8beb19c3bb44d1f269735272b1079a155a976a87
[~shefty/rdma-dev.git] / drivers / md / md.c
1 /*
2    md.c : Multiple Devices driver for Linux
3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
4
5      completely rewritten, based on the MD driver code from Marc Zyngier
6
7    Changes:
8
9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
13    - kmod support by: Cyrus Durgin
14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
16
17    - lots of fixes and improvements to the RAID1/RAID5 and generic
18      RAID code (such as request based resynchronization):
19
20      Neil Brown <neilb@cse.unsw.edu.au>.
21
22    - persistent bitmap code
23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
25    This program is free software; you can redistribute it and/or modify
26    it under the terms of the GNU General Public License as published by
27    the Free Software Foundation; either version 2, or (at your option)
28    any later version.
29
30    You should have received a copy of the GNU General Public License
31    (for example /usr/src/linux/COPYING); if not, write to the Free
32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
33 */
34
35 #include <linux/kthread.h>
36 #include <linux/blkdev.h>
37 #include <linux/sysctl.h>
38 #include <linux/seq_file.h>
39 #include <linux/fs.h>
40 #include <linux/poll.h>
41 #include <linux/ctype.h>
42 #include <linux/string.h>
43 #include <linux/hdreg.h>
44 #include <linux/proc_fs.h>
45 #include <linux/random.h>
46 #include <linux/module.h>
47 #include <linux/reboot.h>
48 #include <linux/file.h>
49 #include <linux/compat.h>
50 #include <linux/delay.h>
51 #include <linux/raid/md_p.h>
52 #include <linux/raid/md_u.h>
53 #include <linux/slab.h>
54 #include "md.h"
55 #include "bitmap.h"
56
57 #ifndef MODULE
58 static void autostart_arrays(int part);
59 #endif
60
61 /* pers_list is a list of registered personalities protected
62  * by pers_lock.
63  * pers_lock does extra service to protect accesses to
64  * mddev->thread when the mutex cannot be held.
65  */
66 static LIST_HEAD(pers_list);
67 static DEFINE_SPINLOCK(pers_lock);
68
69 static void md_print_devices(void);
70
71 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
72 static struct workqueue_struct *md_wq;
73 static struct workqueue_struct *md_misc_wq;
74
75 #define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
76
77 /*
78  * Default number of read corrections we'll attempt on an rdev
79  * before ejecting it from the array. We divide the read error
80  * count by 2 for every hour elapsed between read errors.
81  */
82 #define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
83 /*
84  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85  * is 1000 KB/sec, so the extra system load does not show up that much.
86  * Increase it if you want to have more _guaranteed_ speed. Note that
87  * the RAID driver will use the maximum available bandwidth if the IO
88  * subsystem is idle. There is also an 'absolute maximum' reconstruction
89  * speed limit - in case reconstruction slows down your system despite
90  * idle IO detection.
91  *
92  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
93  * or /sys/block/mdX/md/sync_speed_{min,max}
94  */
95
96 static int sysctl_speed_limit_min = 1000;
97 static int sysctl_speed_limit_max = 200000;
98 static inline int speed_min(struct mddev *mddev)
99 {
100         return mddev->sync_speed_min ?
101                 mddev->sync_speed_min : sysctl_speed_limit_min;
102 }
103
104 static inline int speed_max(struct mddev *mddev)
105 {
106         return mddev->sync_speed_max ?
107                 mddev->sync_speed_max : sysctl_speed_limit_max;
108 }
109
110 static struct ctl_table_header *raid_table_header;
111
112 static ctl_table raid_table[] = {
113         {
114                 .procname       = "speed_limit_min",
115                 .data           = &sysctl_speed_limit_min,
116                 .maxlen         = sizeof(int),
117                 .mode           = S_IRUGO|S_IWUSR,
118                 .proc_handler   = proc_dointvec,
119         },
120         {
121                 .procname       = "speed_limit_max",
122                 .data           = &sysctl_speed_limit_max,
123                 .maxlen         = sizeof(int),
124                 .mode           = S_IRUGO|S_IWUSR,
125                 .proc_handler   = proc_dointvec,
126         },
127         { }
128 };
129
130 static ctl_table raid_dir_table[] = {
131         {
132                 .procname       = "raid",
133                 .maxlen         = 0,
134                 .mode           = S_IRUGO|S_IXUGO,
135                 .child          = raid_table,
136         },
137         { }
138 };
139
140 static ctl_table raid_root_table[] = {
141         {
142                 .procname       = "dev",
143                 .maxlen         = 0,
144                 .mode           = 0555,
145                 .child          = raid_dir_table,
146         },
147         {  }
148 };
149
150 static const struct block_device_operations md_fops;
151
152 static int start_readonly;
153
154 /* bio_clone_mddev
155  * like bio_clone, but with a local bio set
156  */
157
158 static void mddev_bio_destructor(struct bio *bio)
159 {
160         struct mddev *mddev, **mddevp;
161
162         mddevp = (void*)bio;
163         mddev = mddevp[-1];
164
165         bio_free(bio, mddev->bio_set);
166 }
167
168 struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
169                             struct mddev *mddev)
170 {
171         struct bio *b;
172         struct mddev **mddevp;
173
174         if (!mddev || !mddev->bio_set)
175                 return bio_alloc(gfp_mask, nr_iovecs);
176
177         b = bio_alloc_bioset(gfp_mask, nr_iovecs,
178                              mddev->bio_set);
179         if (!b)
180                 return NULL;
181         mddevp = (void*)b;
182         mddevp[-1] = mddev;
183         b->bi_destructor = mddev_bio_destructor;
184         return b;
185 }
186 EXPORT_SYMBOL_GPL(bio_alloc_mddev);
187
188 struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
189                             struct mddev *mddev)
190 {
191         struct bio *b;
192         struct mddev **mddevp;
193
194         if (!mddev || !mddev->bio_set)
195                 return bio_clone(bio, gfp_mask);
196
197         b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
198                              mddev->bio_set);
199         if (!b)
200                 return NULL;
201         mddevp = (void*)b;
202         mddevp[-1] = mddev;
203         b->bi_destructor = mddev_bio_destructor;
204         __bio_clone(b, bio);
205         if (bio_integrity(bio)) {
206                 int ret;
207
208                 ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
209
210                 if (ret < 0) {
211                         bio_put(b);
212                         return NULL;
213                 }
214         }
215
216         return b;
217 }
218 EXPORT_SYMBOL_GPL(bio_clone_mddev);
219
220 void md_trim_bio(struct bio *bio, int offset, int size)
221 {
222         /* 'bio' is a cloned bio which we need to trim to match
223          * the given offset and size.
224          * This requires adjusting bi_sector, bi_size, and bi_io_vec
225          */
226         int i;
227         struct bio_vec *bvec;
228         int sofar = 0;
229
230         size <<= 9;
231         if (offset == 0 && size == bio->bi_size)
232                 return;
233
234         bio->bi_sector += offset;
235         bio->bi_size = size;
236         offset <<= 9;
237         clear_bit(BIO_SEG_VALID, &bio->bi_flags);
238
239         while (bio->bi_idx < bio->bi_vcnt &&
240                bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
241                 /* remove this whole bio_vec */
242                 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
243                 bio->bi_idx++;
244         }
245         if (bio->bi_idx < bio->bi_vcnt) {
246                 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
247                 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
248         }
249         /* avoid any complications with bi_idx being non-zero*/
250         if (bio->bi_idx) {
251                 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
252                         (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
253                 bio->bi_vcnt -= bio->bi_idx;
254                 bio->bi_idx = 0;
255         }
256         /* Make sure vcnt and last bv are not too big */
257         bio_for_each_segment(bvec, bio, i) {
258                 if (sofar + bvec->bv_len > size)
259                         bvec->bv_len = size - sofar;
260                 if (bvec->bv_len == 0) {
261                         bio->bi_vcnt = i;
262                         break;
263                 }
264                 sofar += bvec->bv_len;
265         }
266 }
267 EXPORT_SYMBOL_GPL(md_trim_bio);
268
269 /*
270  * We have a system wide 'event count' that is incremented
271  * on any 'interesting' event, and readers of /proc/mdstat
272  * can use 'poll' or 'select' to find out when the event
273  * count increases.
274  *
275  * Events are:
276  *  start array, stop array, error, add device, remove device,
277  *  start build, activate spare
278  */
279 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
280 static atomic_t md_event_count;
281 void md_new_event(struct mddev *mddev)
282 {
283         atomic_inc(&md_event_count);
284         wake_up(&md_event_waiters);
285 }
286 EXPORT_SYMBOL_GPL(md_new_event);
287
288 /* Alternate version that can be called from interrupts
289  * when calling sysfs_notify isn't needed.
290  */
291 static void md_new_event_inintr(struct mddev *mddev)
292 {
293         atomic_inc(&md_event_count);
294         wake_up(&md_event_waiters);
295 }
296
297 /*
298  * Enables to iterate over all existing md arrays
299  * all_mddevs_lock protects this list.
300  */
301 static LIST_HEAD(all_mddevs);
302 static DEFINE_SPINLOCK(all_mddevs_lock);
303
304
305 /*
306  * iterates through all used mddevs in the system.
307  * We take care to grab the all_mddevs_lock whenever navigating
308  * the list, and to always hold a refcount when unlocked.
309  * Any code which breaks out of this loop while own
310  * a reference to the current mddev and must mddev_put it.
311  */
312 #define for_each_mddev(_mddev,_tmp)                                     \
313                                                                         \
314         for (({ spin_lock(&all_mddevs_lock);                            \
315                 _tmp = all_mddevs.next;                                 \
316                 _mddev = NULL;});                                       \
317              ({ if (_tmp != &all_mddevs)                                \
318                         mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
319                 spin_unlock(&all_mddevs_lock);                          \
320                 if (_mddev) mddev_put(_mddev);                          \
321                 _mddev = list_entry(_tmp, struct mddev, all_mddevs);    \
322                 _tmp != &all_mddevs;});                                 \
323              ({ spin_lock(&all_mddevs_lock);                            \
324                 _tmp = _tmp->next;})                                    \
325                 )
326
327
328 /* Rather than calling directly into the personality make_request function,
329  * IO requests come here first so that we can check if the device is
330  * being suspended pending a reconfiguration.
331  * We hold a refcount over the call to ->make_request.  By the time that
332  * call has finished, the bio has been linked into some internal structure
333  * and so is visible to ->quiesce(), so we don't need the refcount any more.
334  */
335 static void md_make_request(struct request_queue *q, struct bio *bio)
336 {
337         const int rw = bio_data_dir(bio);
338         struct mddev *mddev = q->queuedata;
339         int cpu;
340         unsigned int sectors;
341
342         if (mddev == NULL || mddev->pers == NULL
343             || !mddev->ready) {
344                 bio_io_error(bio);
345                 return;
346         }
347         smp_rmb(); /* Ensure implications of  'active' are visible */
348         rcu_read_lock();
349         if (mddev->suspended) {
350                 DEFINE_WAIT(__wait);
351                 for (;;) {
352                         prepare_to_wait(&mddev->sb_wait, &__wait,
353                                         TASK_UNINTERRUPTIBLE);
354                         if (!mddev->suspended)
355                                 break;
356                         rcu_read_unlock();
357                         schedule();
358                         rcu_read_lock();
359                 }
360                 finish_wait(&mddev->sb_wait, &__wait);
361         }
362         atomic_inc(&mddev->active_io);
363         rcu_read_unlock();
364
365         /*
366          * save the sectors now since our bio can
367          * go away inside make_request
368          */
369         sectors = bio_sectors(bio);
370         mddev->pers->make_request(mddev, bio);
371
372         cpu = part_stat_lock();
373         part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
374         part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
375         part_stat_unlock();
376
377         if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
378                 wake_up(&mddev->sb_wait);
379 }
380
381 /* mddev_suspend makes sure no new requests are submitted
382  * to the device, and that any requests that have been submitted
383  * are completely handled.
384  * Once ->stop is called and completes, the module will be completely
385  * unused.
386  */
387 void mddev_suspend(struct mddev *mddev)
388 {
389         BUG_ON(mddev->suspended);
390         mddev->suspended = 1;
391         synchronize_rcu();
392         wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
393         mddev->pers->quiesce(mddev, 1);
394 }
395 EXPORT_SYMBOL_GPL(mddev_suspend);
396
397 void mddev_resume(struct mddev *mddev)
398 {
399         mddev->suspended = 0;
400         wake_up(&mddev->sb_wait);
401         mddev->pers->quiesce(mddev, 0);
402
403         md_wakeup_thread(mddev->thread);
404         md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
405 }
406 EXPORT_SYMBOL_GPL(mddev_resume);
407
408 int mddev_congested(struct mddev *mddev, int bits)
409 {
410         return mddev->suspended;
411 }
412 EXPORT_SYMBOL(mddev_congested);
413
414 /*
415  * Generic flush handling for md
416  */
417
418 static void md_end_flush(struct bio *bio, int err)
419 {
420         struct md_rdev *rdev = bio->bi_private;
421         struct mddev *mddev = rdev->mddev;
422
423         rdev_dec_pending(rdev, mddev);
424
425         if (atomic_dec_and_test(&mddev->flush_pending)) {
426                 /* The pre-request flush has finished */
427                 queue_work(md_wq, &mddev->flush_work);
428         }
429         bio_put(bio);
430 }
431
432 static void md_submit_flush_data(struct work_struct *ws);
433
434 static void submit_flushes(struct work_struct *ws)
435 {
436         struct mddev *mddev = container_of(ws, struct mddev, flush_work);
437         struct md_rdev *rdev;
438
439         INIT_WORK(&mddev->flush_work, md_submit_flush_data);
440         atomic_set(&mddev->flush_pending, 1);
441         rcu_read_lock();
442         rdev_for_each_rcu(rdev, mddev)
443                 if (rdev->raid_disk >= 0 &&
444                     !test_bit(Faulty, &rdev->flags)) {
445                         /* Take two references, one is dropped
446                          * when request finishes, one after
447                          * we reclaim rcu_read_lock
448                          */
449                         struct bio *bi;
450                         atomic_inc(&rdev->nr_pending);
451                         atomic_inc(&rdev->nr_pending);
452                         rcu_read_unlock();
453                         bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
454                         bi->bi_end_io = md_end_flush;
455                         bi->bi_private = rdev;
456                         bi->bi_bdev = rdev->bdev;
457                         atomic_inc(&mddev->flush_pending);
458                         submit_bio(WRITE_FLUSH, bi);
459                         rcu_read_lock();
460                         rdev_dec_pending(rdev, mddev);
461                 }
462         rcu_read_unlock();
463         if (atomic_dec_and_test(&mddev->flush_pending))
464                 queue_work(md_wq, &mddev->flush_work);
465 }
466
467 static void md_submit_flush_data(struct work_struct *ws)
468 {
469         struct mddev *mddev = container_of(ws, struct mddev, flush_work);
470         struct bio *bio = mddev->flush_bio;
471
472         if (bio->bi_size == 0)
473                 /* an empty barrier - all done */
474                 bio_endio(bio, 0);
475         else {
476                 bio->bi_rw &= ~REQ_FLUSH;
477                 mddev->pers->make_request(mddev, bio);
478         }
479
480         mddev->flush_bio = NULL;
481         wake_up(&mddev->sb_wait);
482 }
483
484 void md_flush_request(struct mddev *mddev, struct bio *bio)
485 {
486         spin_lock_irq(&mddev->write_lock);
487         wait_event_lock_irq(mddev->sb_wait,
488                             !mddev->flush_bio,
489                             mddev->write_lock, /*nothing*/);
490         mddev->flush_bio = bio;
491         spin_unlock_irq(&mddev->write_lock);
492
493         INIT_WORK(&mddev->flush_work, submit_flushes);
494         queue_work(md_wq, &mddev->flush_work);
495 }
496 EXPORT_SYMBOL(md_flush_request);
497
498 /* Support for plugging.
499  * This mirrors the plugging support in request_queue, but does not
500  * require having a whole queue or request structures.
501  * We allocate an md_plug_cb for each md device and each thread it gets
502  * plugged on.  This links tot the private plug_handle structure in the
503  * personality data where we keep a count of the number of outstanding
504  * plugs so other code can see if a plug is active.
505  */
506 struct md_plug_cb {
507         struct blk_plug_cb cb;
508         struct mddev *mddev;
509 };
510
511 static void plugger_unplug(struct blk_plug_cb *cb)
512 {
513         struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
514         if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
515                 md_wakeup_thread(mdcb->mddev->thread);
516         kfree(mdcb);
517 }
518
519 /* Check that an unplug wakeup will come shortly.
520  * If not, wakeup the md thread immediately
521  */
522 int mddev_check_plugged(struct mddev *mddev)
523 {
524         struct blk_plug *plug = current->plug;
525         struct md_plug_cb *mdcb;
526
527         if (!plug)
528                 return 0;
529
530         list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
531                 if (mdcb->cb.callback == plugger_unplug &&
532                     mdcb->mddev == mddev) {
533                         /* Already on the list, move to top */
534                         if (mdcb != list_first_entry(&plug->cb_list,
535                                                     struct md_plug_cb,
536                                                     cb.list))
537                                 list_move(&mdcb->cb.list, &plug->cb_list);
538                         return 1;
539                 }
540         }
541         /* Not currently on the callback list */
542         mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
543         if (!mdcb)
544                 return 0;
545
546         mdcb->mddev = mddev;
547         mdcb->cb.callback = plugger_unplug;
548         atomic_inc(&mddev->plug_cnt);
549         list_add(&mdcb->cb.list, &plug->cb_list);
550         return 1;
551 }
552 EXPORT_SYMBOL_GPL(mddev_check_plugged);
553
554 static inline struct mddev *mddev_get(struct mddev *mddev)
555 {
556         atomic_inc(&mddev->active);
557         return mddev;
558 }
559
560 static void mddev_delayed_delete(struct work_struct *ws);
561
562 static void mddev_put(struct mddev *mddev)
563 {
564         struct bio_set *bs = NULL;
565
566         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
567                 return;
568         if (!mddev->raid_disks && list_empty(&mddev->disks) &&
569             mddev->ctime == 0 && !mddev->hold_active) {
570                 /* Array is not configured at all, and not held active,
571                  * so destroy it */
572                 list_del_init(&mddev->all_mddevs);
573                 bs = mddev->bio_set;
574                 mddev->bio_set = NULL;
575                 if (mddev->gendisk) {
576                         /* We did a probe so need to clean up.  Call
577                          * queue_work inside the spinlock so that
578                          * flush_workqueue() after mddev_find will
579                          * succeed in waiting for the work to be done.
580                          */
581                         INIT_WORK(&mddev->del_work, mddev_delayed_delete);
582                         queue_work(md_misc_wq, &mddev->del_work);
583                 } else
584                         kfree(mddev);
585         }
586         spin_unlock(&all_mddevs_lock);
587         if (bs)
588                 bioset_free(bs);
589 }
590
591 void mddev_init(struct mddev *mddev)
592 {
593         mutex_init(&mddev->open_mutex);
594         mutex_init(&mddev->reconfig_mutex);
595         mutex_init(&mddev->bitmap_info.mutex);
596         INIT_LIST_HEAD(&mddev->disks);
597         INIT_LIST_HEAD(&mddev->all_mddevs);
598         init_timer(&mddev->safemode_timer);
599         atomic_set(&mddev->active, 1);
600         atomic_set(&mddev->openers, 0);
601         atomic_set(&mddev->active_io, 0);
602         atomic_set(&mddev->plug_cnt, 0);
603         spin_lock_init(&mddev->write_lock);
604         atomic_set(&mddev->flush_pending, 0);
605         init_waitqueue_head(&mddev->sb_wait);
606         init_waitqueue_head(&mddev->recovery_wait);
607         mddev->reshape_position = MaxSector;
608         mddev->resync_min = 0;
609         mddev->resync_max = MaxSector;
610         mddev->level = LEVEL_NONE;
611 }
612 EXPORT_SYMBOL_GPL(mddev_init);
613
614 static struct mddev * mddev_find(dev_t unit)
615 {
616         struct mddev *mddev, *new = NULL;
617
618         if (unit && MAJOR(unit) != MD_MAJOR)
619                 unit &= ~((1<<MdpMinorShift)-1);
620
621  retry:
622         spin_lock(&all_mddevs_lock);
623
624         if (unit) {
625                 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
626                         if (mddev->unit == unit) {
627                                 mddev_get(mddev);
628                                 spin_unlock(&all_mddevs_lock);
629                                 kfree(new);
630                                 return mddev;
631                         }
632
633                 if (new) {
634                         list_add(&new->all_mddevs, &all_mddevs);
635                         spin_unlock(&all_mddevs_lock);
636                         new->hold_active = UNTIL_IOCTL;
637                         return new;
638                 }
639         } else if (new) {
640                 /* find an unused unit number */
641                 static int next_minor = 512;
642                 int start = next_minor;
643                 int is_free = 0;
644                 int dev = 0;
645                 while (!is_free) {
646                         dev = MKDEV(MD_MAJOR, next_minor);
647                         next_minor++;
648                         if (next_minor > MINORMASK)
649                                 next_minor = 0;
650                         if (next_minor == start) {
651                                 /* Oh dear, all in use. */
652                                 spin_unlock(&all_mddevs_lock);
653                                 kfree(new);
654                                 return NULL;
655                         }
656                                 
657                         is_free = 1;
658                         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
659                                 if (mddev->unit == dev) {
660                                         is_free = 0;
661                                         break;
662                                 }
663                 }
664                 new->unit = dev;
665                 new->md_minor = MINOR(dev);
666                 new->hold_active = UNTIL_STOP;
667                 list_add(&new->all_mddevs, &all_mddevs);
668                 spin_unlock(&all_mddevs_lock);
669                 return new;
670         }
671         spin_unlock(&all_mddevs_lock);
672
673         new = kzalloc(sizeof(*new), GFP_KERNEL);
674         if (!new)
675                 return NULL;
676
677         new->unit = unit;
678         if (MAJOR(unit) == MD_MAJOR)
679                 new->md_minor = MINOR(unit);
680         else
681                 new->md_minor = MINOR(unit) >> MdpMinorShift;
682
683         mddev_init(new);
684
685         goto retry;
686 }
687
688 static inline int mddev_lock(struct mddev * mddev)
689 {
690         return mutex_lock_interruptible(&mddev->reconfig_mutex);
691 }
692
693 static inline int mddev_is_locked(struct mddev *mddev)
694 {
695         return mutex_is_locked(&mddev->reconfig_mutex);
696 }
697
698 static inline int mddev_trylock(struct mddev * mddev)
699 {
700         return mutex_trylock(&mddev->reconfig_mutex);
701 }
702
703 static struct attribute_group md_redundancy_group;
704
705 static void mddev_unlock(struct mddev * mddev)
706 {
707         if (mddev->to_remove) {
708                 /* These cannot be removed under reconfig_mutex as
709                  * an access to the files will try to take reconfig_mutex
710                  * while holding the file unremovable, which leads to
711                  * a deadlock.
712                  * So hold set sysfs_active while the remove in happeing,
713                  * and anything else which might set ->to_remove or my
714                  * otherwise change the sysfs namespace will fail with
715                  * -EBUSY if sysfs_active is still set.
716                  * We set sysfs_active under reconfig_mutex and elsewhere
717                  * test it under the same mutex to ensure its correct value
718                  * is seen.
719                  */
720                 struct attribute_group *to_remove = mddev->to_remove;
721                 mddev->to_remove = NULL;
722                 mddev->sysfs_active = 1;
723                 mutex_unlock(&mddev->reconfig_mutex);
724
725                 if (mddev->kobj.sd) {
726                         if (to_remove != &md_redundancy_group)
727                                 sysfs_remove_group(&mddev->kobj, to_remove);
728                         if (mddev->pers == NULL ||
729                             mddev->pers->sync_request == NULL) {
730                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
731                                 if (mddev->sysfs_action)
732                                         sysfs_put(mddev->sysfs_action);
733                                 mddev->sysfs_action = NULL;
734                         }
735                 }
736                 mddev->sysfs_active = 0;
737         } else
738                 mutex_unlock(&mddev->reconfig_mutex);
739
740         /* As we've dropped the mutex we need a spinlock to
741          * make sure the thread doesn't disappear
742          */
743         spin_lock(&pers_lock);
744         md_wakeup_thread(mddev->thread);
745         spin_unlock(&pers_lock);
746 }
747
748 static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
749 {
750         struct md_rdev *rdev;
751
752         rdev_for_each(rdev, mddev)
753                 if (rdev->desc_nr == nr)
754                         return rdev;
755
756         return NULL;
757 }
758
759 static struct md_rdev * find_rdev(struct mddev * mddev, dev_t dev)
760 {
761         struct md_rdev *rdev;
762
763         rdev_for_each(rdev, mddev)
764                 if (rdev->bdev->bd_dev == dev)
765                         return rdev;
766
767         return NULL;
768 }
769
770 static struct md_personality *find_pers(int level, char *clevel)
771 {
772         struct md_personality *pers;
773         list_for_each_entry(pers, &pers_list, list) {
774                 if (level != LEVEL_NONE && pers->level == level)
775                         return pers;
776                 if (strcmp(pers->name, clevel)==0)
777                         return pers;
778         }
779         return NULL;
780 }
781
782 /* return the offset of the super block in 512byte sectors */
783 static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
784 {
785         sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
786         return MD_NEW_SIZE_SECTORS(num_sectors);
787 }
788
789 static int alloc_disk_sb(struct md_rdev * rdev)
790 {
791         if (rdev->sb_page)
792                 MD_BUG();
793
794         rdev->sb_page = alloc_page(GFP_KERNEL);
795         if (!rdev->sb_page) {
796                 printk(KERN_ALERT "md: out of memory.\n");
797                 return -ENOMEM;
798         }
799
800         return 0;
801 }
802
803 static void free_disk_sb(struct md_rdev * rdev)
804 {
805         if (rdev->sb_page) {
806                 put_page(rdev->sb_page);
807                 rdev->sb_loaded = 0;
808                 rdev->sb_page = NULL;
809                 rdev->sb_start = 0;
810                 rdev->sectors = 0;
811         }
812         if (rdev->bb_page) {
813                 put_page(rdev->bb_page);
814                 rdev->bb_page = NULL;
815         }
816 }
817
818
819 static void super_written(struct bio *bio, int error)
820 {
821         struct md_rdev *rdev = bio->bi_private;
822         struct mddev *mddev = rdev->mddev;
823
824         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
825                 printk("md: super_written gets error=%d, uptodate=%d\n",
826                        error, test_bit(BIO_UPTODATE, &bio->bi_flags));
827                 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
828                 md_error(mddev, rdev);
829         }
830
831         if (atomic_dec_and_test(&mddev->pending_writes))
832                 wake_up(&mddev->sb_wait);
833         bio_put(bio);
834 }
835
836 void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
837                    sector_t sector, int size, struct page *page)
838 {
839         /* write first size bytes of page to sector of rdev
840          * Increment mddev->pending_writes before returning
841          * and decrement it on completion, waking up sb_wait
842          * if zero is reached.
843          * If an error occurred, call md_error
844          */
845         struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
846
847         bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
848         bio->bi_sector = sector;
849         bio_add_page(bio, page, size, 0);
850         bio->bi_private = rdev;
851         bio->bi_end_io = super_written;
852
853         atomic_inc(&mddev->pending_writes);
854         submit_bio(WRITE_FLUSH_FUA, bio);
855 }
856
857 void md_super_wait(struct mddev *mddev)
858 {
859         /* wait for all superblock writes that were scheduled to complete */
860         DEFINE_WAIT(wq);
861         for(;;) {
862                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
863                 if (atomic_read(&mddev->pending_writes)==0)
864                         break;
865                 schedule();
866         }
867         finish_wait(&mddev->sb_wait, &wq);
868 }
869
870 static void bi_complete(struct bio *bio, int error)
871 {
872         complete((struct completion*)bio->bi_private);
873 }
874
875 int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
876                  struct page *page, int rw, bool metadata_op)
877 {
878         struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
879         struct completion event;
880         int ret;
881
882         rw |= REQ_SYNC;
883
884         bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
885                 rdev->meta_bdev : rdev->bdev;
886         if (metadata_op)
887                 bio->bi_sector = sector + rdev->sb_start;
888         else
889                 bio->bi_sector = sector + rdev->data_offset;
890         bio_add_page(bio, page, size, 0);
891         init_completion(&event);
892         bio->bi_private = &event;
893         bio->bi_end_io = bi_complete;
894         submit_bio(rw, bio);
895         wait_for_completion(&event);
896
897         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
898         bio_put(bio);
899         return ret;
900 }
901 EXPORT_SYMBOL_GPL(sync_page_io);
902
903 static int read_disk_sb(struct md_rdev * rdev, int size)
904 {
905         char b[BDEVNAME_SIZE];
906         if (!rdev->sb_page) {
907                 MD_BUG();
908                 return -EINVAL;
909         }
910         if (rdev->sb_loaded)
911                 return 0;
912
913
914         if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
915                 goto fail;
916         rdev->sb_loaded = 1;
917         return 0;
918
919 fail:
920         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
921                 bdevname(rdev->bdev,b));
922         return -EINVAL;
923 }
924
925 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
926 {
927         return  sb1->set_uuid0 == sb2->set_uuid0 &&
928                 sb1->set_uuid1 == sb2->set_uuid1 &&
929                 sb1->set_uuid2 == sb2->set_uuid2 &&
930                 sb1->set_uuid3 == sb2->set_uuid3;
931 }
932
933 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
934 {
935         int ret;
936         mdp_super_t *tmp1, *tmp2;
937
938         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
939         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
940
941         if (!tmp1 || !tmp2) {
942                 ret = 0;
943                 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
944                 goto abort;
945         }
946
947         *tmp1 = *sb1;
948         *tmp2 = *sb2;
949
950         /*
951          * nr_disks is not constant
952          */
953         tmp1->nr_disks = 0;
954         tmp2->nr_disks = 0;
955
956         ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
957 abort:
958         kfree(tmp1);
959         kfree(tmp2);
960         return ret;
961 }
962
963
964 static u32 md_csum_fold(u32 csum)
965 {
966         csum = (csum & 0xffff) + (csum >> 16);
967         return (csum & 0xffff) + (csum >> 16);
968 }
969
970 static unsigned int calc_sb_csum(mdp_super_t * sb)
971 {
972         u64 newcsum = 0;
973         u32 *sb32 = (u32*)sb;
974         int i;
975         unsigned int disk_csum, csum;
976
977         disk_csum = sb->sb_csum;
978         sb->sb_csum = 0;
979
980         for (i = 0; i < MD_SB_BYTES/4 ; i++)
981                 newcsum += sb32[i];
982         csum = (newcsum & 0xffffffff) + (newcsum>>32);
983
984
985 #ifdef CONFIG_ALPHA
986         /* This used to use csum_partial, which was wrong for several
987          * reasons including that different results are returned on
988          * different architectures.  It isn't critical that we get exactly
989          * the same return value as before (we always csum_fold before
990          * testing, and that removes any differences).  However as we
991          * know that csum_partial always returned a 16bit value on
992          * alphas, do a fold to maximise conformity to previous behaviour.
993          */
994         sb->sb_csum = md_csum_fold(disk_csum);
995 #else
996         sb->sb_csum = disk_csum;
997 #endif
998         return csum;
999 }
1000
1001
1002 /*
1003  * Handle superblock details.
1004  * We want to be able to handle multiple superblock formats
1005  * so we have a common interface to them all, and an array of
1006  * different handlers.
1007  * We rely on user-space to write the initial superblock, and support
1008  * reading and updating of superblocks.
1009  * Interface methods are:
1010  *   int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
1011  *      loads and validates a superblock on dev.
1012  *      if refdev != NULL, compare superblocks on both devices
1013  *    Return:
1014  *      0 - dev has a superblock that is compatible with refdev
1015  *      1 - dev has a superblock that is compatible and newer than refdev
1016  *          so dev should be used as the refdev in future
1017  *     -EINVAL superblock incompatible or invalid
1018  *     -othererror e.g. -EIO
1019  *
1020  *   int validate_super(struct mddev *mddev, struct md_rdev *dev)
1021  *      Verify that dev is acceptable into mddev.
1022  *       The first time, mddev->raid_disks will be 0, and data from
1023  *       dev should be merged in.  Subsequent calls check that dev
1024  *       is new enough.  Return 0 or -EINVAL
1025  *
1026  *   void sync_super(struct mddev *mddev, struct md_rdev *dev)
1027  *     Update the superblock for rdev with data in mddev
1028  *     This does not write to disc.
1029  *
1030  */
1031
1032 struct super_type  {
1033         char                *name;
1034         struct module       *owner;
1035         int                 (*load_super)(struct md_rdev *rdev, struct md_rdev *refdev,
1036                                           int minor_version);
1037         int                 (*validate_super)(struct mddev *mddev, struct md_rdev *rdev);
1038         void                (*sync_super)(struct mddev *mddev, struct md_rdev *rdev);
1039         unsigned long long  (*rdev_size_change)(struct md_rdev *rdev,
1040                                                 sector_t num_sectors);
1041 };
1042
1043 /*
1044  * Check that the given mddev has no bitmap.
1045  *
1046  * This function is called from the run method of all personalities that do not
1047  * support bitmaps. It prints an error message and returns non-zero if mddev
1048  * has a bitmap. Otherwise, it returns 0.
1049  *
1050  */
1051 int md_check_no_bitmap(struct mddev *mddev)
1052 {
1053         if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1054                 return 0;
1055         printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1056                 mdname(mddev), mddev->pers->name);
1057         return 1;
1058 }
1059 EXPORT_SYMBOL(md_check_no_bitmap);
1060
1061 /*
1062  * load_super for 0.90.0 
1063  */
1064 static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1065 {
1066         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1067         mdp_super_t *sb;
1068         int ret;
1069
1070         /*
1071          * Calculate the position of the superblock (512byte sectors),
1072          * it's at the end of the disk.
1073          *
1074          * It also happens to be a multiple of 4Kb.
1075          */
1076         rdev->sb_start = calc_dev_sboffset(rdev);
1077
1078         ret = read_disk_sb(rdev, MD_SB_BYTES);
1079         if (ret) return ret;
1080
1081         ret = -EINVAL;
1082
1083         bdevname(rdev->bdev, b);
1084         sb = page_address(rdev->sb_page);
1085
1086         if (sb->md_magic != MD_SB_MAGIC) {
1087                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1088                        b);
1089                 goto abort;
1090         }
1091
1092         if (sb->major_version != 0 ||
1093             sb->minor_version < 90 ||
1094             sb->minor_version > 91) {
1095                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1096                         sb->major_version, sb->minor_version,
1097                         b);
1098                 goto abort;
1099         }
1100
1101         if (sb->raid_disks <= 0)
1102                 goto abort;
1103
1104         if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1105                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1106                         b);
1107                 goto abort;
1108         }
1109
1110         rdev->preferred_minor = sb->md_minor;
1111         rdev->data_offset = 0;
1112         rdev->sb_size = MD_SB_BYTES;
1113         rdev->badblocks.shift = -1;
1114
1115         if (sb->level == LEVEL_MULTIPATH)
1116                 rdev->desc_nr = -1;
1117         else
1118                 rdev->desc_nr = sb->this_disk.number;
1119
1120         if (!refdev) {
1121                 ret = 1;
1122         } else {
1123                 __u64 ev1, ev2;
1124                 mdp_super_t *refsb = page_address(refdev->sb_page);
1125                 if (!uuid_equal(refsb, sb)) {
1126                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
1127                                 b, bdevname(refdev->bdev,b2));
1128                         goto abort;
1129                 }
1130                 if (!sb_equal(refsb, sb)) {
1131                         printk(KERN_WARNING "md: %s has same UUID"
1132                                " but different superblock to %s\n",
1133                                b, bdevname(refdev->bdev, b2));
1134                         goto abort;
1135                 }
1136                 ev1 = md_event(sb);
1137                 ev2 = md_event(refsb);
1138                 if (ev1 > ev2)
1139                         ret = 1;
1140                 else 
1141                         ret = 0;
1142         }
1143         rdev->sectors = rdev->sb_start;
1144         /* Limit to 4TB as metadata cannot record more than that */
1145         if (rdev->sectors >= (2ULL << 32))
1146                 rdev->sectors = (2ULL << 32) - 2;
1147
1148         if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1149                 /* "this cannot possibly happen" ... */
1150                 ret = -EINVAL;
1151
1152  abort:
1153         return ret;
1154 }
1155
1156 /*
1157  * validate_super for 0.90.0
1158  */
1159 static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
1160 {
1161         mdp_disk_t *desc;
1162         mdp_super_t *sb = page_address(rdev->sb_page);
1163         __u64 ev1 = md_event(sb);
1164
1165         rdev->raid_disk = -1;
1166         clear_bit(Faulty, &rdev->flags);
1167         clear_bit(In_sync, &rdev->flags);
1168         clear_bit(WriteMostly, &rdev->flags);
1169
1170         if (mddev->raid_disks == 0) {
1171                 mddev->major_version = 0;
1172                 mddev->minor_version = sb->minor_version;
1173                 mddev->patch_version = sb->patch_version;
1174                 mddev->external = 0;
1175                 mddev->chunk_sectors = sb->chunk_size >> 9;
1176                 mddev->ctime = sb->ctime;
1177                 mddev->utime = sb->utime;
1178                 mddev->level = sb->level;
1179                 mddev->clevel[0] = 0;
1180                 mddev->layout = sb->layout;
1181                 mddev->raid_disks = sb->raid_disks;
1182                 mddev->dev_sectors = ((sector_t)sb->size) * 2;
1183                 mddev->events = ev1;
1184                 mddev->bitmap_info.offset = 0;
1185                 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
1186
1187                 if (mddev->minor_version >= 91) {
1188                         mddev->reshape_position = sb->reshape_position;
1189                         mddev->delta_disks = sb->delta_disks;
1190                         mddev->new_level = sb->new_level;
1191                         mddev->new_layout = sb->new_layout;
1192                         mddev->new_chunk_sectors = sb->new_chunk >> 9;
1193                 } else {
1194                         mddev->reshape_position = MaxSector;
1195                         mddev->delta_disks = 0;
1196                         mddev->new_level = mddev->level;
1197                         mddev->new_layout = mddev->layout;
1198                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1199                 }
1200
1201                 if (sb->state & (1<<MD_SB_CLEAN))
1202                         mddev->recovery_cp = MaxSector;
1203                 else {
1204                         if (sb->events_hi == sb->cp_events_hi && 
1205                                 sb->events_lo == sb->cp_events_lo) {
1206                                 mddev->recovery_cp = sb->recovery_cp;
1207                         } else
1208                                 mddev->recovery_cp = 0;
1209                 }
1210
1211                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1212                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1213                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1214                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1215
1216                 mddev->max_disks = MD_SB_DISKS;
1217
1218                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1219                     mddev->bitmap_info.file == NULL)
1220                         mddev->bitmap_info.offset =
1221                                 mddev->bitmap_info.default_offset;
1222
1223         } else if (mddev->pers == NULL) {
1224                 /* Insist on good event counter while assembling, except
1225                  * for spares (which don't need an event count) */
1226                 ++ev1;
1227                 if (sb->disks[rdev->desc_nr].state & (
1228                             (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1229                         if (ev1 < mddev->events) 
1230                                 return -EINVAL;
1231         } else if (mddev->bitmap) {
1232                 /* if adding to array with a bitmap, then we can accept an
1233                  * older device ... but not too old.
1234                  */
1235                 if (ev1 < mddev->bitmap->events_cleared)
1236                         return 0;
1237         } else {
1238                 if (ev1 < mddev->events)
1239                         /* just a hot-add of a new device, leave raid_disk at -1 */
1240                         return 0;
1241         }
1242
1243         if (mddev->level != LEVEL_MULTIPATH) {
1244                 desc = sb->disks + rdev->desc_nr;
1245
1246                 if (desc->state & (1<<MD_DISK_FAULTY))
1247                         set_bit(Faulty, &rdev->flags);
1248                 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1249                             desc->raid_disk < mddev->raid_disks */) {
1250                         set_bit(In_sync, &rdev->flags);
1251                         rdev->raid_disk = desc->raid_disk;
1252                 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1253                         /* active but not in sync implies recovery up to
1254                          * reshape position.  We don't know exactly where
1255                          * that is, so set to zero for now */
1256                         if (mddev->minor_version >= 91) {
1257                                 rdev->recovery_offset = 0;
1258                                 rdev->raid_disk = desc->raid_disk;
1259                         }
1260                 }
1261                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1262                         set_bit(WriteMostly, &rdev->flags);
1263         } else /* MULTIPATH are always insync */
1264                 set_bit(In_sync, &rdev->flags);
1265         return 0;
1266 }
1267
1268 /*
1269  * sync_super for 0.90.0
1270  */
1271 static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
1272 {
1273         mdp_super_t *sb;
1274         struct md_rdev *rdev2;
1275         int next_spare = mddev->raid_disks;
1276
1277
1278         /* make rdev->sb match mddev data..
1279          *
1280          * 1/ zero out disks
1281          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1282          * 3/ any empty disks < next_spare become removed
1283          *
1284          * disks[0] gets initialised to REMOVED because
1285          * we cannot be sure from other fields if it has
1286          * been initialised or not.
1287          */
1288         int i;
1289         int active=0, working=0,failed=0,spare=0,nr_disks=0;
1290
1291         rdev->sb_size = MD_SB_BYTES;
1292
1293         sb = page_address(rdev->sb_page);
1294
1295         memset(sb, 0, sizeof(*sb));
1296
1297         sb->md_magic = MD_SB_MAGIC;
1298         sb->major_version = mddev->major_version;
1299         sb->patch_version = mddev->patch_version;
1300         sb->gvalid_words  = 0; /* ignored */
1301         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1302         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1303         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1304         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1305
1306         sb->ctime = mddev->ctime;
1307         sb->level = mddev->level;
1308         sb->size = mddev->dev_sectors / 2;
1309         sb->raid_disks = mddev->raid_disks;
1310         sb->md_minor = mddev->md_minor;
1311         sb->not_persistent = 0;
1312         sb->utime = mddev->utime;
1313         sb->state = 0;
1314         sb->events_hi = (mddev->events>>32);
1315         sb->events_lo = (u32)mddev->events;
1316
1317         if (mddev->reshape_position == MaxSector)
1318                 sb->minor_version = 90;
1319         else {
1320                 sb->minor_version = 91;
1321                 sb->reshape_position = mddev->reshape_position;
1322                 sb->new_level = mddev->new_level;
1323                 sb->delta_disks = mddev->delta_disks;
1324                 sb->new_layout = mddev->new_layout;
1325                 sb->new_chunk = mddev->new_chunk_sectors << 9;
1326         }
1327         mddev->minor_version = sb->minor_version;
1328         if (mddev->in_sync)
1329         {
1330                 sb->recovery_cp = mddev->recovery_cp;
1331                 sb->cp_events_hi = (mddev->events>>32);
1332                 sb->cp_events_lo = (u32)mddev->events;
1333                 if (mddev->recovery_cp == MaxSector)
1334                         sb->state = (1<< MD_SB_CLEAN);
1335         } else
1336                 sb->recovery_cp = 0;
1337
1338         sb->layout = mddev->layout;
1339         sb->chunk_size = mddev->chunk_sectors << 9;
1340
1341         if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1342                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1343
1344         sb->disks[0].state = (1<<MD_DISK_REMOVED);
1345         rdev_for_each(rdev2, mddev) {
1346                 mdp_disk_t *d;
1347                 int desc_nr;
1348                 int is_active = test_bit(In_sync, &rdev2->flags);
1349
1350                 if (rdev2->raid_disk >= 0 &&
1351                     sb->minor_version >= 91)
1352                         /* we have nowhere to store the recovery_offset,
1353                          * but if it is not below the reshape_position,
1354                          * we can piggy-back on that.
1355                          */
1356                         is_active = 1;
1357                 if (rdev2->raid_disk < 0 ||
1358                     test_bit(Faulty, &rdev2->flags))
1359                         is_active = 0;
1360                 if (is_active)
1361                         desc_nr = rdev2->raid_disk;
1362                 else
1363                         desc_nr = next_spare++;
1364                 rdev2->desc_nr = desc_nr;
1365                 d = &sb->disks[rdev2->desc_nr];
1366                 nr_disks++;
1367                 d->number = rdev2->desc_nr;
1368                 d->major = MAJOR(rdev2->bdev->bd_dev);
1369                 d->minor = MINOR(rdev2->bdev->bd_dev);
1370                 if (is_active)
1371                         d->raid_disk = rdev2->raid_disk;
1372                 else
1373                         d->raid_disk = rdev2->desc_nr; /* compatibility */
1374                 if (test_bit(Faulty, &rdev2->flags))
1375                         d->state = (1<<MD_DISK_FAULTY);
1376                 else if (is_active) {
1377                         d->state = (1<<MD_DISK_ACTIVE);
1378                         if (test_bit(In_sync, &rdev2->flags))
1379                                 d->state |= (1<<MD_DISK_SYNC);
1380                         active++;
1381                         working++;
1382                 } else {
1383                         d->state = 0;
1384                         spare++;
1385                         working++;
1386                 }
1387                 if (test_bit(WriteMostly, &rdev2->flags))
1388                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
1389         }
1390         /* now set the "removed" and "faulty" bits on any missing devices */
1391         for (i=0 ; i < mddev->raid_disks ; i++) {
1392                 mdp_disk_t *d = &sb->disks[i];
1393                 if (d->state == 0 && d->number == 0) {
1394                         d->number = i;
1395                         d->raid_disk = i;
1396                         d->state = (1<<MD_DISK_REMOVED);
1397                         d->state |= (1<<MD_DISK_FAULTY);
1398                         failed++;
1399                 }
1400         }
1401         sb->nr_disks = nr_disks;
1402         sb->active_disks = active;
1403         sb->working_disks = working;
1404         sb->failed_disks = failed;
1405         sb->spare_disks = spare;
1406
1407         sb->this_disk = sb->disks[rdev->desc_nr];
1408         sb->sb_csum = calc_sb_csum(sb);
1409 }
1410
1411 /*
1412  * rdev_size_change for 0.90.0
1413  */
1414 static unsigned long long
1415 super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1416 {
1417         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1418                 return 0; /* component must fit device */
1419         if (rdev->mddev->bitmap_info.offset)
1420                 return 0; /* can't move bitmap */
1421         rdev->sb_start = calc_dev_sboffset(rdev);
1422         if (!num_sectors || num_sectors > rdev->sb_start)
1423                 num_sectors = rdev->sb_start;
1424         /* Limit to 4TB as metadata cannot record more than that.
1425          * 4TB == 2^32 KB, or 2*2^32 sectors.
1426          */
1427         if (num_sectors >= (2ULL << 32))
1428                 num_sectors = (2ULL << 32) - 2;
1429         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1430                        rdev->sb_page);
1431         md_super_wait(rdev->mddev);
1432         return num_sectors;
1433 }
1434
1435
1436 /*
1437  * version 1 superblock
1438  */
1439
1440 static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1441 {
1442         __le32 disk_csum;
1443         u32 csum;
1444         unsigned long long newcsum;
1445         int size = 256 + le32_to_cpu(sb->max_dev)*2;
1446         __le32 *isuper = (__le32*)sb;
1447         int i;
1448
1449         disk_csum = sb->sb_csum;
1450         sb->sb_csum = 0;
1451         newcsum = 0;
1452         for (i=0; size>=4; size -= 4 )
1453                 newcsum += le32_to_cpu(*isuper++);
1454
1455         if (size == 2)
1456                 newcsum += le16_to_cpu(*(__le16*) isuper);
1457
1458         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1459         sb->sb_csum = disk_csum;
1460         return cpu_to_le32(csum);
1461 }
1462
1463 static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1464                             int acknowledged);
1465 static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
1466 {
1467         struct mdp_superblock_1 *sb;
1468         int ret;
1469         sector_t sb_start;
1470         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1471         int bmask;
1472
1473         /*
1474          * Calculate the position of the superblock in 512byte sectors.
1475          * It is always aligned to a 4K boundary and
1476          * depeding on minor_version, it can be:
1477          * 0: At least 8K, but less than 12K, from end of device
1478          * 1: At start of device
1479          * 2: 4K from start of device.
1480          */
1481         switch(minor_version) {
1482         case 0:
1483                 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1484                 sb_start -= 8*2;
1485                 sb_start &= ~(sector_t)(4*2-1);
1486                 break;
1487         case 1:
1488                 sb_start = 0;
1489                 break;
1490         case 2:
1491                 sb_start = 8;
1492                 break;
1493         default:
1494                 return -EINVAL;
1495         }
1496         rdev->sb_start = sb_start;
1497
1498         /* superblock is rarely larger than 1K, but it can be larger,
1499          * and it is safe to read 4k, so we do that
1500          */
1501         ret = read_disk_sb(rdev, 4096);
1502         if (ret) return ret;
1503
1504
1505         sb = page_address(rdev->sb_page);
1506
1507         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1508             sb->major_version != cpu_to_le32(1) ||
1509             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1510             le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1511             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1512                 return -EINVAL;
1513
1514         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1515                 printk("md: invalid superblock checksum on %s\n",
1516                         bdevname(rdev->bdev,b));
1517                 return -EINVAL;
1518         }
1519         if (le64_to_cpu(sb->data_size) < 10) {
1520                 printk("md: data_size too small on %s\n",
1521                        bdevname(rdev->bdev,b));
1522                 return -EINVAL;
1523         }
1524
1525         rdev->preferred_minor = 0xffff;
1526         rdev->data_offset = le64_to_cpu(sb->data_offset);
1527         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1528
1529         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1530         bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1531         if (rdev->sb_size & bmask)
1532                 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1533
1534         if (minor_version
1535             && rdev->data_offset < sb_start + (rdev->sb_size/512))
1536                 return -EINVAL;
1537
1538         if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1539                 rdev->desc_nr = -1;
1540         else
1541                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1542
1543         if (!rdev->bb_page) {
1544                 rdev->bb_page = alloc_page(GFP_KERNEL);
1545                 if (!rdev->bb_page)
1546                         return -ENOMEM;
1547         }
1548         if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1549             rdev->badblocks.count == 0) {
1550                 /* need to load the bad block list.
1551                  * Currently we limit it to one page.
1552                  */
1553                 s32 offset;
1554                 sector_t bb_sector;
1555                 u64 *bbp;
1556                 int i;
1557                 int sectors = le16_to_cpu(sb->bblog_size);
1558                 if (sectors > (PAGE_SIZE / 512))
1559                         return -EINVAL;
1560                 offset = le32_to_cpu(sb->bblog_offset);
1561                 if (offset == 0)
1562                         return -EINVAL;
1563                 bb_sector = (long long)offset;
1564                 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1565                                   rdev->bb_page, READ, true))
1566                         return -EIO;
1567                 bbp = (u64 *)page_address(rdev->bb_page);
1568                 rdev->badblocks.shift = sb->bblog_shift;
1569                 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1570                         u64 bb = le64_to_cpu(*bbp);
1571                         int count = bb & (0x3ff);
1572                         u64 sector = bb >> 10;
1573                         sector <<= sb->bblog_shift;
1574                         count <<= sb->bblog_shift;
1575                         if (bb + 1 == 0)
1576                                 break;
1577                         if (md_set_badblocks(&rdev->badblocks,
1578                                              sector, count, 1) == 0)
1579                                 return -EINVAL;
1580                 }
1581         } else if (sb->bblog_offset == 0)
1582                 rdev->badblocks.shift = -1;
1583
1584         if (!refdev) {
1585                 ret = 1;
1586         } else {
1587                 __u64 ev1, ev2;
1588                 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1589
1590                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1591                     sb->level != refsb->level ||
1592                     sb->layout != refsb->layout ||
1593                     sb->chunksize != refsb->chunksize) {
1594                         printk(KERN_WARNING "md: %s has strangely different"
1595                                 " superblock to %s\n",
1596                                 bdevname(rdev->bdev,b),
1597                                 bdevname(refdev->bdev,b2));
1598                         return -EINVAL;
1599                 }
1600                 ev1 = le64_to_cpu(sb->events);
1601                 ev2 = le64_to_cpu(refsb->events);
1602
1603                 if (ev1 > ev2)
1604                         ret = 1;
1605                 else
1606                         ret = 0;
1607         }
1608         if (minor_version)
1609                 rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1610                         le64_to_cpu(sb->data_offset);
1611         else
1612                 rdev->sectors = rdev->sb_start;
1613         if (rdev->sectors < le64_to_cpu(sb->data_size))
1614                 return -EINVAL;
1615         rdev->sectors = le64_to_cpu(sb->data_size);
1616         if (le64_to_cpu(sb->size) > rdev->sectors)
1617                 return -EINVAL;
1618         return ret;
1619 }
1620
1621 static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
1622 {
1623         struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1624         __u64 ev1 = le64_to_cpu(sb->events);
1625
1626         rdev->raid_disk = -1;
1627         clear_bit(Faulty, &rdev->flags);
1628         clear_bit(In_sync, &rdev->flags);
1629         clear_bit(WriteMostly, &rdev->flags);
1630
1631         if (mddev->raid_disks == 0) {
1632                 mddev->major_version = 1;
1633                 mddev->patch_version = 0;
1634                 mddev->external = 0;
1635                 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1636                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1637                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1638                 mddev->level = le32_to_cpu(sb->level);
1639                 mddev->clevel[0] = 0;
1640                 mddev->layout = le32_to_cpu(sb->layout);
1641                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1642                 mddev->dev_sectors = le64_to_cpu(sb->size);
1643                 mddev->events = ev1;
1644                 mddev->bitmap_info.offset = 0;
1645                 mddev->bitmap_info.default_offset = 1024 >> 9;
1646                 
1647                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1648                 memcpy(mddev->uuid, sb->set_uuid, 16);
1649
1650                 mddev->max_disks =  (4096-256)/2;
1651
1652                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1653                     mddev->bitmap_info.file == NULL )
1654                         mddev->bitmap_info.offset =
1655                                 (__s32)le32_to_cpu(sb->bitmap_offset);
1656
1657                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1658                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1659                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1660                         mddev->new_level = le32_to_cpu(sb->new_level);
1661                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1662                         mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
1663                 } else {
1664                         mddev->reshape_position = MaxSector;
1665                         mddev->delta_disks = 0;
1666                         mddev->new_level = mddev->level;
1667                         mddev->new_layout = mddev->layout;
1668                         mddev->new_chunk_sectors = mddev->chunk_sectors;
1669                 }
1670
1671         } else if (mddev->pers == NULL) {
1672                 /* Insist of good event counter while assembling, except for
1673                  * spares (which don't need an event count) */
1674                 ++ev1;
1675                 if (rdev->desc_nr >= 0 &&
1676                     rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1677                     le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1678                         if (ev1 < mddev->events)
1679                                 return -EINVAL;
1680         } else if (mddev->bitmap) {
1681                 /* If adding to array with a bitmap, then we can accept an
1682                  * older device, but not too old.
1683                  */
1684                 if (ev1 < mddev->bitmap->events_cleared)
1685                         return 0;
1686         } else {
1687                 if (ev1 < mddev->events)
1688                         /* just a hot-add of a new device, leave raid_disk at -1 */
1689                         return 0;
1690         }
1691         if (mddev->level != LEVEL_MULTIPATH) {
1692                 int role;
1693                 if (rdev->desc_nr < 0 ||
1694                     rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1695                         role = 0xffff;
1696                         rdev->desc_nr = -1;
1697                 } else
1698                         role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1699                 switch(role) {
1700                 case 0xffff: /* spare */
1701                         break;
1702                 case 0xfffe: /* faulty */
1703                         set_bit(Faulty, &rdev->flags);
1704                         break;
1705                 default:
1706                         if ((le32_to_cpu(sb->feature_map) &
1707                              MD_FEATURE_RECOVERY_OFFSET))
1708                                 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1709                         else
1710                                 set_bit(In_sync, &rdev->flags);
1711                         rdev->raid_disk = role;
1712                         break;
1713                 }
1714                 if (sb->devflags & WriteMostly1)
1715                         set_bit(WriteMostly, &rdev->flags);
1716                 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1717                         set_bit(Replacement, &rdev->flags);
1718         } else /* MULTIPATH are always insync */
1719                 set_bit(In_sync, &rdev->flags);
1720
1721         return 0;
1722 }
1723
1724 static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
1725 {
1726         struct mdp_superblock_1 *sb;
1727         struct md_rdev *rdev2;
1728         int max_dev, i;
1729         /* make rdev->sb match mddev and rdev data. */
1730
1731         sb = page_address(rdev->sb_page);
1732
1733         sb->feature_map = 0;
1734         sb->pad0 = 0;
1735         sb->recovery_offset = cpu_to_le64(0);
1736         memset(sb->pad1, 0, sizeof(sb->pad1));
1737         memset(sb->pad3, 0, sizeof(sb->pad3));
1738
1739         sb->utime = cpu_to_le64((__u64)mddev->utime);
1740         sb->events = cpu_to_le64(mddev->events);
1741         if (mddev->in_sync)
1742                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1743         else
1744                 sb->resync_offset = cpu_to_le64(0);
1745
1746         sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1747
1748         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1749         sb->size = cpu_to_le64(mddev->dev_sectors);
1750         sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1751         sb->level = cpu_to_le32(mddev->level);
1752         sb->layout = cpu_to_le32(mddev->layout);
1753
1754         if (test_bit(WriteMostly, &rdev->flags))
1755                 sb->devflags |= WriteMostly1;
1756         else
1757                 sb->devflags &= ~WriteMostly1;
1758
1759         if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1760                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1761                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1762         }
1763
1764         if (rdev->raid_disk >= 0 &&
1765             !test_bit(In_sync, &rdev->flags)) {
1766                 sb->feature_map |=
1767                         cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1768                 sb->recovery_offset =
1769                         cpu_to_le64(rdev->recovery_offset);
1770         }
1771         if (test_bit(Replacement, &rdev->flags))
1772                 sb->feature_map |=
1773                         cpu_to_le32(MD_FEATURE_REPLACEMENT);
1774
1775         if (mddev->reshape_position != MaxSector) {
1776                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1777                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1778                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1779                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1780                 sb->new_level = cpu_to_le32(mddev->new_level);
1781                 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
1782         }
1783
1784         if (rdev->badblocks.count == 0)
1785                 /* Nothing to do for bad blocks*/ ;
1786         else if (sb->bblog_offset == 0)
1787                 /* Cannot record bad blocks on this device */
1788                 md_error(mddev, rdev);
1789         else {
1790                 struct badblocks *bb = &rdev->badblocks;
1791                 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1792                 u64 *p = bb->page;
1793                 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1794                 if (bb->changed) {
1795                         unsigned seq;
1796
1797 retry:
1798                         seq = read_seqbegin(&bb->lock);
1799
1800                         memset(bbp, 0xff, PAGE_SIZE);
1801
1802                         for (i = 0 ; i < bb->count ; i++) {
1803                                 u64 internal_bb = *p++;
1804                                 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1805                                                 | BB_LEN(internal_bb));
1806                                 *bbp++ = cpu_to_le64(store_bb);
1807                         }
1808                         bb->changed = 0;
1809                         if (read_seqretry(&bb->lock, seq))
1810                                 goto retry;
1811
1812                         bb->sector = (rdev->sb_start +
1813                                       (int)le32_to_cpu(sb->bblog_offset));
1814                         bb->size = le16_to_cpu(sb->bblog_size);
1815                 }
1816         }
1817
1818         max_dev = 0;
1819         rdev_for_each(rdev2, mddev)
1820                 if (rdev2->desc_nr+1 > max_dev)
1821                         max_dev = rdev2->desc_nr+1;
1822
1823         if (max_dev > le32_to_cpu(sb->max_dev)) {
1824                 int bmask;
1825                 sb->max_dev = cpu_to_le32(max_dev);
1826                 rdev->sb_size = max_dev * 2 + 256;
1827                 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1828                 if (rdev->sb_size & bmask)
1829                         rdev->sb_size = (rdev->sb_size | bmask) + 1;
1830         } else
1831                 max_dev = le32_to_cpu(sb->max_dev);
1832
1833         for (i=0; i<max_dev;i++)
1834                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1835         
1836         rdev_for_each(rdev2, mddev) {
1837                 i = rdev2->desc_nr;
1838                 if (test_bit(Faulty, &rdev2->flags))
1839                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1840                 else if (test_bit(In_sync, &rdev2->flags))
1841                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1842                 else if (rdev2->raid_disk >= 0)
1843                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1844                 else
1845                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1846         }
1847
1848         sb->sb_csum = calc_sb_1_csum(sb);
1849 }
1850
1851 static unsigned long long
1852 super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
1853 {
1854         struct mdp_superblock_1 *sb;
1855         sector_t max_sectors;
1856         if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1857                 return 0; /* component must fit device */
1858         if (rdev->sb_start < rdev->data_offset) {
1859                 /* minor versions 1 and 2; superblock before data */
1860                 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1861                 max_sectors -= rdev->data_offset;
1862                 if (!num_sectors || num_sectors > max_sectors)
1863                         num_sectors = max_sectors;
1864         } else if (rdev->mddev->bitmap_info.offset) {
1865                 /* minor version 0 with bitmap we can't move */
1866                 return 0;
1867         } else {
1868                 /* minor version 0; superblock after data */
1869                 sector_t sb_start;
1870                 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1871                 sb_start &= ~(sector_t)(4*2 - 1);
1872                 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1873                 if (!num_sectors || num_sectors > max_sectors)
1874                         num_sectors = max_sectors;
1875                 rdev->sb_start = sb_start;
1876         }
1877         sb = page_address(rdev->sb_page);
1878         sb->data_size = cpu_to_le64(num_sectors);
1879         sb->super_offset = rdev->sb_start;
1880         sb->sb_csum = calc_sb_1_csum(sb);
1881         md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1882                        rdev->sb_page);
1883         md_super_wait(rdev->mddev);
1884         return num_sectors;
1885 }
1886
1887 static struct super_type super_types[] = {
1888         [0] = {
1889                 .name   = "0.90.0",
1890                 .owner  = THIS_MODULE,
1891                 .load_super         = super_90_load,
1892                 .validate_super     = super_90_validate,
1893                 .sync_super         = super_90_sync,
1894                 .rdev_size_change   = super_90_rdev_size_change,
1895         },
1896         [1] = {
1897                 .name   = "md-1",
1898                 .owner  = THIS_MODULE,
1899                 .load_super         = super_1_load,
1900                 .validate_super     = super_1_validate,
1901                 .sync_super         = super_1_sync,
1902                 .rdev_size_change   = super_1_rdev_size_change,
1903         },
1904 };
1905
1906 static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
1907 {
1908         if (mddev->sync_super) {
1909                 mddev->sync_super(mddev, rdev);
1910                 return;
1911         }
1912
1913         BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1914
1915         super_types[mddev->major_version].sync_super(mddev, rdev);
1916 }
1917
1918 static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
1919 {
1920         struct md_rdev *rdev, *rdev2;
1921
1922         rcu_read_lock();
1923         rdev_for_each_rcu(rdev, mddev1)
1924                 rdev_for_each_rcu(rdev2, mddev2)
1925                         if (rdev->bdev->bd_contains ==
1926                             rdev2->bdev->bd_contains) {
1927                                 rcu_read_unlock();
1928                                 return 1;
1929                         }
1930         rcu_read_unlock();
1931         return 0;
1932 }
1933
1934 static LIST_HEAD(pending_raid_disks);
1935
1936 /*
1937  * Try to register data integrity profile for an mddev
1938  *
1939  * This is called when an array is started and after a disk has been kicked
1940  * from the array. It only succeeds if all working and active component devices
1941  * are integrity capable with matching profiles.
1942  */
1943 int md_integrity_register(struct mddev *mddev)
1944 {
1945         struct md_rdev *rdev, *reference = NULL;
1946
1947         if (list_empty(&mddev->disks))
1948                 return 0; /* nothing to do */
1949         if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1950                 return 0; /* shouldn't register, or already is */
1951         rdev_for_each(rdev, mddev) {
1952                 /* skip spares and non-functional disks */
1953                 if (test_bit(Faulty, &rdev->flags))
1954                         continue;
1955                 if (rdev->raid_disk < 0)
1956                         continue;
1957                 if (!reference) {
1958                         /* Use the first rdev as the reference */
1959                         reference = rdev;
1960                         continue;
1961                 }
1962                 /* does this rdev's profile match the reference profile? */
1963                 if (blk_integrity_compare(reference->bdev->bd_disk,
1964                                 rdev->bdev->bd_disk) < 0)
1965                         return -EINVAL;
1966         }
1967         if (!reference || !bdev_get_integrity(reference->bdev))
1968                 return 0;
1969         /*
1970          * All component devices are integrity capable and have matching
1971          * profiles, register the common profile for the md device.
1972          */
1973         if (blk_integrity_register(mddev->gendisk,
1974                         bdev_get_integrity(reference->bdev)) != 0) {
1975                 printk(KERN_ERR "md: failed to register integrity for %s\n",
1976                         mdname(mddev));
1977                 return -EINVAL;
1978         }
1979         printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1980         if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1981                 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1982                        mdname(mddev));
1983                 return -EINVAL;
1984         }
1985         return 0;
1986 }
1987 EXPORT_SYMBOL(md_integrity_register);
1988
1989 /* Disable data integrity if non-capable/non-matching disk is being added */
1990 void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
1991 {
1992         struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1993         struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1994
1995         if (!bi_mddev) /* nothing to do */
1996                 return;
1997         if (rdev->raid_disk < 0) /* skip spares */
1998                 return;
1999         if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2000                                              rdev->bdev->bd_disk) >= 0)
2001                 return;
2002         printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2003         blk_integrity_unregister(mddev->gendisk);
2004 }
2005 EXPORT_SYMBOL(md_integrity_add_rdev);
2006
2007 static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
2008 {
2009         char b[BDEVNAME_SIZE];
2010         struct kobject *ko;
2011         char *s;
2012         int err;
2013
2014         if (rdev->mddev) {
2015                 MD_BUG();
2016                 return -EINVAL;
2017         }
2018
2019         /* prevent duplicates */
2020         if (find_rdev(mddev, rdev->bdev->bd_dev))
2021                 return -EEXIST;
2022
2023         /* make sure rdev->sectors exceeds mddev->dev_sectors */
2024         if (rdev->sectors && (mddev->dev_sectors == 0 ||
2025                         rdev->sectors < mddev->dev_sectors)) {
2026                 if (mddev->pers) {
2027                         /* Cannot change size, so fail
2028                          * If mddev->level <= 0, then we don't care
2029                          * about aligning sizes (e.g. linear)
2030                          */
2031                         if (mddev->level > 0)
2032                                 return -ENOSPC;
2033                 } else
2034                         mddev->dev_sectors = rdev->sectors;
2035         }
2036
2037         /* Verify rdev->desc_nr is unique.
2038          * If it is -1, assign a free number, else
2039          * check number is not in use
2040          */
2041         if (rdev->desc_nr < 0) {
2042                 int choice = 0;
2043                 if (mddev->pers) choice = mddev->raid_disks;
2044                 while (find_rdev_nr(mddev, choice))
2045                         choice++;
2046                 rdev->desc_nr = choice;
2047         } else {
2048                 if (find_rdev_nr(mddev, rdev->desc_nr))
2049                         return -EBUSY;
2050         }
2051         if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2052                 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2053                        mdname(mddev), mddev->max_disks);
2054                 return -EBUSY;
2055         }
2056         bdevname(rdev->bdev,b);
2057         while ( (s=strchr(b, '/')) != NULL)
2058                 *s = '!';
2059
2060         rdev->mddev = mddev;
2061         printk(KERN_INFO "md: bind<%s>\n", b);
2062
2063         if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2064                 goto fail;
2065
2066         ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2067         if (sysfs_create_link(&rdev->kobj, ko, "block"))
2068                 /* failure here is OK */;
2069         rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2070
2071         list_add_rcu(&rdev->same_set, &mddev->disks);
2072         bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2073
2074         /* May as well allow recovery to be retried once */
2075         mddev->recovery_disabled++;
2076
2077         return 0;
2078
2079  fail:
2080         printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2081                b, mdname(mddev));
2082         return err;
2083 }
2084
2085 static void md_delayed_delete(struct work_struct *ws)
2086 {
2087         struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
2088         kobject_del(&rdev->kobj);
2089         kobject_put(&rdev->kobj);
2090 }
2091
2092 static void unbind_rdev_from_array(struct md_rdev * rdev)
2093 {
2094         char b[BDEVNAME_SIZE];
2095         if (!rdev->mddev) {
2096                 MD_BUG();
2097                 return;
2098         }
2099         bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2100         list_del_rcu(&rdev->same_set);
2101         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2102         rdev->mddev = NULL;
2103         sysfs_remove_link(&rdev->kobj, "block");
2104         sysfs_put(rdev->sysfs_state);
2105         rdev->sysfs_state = NULL;
2106         kfree(rdev->badblocks.page);
2107         rdev->badblocks.count = 0;
2108         rdev->badblocks.page = NULL;
2109         /* We need to delay this, otherwise we can deadlock when
2110          * writing to 'remove' to "dev/state".  We also need
2111          * to delay it due to rcu usage.
2112          */
2113         synchronize_rcu();
2114         INIT_WORK(&rdev->del_work, md_delayed_delete);
2115         kobject_get(&rdev->kobj);
2116         queue_work(md_misc_wq, &rdev->del_work);
2117 }
2118
2119 /*
2120  * prevent the device from being mounted, repartitioned or
2121  * otherwise reused by a RAID array (or any other kernel
2122  * subsystem), by bd_claiming the device.
2123  */
2124 static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
2125 {
2126         int err = 0;
2127         struct block_device *bdev;
2128         char b[BDEVNAME_SIZE];
2129
2130         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2131                                  shared ? (struct md_rdev *)lock_rdev : rdev);
2132         if (IS_ERR(bdev)) {
2133                 printk(KERN_ERR "md: could not open %s.\n",
2134                         __bdevname(dev, b));
2135                 return PTR_ERR(bdev);
2136         }
2137         rdev->bdev = bdev;
2138         return err;
2139 }
2140
2141 static void unlock_rdev(struct md_rdev *rdev)
2142 {
2143         struct block_device *bdev = rdev->bdev;
2144         rdev->bdev = NULL;
2145         if (!bdev)
2146                 MD_BUG();
2147         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2148 }
2149
2150 void md_autodetect_dev(dev_t dev);
2151
2152 static void export_rdev(struct md_rdev * rdev)
2153 {
2154         char b[BDEVNAME_SIZE];
2155         printk(KERN_INFO "md: export_rdev(%s)\n",
2156                 bdevname(rdev->bdev,b));
2157         if (rdev->mddev)
2158                 MD_BUG();
2159         free_disk_sb(rdev);
2160 #ifndef MODULE
2161         if (test_bit(AutoDetected, &rdev->flags))
2162                 md_autodetect_dev(rdev->bdev->bd_dev);
2163 #endif
2164         unlock_rdev(rdev);
2165         kobject_put(&rdev->kobj);
2166 }
2167
2168 static void kick_rdev_from_array(struct md_rdev * rdev)
2169 {
2170         unbind_rdev_from_array(rdev);
2171         export_rdev(rdev);
2172 }
2173
2174 static void export_array(struct mddev *mddev)
2175 {
2176         struct md_rdev *rdev, *tmp;
2177
2178         rdev_for_each_safe(rdev, tmp, mddev) {
2179                 if (!rdev->mddev) {
2180                         MD_BUG();
2181                         continue;
2182                 }
2183                 kick_rdev_from_array(rdev);
2184         }
2185         if (!list_empty(&mddev->disks))
2186                 MD_BUG();
2187         mddev->raid_disks = 0;
2188         mddev->major_version = 0;
2189 }
2190
2191 static void print_desc(mdp_disk_t *desc)
2192 {
2193         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2194                 desc->major,desc->minor,desc->raid_disk,desc->state);
2195 }
2196
2197 static void print_sb_90(mdp_super_t *sb)
2198 {
2199         int i;
2200
2201         printk(KERN_INFO 
2202                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2203                 sb->major_version, sb->minor_version, sb->patch_version,
2204                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2205                 sb->ctime);
2206         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2207                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2208                 sb->md_minor, sb->layout, sb->chunk_size);
2209         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
2210                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2211                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2212                 sb->failed_disks, sb->spare_disks,
2213                 sb->sb_csum, (unsigned long)sb->events_lo);
2214
2215         printk(KERN_INFO);
2216         for (i = 0; i < MD_SB_DISKS; i++) {
2217                 mdp_disk_t *desc;
2218
2219                 desc = sb->disks + i;
2220                 if (desc->number || desc->major || desc->minor ||
2221                     desc->raid_disk || (desc->state && (desc->state != 4))) {
2222                         printk("     D %2d: ", i);
2223                         print_desc(desc);
2224                 }
2225         }
2226         printk(KERN_INFO "md:     THIS: ");
2227         print_desc(&sb->this_disk);
2228 }
2229
2230 static void print_sb_1(struct mdp_superblock_1 *sb)
2231 {
2232         __u8 *uuid;
2233
2234         uuid = sb->set_uuid;
2235         printk(KERN_INFO
2236                "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2237                "md:    Name: \"%s\" CT:%llu\n",
2238                 le32_to_cpu(sb->major_version),
2239                 le32_to_cpu(sb->feature_map),
2240                 uuid,
2241                 sb->set_name,
2242                 (unsigned long long)le64_to_cpu(sb->ctime)
2243                        & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2244
2245         uuid = sb->device_uuid;
2246         printk(KERN_INFO
2247                "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2248                         " RO:%llu\n"
2249                "md:     Dev:%08x UUID: %pU\n"
2250                "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2251                "md:         (MaxDev:%u) \n",
2252                 le32_to_cpu(sb->level),
2253                 (unsigned long long)le64_to_cpu(sb->size),
2254                 le32_to_cpu(sb->raid_disks),
2255                 le32_to_cpu(sb->layout),
2256                 le32_to_cpu(sb->chunksize),
2257                 (unsigned long long)le64_to_cpu(sb->data_offset),
2258                 (unsigned long long)le64_to_cpu(sb->data_size),
2259                 (unsigned long long)le64_to_cpu(sb->super_offset),
2260                 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2261                 le32_to_cpu(sb->dev_number),
2262                 uuid,
2263                 sb->devflags,
2264                 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2265                 (unsigned long long)le64_to_cpu(sb->events),
2266                 (unsigned long long)le64_to_cpu(sb->resync_offset),
2267                 le32_to_cpu(sb->sb_csum),
2268                 le32_to_cpu(sb->max_dev)
2269                 );
2270 }
2271
2272 static void print_rdev(struct md_rdev *rdev, int major_version)
2273 {
2274         char b[BDEVNAME_SIZE];
2275         printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2276                 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2277                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2278                 rdev->desc_nr);
2279         if (rdev->sb_loaded) {
2280                 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2281                 switch (major_version) {
2282                 case 0:
2283                         print_sb_90(page_address(rdev->sb_page));
2284                         break;
2285                 case 1:
2286                         print_sb_1(page_address(rdev->sb_page));
2287                         break;
2288                 }
2289         } else
2290                 printk(KERN_INFO "md: no rdev superblock!\n");
2291 }
2292
2293 static void md_print_devices(void)
2294 {
2295         struct list_head *tmp;
2296         struct md_rdev *rdev;
2297         struct mddev *mddev;
2298         char b[BDEVNAME_SIZE];
2299
2300         printk("\n");
2301         printk("md:     **********************************\n");
2302         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
2303         printk("md:     **********************************\n");
2304         for_each_mddev(mddev, tmp) {
2305
2306                 if (mddev->bitmap)
2307                         bitmap_print_sb(mddev->bitmap);
2308                 else
2309                         printk("%s: ", mdname(mddev));
2310                 rdev_for_each(rdev, mddev)
2311                         printk("<%s>", bdevname(rdev->bdev,b));
2312                 printk("\n");
2313
2314                 rdev_for_each(rdev, mddev)
2315                         print_rdev(rdev, mddev->major_version);
2316         }
2317         printk("md:     **********************************\n");
2318         printk("\n");
2319 }
2320
2321
2322 static void sync_sbs(struct mddev * mddev, int nospares)
2323 {
2324         /* Update each superblock (in-memory image), but
2325          * if we are allowed to, skip spares which already
2326          * have the right event counter, or have one earlier
2327          * (which would mean they aren't being marked as dirty
2328          * with the rest of the array)
2329          */
2330         struct md_rdev *rdev;
2331         rdev_for_each(rdev, mddev) {
2332                 if (rdev->sb_events == mddev->events ||
2333                     (nospares &&
2334                      rdev->raid_disk < 0 &&
2335                      rdev->sb_events+1 == mddev->events)) {
2336                         /* Don't update this superblock */
2337                         rdev->sb_loaded = 2;
2338                 } else {
2339                         sync_super(mddev, rdev);
2340                         rdev->sb_loaded = 1;
2341                 }
2342         }
2343 }
2344
2345 static void md_update_sb(struct mddev * mddev, int force_change)
2346 {
2347         struct md_rdev *rdev;
2348         int sync_req;
2349         int nospares = 0;
2350         int any_badblocks_changed = 0;
2351
2352 repeat:
2353         /* First make sure individual recovery_offsets are correct */
2354         rdev_for_each(rdev, mddev) {
2355                 if (rdev->raid_disk >= 0 &&
2356                     mddev->delta_disks >= 0 &&
2357                     !test_bit(In_sync, &rdev->flags) &&
2358                     mddev->curr_resync_completed > rdev->recovery_offset)
2359                                 rdev->recovery_offset = mddev->curr_resync_completed;
2360
2361         }       
2362         if (!mddev->persistent) {
2363                 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2364                 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2365                 if (!mddev->external) {
2366                         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2367                         rdev_for_each(rdev, mddev) {
2368                                 if (rdev->badblocks.changed) {
2369                                         rdev->badblocks.changed = 0;
2370                                         md_ack_all_badblocks(&rdev->badblocks);
2371                                         md_error(mddev, rdev);
2372                                 }
2373                                 clear_bit(Blocked, &rdev->flags);
2374                                 clear_bit(BlockedBadBlocks, &rdev->flags);
2375                                 wake_up(&rdev->blocked_wait);
2376                         }
2377                 }
2378                 wake_up(&mddev->sb_wait);
2379                 return;
2380         }
2381
2382         spin_lock_irq(&mddev->write_lock);
2383
2384         mddev->utime = get_seconds();
2385
2386         if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2387                 force_change = 1;
2388         if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2389                 /* just a clean<-> dirty transition, possibly leave spares alone,
2390                  * though if events isn't the right even/odd, we will have to do
2391                  * spares after all
2392                  */
2393                 nospares = 1;
2394         if (force_change)
2395                 nospares = 0;
2396         if (mddev->degraded)
2397                 /* If the array is degraded, then skipping spares is both
2398                  * dangerous and fairly pointless.
2399                  * Dangerous because a device that was removed from the array
2400                  * might have a event_count that still looks up-to-date,
2401                  * so it can be re-added without a resync.
2402                  * Pointless because if there are any spares to skip,
2403                  * then a recovery will happen and soon that array won't
2404                  * be degraded any more and the spare can go back to sleep then.
2405                  */
2406                 nospares = 0;
2407
2408         sync_req = mddev->in_sync;
2409
2410         /* If this is just a dirty<->clean transition, and the array is clean
2411          * and 'events' is odd, we can roll back to the previous clean state */
2412         if (nospares
2413             && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2414             && mddev->can_decrease_events
2415             && mddev->events != 1) {
2416                 mddev->events--;
2417                 mddev->can_decrease_events = 0;
2418         } else {
2419                 /* otherwise we have to go forward and ... */
2420                 mddev->events ++;
2421                 mddev->can_decrease_events = nospares;
2422         }
2423
2424         if (!mddev->events) {
2425                 /*
2426                  * oops, this 64-bit counter should never wrap.
2427                  * Either we are in around ~1 trillion A.C., assuming
2428                  * 1 reboot per second, or we have a bug:
2429                  */
2430                 MD_BUG();
2431                 mddev->events --;
2432         }
2433
2434         rdev_for_each(rdev, mddev) {
2435                 if (rdev->badblocks.changed)
2436                         any_badblocks_changed++;
2437                 if (test_bit(Faulty, &rdev->flags))
2438                         set_bit(FaultRecorded, &rdev->flags);
2439         }
2440
2441         sync_sbs(mddev, nospares);
2442         spin_unlock_irq(&mddev->write_lock);
2443
2444         pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2445                  mdname(mddev), mddev->in_sync);
2446
2447         bitmap_update_sb(mddev->bitmap);
2448         rdev_for_each(rdev, mddev) {
2449                 char b[BDEVNAME_SIZE];
2450
2451                 if (rdev->sb_loaded != 1)
2452                         continue; /* no noise on spare devices */
2453
2454                 if (!test_bit(Faulty, &rdev->flags) &&
2455                     rdev->saved_raid_disk == -1) {
2456                         md_super_write(mddev,rdev,
2457                                        rdev->sb_start, rdev->sb_size,
2458                                        rdev->sb_page);
2459                         pr_debug("md: (write) %s's sb offset: %llu\n",
2460                                  bdevname(rdev->bdev, b),
2461                                  (unsigned long long)rdev->sb_start);
2462                         rdev->sb_events = mddev->events;
2463                         if (rdev->badblocks.size) {
2464                                 md_super_write(mddev, rdev,
2465                                                rdev->badblocks.sector,
2466                                                rdev->badblocks.size << 9,
2467                                                rdev->bb_page);
2468                                 rdev->badblocks.size = 0;
2469                         }
2470
2471                 } else if (test_bit(Faulty, &rdev->flags))
2472                         pr_debug("md: %s (skipping faulty)\n",
2473                                  bdevname(rdev->bdev, b));
2474                 else
2475                         pr_debug("(skipping incremental s/r ");
2476
2477                 if (mddev->level == LEVEL_MULTIPATH)
2478                         /* only need to write one superblock... */
2479                         break;
2480         }
2481         md_super_wait(mddev);
2482         /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
2483
2484         spin_lock_irq(&mddev->write_lock);
2485         if (mddev->in_sync != sync_req ||
2486             test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
2487                 /* have to write it out again */
2488                 spin_unlock_irq(&mddev->write_lock);
2489                 goto repeat;
2490         }
2491         clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2492         spin_unlock_irq(&mddev->write_lock);
2493         wake_up(&mddev->sb_wait);
2494         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2495                 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
2496
2497         rdev_for_each(rdev, mddev) {
2498                 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2499                         clear_bit(Blocked, &rdev->flags);
2500
2501                 if (any_badblocks_changed)
2502                         md_ack_all_badblocks(&rdev->badblocks);
2503                 clear_bit(BlockedBadBlocks, &rdev->flags);
2504                 wake_up(&rdev->blocked_wait);
2505         }
2506 }
2507
2508 /* words written to sysfs files may, or may not, be \n terminated.
2509  * We want to accept with case. For this we use cmd_match.
2510  */
2511 static int cmd_match(const char *cmd, const char *str)
2512 {
2513         /* See if cmd, written into a sysfs file, matches
2514          * str.  They must either be the same, or cmd can
2515          * have a trailing newline
2516          */
2517         while (*cmd && *str && *cmd == *str) {
2518                 cmd++;
2519                 str++;
2520         }
2521         if (*cmd == '\n')
2522                 cmd++;
2523         if (*str || *cmd)
2524                 return 0;
2525         return 1;
2526 }
2527
2528 struct rdev_sysfs_entry {
2529         struct attribute attr;
2530         ssize_t (*show)(struct md_rdev *, char *);
2531         ssize_t (*store)(struct md_rdev *, const char *, size_t);
2532 };
2533
2534 static ssize_t
2535 state_show(struct md_rdev *rdev, char *page)
2536 {
2537         char *sep = "";
2538         size_t len = 0;
2539
2540         if (test_bit(Faulty, &rdev->flags) ||
2541             rdev->badblocks.unacked_exist) {
2542                 len+= sprintf(page+len, "%sfaulty",sep);
2543                 sep = ",";
2544         }
2545         if (test_bit(In_sync, &rdev->flags)) {
2546                 len += sprintf(page+len, "%sin_sync",sep);
2547                 sep = ",";
2548         }
2549         if (test_bit(WriteMostly, &rdev->flags)) {
2550                 len += sprintf(page+len, "%swrite_mostly",sep);
2551                 sep = ",";
2552         }
2553         if (test_bit(Blocked, &rdev->flags) ||
2554             (rdev->badblocks.unacked_exist
2555              && !test_bit(Faulty, &rdev->flags))) {
2556                 len += sprintf(page+len, "%sblocked", sep);
2557                 sep = ",";
2558         }
2559         if (!test_bit(Faulty, &rdev->flags) &&
2560             !test_bit(In_sync, &rdev->flags)) {
2561                 len += sprintf(page+len, "%sspare", sep);
2562                 sep = ",";
2563         }
2564         if (test_bit(WriteErrorSeen, &rdev->flags)) {
2565                 len += sprintf(page+len, "%swrite_error", sep);
2566                 sep = ",";
2567         }
2568         if (test_bit(WantReplacement, &rdev->flags)) {
2569                 len += sprintf(page+len, "%swant_replacement", sep);
2570                 sep = ",";
2571         }
2572         if (test_bit(Replacement, &rdev->flags)) {
2573                 len += sprintf(page+len, "%sreplacement", sep);
2574                 sep = ",";
2575         }
2576
2577         return len+sprintf(page+len, "\n");
2578 }
2579
2580 static ssize_t
2581 state_store(struct md_rdev *rdev, const char *buf, size_t len)
2582 {
2583         /* can write
2584          *  faulty  - simulates an error
2585          *  remove  - disconnects the device
2586          *  writemostly - sets write_mostly
2587          *  -writemostly - clears write_mostly
2588          *  blocked - sets the Blocked flags
2589          *  -blocked - clears the Blocked and possibly simulates an error
2590          *  insync - sets Insync providing device isn't active
2591          *  write_error - sets WriteErrorSeen
2592          *  -write_error - clears WriteErrorSeen
2593          */
2594         int err = -EINVAL;
2595         if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2596                 md_error(rdev->mddev, rdev);
2597                 if (test_bit(Faulty, &rdev->flags))
2598                         err = 0;
2599                 else
2600                         err = -EBUSY;
2601         } else if (cmd_match(buf, "remove")) {
2602                 if (rdev->raid_disk >= 0)
2603                         err = -EBUSY;
2604                 else {
2605                         struct mddev *mddev = rdev->mddev;
2606                         kick_rdev_from_array(rdev);
2607                         if (mddev->pers)
2608                                 md_update_sb(mddev, 1);
2609                         md_new_event(mddev);
2610                         err = 0;
2611                 }
2612         } else if (cmd_match(buf, "writemostly")) {
2613                 set_bit(WriteMostly, &rdev->flags);
2614                 err = 0;
2615         } else if (cmd_match(buf, "-writemostly")) {
2616                 clear_bit(WriteMostly, &rdev->flags);
2617                 err = 0;
2618         } else if (cmd_match(buf, "blocked")) {
2619                 set_bit(Blocked, &rdev->flags);
2620                 err = 0;
2621         } else if (cmd_match(buf, "-blocked")) {
2622                 if (!test_bit(Faulty, &rdev->flags) &&
2623                     rdev->badblocks.unacked_exist) {
2624                         /* metadata handler doesn't understand badblocks,
2625                          * so we need to fail the device
2626                          */
2627                         md_error(rdev->mddev, rdev);
2628                 }
2629                 clear_bit(Blocked, &rdev->flags);
2630                 clear_bit(BlockedBadBlocks, &rdev->flags);
2631                 wake_up(&rdev->blocked_wait);
2632                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2633                 md_wakeup_thread(rdev->mddev->thread);
2634
2635                 err = 0;
2636         } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2637                 set_bit(In_sync, &rdev->flags);
2638                 err = 0;
2639         } else if (cmd_match(buf, "write_error")) {
2640                 set_bit(WriteErrorSeen, &rdev->flags);
2641                 err = 0;
2642         } else if (cmd_match(buf, "-write_error")) {
2643                 clear_bit(WriteErrorSeen, &rdev->flags);
2644                 err = 0;
2645         } else if (cmd_match(buf, "want_replacement")) {
2646                 /* Any non-spare device that is not a replacement can
2647                  * become want_replacement at any time, but we then need to
2648                  * check if recovery is needed.
2649                  */
2650                 if (rdev->raid_disk >= 0 &&
2651                     !test_bit(Replacement, &rdev->flags))
2652                         set_bit(WantReplacement, &rdev->flags);
2653                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2654                 md_wakeup_thread(rdev->mddev->thread);
2655                 err = 0;
2656         } else if (cmd_match(buf, "-want_replacement")) {
2657                 /* Clearing 'want_replacement' is always allowed.
2658                  * Once replacements starts it is too late though.
2659                  */
2660                 err = 0;
2661                 clear_bit(WantReplacement, &rdev->flags);
2662         } else if (cmd_match(buf, "replacement")) {
2663                 /* Can only set a device as a replacement when array has not
2664                  * yet been started.  Once running, replacement is automatic
2665                  * from spares, or by assigning 'slot'.
2666                  */
2667                 if (rdev->mddev->pers)
2668                         err = -EBUSY;
2669                 else {
2670                         set_bit(Replacement, &rdev->flags);
2671                         err = 0;
2672                 }
2673         } else if (cmd_match(buf, "-replacement")) {
2674                 /* Similarly, can only clear Replacement before start */
2675                 if (rdev->mddev->pers)
2676                         err = -EBUSY;
2677                 else {
2678                         clear_bit(Replacement, &rdev->flags);
2679                         err = 0;
2680                 }
2681         }
2682         if (!err)
2683                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2684         return err ? err : len;
2685 }
2686 static struct rdev_sysfs_entry rdev_state =
2687 __ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2688
2689 static ssize_t
2690 errors_show(struct md_rdev *rdev, char *page)
2691 {
2692         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2693 }
2694
2695 static ssize_t
2696 errors_store(struct md_rdev *rdev, const char *buf, size_t len)
2697 {
2698         char *e;
2699         unsigned long n = simple_strtoul(buf, &e, 10);
2700         if (*buf && (*e == 0 || *e == '\n')) {
2701                 atomic_set(&rdev->corrected_errors, n);
2702                 return len;
2703         }
2704         return -EINVAL;
2705 }
2706 static struct rdev_sysfs_entry rdev_errors =
2707 __ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2708
2709 static ssize_t
2710 slot_show(struct md_rdev *rdev, char *page)
2711 {
2712         if (rdev->raid_disk < 0)
2713                 return sprintf(page, "none\n");
2714         else
2715                 return sprintf(page, "%d\n", rdev->raid_disk);
2716 }
2717
2718 static ssize_t
2719 slot_store(struct md_rdev *rdev, const char *buf, size_t len)
2720 {
2721         char *e;
2722         int err;
2723         int slot = simple_strtoul(buf, &e, 10);
2724         if (strncmp(buf, "none", 4)==0)
2725                 slot = -1;
2726         else if (e==buf || (*e && *e!= '\n'))
2727                 return -EINVAL;
2728         if (rdev->mddev->pers && slot == -1) {
2729                 /* Setting 'slot' on an active array requires also
2730                  * updating the 'rd%d' link, and communicating
2731                  * with the personality with ->hot_*_disk.
2732                  * For now we only support removing
2733                  * failed/spare devices.  This normally happens automatically,
2734                  * but not when the metadata is externally managed.
2735                  */
2736                 if (rdev->raid_disk == -1)
2737                         return -EEXIST;
2738                 /* personality does all needed checks */
2739                 if (rdev->mddev->pers->hot_remove_disk == NULL)
2740                         return -EINVAL;
2741                 err = rdev->mddev->pers->
2742                         hot_remove_disk(rdev->mddev, rdev);
2743                 if (err)
2744                         return err;
2745                 sysfs_unlink_rdev(rdev->mddev, rdev);
2746                 rdev->raid_disk = -1;
2747                 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2748                 md_wakeup_thread(rdev->mddev->thread);
2749         } else if (rdev->mddev->pers) {
2750                 /* Activating a spare .. or possibly reactivating
2751                  * if we ever get bitmaps working here.
2752                  */
2753
2754                 if (rdev->raid_disk != -1)
2755                         return -EBUSY;
2756
2757                 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2758                         return -EBUSY;
2759
2760                 if (rdev->mddev->pers->hot_add_disk == NULL)
2761                         return -EINVAL;
2762
2763                 if (slot >= rdev->mddev->raid_disks &&
2764                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2765                         return -ENOSPC;
2766
2767                 rdev->raid_disk = slot;
2768                 if (test_bit(In_sync, &rdev->flags))
2769                         rdev->saved_raid_disk = slot;
2770                 else
2771                         rdev->saved_raid_disk = -1;
2772                 clear_bit(In_sync, &rdev->flags);
2773                 err = rdev->mddev->pers->
2774                         hot_add_disk(rdev->mddev, rdev);
2775                 if (err) {
2776                         rdev->raid_disk = -1;
2777                         return err;
2778                 } else
2779                         sysfs_notify_dirent_safe(rdev->sysfs_state);
2780                 if (sysfs_link_rdev(rdev->mddev, rdev))
2781                         /* failure here is OK */;
2782                 /* don't wakeup anyone, leave that to userspace. */
2783         } else {
2784                 if (slot >= rdev->mddev->raid_disks &&
2785                     slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2786                         return -ENOSPC;
2787                 rdev->raid_disk = slot;
2788                 /* assume it is working */
2789                 clear_bit(Faulty, &rdev->flags);
2790                 clear_bit(WriteMostly, &rdev->flags);
2791                 set_bit(In_sync, &rdev->flags);
2792                 sysfs_notify_dirent_safe(rdev->sysfs_state);
2793         }
2794         return len;
2795 }
2796
2797
2798 static struct rdev_sysfs_entry rdev_slot =
2799 __ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2800
2801 static ssize_t
2802 offset_show(struct md_rdev *rdev, char *page)
2803 {
2804         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2805 }
2806
2807 static ssize_t
2808 offset_store(struct md_rdev *rdev, const char *buf, size_t len)
2809 {
2810         char *e;
2811         unsigned long long offset = simple_strtoull(buf, &e, 10);
2812         if (e==buf || (*e && *e != '\n'))
2813                 return -EINVAL;
2814         if (rdev->mddev->pers && rdev->raid_disk >= 0)
2815                 return -EBUSY;
2816         if (rdev->sectors && rdev->mddev->external)
2817                 /* Must set offset before size, so overlap checks
2818                  * can be sane */
2819                 return -EBUSY;
2820         rdev->data_offset = offset;
2821         return len;
2822 }
2823
2824 static struct rdev_sysfs_entry rdev_offset =
2825 __ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2826
2827 static ssize_t
2828 rdev_size_show(struct md_rdev *rdev, char *page)
2829 {
2830         return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2831 }
2832
2833 static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2834 {
2835         /* check if two start/length pairs overlap */
2836         if (s1+l1 <= s2)
2837                 return 0;
2838         if (s2+l2 <= s1)
2839                 return 0;
2840         return 1;
2841 }
2842
2843 static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2844 {
2845         unsigned long long blocks;
2846         sector_t new;
2847
2848         if (strict_strtoull(buf, 10, &blocks) < 0)
2849                 return -EINVAL;
2850
2851         if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2852                 return -EINVAL; /* sector conversion overflow */
2853
2854         new = blocks * 2;
2855         if (new != blocks * 2)
2856                 return -EINVAL; /* unsigned long long to sector_t overflow */
2857
2858         *sectors = new;
2859         return 0;
2860 }
2861
2862 static ssize_t
2863 rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
2864 {
2865         struct mddev *my_mddev = rdev->mddev;
2866         sector_t oldsectors = rdev->sectors;
2867         sector_t sectors;
2868
2869         if (strict_blocks_to_sectors(buf, &sectors) < 0)
2870                 return -EINVAL;
2871         if (my_mddev->pers && rdev->raid_disk >= 0) {
2872                 if (my_mddev->persistent) {
2873                         sectors = super_types[my_mddev->major_version].
2874                                 rdev_size_change(rdev, sectors);
2875                         if (!sectors)
2876                                 return -EBUSY;
2877                 } else if (!sectors)
2878                         sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2879                                 rdev->data_offset;
2880         }
2881         if (sectors < my_mddev->dev_sectors)
2882                 return -EINVAL; /* component must fit device */
2883
2884         rdev->sectors = sectors;
2885         if (sectors > oldsectors && my_mddev->external) {
2886                 /* need to check that all other rdevs with the same ->bdev
2887                  * do not overlap.  We need to unlock the mddev to avoid
2888                  * a deadlock.  We have already changed rdev->sectors, and if
2889                  * we have to change it back, we will have the lock again.
2890                  */
2891                 struct mddev *mddev;
2892                 int overlap = 0;
2893                 struct list_head *tmp;
2894
2895                 mddev_unlock(my_mddev);
2896                 for_each_mddev(mddev, tmp) {
2897                         struct md_rdev *rdev2;
2898
2899                         mddev_lock(mddev);
2900                         rdev_for_each(rdev2, mddev)
2901                                 if (rdev->bdev == rdev2->bdev &&
2902                                     rdev != rdev2 &&
2903                                     overlaps(rdev->data_offset, rdev->sectors,
2904                                              rdev2->data_offset,
2905                                              rdev2->sectors)) {
2906                                         overlap = 1;
2907                                         break;
2908                                 }
2909                         mddev_unlock(mddev);
2910                         if (overlap) {
2911                                 mddev_put(mddev);
2912                                 break;
2913                         }
2914                 }
2915                 mddev_lock(my_mddev);
2916                 if (overlap) {
2917                         /* Someone else could have slipped in a size
2918                          * change here, but doing so is just silly.
2919                          * We put oldsectors back because we *know* it is
2920                          * safe, and trust userspace not to race with
2921                          * itself
2922                          */
2923                         rdev->sectors = oldsectors;
2924                         return -EBUSY;
2925                 }
2926         }
2927         return len;
2928 }
2929
2930 static struct rdev_sysfs_entry rdev_size =
2931 __ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2932
2933
2934 static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
2935 {
2936         unsigned long long recovery_start = rdev->recovery_offset;
2937
2938         if (test_bit(In_sync, &rdev->flags) ||
2939             recovery_start == MaxSector)
2940                 return sprintf(page, "none\n");
2941
2942         return sprintf(page, "%llu\n", recovery_start);
2943 }
2944
2945 static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
2946 {
2947         unsigned long long recovery_start;
2948
2949         if (cmd_match(buf, "none"))
2950                 recovery_start = MaxSector;
2951         else if (strict_strtoull(buf, 10, &recovery_start))
2952                 return -EINVAL;
2953
2954         if (rdev->mddev->pers &&
2955             rdev->raid_disk >= 0)
2956                 return -EBUSY;
2957
2958         rdev->recovery_offset = recovery_start;
2959         if (recovery_start == MaxSector)
2960                 set_bit(In_sync, &rdev->flags);
2961         else
2962                 clear_bit(In_sync, &rdev->flags);
2963         return len;
2964 }
2965
2966 static struct rdev_sysfs_entry rdev_recovery_start =
2967 __ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2968
2969
2970 static ssize_t
2971 badblocks_show(struct badblocks *bb, char *page, int unack);
2972 static ssize_t
2973 badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2974
2975 static ssize_t bb_show(struct md_rdev *rdev, char *page)
2976 {
2977         return badblocks_show(&rdev->badblocks, page, 0);
2978 }
2979 static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
2980 {
2981         int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2982         /* Maybe that ack was all we needed */
2983         if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2984                 wake_up(&rdev->blocked_wait);
2985         return rv;
2986 }
2987 static struct rdev_sysfs_entry rdev_bad_blocks =
2988 __ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2989
2990
2991 static ssize_t ubb_show(struct md_rdev *rdev, char *page)
2992 {
2993         return badblocks_show(&rdev->badblocks, page, 1);
2994 }
2995 static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
2996 {
2997         return badblocks_store(&rdev->badblocks, page, len, 1);
2998 }
2999 static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3000 __ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3001
3002 static struct attribute *rdev_default_attrs[] = {
3003         &rdev_state.attr,
3004         &rdev_errors.attr,
3005         &rdev_slot.attr,
3006         &rdev_offset.attr,
3007         &rdev_size.attr,
3008         &rdev_recovery_start.attr,
3009         &rdev_bad_blocks.attr,
3010         &rdev_unack_bad_blocks.attr,
3011         NULL,
3012 };
3013 static ssize_t
3014 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3015 {
3016         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3017         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3018         struct mddev *mddev = rdev->mddev;
3019         ssize_t rv;
3020
3021         if (!entry->show)
3022                 return -EIO;
3023
3024         rv = mddev ? mddev_lock(mddev) : -EBUSY;
3025         if (!rv) {
3026                 if (rdev->mddev == NULL)
3027                         rv = -EBUSY;
3028                 else
3029                         rv = entry->show(rdev, page);
3030                 mddev_unlock(mddev);
3031         }
3032         return rv;
3033 }
3034
3035 static ssize_t
3036 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3037               const char *page, size_t length)
3038 {
3039         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
3040         struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
3041         ssize_t rv;
3042         struct mddev *mddev = rdev->mddev;
3043
3044         if (!entry->store)
3045                 return -EIO;
3046         if (!capable(CAP_SYS_ADMIN))
3047                 return -EACCES;
3048         rv = mddev ? mddev_lock(mddev): -EBUSY;
3049         if (!rv) {
3050                 if (rdev->mddev == NULL)
3051                         rv = -EBUSY;
3052                 else
3053                         rv = entry->store(rdev, page, length);
3054                 mddev_unlock(mddev);
3055         }
3056         return rv;
3057 }
3058
3059 static void rdev_free(struct kobject *ko)
3060 {
3061         struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
3062         kfree(rdev);
3063 }
3064 static const struct sysfs_ops rdev_sysfs_ops = {
3065         .show           = rdev_attr_show,
3066         .store          = rdev_attr_store,
3067 };
3068 static struct kobj_type rdev_ktype = {
3069         .release        = rdev_free,
3070         .sysfs_ops      = &rdev_sysfs_ops,
3071         .default_attrs  = rdev_default_attrs,
3072 };
3073
3074 int md_rdev_init(struct md_rdev *rdev)
3075 {
3076         rdev->desc_nr = -1;
3077         rdev->saved_raid_disk = -1;
3078         rdev->raid_disk = -1;
3079         rdev->flags = 0;
3080         rdev->data_offset = 0;
3081         rdev->sb_events = 0;
3082         rdev->last_read_error.tv_sec  = 0;
3083         rdev->last_read_error.tv_nsec = 0;
3084         rdev->sb_loaded = 0;
3085         rdev->bb_page = NULL;
3086         atomic_set(&rdev->nr_pending, 0);
3087         atomic_set(&rdev->read_errors, 0);
3088         atomic_set(&rdev->corrected_errors, 0);
3089
3090         INIT_LIST_HEAD(&rdev->same_set);
3091         init_waitqueue_head(&rdev->blocked_wait);
3092
3093         /* Add space to store bad block list.
3094          * This reserves the space even on arrays where it cannot
3095          * be used - I wonder if that matters
3096          */
3097         rdev->badblocks.count = 0;
3098         rdev->badblocks.shift = 0;
3099         rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3100         seqlock_init(&rdev->badblocks.lock);
3101         if (rdev->badblocks.page == NULL)
3102                 return -ENOMEM;
3103
3104         return 0;
3105 }
3106 EXPORT_SYMBOL_GPL(md_rdev_init);
3107 /*
3108  * Import a device. If 'super_format' >= 0, then sanity check the superblock
3109  *
3110  * mark the device faulty if:
3111  *
3112  *   - the device is nonexistent (zero size)
3113  *   - the device has no valid superblock
3114  *
3115  * a faulty rdev _never_ has rdev->sb set.
3116  */
3117 static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
3118 {
3119         char b[BDEVNAME_SIZE];
3120         int err;
3121         struct md_rdev *rdev;
3122         sector_t size;
3123
3124         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3125         if (!rdev) {
3126                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3127                 return ERR_PTR(-ENOMEM);
3128         }
3129
3130         err = md_rdev_init(rdev);
3131         if (err)
3132                 goto abort_free;
3133         err = alloc_disk_sb(rdev);
3134         if (err)
3135                 goto abort_free;
3136
3137         err = lock_rdev(rdev, newdev, super_format == -2);
3138         if (err)
3139                 goto abort_free;
3140
3141         kobject_init(&rdev->kobj, &rdev_ktype);
3142
3143         size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3144         if (!size) {
3145                 printk(KERN_WARNING 
3146                         "md: %s has zero or unknown size, marking faulty!\n",
3147                         bdevname(rdev->bdev,b));
3148                 err = -EINVAL;
3149                 goto abort_free;
3150         }
3151
3152         if (super_format >= 0) {
3153                 err = super_types[super_format].
3154                         load_super(rdev, NULL, super_minor);
3155                 if (err == -EINVAL) {
3156                         printk(KERN_WARNING
3157                                 "md: %s does not have a valid v%d.%d "
3158                                "superblock, not importing!\n",
3159                                 bdevname(rdev->bdev,b),
3160                                super_format, super_minor);
3161                         goto abort_free;
3162                 }
3163                 if (err < 0) {
3164                         printk(KERN_WARNING 
3165                                 "md: could not read %s's sb, not importing!\n",
3166                                 bdevname(rdev->bdev,b));
3167                         goto abort_free;
3168                 }
3169         }
3170         if (super_format == -1)
3171                 /* hot-add for 0.90, or non-persistent: so no badblocks */
3172                 rdev->badblocks.shift = -1;
3173
3174         return rdev;
3175
3176 abort_free:
3177         if (rdev->bdev)
3178                 unlock_rdev(rdev);
3179         free_disk_sb(rdev);
3180         kfree(rdev->badblocks.page);
3181         kfree(rdev);
3182         return ERR_PTR(err);
3183 }
3184
3185 /*
3186  * Check a full RAID array for plausibility
3187  */
3188
3189
3190 static void analyze_sbs(struct mddev * mddev)
3191 {
3192         int i;
3193         struct md_rdev *rdev, *freshest, *tmp;
3194         char b[BDEVNAME_SIZE];
3195
3196         freshest = NULL;
3197         rdev_for_each_safe(rdev, tmp, mddev)
3198                 switch (super_types[mddev->major_version].
3199                         load_super(rdev, freshest, mddev->minor_version)) {
3200                 case 1:
3201                         freshest = rdev;
3202                         break;
3203                 case 0:
3204                         break;
3205                 default:
3206                         printk( KERN_ERR \
3207                                 "md: fatal superblock inconsistency in %s"
3208                                 " -- removing from array\n", 
3209                                 bdevname(rdev->bdev,b));
3210                         kick_rdev_from_array(rdev);
3211                 }
3212
3213
3214         super_types[mddev->major_version].
3215                 validate_super(mddev, freshest);
3216
3217         i = 0;
3218         rdev_for_each_safe(rdev, tmp, mddev) {
3219                 if (mddev->max_disks &&
3220                     (rdev->desc_nr >= mddev->max_disks ||
3221                      i > mddev->max_disks)) {
3222                         printk(KERN_WARNING
3223                                "md: %s: %s: only %d devices permitted\n",
3224                                mdname(mddev), bdevname(rdev->bdev, b),
3225                                mddev->max_disks);
3226                         kick_rdev_from_array(rdev);
3227                         continue;
3228                 }
3229                 if (rdev != freshest)
3230                         if (super_types[mddev->major_version].
3231                             validate_super(mddev, rdev)) {
3232                                 printk(KERN_WARNING "md: kicking non-fresh %s"
3233                                         " from array!\n",
3234                                         bdevname(rdev->bdev,b));
3235                                 kick_rdev_from_array(rdev);
3236                                 continue;
3237                         }
3238                 if (mddev->level == LEVEL_MULTIPATH) {
3239                         rdev->desc_nr = i++;
3240                         rdev->raid_disk = rdev->desc_nr;
3241                         set_bit(In_sync, &rdev->flags);
3242                 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3243                         rdev->raid_disk = -1;
3244                         clear_bit(In_sync, &rdev->flags);
3245                 }
3246         }
3247 }
3248
3249 /* Read a fixed-point number.
3250  * Numbers in sysfs attributes should be in "standard" units where
3251  * possible, so time should be in seconds.
3252  * However we internally use a a much smaller unit such as 
3253  * milliseconds or jiffies.
3254  * This function takes a decimal number with a possible fractional
3255  * component, and produces an integer which is the result of
3256  * multiplying that number by 10^'scale'.
3257  * all without any floating-point arithmetic.
3258  */
3259 int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3260 {
3261         unsigned long result = 0;
3262         long decimals = -1;
3263         while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3264                 if (*cp == '.')
3265                         decimals = 0;
3266                 else if (decimals < scale) {
3267                         unsigned int value;
3268                         value = *cp - '0';
3269                         result = result * 10 + value;
3270                         if (decimals >= 0)
3271                                 decimals++;
3272                 }
3273                 cp++;
3274         }
3275         if (*cp == '\n')
3276                 cp++;
3277         if (*cp)
3278                 return -EINVAL;
3279         if (decimals < 0)
3280                 decimals = 0;
3281         while (decimals < scale) {
3282                 result *= 10;
3283                 decimals ++;
3284         }
3285         *res = result;
3286         return 0;
3287 }
3288
3289
3290 static void md_safemode_timeout(unsigned long data);
3291
3292 static ssize_t
3293 safe_delay_show(struct mddev *mddev, char *page)
3294 {
3295         int msec = (mddev->safemode_delay*1000)/HZ;
3296         return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3297 }
3298 static ssize_t
3299 safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
3300 {
3301         unsigned long msec;
3302
3303         if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3304                 return -EINVAL;
3305         if (msec == 0)
3306                 mddev->safemode_delay = 0;
3307         else {
3308                 unsigned long old_delay = mddev->safemode_delay;
3309                 mddev->safemode_delay = (msec*HZ)/1000;
3310                 if (mddev->safemode_delay == 0)
3311                         mddev->safemode_delay = 1;
3312                 if (mddev->safemode_delay < old_delay)
3313                         md_safemode_timeout((unsigned long)mddev);
3314         }
3315         return len;
3316 }
3317 static struct md_sysfs_entry md_safe_delay =
3318 __ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3319
3320 static ssize_t
3321 level_show(struct mddev *mddev, char *page)
3322 {
3323         struct md_personality *p = mddev->pers;
3324         if (p)
3325                 return sprintf(page, "%s\n", p->name);
3326         else if (mddev->clevel[0])
3327                 return sprintf(page, "%s\n", mddev->clevel);
3328         else if (mddev->level != LEVEL_NONE)
3329                 return sprintf(page, "%d\n", mddev->level);
3330         else
3331                 return 0;
3332 }
3333
3334 static ssize_t
3335 level_store(struct mddev *mddev, const char *buf, size_t len)
3336 {
3337         char clevel[16];
3338         ssize_t rv = len;
3339         struct md_personality *pers;
3340         long level;
3341         void *priv;
3342         struct md_rdev *rdev;
3343
3344         if (mddev->pers == NULL) {
3345                 if (len == 0)
3346                         return 0;