4871
4872         return 0;
4873 busy:
4874         printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
4875                 bdevname(rdev->bdev,b), mdname(mddev));
4876         return -EBUSY;
4877 }
4878
4879 static int hot_add_disk(mddev_t * mddev, dev_t dev)
4880 {
4881         char b[BDEVNAME_SIZE];
4882         int err;
4883         mdk_rdev_t *rdev;
4884
4885         if (!mddev->pers)
4886                 return -ENODEV;
4887
4888         if (mddev->major_version != 0) {
4889                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
4890                         " version-0 superblocks.\n",
4891                         mdname(mddev));
4892                 return -EINVAL;
4893         }
4894         if (!mddev->pers->hot_add_disk) {
4895                 printk(KERN_WARNING 
4896                         "%s: personality does not support diskops!\n",
4897                         mdname(mddev));
4898                 return -EINVAL;
4899         }
4900
4901         rdev = md_import_device(dev, -1, 0);
4902         if (IS_ERR(rdev)) {
4903                 printk(KERN_WARNING 
4904                         "md: error, md_import_device() returned %ld\n",
4905                         PTR_ERR(rdev));
4906                 return -EINVAL;
4907         }
4908
4909         if (mddev->persistent)
4910                 rdev->sb_start = calc_dev_sboffset(rdev->bdev);
4911         else
4912                 rdev->sb_start = rdev->bdev->bd_inode->i_size / 512;
4913
4914         rdev->sectors = rdev->sb_start;
4915
4916         if (test_bit(Faulty, &rdev->flags)) {
4917                 printk(KERN_WARNING 
4918                         "md: can not hot-add faulty %s disk to %s!\n",
4919                         bdevname(rdev->bdev,b), mdname(mddev));
4920                 err = -EINVAL;
4921                 goto abort_export;
4922         }
4923         clear_bit(In_sync, &rdev->flags);
4924         rdev->desc_nr = -1;
4925         rdev->saved_raid_disk = -1;
4926         err = bind_rdev_to_array(rdev, mddev);
4927         if (err)
4928                 goto abort_export;
4929
4930         /*
4931          * The rest should better be atomic, we can have disk failures
4932          * noticed in interrupt contexts ...
4933          */
4934
4935         rdev->raid_disk = -1;
4936
4937         md_update_sb(mddev, 1);
4938
4939         /*
4940          * Kick recovery, maybe this spare has to be added to the
4941          * array immediately.
4942          */
4943         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4944         md_wakeup_thread(mddev->thread);
4945         md_new_event(mddev);
4946         return 0;
4947
4948 abort_export:
4949         export_rdev(rdev);
4950         return err;
4951 }
4952
4953 static int set_bitmap_file(mddev_t *mddev, int fd)
4954 {
4955         int err;
4956
4957         if (mddev->pers) {
4958                 if (!mddev->pers->quiesce)
4959                         return -EBUSY;
4960                 if (mddev->recovery || mddev->sync_thread)
4961                         return -EBUSY;
4962                 /* we should be able to change the bitmap.. */
4963         }
4964
4965
4966         if (fd >= 0) {
4967                 if (mddev->bitmap)
4968                         return -EEXIST; /* cannot add when bitmap is present */
4969                 mddev->bitmap_file = fget(fd);
4970
4971                 if (mddev->bitmap_file == NULL) {
4972                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
4973                                mdname(mddev));
4974                         return -EBADF;
4975                 }
4976
4977                 err = deny_bitmap_write_access(mddev->bitmap_file);
4978                 if (err) {
4979                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
4980                                mdname(mddev));
4981                         fput(mddev->bitmap_file);
4982                         mddev->bitmap_file = NULL;
4983                         return err;
4984                 }
4985                 mddev->bitmap_offset = 0; /* file overrides offset */
4986         } else if (mddev->bitmap == NULL)
4987                 return -ENOENT; /* cannot remove what isn't there */
4988         err = 0;
4989         if (mddev->pers) {
4990                 mddev->pers->quiesce(mddev, 1);
4991                 if (fd >= 0)
4992                         err = bitmap_create(mddev);
4993                 if (fd < 0 || err) {
4994                         bitmap_destroy(mddev);
4995                         fd = -1; /* make sure to put the file */
4996                 }
4997                 mddev->pers->quiesce(mddev, 0);
4998         }
4999         if (fd < 0) {
5000                 if (mddev->bitmap_file) {
5001                         restore_bitmap_write_access(mddev->bitmap_file);
5002                         fput(mddev->bitmap_file);
5003                 }
5004                 mddev->bitmap_file = NULL;
5005         }
5006
5007         return err;
5008 }
5009
5010 /*
5011  * set_array_info is used two different ways
5012  * The original usage is when creating a new array.
5013  * In this usage, raid_disks is > 0 and it together with
5014  *  level, size, not_persistent,layout,chunksize determine the
5015  *  shape of the array.
5016  *  This will always create an array with a type-0.90.0 superblock.
5017  * The newer usage is when assembling an array.
5018  *  In this case raid_disks will be 0, and the major_version field is
5019  *  use to determine which style super-blocks are to be found on the devices.
5020  *  The minor and patch _version numbers are also kept incase the
5021  *  super_block handler wishes to interpret them.
5022  */
5023 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5024 {
5025
5026         if (info->raid_disks == 0) {
5027                 /* just setting version number for superblock loading */
5028                 if (info->major_version < 0 ||
5029                     info->major_version >= ARRAY_SIZE(super_types) ||
5030                     super_types[info->major_version].name == NULL) {
5031                         /* maybe try to auto-load a module? */
5032                         printk(KERN_INFO 
5033                                 "md: superblock version %d not known\n",
5034                                 info->major_version);
5035                         return -EINVAL;
5036                 }
5037                 mddev->major_version = info->major_version;
5038                 mddev->minor_version = info->minor_version;
5039                 mddev->patch_version = info->patch_version;
5040                 mddev->persistent = !info->not_persistent;
5041                 return 0;
5042         }
5043         mddev->major_version = MD_MAJOR_VERSION;
5044         mddev->minor_version = MD_MINOR_VERSION;
5045         mddev->patch_version = MD_PATCHLEVEL_VERSION;
5046         mddev->ctime         = get_seconds();
5047
5048         mddev->level         = info->level;
5049         mddev->clevel[0]     = 0;
5050         mddev->dev_sectors   = 2 * (sector_t)info->size;
5051         mddev->raid_disks    = info->raid_disks;
5052         /* don't set md_minor, it is determined by which /dev/md* was
5053          * openned
5054          */
5055         if (info->state & (1<<MD_SB_CLEAN))
5056                 mddev->recovery_cp = MaxSector;
5057         else
5058                 mddev->recovery_cp = 0;
5059         mddev->persistent    = ! info->not_persistent;
5060         mddev->external      = 0;
5061
5062         mddev->layout        = info->layout;
5063         mddev->chunk_sectors = info->chunk_size >> 9;
5064
5065         mddev->max_disks     = MD_SB_DISKS;
5066
5067         if (mddev->persistent)
5068                 mddev->flags         = 0;
5069         set_bit(MD_CHANGE_DEVS, &mddev->flags);
5070
5071         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
5072         mddev->bitmap_offset = 0;
5073
5074         mddev->reshape_position = MaxSector;
5075
5076         /*
5077          * Generate a 128 bit UUID
5078          */
5079         get_random_bytes(mddev->uuid, 16);
5080
5081         mddev->new_level = mddev->level;
5082         mddev->new_chunk_sectors = mddev->chunk_sectors;
5083         mddev->new_layout = mddev->layout;
5084         mddev->delta_disks = 0;
5085
5086         return 0;
5087 }
5088
5089 void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5090 {
5091         WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5092
5093         if (mddev->external_size)
5094                 return;
5095
5096         mddev->array_sectors = array_sectors;
5097 }
5098 EXPORT_SYMBOL(md_set_array_sectors);
5099
5100 static int update_size(mddev_t *mddev, sector_t num_sectors)
5101 {
5102         mdk_rdev_t *rdev;
5103         int rv;
5104         int fit = (num_sectors == 0);
5105
5106         if (mddev->pers->resize == NULL)
5107                 return -EINVAL;
5108         /* The "num_sectors" is the number of sectors of each device that
5109          * is used.  This can only make sense for arrays with redundancy.
5110          * linear and raid0 always use whatever space is available. We can only
5111          * consider changing this number if no resync or reconstruction is
5112          * happening, and if the new size is acceptable. It must fit before the
5113          * sb_start or, if that is <data_offset, it must fit before the size
5114          * of each device.  If num_sectors is zero, we find the largest size
5115          * that fits.
5116
5117          */
5118         if (mddev->sync_thread)
5119                 return -EBUSY;
5120         if (mddev->bitmap)
5121                 /* Sorry, cannot grow a bitmap yet, just remove it,
5122                  * grow, and re-add.
5123                  */
5124                 return -EBUSY;
5125         list_for_each_entry(rdev, &mddev->disks, same_set) {
5126                 sector_t avail = rdev->sectors;
5127
5128                 if (fit && (num_sectors == 0 || num_sectors > avail))
5129                         num_sectors = avail;
5130                 if (avail < num_sectors)
5131                         return -ENOSPC;
5132         }
5133         rv = mddev->pers->resize(mddev, num_sectors);
5134         if (!rv)
5135                 revalidate_disk(mddev->gendisk);
5136         return rv;
5137 }
5138
5139 static int update_raid_disks(mddev_t *mddev, int raid_disks)
5140 {
5141         int rv;
5142         /* change the number of raid disks */
5143         if (mddev->pers->check_reshape == NULL)
5144                 return -EINVAL;
5145         if (raid_disks <= 0 ||
5146             raid_disks >= mddev->max_disks)
5147                 return -EINVAL;
5148         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5149                 return -EBUSY;
5150         mddev->delta_disks = raid_disks - mddev->raid_disks;
5151
5152         rv = mddev->pers->check_reshape(mddev);
5153         return rv;
5154 }
5155
5156
5157 /*
5158  * update_array_info is used to change the configuration of an
5159  * on-line array.
5160  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5161  * fields in the info are checked against the array.
5162  * Any differences that cannot be handled will cause an error.
5163  * Normally, only one change can be managed at a time.
5164  */
5165 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5166 {
5167         int rv = 0;
5168         int cnt = 0;
5169         int state = 0;
5170
5171         /* calculate expected state,ignoring low bits */
5172         if (mddev->bitmap && mddev->bitmap_offset)
5173                 state |= (1 << MD_SB_BITMAP_PRESENT);
5174
5175         if (mddev->major_version != info->major_version ||
5176             mddev->minor_version != info->minor_version ||
5177 /*          mddev->patch_version != info->patch_version || */
5178             mddev->ctime         != info->ctime         ||
5179             mddev->level         != info->level         ||
5180 /*          mddev->layout        != info->layout        || */
5181             !mddev->persistent   != info->not_persistent||
5182             mddev->chunk_sectors != info->chunk_size >> 9 ||
5183             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5184             ((state^info->state) & 0xfffffe00)
5185                 )
5186                 return -EINVAL;
5187         /* Check there is only one change */
5188         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5189                 cnt++;
5190         if (mddev->raid_disks != info->raid_disks)
5191                 cnt++;
5192         if (mddev->layout != info->layout)
5193                 cnt++;
5194         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5195                 cnt++;
5196         if (cnt == 0)
5197                 return 0;
5198         if (cnt > 1)
5199                 return -EINVAL;
5200
5201         if (mddev->layout != info->layout) {
5202                 /* Change layout
5203                  * we don't need to do anything at the md level, the
5204                  * personality will take care of it all.
5205                  */
5206                 if (mddev->pers->check_reshape == NULL)
5207                         return -EINVAL;
5208                 else {
5209                         mddev->new_layout = info->layout;
5210                         rv = mddev->pers->check_reshape(mddev);
5211                         if (rv)
5212                                 mddev->new_layout = mddev->layout;
5213                         return rv;
5214                 }
5215         }
5216         if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5217                 rv = update_size(mddev, (sector_t)info->size * 2);
5218
5219         if (mddev->raid_disks    != info->raid_disks)
5220                 rv = update_raid_disks(mddev, info->raid_disks);
5221
5222         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5223                 if (mddev->pers->quiesce == NULL)
5224                         return -EINVAL;
5225                 if (mddev->recovery || mddev->sync_thread)
5226                         return -EBUSY;
5227                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5228                         /* add the bitmap */
5229                         if (mddev->bitmap)
5230                                 return -EEXIST;
5231                         if (mddev->default_bitmap_offset == 0)
5232                                 return -EINVAL;
5233                         mddev->bitmap_offset = mddev->default_bitmap_offset;
5234                         mddev->pers->quiesce(mddev, 1);
5235                         rv = bitmap_create(mddev);
5236                         if (rv)
5237                                 bitmap_destroy(mddev);
5238                         mddev->pers->quiesce(mddev, 0);
5239                 } else {
5240                         /* remove the bitmap */
5241                         if (!mddev->bitmap)
5242                                 return -ENOENT;
5243                         if (mddev->bitmap->file)
5244                                 return -EINVAL;
5245                         mddev->pers->quiesce(mddev, 1);
5246                         bitmap_destroy(mddev);
5247                         mddev->pers->quiesce(mddev, 0);
5248                         mddev->bitmap_offset = 0;
5249                 }
5250         }
5251         md_update_sb(mddev, 1);
5252         return rv;
5253 }
5254
5255 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
5256 {
5257         mdk_rdev_t *rdev;
5258
5259         if (mddev->pers == NULL)
5260                 return -ENODEV;
5261
5262         rdev = find_rdev(mddev, dev);
5263         if (!rdev)
5264                 return -ENODEV;
5265
5266         md_error(mddev, rdev);
5267         return 0;
5268 }
5269
5270 /*
5271  * We have a problem here : there is no easy way to give a CHS
5272  * virtual geometry. We currently pretend that we have a 2 heads
5273  * 4 sectors (with a BIG number of cylinders...). This drives
5274  * dosfs just mad... ;-)
5275  */
5276 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
5277 {
5278         mddev_t *mddev = bdev->bd_disk->private_data;
5279
5280         geo->heads = 2;
5281         geo->sectors = 4;
5282         geo->cylinders = get_capacity(mddev->gendisk) / 8;
5283         return 0;
5284 }
5285
5286 static int md_ioctl(struct block_device *bdev, fmode_t mode,
5287                         unsigned int cmd, unsigned long arg)
5288 {
5289         int err = 0;
5290         void __user *argp = (void __user *)arg;
5291         mddev_t *mddev = NULL;
5292
5293         if (!capable(CAP_SYS_ADMIN))
5294                 return -EACCES;
5295
5296         /*
5297          * Commands dealing with the RAID driver but not any
5298          * particular array:
5299          */
5300         switch (cmd)
5301         {
5302                 case RAID_VERSION:
5303                         err = get_version(argp);
5304                         goto done;
5305
5306                 case PRINT_RAID_DEBUG:
5307                         err = 0;
5308                         md_print_devices();
5309                         goto done;
5310
5311 #ifndef MODULE
5312                 case RAID_AUTORUN:
5313                         err = 0;
5314                         autostart_arrays(arg);
5315                         goto done;
5316 #endif
5317                 default:;
5318         }
5319
5320         /*
5321          * Commands creating/starting a new array:
5322          */
5323
5324         mddev = bdev->bd_disk->private_data;
5325
5326         if (!mddev) {
5327                 BUG();
5328                 goto abort;
5329         }
5330
5331         err = mddev_lock(mddev);
5332         if (err) {
5333                 printk(KERN_INFO 
5334                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
5335                         err, cmd);
5336                 goto abort;
5337         }
5338
5339         switch (cmd)
5340         {
5341                 case SET_ARRAY_INFO:
5342                         {
5343                                 mdu_array_info_t info;
5344                                 if (!arg)
5345                                         memset(&info, 0, sizeof(info));
5346                                 else if (copy_from_user(&info, argp, sizeof(info))) {
5347                                         err = -EFAULT;
5348                                         goto abort_unlock;
5349                                 }
5350                                 if (mddev->pers) {
5351                                         err = update_array_info(mddev, &info);
5352                                         if (err) {
5353                                                 printk(KERN_WARNING "md: couldn't update"
5354                                                        " array info. %d\n", err);
5355                                                 goto abort_unlock;
5356                                         }
5357                                         goto done_unlock;
5358                                 }
5359                                 if (!list_empty(&mddev->disks)) {
5360                                         printk(KERN_WARNING
5361                                                "md: array %s already has disks!\n",
5362                                                mdname(mddev));
5363                                         err = -EBUSY;
5364                                         goto abort_unlock;
5365                                 }
5366                                 if (mddev->raid_disks) {
5367                                         printk(KERN_WARNING
5368                                                "md: array %s already initialised!\n",
5369                                                mdname(mddev));
5370                                         err = -EBUSY;
5371                                         goto abort_unlock;
5372                                 }
5373                                 err = set_array_info(mddev, &info);
5374                                 if (err) {
5375                                         printk(KERN_WARNING "md: couldn't set"
5376                                                " array info. %d\n", err);
5377                                         goto abort_unlock;
5378                                 }
5379                         }
5380                         goto done_unlock;
5381
5382                 default:;
5383         }
5384
5385         /*
5386          * Commands querying/configuring an existing array:
5387          */
5388         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
5389          * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
5390         if ((!mddev->raid_disks && !mddev->external)
5391             && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
5392             && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
5393             && cmd != GET_BITMAP_FILE) {
5394                 err = -ENODEV;
5395                 goto abort_unlock;
5396         }
5397
5398         /*
5399          * Commands even a read-only array can execute:
5400          */
5401         switch (cmd)
5402         {
5403                 case GET_ARRAY_INFO:
5404                         err = get_array_info(mddev, argp);
5405                         goto done_unlock;
5406
5407                 case GET_BITMAP_FILE:
5408                         err = get_bitmap_file(mddev, argp);
5409                         goto done_unlock;
5410
5411                 case GET_DISK_INFO:
5412                         err = get_disk_info(mddev, argp);
5413                         goto done_unlock;
5414
5415                 case RESTART_ARRAY_RW:
5416                         err = restart_array(mddev);
5417                         goto done_unlock;
5418
5419                 case STOP_ARRAY:
5420                         err = do_md_stop(mddev, 0, 1);
5421                         goto done_unlock;
5422
5423                 case STOP_ARRAY_RO:
5424                         err = do_md_stop(mddev, 1, 1);
5425                         goto done_unlock;
5426
5427         }
5428
5429         /*
5430          * The remaining ioctls are changing the state of the
5431          * superblock, so we do not allow them on read-only arrays.
5432          * However non-MD ioctls (e.g. get-size) will still come through
5433          * here and hit the 'default' below, so only disallow
5434          * 'md' ioctls, and switch to rw mode if started auto-readonly.
5435          */
5436         if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
5437                 if (mddev->ro == 2) {
5438                         mddev->ro = 0;
5439                         sysfs_notify_dirent(mddev->sysfs_state);
5440                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5441                         md_wakeup_thread(mddev->thread);
5442                 } else {
5443                         err = -EROFS;
5444                         goto abort_unlock;
5445                 }
5446         }
5447
5448         switch (cmd)
5449         {
5450                 case ADD_NEW_DISK:
5451                 {
5452                         mdu_disk_info_t info;
5453                         if (copy_from_user(&info, argp, sizeof(info)))
5454                                 err = -EFAULT;
5455                         else
5456                                 err = add_new_disk(mddev, &info);
5457                         goto done_unlock;
5458                 }
5459
5460                 case HOT_REMOVE_DISK:
5461                         err = hot_remove_disk(mddev, new_decode_dev(arg));
5462                         goto done_unlock;
5463
5464                 case HOT_ADD_DISK:
5465                         err = hot_add_disk(mddev, new_decode_dev(arg));
5466                         goto done_unlock;
5467
5468                 case SET_DISK_FAULTY:
5469                         err = set_disk_faulty(mddev, new_decode_dev(arg));
5470                         goto done_unlock;
5471
5472                 case RUN_ARRAY:
5473                         err = do_md_run(mddev);
5474                         goto done_unlock;
5475
5476                 case SET_BITMAP_FILE:
5477                         err = set_bitmap_file(mddev, (int)arg);
5478                         goto done_unlock;
5479
5480                 default:
5481                         err = -EINVAL;
5482                         goto abort_unlock;
5483         }
5484
5485 done_unlock:
5486 abort_unlock:
5487         if (mddev->hold_active == UNTIL_IOCTL &&
5488             err != -EINVAL)
5489                 mddev->hold_active = 0;
5490         mddev_unlock(mddev);
5491
5492         return err;
5493 done:
5494         if (err)
5495                 MD_BUG();
5496 abort:
5497         return err;
5498 }
5499
5500 static int md_open(struct block_device *bdev, fmode_t mode)
5501 {
5502         /*
5503          * Succeed if we can lock the mddev, which confirms that
5504          * it isn't being stopped right now.
5505          */
5506         mddev_t *mddev = mddev_find(bdev->bd_dev);
5507         int err;
5508
5509         if (mddev->gendisk != bdev->bd_disk) {
5510                 /* we are racing with mddev_put which is discarding this
5511                  * bd_disk.
5512                  */
5513                 mddev_put(mddev);
5514                 /* Wait until bdev->bd_disk is definitely gone */
5515                 flush_scheduled_work();
5516                 /* Then retry the open from the top */
5517                 return -ERESTARTSYS;
5518         }
5519         BUG_ON(mddev != bdev->bd_disk->private_data);
5520
5521         if ((err = mutex_lock_interruptible_nested(&mddev->reconfig_mutex, 1)))
5522                 goto out;
5523
5524         err = 0;
5525         atomic_inc(&mddev->openers);
5526         mddev_unlock(mddev);
5527
5528         check_disk_change(bdev);
5529  out:
5530         return err;
5531 }
5532
5533 static int md_release(struct gendisk *disk, fmode_t mode)
5534 {
5535         mddev_t *mddev = disk->private_data;
5536
5537         BUG_ON(!mddev);
5538         atomic_dec(&mddev->openers);
5539         mddev_put(mddev);
5540
5541         return 0;
5542 }
5543
5544 static int md_media_changed(struct gendisk *disk)
5545 {
5546         mddev_t *mddev = disk->private_data;
5547
5548         return mddev->changed;
5549 }
5550
5551 static int md_revalidate(struct gendisk *disk)
5552 {
5553         mddev_t *mddev = disk->private_data;
5554
5555         mddev->changed = 0;
5556         return 0;
5557 }
5558 static struct block_device_operations md_fops =
5559 {
5560         .owner          = THIS_MODULE,
5561         .open           = md_open,
5562         .release        = md_release,
5563         .ioctl          = md_ioctl,
5564         .getgeo         = md_getgeo,
5565         .media_changed  = md_media_changed,
5566         .revalidate_disk= md_revalidate,
5567 };
5568
5569 static int md_thread(void * arg)
5570 {
5571         mdk_thread_t *thread = arg;
5572
5573         /*
5574          * md_thread is a 'system-thread', it's priority should be very
5575          * high. We avoid resource deadlocks individually in each
5576          * raid personality. (RAID5 does preallocation) We also use RR and
5577          * the very same RT priority as kswapd, thus we will never get
5578          * into a priority inversion deadlock.
5579          *
5580          * we definitely have to have equal or higher priority than
5581          * bdflush, otherwise bdflush will deadlock if there are too
5582          * many dirty RAID5 blocks.
5583          */
5584
5585         allow_signal(SIGKILL);
5586         while (!kthread_should_stop()) {
5587
5588                 /* We need to wait INTERRUPTIBLE so that
5589                  * we don't add to the load-average.
5590                  * That means we need to be sure no signals are
5591                  * pending
5592                  */
5593                 if (signal_pending(current))
5594                         flush_signals(current);
5595
5596                 wait_event_interruptible_timeout
5597                         (thread->wqueue,
5598                          test_bit(THREAD_WAKEUP, &thread->flags)
5599                          || kthread_should_stop(),
5600                          thread->timeout);
5601
5602                 clear_bit(THREAD_WAKEUP, &thread->flags);
5603
5604                 thread->run(thread->mddev);
5605         }
5606
5607         return 0;
5608 }
5609
5610 void md_wakeup_thread(mdk_thread_t *thread)
5611 {
5612         if (thread) {
5613                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
5614                 set_bit(THREAD_WAKEUP, &thread->flags);
5615                 wake_up(&thread->wqueue);
5616         }
5617 }
5618
5619 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
5620                                  const char *name)
5621 {
5622         mdk_thread_t *thread;
5623
5624         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
5625         if (!thread)
5626                 return NULL;
5627
5628         init_waitqueue_head(&thread->wqueue);
5629
5630         thread->run = run;
5631         thread->mddev = mddev;
5632         thread->timeout = MAX_SCHEDULE_TIMEOUT;
5633         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
5634         if (IS_ERR(thread->tsk)) {
5635                 kfree(thread);
5636                 return NULL;
5637         }
5638         return thread;
5639 }
5640
5641 void md_unregister_thread(mdk_thread_t *thread)
5642 {
5643         if (!thread)
5644                 return;
5645         dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
5646
5647         kthread_stop(thread->tsk);
5648         kfree(thread);
5649 }
5650
5651 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
5652 {
5653         if (!mddev) {
5654                 MD_BUG();
5655                 return;
5656         }
5657
5658         if (!rdev || test_bit(Faulty, &rdev->flags))
5659                 return;
5660
5661         if (mddev->external)
5662                 set_bit(Blocked, &rdev->flags);
5663 /*
5664         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
5665                 mdname(mddev),
5666                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
5667                 __builtin_return_address(0),__builtin_return_address(1),
5668                 __builtin_return_address(2),__builtin_return_address(3));
5669 */
5670         if (!mddev->pers)
5671                 return;
5672         if (!mddev->pers->error_handler)
5673                 return;
5674         mddev->pers->error_handler(mddev,rdev);
5675         if (mddev->degraded)
5676                 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5677         set_bit(StateChanged, &rdev->flags);
5678         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5679         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5680         md_wakeup_thread(mddev->thread);
5681         md_new_event_inintr(mddev);
5682 }
5683
5684 /* seq_file implementation /proc/mdstat */
5685
5686 static void status_unused(struct seq_file *seq)
5687 {
5688         int i = 0;
5689         mdk_rdev_t *rdev;
5690
5691         seq_printf(seq, "unused devices: ");
5692
5693         list_for_each_entry(rdev, &pending_raid_disks, same_set) {
5694                 char b[BDEVNAME_SIZE];
5695                 i++;
5696                 seq_printf(seq, "%s ",
5697                               bdevname(rdev->bdev,b));
5698         }
5699         if (!i)
5700                 seq_printf(seq, "<none>");
5701
5702         seq_printf(seq, "\n");
5703 }
5704
5705
5706 static void status_resync(struct seq_file *seq, mddev_t * mddev)
5707 {
5708         sector_t max_sectors, resync, res;
5709         unsigned long dt, db;
5710         sector_t rt;
5711         int scale;
5712         unsigned int per_milli;
5713
5714         resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
5715
5716         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
5717                 max_sectors = mddev->resync_max_sectors;
5718         else
5719                 max_sectors = mddev->dev_sectors;
5720
5721         /*
5722          * Should not happen.
5723          */
5724         if (!max_sectors) {
5725                 MD_BUG();
5726                 return;
5727         }
5728         /* Pick 'scale' such that (resync>>scale)*1000 will fit
5729          * in a sector_t, and (max_sectors>>scale) will fit in a
5730          * u32, as those are the requirements for sector_div.
5731          * Thus 'scale' must be at least 10
5732          */
5733         scale = 10;
5734         if (sizeof(sector_t) > sizeof(unsigned long)) {
5735                 while ( max_sectors/2 > (1ULL<<(scale+32)))
5736                         scale++;
5737         }
5738         res = (resync>>scale)*1000;
5739         sector_div(res, (u32)((max_sectors>>scale)+1));
5740
5741         per_milli = res;
5742         {
5743                 int i, x = per_milli/50, y = 20-x;
5744                 seq_printf(seq, "[");
5745                 for (i = 0; i < x; i++)
5746                         seq_printf(seq, "=");
5747                 seq_printf(seq, ">");
5748                 for (i = 0; i < y; i++)
5749                         seq_printf(seq, ".");
5750                 seq_printf(seq, "] ");
5751         }
5752         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
5753                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
5754                     "reshape" :
5755                     (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
5756                      "check" :
5757                      (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
5758                       "resync" : "recovery"))),
5759                    per_milli/10, per_milli % 10,
5760                    (unsigned long long) resync/2,
5761                    (unsigned long long) max_sectors/2);
5762
5763         /*
5764          * dt: time from mark until now
5765          * db: blocks written from mark until now
5766          * rt: remaining time
5767          *
5768          * rt is a sector_t, so could be 32bit or 64bit.
5769          * So we divide before multiply in case it is 32bit and close
5770          * to the limit.
5771          * We scale the divisor (db) by 32 to avoid loosing precision
5772          * near the end of resync when the number of remaining sectors
5773          * is close to 'db'.
5774          * We then divide rt by 32 after multiplying by db to compensate.
5775          * The '+1' avoids division by zero if db is very small.
5776          */
5777         dt = ((jiffies - mddev->resync_mark) / HZ);
5778         if (!dt) dt++;
5779         db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
5780                 - mddev->resync_mark_cnt;
5781
5782         rt = max_sectors - resync;    /* number of remaining sectors */
5783         sector_div(rt, db/32+1);
5784         rt *= dt;
5785         rt >>= 5;
5786
5787         seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
5788                    ((unsigned long)rt % 60)/6);
5789
5790         seq_printf(seq, " speed=%ldK/sec", db/2/dt);
5791 }
5792
5793 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
5794 {
5795         struct list_head *tmp;
5796         loff_t l = *pos;
5797         mddev_t *mddev;
5798
5799         if (l >= 0x10000)
5800                 return NULL;
5801         if (!l--)
5802                 /* header */
5803                 return (void*)1;
5804
5805         spin_lock(&all_mddevs_lock);
5806         list_for_each(tmp,&all_mddevs)
5807                 if (!l--) {
5808                         mddev = list_entry(tmp, mddev_t, all_mddevs);
5809                         mddev_get(mddev);
5810                         spin_unlock(&all_mddevs_lock);
5811                         return mddev;
5812                 }
5813         spin_unlock(&all_mddevs_lock);
5814         if (!l--)
5815                 return (void*)2;/* tail */
5816         return NULL;
5817 }
5818
5819 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
5820 {
5821         struct list_head *tmp;
5822         mddev_t *next_mddev, *mddev = v;
5823         
5824         ++*pos;
5825         if (v == (void*)2)
5826                 return NULL;
5827
5828         spin_lock(&all_mddevs_lock);
5829         if (v == (void*)1)
5830                 tmp = all_mddevs.next;
5831         else
5832                 tmp = mddev->all_mddevs.next;
5833         if (tmp != &all_mddevs)
5834                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
5835         else {
5836                 next_mddev = (void*)2;
5837                 *pos = 0x10000;
5838         }               
5839         spin_unlock(&all_mddevs_lock);
5840
5841         if (v != (void*)1)
5842                 mddev_put(mddev);
5843         return next_mddev;
5844
5845 }
5846
5847 static void md_seq_stop(struct seq_file *seq, void *v)
5848 {
5849         mddev_t *mddev = v;
5850
5851         if (mddev && v != (void*)1 && v != (void*)2)
5852                 mddev_put(mddev);
5853 }
5854
5855 struct mdstat_info {
5856         int event;
5857 };
5858
5859 static int md_seq_show(struct seq_file *seq, void *v)
5860 {
5861         mddev_t *mddev = v;
5862         sector_t sectors;
5863         mdk_rdev_t *rdev;
5864         struct mdstat_info *mi = seq->private;
5865         struct bitmap *bitmap;
5866
5867         if (v == (void*)1) {
5868                 struct mdk_personality *pers;
5869                 seq_printf(seq, "Personalities : ");
5870                 spin_lock(&pers_lock);
5871                 list_for_each_entry(pers, &pers_list, list)
5872                         seq_printf(seq, "[%s] ", pers->name);
5873
5874                 spin_unlock(&pers_lock);
5875                 seq_printf(seq, "\n");
5876                 mi->event = atomic_read(&md_event_count);
5877                 return 0;
5878         }
5879         if (v == (void*)2) {
5880                 status_unused(seq);
5881                 return 0;
5882         }
5883
5884         if (mddev_lock(mddev) < 0)
5885                 return -EINTR;
5886
5887         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
5888                 seq_printf(seq, "%s : %sactive", mdname(mddev),
5889                                                 mddev->pers ? "" : "in");
5890                 if (mddev->pers) {
5891                         if (mddev->ro==1)
5892                                 seq_printf(seq, " (read-only)");
5893                         if (mddev->ro==2)
5894                                 seq_printf(seq, " (auto-read-only)");
5895                         seq_printf(seq, " %s", mddev->pers->name);
5896                 }
5897
5898                 sectors = 0;
5899                 list_for_each_entry(rdev, &mddev->disks, same_set) {
5900                         char b[BDEVNAME_SIZE];
5901                         seq_printf(seq, " %s[%d]",
5902                                 bdevname(rdev->bdev,b), rdev->desc_nr);
5903                         if (test_bit(WriteMostly, &rdev->flags))
5904                                 seq_printf(seq, "(W)");
5905                         if (test_bit(Faulty, &rdev->flags)) {
5906                                 seq_printf(seq, "(F)");
5907                                 continue;
5908                         } else if (rdev->raid_disk < 0)
5909                                 seq_printf(seq, "(S)"); /* spare */
5910                         sectors += rdev->sectors;
5911                 }
5912
5913                 if (!list_empty(&mddev->disks)) {
5914                         if (mddev->pers)
5915                                 seq_printf(seq, "\n      %llu blocks",
5916                                            (unsigned long long)
5917                                            mddev->array_sectors / 2);
5918                         else
5919                                 seq_printf(seq, "\n      %llu blocks",
5920                                            (unsigned long long)sectors / 2);
5921                 }
5922                 if (mddev->persistent) {
5923                         if (mddev->major_version != 0 ||
5924                             mddev->minor_version != 90) {
5925                                 seq_printf(seq," super %d.%d",
5926                                            mddev->major_version,
5927                                            mddev->minor_version);
5928                         }
5929                 } else if (mddev->external)
5930                         seq_printf(seq, " super external:%s",
5931                                    mddev->metadata_type);
5932                 else
5933                         seq_printf(seq, " super non-persistent");
5934
5935                 if (mddev->pers) {
5936                         mddev->pers->status(seq, mddev);
5937                         seq_printf(seq, "\n      ");
5938                         if (mddev->pers->sync_request) {
5939                                 if (mddev->curr_resync > 2) {
5940                                         status_resync(seq, mddev);
5941                                         seq_printf(seq, "\n      ");
5942                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
5943                                         seq_printf(seq, "\tresync=DELAYED\n      ");
5944                                 else if (mddev->recovery_cp < MaxSector)
5945                                         seq_printf(seq, "\tresync=PENDING\n      ");
5946                         }
5947                 } else
5948                         seq_printf(seq, "\n       ");
5949
5950                 if ((bitmap = mddev->bitmap)) {
5951                         unsigned long chunk_kb;
5952                         unsigned long flags;
5953                         spin_lock_irqsave(&bitmap->lock, flags);
5954                         chunk_kb = bitmap->chunksize >> 10;
5955                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
5956                                 "%lu%s chunk",
5957                                 bitmap->pages - bitmap->missing_pages,
5958                                 bitmap->pages,
5959                                 (bitmap->pages - bitmap->missing_pages)
5960                                         << (PAGE_SHIFT - 10),
5961                                 chunk_kb ? chunk_kb : bitmap->chunksize,
5962                                 chunk_kb ? "KB" : "B");
5963                         if (bitmap->file) {
5964                                 seq_printf(seq, ", file: ");
5965                                 seq_path(seq, &bitmap->file->f_path, " \t\n");
5966                         }
5967
5968                         seq_printf(seq, "\n");
5969                         spin_unlock_irqrestore(&bitmap->lock, flags);
5970                 }
5971
5972                 seq_printf(seq, "\n");
5973         }
5974         mddev_unlock(mddev);
5975         
5976         return 0;
5977 }
5978
5979 static const struct seq_operations md_seq_ops = {
5980         .start  = md_seq_start,
5981         .next   = md_seq_next,
5982         .stop   = md_seq_stop,
5983         .show   = md_seq_show,
5984 };
5985
5986 static int md_seq_open(struct inode *inode, struct file *file)
5987 {
5988         int error;
5989         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
5990         if (mi == NULL)
5991                 return -ENOMEM;
5992
5993         error = seq_open(file, &md_seq_ops);
5994         if (error)
5995                 kfree(mi);
5996         else {
5997                 struct seq_file *p = file->private_data;
5998                 p->private = mi;
5999                 mi->event = atomic_read(&md_event_count);
6000         }
6001         return error;
6002 }
6003
6004 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6005 {
6006         struct seq_file *m = filp->private_data;
6007         struct mdstat_info *mi = m->private;
6008         int mask;
6009
6010         poll_wait(filp, &md_event_waiters, wait);
6011
6012         /* always allow read */
6013         mask = POLLIN | POLLRDNORM;
6014
6015         if (mi->event != atomic_read(&md_event_count))
6016                 mask |= POLLERR | POLLPRI;
6017         return mask;
6018 }
6019
6020 static const struct file_operations md_seq_fops = {
6021         .owner          = THIS_MODULE,
6022         .open           = md_seq_open,
6023         .read           = seq_read,
6024         .llseek         = seq_lseek,
6025         .release        = seq_release_private,
6026         .poll           = mdstat_poll,
6027 };
6028
6029 int register_md_personality(struct mdk_personality *p)
6030 {
6031         spin_lock(&pers_lock);
6032         list_add_tail(&p->list, &pers_list);
6033         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6034         spin_unlock(&pers_lock);
6035         return 0;
6036 }
6037
6038 int unregister_md_personality(struct mdk_personality *p)
6039 {
6040         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6041         spin_lock(&pers_lock);
6042         list_del_init(&p->list);
6043         spin_unlock(&pers_lock);
6044         return 0;
6045 }
6046
6047 static int is_mddev_idle(mddev_t *mddev, int init)
6048 {
6049         mdk_rdev_t * rdev;
6050         int idle;
6051         int curr_events;
6052
6053         idle = 1;
6054         rcu_read_lock();
6055         rdev_for_each_rcu(rdev, mddev) {
6056                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6057                 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6058                               (int)part_stat_read(&disk->part0, sectors[1]) -
6059                               atomic_read(&disk->sync_io);
6060                 /* sync IO will cause sync_io to increase before the disk_stats
6061                  * as sync_io is counted when a request starts, and
6062                  * disk_stats is counted when it completes.
6063                  * So resync activity will cause curr_events to be smaller than
6064                  * when there was no such activity.
6065                  * non-sync IO will cause disk_stat to increase without
6066                  * increasing sync_io so curr_events will (eventually)
6067                  * be larger than it was before.  Once it becomes
6068                  * substantially larger, the test below will cause
6069                  * the array to appear non-idle, and resync will slow
6070                  * down.
6071                  * If there is a lot of outstanding resync activity when
6072                  * we set last_event to curr_events, then all that activity
6073                  * completing might cause the array to appear non-idle
6074                  * and resync will be slowed down even though there might
6075                  * not have been non-resync activity.  This will only
6076                  * happen once though.  'last_events' will soon reflect
6077                  * the state where there is little or no outstanding
6078                  * resync requests, and further resync activity will
6079                  * always make curr_events less than last_events.
6080                  *
6081                  */
6082                 if (init || curr_events - rdev->last_events > 64) {
6083                         rdev->last_events = curr_events;
6084                         idle = 0;
6085                 }
6086         }
6087         rcu_read_unlock();
6088         return idle;
6089 }
6090
6091 void md_done_sync(mddev_t *mddev, int blocks, int ok)
6092 {
6093         /* another "blocks" (512byte) blocks have been synced */
6094         atomic_sub(blocks, &mddev->recovery_active);
6095         wake_up(&mddev->recovery_wait);
6096         if (!ok) {
6097                 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6098                 md_wakeup_thread(mddev->thread);
6099                 // stop recovery, signal do_sync ....
6100         }
6101 }
6102
6103
6104 /* md_write_start(mddev, bi)
6105  * If we need to update some array metadata (e.g. 'active' flag
6106  * in superblock) before writing, schedule a superblock update
6107  * and wait for it to complete.
6108  */
6109 void md_write_start(mddev_t *mddev, struct bio *bi)
6110 {
6111         int did_change = 0;
6112         if (bio_data_dir(bi) != WRITE)
6113                 return;
6114
6115         BUG_ON(mddev->ro == 1);
6116         if (mddev->ro == 2) {
6117                 /* need to switch to read/write */
6118                 mddev->ro = 0;
6119                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6120                 md_wakeup_thread(mddev->thread);
6121                 md_wakeup_thread(mddev->sync_thread);
6122                 did_change = 1;
6123         }
6124         atomic_inc(&mddev->writes_pending);
6125         if (mddev->safemode == 1)
6126                 mddev->safemode = 0;
6127         if (mddev->in_sync) {
6128                 spin_lock_irq(&mddev->write_lock);
6129                 if (mddev->in_sync) {
6130                         mddev->in_sync = 0;
6131                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6132                         md_wakeup_thread(mddev->thread);
6133                         did_change = 1;
6134                 }
6135                 spin_unlock_irq(&mddev->write_lock);
6136         }
6137         if (did_change)
6138                 sysfs_notify_dirent(mddev->sysfs_state);
6139         wait_event(mddev->sb_wait,
6140                    !test_bit(MD_CHANGE_CLEAN, &mddev->flags) &&
6141                    !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6142 }
6143
6144 void md_write_end(mddev_t *mddev)
6145 {
6146         if (atomic_dec_and_test(&mddev->writes_pending)) {
6147                 if (mddev->safemode == 2)
6148                         md_wakeup_thread(mddev->thread);
6149                 else if (mddev->safemode_delay)
6150                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6151         }
6152 }
6153
6154 /* md_allow_write(mddev)
6155  * Calling this ensures that the array is marked 'active' so that writes
6156  * may proceed without blocking.  It is important to call this before
6157  * attempting a GFP_KERNEL allocation while holding the mddev lock.
6158  * Must be called with mddev_lock held.
6159  *
6160  * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6161  * is dropped, so return -EAGAIN after notifying userspace.
6162  */
6163 int md_allow_write(mddev_t *mddev)
6164 {
6165         if (!mddev->pers)
6166                 return 0;
6167         if (mddev->ro)
6168                 return 0;
6169         if (!mddev->pers->sync_request)
6170                 return 0;
6171
6172         spin_lock_irq(&mddev->write_lock);
6173         if (mddev->in_sync) {
6174                 mddev->in_sync = 0;
6175                 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6176                 if (mddev->safemode_delay &&
6177                     mddev->safemode == 0)
6178                         mddev->safemode = 1;
6179                 spin_unlock_irq(&mddev->write_lock);
6180                 md_update_sb(mddev, 0);
6181                 sysfs_notify_dirent(mddev->sysfs_state);
6182         } else
6183                 spin_unlock_irq(&mddev->write_lock);
6184
6185         if (test_bit(MD_CHANGE_CLEAN, &mddev->flags))
6186                 return -EAGAIN;
6187         else
6188                 return 0;
6189 }
6190 EXPORT_SYMBOL_GPL(md_allow_write);
6191
6192 #define SYNC_MARKS      10
6193 #define SYNC_MARK_STEP  (3*HZ)
6194 void md_do_sync(mddev_t *mddev)
6195 {
6196         mddev_t *mddev2;
6197         unsigned int currspeed = 0,
6198                  window;
6199         sector_t max_sectors,j, io_sectors;
6200         unsigned long mark[SYNC_MARKS];
6201         sector_t mark_cnt[SYNC_MARKS];
6202         int last_mark,m;
6203         struct list_head *tmp;
6204         sector_t last_check;
6205         int skipped = 0;
6206         mdk_rdev_t *rdev;
6207         char *desc;
6208
6209         /* just incase thread restarts... */
6210         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
6211                 return;
6212         if (mddev->ro) /* never try to sync a read-only array */
6213                 return;
6214
6215         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6216                 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
6217                         desc = "data-check";
6218                 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6219                         desc = "requested-resync";
6220                 else
6221                         desc = "resync";
6222         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6223                 desc = "reshape";
6224         else
6225                 desc = "recovery";
6226
6227         /* we overload curr_resync somewhat here.
6228          * 0 == not engaged in resync at all
6229          * 2 == checking that there is no conflict with another sync
6230          * 1 == like 2, but have yielded to allow conflicting resync to
6231          *              commense
6232          * other == active in resync - this many blocks
6233          *
6234          * Before starting a resync we must have set curr_resync to
6235          * 2, and then checked that every "conflicting" array has curr_resync
6236          * less than ours.  When we find one that is the same or higher
6237          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
6238          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
6239          * This will mean we have to start checking from the beginning again.
6240          *
6241          */
6242
6243         do {
6244                 mddev->curr_resync = 2;
6245
6246         try_again:
6247                 if (kthread_should_stop()) {
6248                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6249                         goto skip;
6250                 }
6251                 for_each_mddev(mddev2, tmp) {
6252                         if (mddev2 == mddev)
6253                                 continue;
6254                         if (!mddev->parallel_resync
6255                         &&  mddev2->curr_resync
6256                         &&  match_mddev_units(mddev, mddev2)) {
6257                                 DEFINE_WAIT(wq);
6258                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
6259                                         /* arbitrarily yield */
6260                                         mddev->curr_resync = 1;
6261                                         wake_up(&resync_wait);
6262                                 }
6263                                 if (mddev > mddev2 && mddev->curr_resync == 1)
6264                                         /* no need to wait here, we can wait the next
6265                                          * time 'round when curr_resync == 2
6266                                          */
6267                                         continue;
6268                                 /* We need to wait 'interruptible' so as not to
6269                                  * contribute to the load average, and not to
6270                                  * be caught by 'softlockup'
6271                                  */
6272                                 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
6273                                 if (!kthread_should_stop() &&
6274                                     mddev2->curr_resync >= mddev->curr_resync) {
6275                                         printk(KERN_INFO "md: delaying %s of %s"
6276                                                " until %s has finished (they"
6277                                                " share one or more physical units)\n",
6278                                                desc, mdname(mddev), mdname(mddev2));
6279                                         mddev_put(mddev2);
6280                                         if (signal_pending(current))
6281                                                 flush_signals(current);
6282                                         schedule();
6283                                         finish_wait(&resync_wait, &wq);
6284                                         goto try_again;
6285                                 }
6286                                 finish_wait(&resync_wait, &wq);
6287                         }
6288                 }
6289         } while (mddev->curr_resync < 2);
6290
6291         j = 0;
6292         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6293                 /* resync follows the size requested by the personality,
6294                  * which defaults to physical size, but can be virtual size
6295                  */
6296                 max_sectors = mddev->resync_max_sectors;
6297                 mddev->resync_mismatches = 0;
6298                 /* we don't use the checkpoint if there's a bitmap */
6299                 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
6300                         j = mddev->resync_min;
6301                 else if (!mddev->bitmap)
6302                         j = mddev->recovery_cp;
6303
6304         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
6305                 max_sectors = mddev->dev_sectors;
6306         else {
6307                 /* recovery follows the physical size of devices */
6308                 max_sectors = mddev->dev_sectors;
6309                 j = MaxSector;
6310                 list_for_each_entry(rdev, &mddev->disks, same_set)
6311                         if (rdev->raid_disk >= 0 &&
6312                             !test_bit(Faulty, &rdev->flags) &&
6313                             !test_bit(In_sync, &rdev->flags) &&
6314                             rdev->recovery_offset < j)
6315                                 j = rdev->recovery_offset;
6316         }
6317
6318         printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
6319         printk(KERN_INFO "md: minimum _guaranteed_  speed:"
6320                 " %d KB/sec/disk.\n", speed_min(mddev));
6321         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
6322                "(but not more than %d KB/sec) for %s.\n",
6323                speed_max(mddev), desc);
6324
6325         is_mddev_idle(mddev, 1); /* this initializes IO event counters */
6326
6327         io_sectors = 0;
6328         for (m = 0; m < SYNC_MARKS; m++) {
6329                 mark[m] = jiffies;
6330                 mark_cnt[m] = io_sectors;
6331         }
6332         last_mark = 0;
6333         mddev->resync_mark = mark[last_mark];
6334         mddev->resync_mark_cnt = mark_cnt[last_mark];
6335
6336         /*
6337          * Tune reconstruction:
6338          */
6339         window = 32*(PAGE_SIZE/512);
6340         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
6341                 window/2,(unsigned long long) max_sectors/2);
6342
6343         atomic_set(&mddev->recovery_active, 0);
6344         last_check = 0;
6345
6346         if (j>2) {
6347                 printk(KERN_INFO 
6348                        "md: resuming %s of %s from checkpoint.\n",
6349                        desc, mdname(mddev));
6350                 mddev->curr_resync = j;
6351         }
6352
6353         while (j < max_sectors) {
6354                 sector_t sectors;
6355
6356                 skipped = 0;
6357
6358                 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6359                     ((mddev->curr_resync > mddev->curr_resync_completed &&
6360                       (mddev->curr_resync - mddev->curr_resync_completed)
6361                       > (max_sectors >> 4)) ||
6362                      (j - mddev->curr_resync_completed)*2
6363                      >= mddev->resync_max - mddev->curr_resync_completed
6364                             )) {
6365                         /* time to update curr_resync_completed */
6366                         blk_unplug(mddev->queue);
6367                         wait_event(mddev->recovery_wait,
6368                                    atomic_read(&mddev->recovery_active) == 0);
6369                         mddev->curr_resync_completed =
6370                                 mddev->curr_resync;
6371                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6372                         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6373                 }
6374
6375                 while (j >= mddev->resync_max && !kthread_should_stop()) {
6376                         /* As this condition is controlled by user-space,
6377                          * we can block indefinitely, so use '_interruptible'
6378                          * to avoid triggering warnings.
6379                          */
6380                         flush_signals(current); /* just in case */
6381                         wait_event_interruptible(mddev->recovery_wait,
6382                                                  mddev->resync_max > j
6383                                                  || kthread_should_stop());
6384                 }
6385
6386                 if (kthread_should_stop())
6387                         goto interrupted;
6388
6389                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
6390                                                   currspeed < speed_min(mddev));
6391                 if (sectors == 0) {
6392                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6393                         goto out;
6394                 }
6395
6396                 if (!skipped) { /* actual IO requested */
6397                         io_sectors += sectors;
6398                         atomic_add(sectors, &mddev->recovery_active);
6399                 }
6400
6401                 j += sectors;
6402                 if (j>1) mddev->curr_resync = j;
6403                 mddev->curr_mark_cnt = io_sectors;
6404                 if (last_check == 0)
6405                         /* this is the earliers that rebuilt will be
6406                          * visible in /proc/mdstat
6407                          */
6408                         md_new_event(mddev);
6409
6410                 if (last_check + window > io_sectors || j == max_sectors)
6411                         continue;
6412
6413                 last_check = io_sectors;
6414
6415                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6416                         break;
6417
6418         repeat:
6419                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
6420                         /* step marks */
6421                         int next = (last_mark+1) % SYNC_MARKS;
6422
6423                         mddev->resync_mark = mark[next];
6424                         mddev->resync_mark_cnt = mark_cnt[next];
6425                         mark[next] = jiffies;
6426                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
6427                         last_mark = next;
6428                 }
6429
6430
6431                 if (kthread_should_stop())
6432                         goto interrupted;
6433
6434
6435                 /*
6436                  * this loop exits only if either when we are slower than
6437                  * the 'hard' speed limit, or the system was IO-idle for
6438                  * a jiffy.
6439                  * the system might be non-idle CPU-wise, but we only care
6440                  * about not overloading the IO subsystem. (things like an
6441                  * e2fsck being done on the RAID array should execute fast)
6442                  */
6443                 blk_unplug(mddev->queue);
6444                 cond_resched();
6445
6446                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
6447                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
6448
6449                 if (currspeed > speed_min(mddev)) {
6450                         if ((currspeed > speed_max(mddev)) ||
6451                                         !is_mddev_idle(mddev, 0)) {
6452                                 msleep(500);
6453                                 goto repeat;
6454                         }
6455                 }
6456         }
6457         printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
6458         /*
6459          * this also signals 'finished resyncing' to md_stop
6460          */
6461  out:
6462         blk_unplug(mddev->queue);
6463
6464         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
6465
6466         /* tell personality that we are finished */
6467         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
6468
6469         if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
6470             mddev->curr_resync > 2) {
6471                 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
6472                         if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
6473                                 if (mddev->curr_resync >= mddev->recovery_cp) {
6474                                         printk(KERN_INFO
6475                                                "md: checkpointing %s of %s.\n",
6476                                                desc, mdname(mddev));
6477                                         mddev->recovery_cp = mddev->curr_resync;
6478                                 }
6479                         } else
6480                                 mddev->recovery_cp = MaxSector;
6481                 } else {
6482                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
6483                                 mddev->curr_resync = MaxSector;
6484                         list_for_each_entry(rdev, &mddev->disks, same_set)
6485                                 if (rdev->raid_disk >= 0 &&
6486                                     !test_bit(Faulty, &rdev->flags) &&
6487                                     !test_bit(In_sync, &rdev->flags) &&
6488                                     rdev->recovery_offset < mddev->curr_resync)
6489                                         rdev->recovery_offset = mddev->curr_resync;
6490                 }
6491         }
6492         set_bit(MD_CHANGE_DEVS, &mddev->flags);
6493
6494  skip:
6495         mddev->curr_resync = 0;
6496         mddev->curr_resync_completed = 0;
6497         mddev->resync_min = 0;
6498         mddev->resync_max = MaxSector;
6499         sysfs_notify(&mddev->kobj, NULL, "sync_completed");
6500         wake_up(&resync_wait);
6501         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
6502         md_wakeup_thread(mddev->thread);
6503         return;
6504
6505  interrupted:
6506         /*
6507          * got a signal, exit.
6508          */
6509         printk(KERN_INFO
6510                "md: md_do_sync() got signal ... exiting\n");
6511         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6512         goto out;
6513
6514 }
6515 EXPORT_SYMBOL_GPL(md_do_sync);
6516
6517
6518 static int remove_and_add_spares(mddev_t *mddev)
6519 {
6520         mdk_rdev_t *rdev;
6521         int spares = 0;
6522
6523         mddev->curr_resync_completed = 0;
6524
6525         list_for_each_entry(rdev, &mddev->disks, same_set)
6526                 if (rdev->raid_disk >= 0 &&
6527                     !test_bit(Blocked, &rdev->flags) &&
6528                     (test_bit(Faulty, &rdev->flags) ||
6529                      ! test_bit(In_sync, &rdev->flags)) &&
6530                     atomic_read(&rdev->nr_pending)==0) {
6531                         if (mddev->pers->hot_remove_disk(
6532                                     mddev, rdev->raid_disk)==0) {
6533                                 char nm[20];
6534                                 sprintf(nm,"rd%d", rdev->raid_disk);
6535                                 sysfs_remove_link(&mddev->kobj, nm);
6536                                 rdev->raid_disk = -1;
6537                         }
6538                 }
6539
6540         if (mddev->degraded && ! mddev->ro && !mddev->recovery_disabled) {
6541                 list_for_each_entry(rdev, &mddev->disks, same_set) {
6542                         if (rdev->raid_disk >= 0 &&
6543                             !test_bit(In_sync, &rdev->flags) &&
6544                             !test_bit(Blocked, &rdev->flags))
6545                                 spares++;
6546                         if (rdev->raid_disk < 0
6547                             && !test_bit(Faulty, &rdev->flags)) {
6548                                 rdev->recovery_offset = 0;
6549                                 if (mddev->pers->
6550                                     hot_add_disk(mddev, rdev) == 0) {
6551                                         char nm[20];
6552                                         sprintf(nm, "rd%d", rdev->raid_disk);
6553                                         if (sysfs_create_link(&mddev->kobj,
6554                                                               &rdev->kobj, nm))
6555                                                 printk(KERN_WARNING
6556                                                        "md: cannot register "
6557                                                        "%s for %s\n",
6558                                                        nm, mdname(mddev));
6559                                         spares++;
6560                                         md_new_event(mddev);
6561                                 } else
6562                                         break;
6563                         }
6564                 }
6565         }
6566         return spares;
6567 }
6568 /*
6569  * This routine is regularly called by all per-raid-array threads to
6570  * deal with generic issues like resync and super-block update.
6571  * Raid personalities that don't have a thread (linear/raid0) do not
6572  * need this as they never do any recovery or update the superblock.
6573  *
6574  * It does not do any resync itself, but rather "forks" off other threads
6575  * to do that as needed.
6576  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
6577  * "->recovery" and create a thread at ->sync_thread.
6578  * When the thread finishes it sets MD_RECOVERY_DONE
6579  * and wakeups up this thread which will reap the thread and finish up.
6580  * This thread also removes any faulty devices (with nr_pending == 0).
6581  *
6582  * The overall approach is:
6583  *  1/ if the superblock needs updating, update it.
6584  *  2/ If a recovery thread is running, don't do anything else.
6585  *  3/ If recovery has finished, clean up, possibly marking spares active.
6586  *  4/ If there are any faulty devices, remove them.
6587  *  5/ If array is degraded, try to add spares devices
6588  *  6/ If array has spares or is not in-sync, start a resync thread.
6589  */
6590 void md_check_recovery(mddev_t *mddev)
6591 {
6592         mdk_rdev_t *rdev;
6593
6594
6595         if (mddev->bitmap)
6596                 bitmap_daemon_work(mddev->bitmap);
6597
6598         if (mddev->ro)
6599                 return;
6600
6601         if (signal_pending(current)) {
6602                 if (mddev->pers->sync_request && !mddev->external) {
6603                         printk(KERN_INFO "md: %s in immediate safe mode\n",
6604                                mdname(mddev));
6605                         mddev->safemode = 2;
6606                 }
6607                 flush_signals(current);
6608         }
6609
6610         if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
6611                 return;
6612         if ( ! (
6613                 (mddev->flags && !mddev->external) ||
6614                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
6615                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
6616                 (mddev->external == 0 && mddev->safemode == 1) ||
6617                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
6618                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
6619                 ))
6620                 return;
6621
6622         if (mddev_trylock(mddev)) {
6623                 int spares = 0;
6624
6625                 if (mddev->ro) {
6626                         /* Only thing we do on a ro array is remove
6627                          * failed devices.
6628                          */
6629                         remove_and_add_spares(mddev);
6630                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6631                         goto unlock;
6632                 }
6633
6634                 if (!mddev->external) {
6635                         int did_change = 0;
6636                         spin_lock_irq(&mddev->write_lock);
6637                         if (mddev->safemode &&
6638                             !atomic_read(&mddev->writes_pending) &&
6639                             !mddev->in_sync &&
6640                             mddev->recovery_cp == MaxSector) {
6641                                 mddev->in_sync = 1;
6642                                 did_change = 1;
6643                                 if (mddev->persistent)
6644                                         set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6645                         }
6646                         if (mddev->safemode == 1)
6647                                 mddev->safemode = 0;
6648                         spin_unlock_irq(&mddev->write_lock);
6649                         if (did_change)
6650                                 sysfs_notify_dirent(mddev->sysfs_state);
6651                 }
6652
6653                 if (mddev->flags)
6654                         md_update_sb(mddev, 0);
6655
6656                 list_for_each_entry(rdev, &mddev->disks, same_set)
6657                         if (test_and_clear_bit(StateChanged, &rdev->flags))
6658                                 sysfs_notify_dirent(rdev->sysfs_state);
6659
6660
6661                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
6662                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
6663                         /* resync/recovery still happening */
6664                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6665                         goto unlock;
6666                 }
6667                 if (mddev->sync_thread) {
6668                         /* resync has finished, collect result */
6669                         md_unregister_thread(mddev->sync_thread);
6670                         mddev->sync_thread = NULL;
6671                         if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
6672                             !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
6673                                 /* success...*/
6674                                 /* activate any spares */
6675                                 if (mddev->pers->spare_active(mddev))
6676                                         sysfs_notify(&mddev->kobj, NULL,
6677                                                      "degraded");
6678                         }
6679                         if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
6680                             mddev->pers->finish_reshape)
6681                                 mddev->pers->finish_reshape(mddev);
6682                         md_update_sb(mddev, 1);
6683
6684                         /* if array is no-longer degraded, then any saved_raid_disk
6685                          * information must be scrapped
6686                          */
6687                         if (!mddev->degraded)
6688                                 list_for_each_entry(rdev, &mddev->disks, same_set)
6689                                         rdev->saved_raid_disk = -1;
6690
6691                         mddev->recovery = 0;
6692                         /* flag recovery needed just to double check */
6693                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6694                         sysfs_notify_dirent(mddev->sysfs_action);
6695                         md_new_event(mddev);
6696                         goto unlock;
6697                 }
6698                 /* Set RUNNING before clearing NEEDED to avoid
6699                  * any transients in the value of "sync_action".
6700                  */
6701                 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6702                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6703                 /* Clear some bits that don't mean anything, but
6704                  * might be left set
6705                  */
6706                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
6707                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
6708
6709                 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
6710                         goto unlock;
6711                 /* no recovery is running.
6712                  * remove any failed drives, then
6713                  * add spares if possible.
6714                  * Spare are also removed and re-added, to allow
6715                  * the personality to fail the re-add.
6716                  */
6717
6718                 if (mddev->reshape_position != MaxSector) {
6719                         if (mddev->pers->check_reshape == NULL ||
6720                             mddev->pers->check_reshape(mddev) != 0)
6721                                 /* Cannot proceed */
6722                                 goto unlock;
6723                         set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
6724                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6725                 } else if ((spares = remove_and_add_spares(mddev))) {
6726                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6727                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
6728                         clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
6729                         set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6730                 } else if (mddev->recovery_cp < MaxSector) {
6731                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
6732                         clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6733                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
6734                         /* nothing to be done ... */
6735                         goto unlock;
6736
6737                 if (mddev->pers->sync_request) {
6738                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
6739                                 /* We are adding a device or devices to an array
6740                                  * which has the bitmap stored on all devices.
6741                                  * So make sure all bitmap pages get written
6742                                  */
6743                                 bitmap_write_all(mddev->bitmap);
6744                         }
6745                         mddev->sync_thread = md_register_thread(md_do_sync,
6746                                                                 mddev,
6747                                                                 "%s_resync");
6748                         if (!mddev->sync_thread) {
6749                                 printk(KERN_ERR "%s: could not start resync"
6750                                         " thread...\n", 
6751                                         mdname(mddev));
6752                                 /* leave the spares where they are, it shouldn't hurt */
6753                                 mddev->recovery = 0;
6754                         } else
6755                                 md_wakeup_thread(mddev->sync_thread);
6756                         sysfs_notify_dirent(mddev->sysfs_action);
6757                         md_new_event(mddev);
6758                 }
6759         unlock:
6760                 if (!mddev->sync_thread) {
6761                         clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
6762                         if (test_and_clear_bit(MD_RECOVERY_RECOVER,
6763                                                &mddev->recovery))
6764                                 if (mddev->sysfs_action)
6765                                         sysfs_notify_dirent(mddev->sysfs_action);
6766                 }
6767                 mddev_unlock(mddev);
6768         }
6769 }
6770
6771 void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
6772 {
6773         sysfs_notify_dirent(rdev->sysfs_state);
6774         wait_event_timeout(rdev->blocked_wait,
6775                            !test_bit(Blocked, &rdev->flags),
6776                            msecs_to_jiffies(5000));
6777         rdev_dec_pending(rdev, mddev);
6778 }
6779 EXPORT_SYMBOL(md_wait_for_blocked_rdev);
6780
6781 static int md_notify_reboot(struct notifier_block *this,
6782                             unsigned long code, void *x)
6783 {
6784         struct list_head *tmp;
6785         mddev_t *mddev;
6786
6787         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
6788
6789                 printk(KERN_INFO "md: stopping all md devices.\n");
6790
6791                 for_each_mddev(mddev, tmp)
6792                         if (mddev_trylock(mddev)) {
6793                                 /* Force a switch to readonly even array
6794                                  * appears to still be in use.  Hence
6795                                  * the '100'.
6796                                  */
6797                                 do_md_stop(mddev, 1, 100);
6798                                 mddev_unlock(mddev);
6799                         }
6800                 /*
6801                  * certain more exotic SCSI devices are known to be
6802                  * volatile wrt too early system reboots. While the
6803                  * right place to handle this issue is the given
6804                  * driver, we do want to have a safe RAID driver ...
6805                  */
6806                 mdelay(1000*1);
6807         }
6808         return NOTIFY_DONE;
6809 }
6810
6811 static struct notifier_block md_notifier = {
6812         .notifier_call  = md_notify_reboot,
6813         .next           = NULL,
6814         .priority       = INT_MAX, /* before any real devices */
6815 };
6816
6817 static void md_geninit(void)
6818 {
6819         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
6820
6821         proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
6822 }
6823
6824 static int __init md_init(void)
6825 {
6826         if (register_blkdev(MD_MAJOR, "md"))
6827                 return -1;
6828         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
6829                 unregister_blkdev(MD_MAJOR, "md");
6830                 return -1;
6831         }
6832         blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
6833                             md_probe, NULL, NULL);
6834         blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
6835                             md_probe, NULL, NULL);
6836
6837         register_reboot_notifier(&md_notifier);
6838         raid_table_header = register_sysctl_table(raid_root_table);
6839
6840         md_geninit();
6841         return 0;
6842 }
6843
6844
6845 #ifndef MODULE
6846
6847 /*
6848  * Searches all registered partitions for autorun RAID arrays
6849  * at boot time.
6850  */
6851
6852 static LIST_HEAD(all_detected_devices);
6853 struct detected_devices_node {
6854         struct list_head list;
6855         dev_t dev;
6856 };
6857
6858 void md_autodetect_dev(dev_t dev)
6859 {
6860         struct detected_devices_node *node_detected_dev;
6861
6862         node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
6863         if (node_detected_dev) {
6864                 node_detected_dev->dev = dev;
6865                 list_add_tail(&node_detected_dev->list, &all_detected_devices);
6866         } else {
6867                 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
6868                         ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
6869         }
6870 }
6871
6872
6873 static void autostart_arrays(int part)
6874 {
6875         mdk_rdev_t *rdev;
6876         struct detected_devices_node *node_detected_dev;
6877         dev_t dev;
6878         int i_scanned, i_passed;
6879
6880         i_scanned = 0;
6881         i_passed = 0;
6882
6883         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
6884
6885         while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
6886                 i_scanned++;
6887                 node_detected_dev = list_entry(all_detected_devices.next,
6888                                         struct detected_devices_node, list);
6889                 list_del(&node_detected_dev->list);
6890                 dev = node_detected_dev->dev;
6891                 kfree(node_detected_dev);
6892                 rdev = md_import_device(dev,0, 90);
6893                 if (IS_ERR(rdev))
6894                         continue;
6895
6896                 if (test_bit(Faulty, &rdev->flags)) {
6897                         MD_BUG();
6898                         continue;
6899                 }
6900                 set_bit(AutoDetected, &rdev->flags);
6901                 list_add(&rdev->same_set, &pending_raid_disks);
6902                 i_passed++;
6903         }
6904
6905         printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
6906                                                 i_scanned, i_passed);
6907
6908         autorun_devices(part);
6909 }
6910
6911 #endif /* !MODULE */
6912
6913 static __exit void md_exit(void)
6914 {
6915         mddev_t *mddev;
6916         struct list_head *tmp;
6917
6918         blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
6919         blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
6920
6921         unregister_blkdev(MD_MAJOR,"md");
6922         unregister_blkdev(mdp_major, "mdp");
6923         unregister_reboot_notifier(&md_notifier);
6924         unregister_sysctl_table(raid_table_header);
6925         remove_proc_entry("mdstat", NULL);
6926         for_each_mddev(mddev, tmp) {
6927                 export_array(mddev);
6928                 mddev->hold_active = 0;
6929         }
6930 }
6931
6932 subsys_initcall(md_init);
6933 module_exit(md_exit)
6934
6935 static int get_ro(char *buffer, struct kernel_param *kp)
6936 {
6937         return sprintf(buffer, "%d", start_readonly);
6938 }
6939 static int set_ro(const char *val, struct kernel_param *kp)
6940 {
6941         char *e;
6942         int num = simple_strtoul(val, &e, 10);
6943         if (*val && (*e == '\0' || *e == '\n')) {
6944                 start_readonly = num;
6945                 return 0;
6946         }
6947         return -EINVAL;
6948 }
6949
6950 module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
6951 module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
6952
6953 module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
6954
6955 EXPORT_SYMBOL(register_md_personality);
6956 EXPORT_SYMBOL(unregister_md_personality);
6957 EXPORT_SYMBOL(md_error);
6958 EXPORT_SYMBOL(md_done_sync);
6959 EXPORT_SYMBOL(md_write_start);
6960 EXPORT_SYMBOL(md_write_end);
6961 EXPORT_SYMBOL(md_register_thread);
6962 EXPORT_SYMBOL(md_unregister_thread);
6963 EXPORT_SYMBOL(md_wakeup_thread);
6964 EXPORT_SYMBOL(md_check_recovery);
6965 MODULE_LICENSE("GPL");
6966 MODULE_ALIAS("md");
6967 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);