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/mutex.h>
  40#include <linux/buffer_head.h> /* for invalidate_bdev */
  41#include <linux/poll.h>
  42#include <linux/ctype.h>
  43#include <linux/string.h>
  44#include <linux/hdreg.h>
  45#include <linux/proc_fs.h>
  46#include <linux/random.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#define DEBUG 0
  58#define dprintk(x...) ((void)(DEBUG && printk(x)))
  59
  60#ifndef MODULE
  61static void autostart_arrays(int part);
  62#endif
  63
  64/* pers_list is a list of registered personalities protected
  65 * by pers_lock.
  66 * pers_lock does extra service to protect accesses to
  67 * mddev->thread when the mutex cannot be held.
  68 */
  69static LIST_HEAD(pers_list);
  70static DEFINE_SPINLOCK(pers_lock);
  71
  72static void md_print_devices(void);
  73
  74static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
  75static struct workqueue_struct *md_wq;
  76static struct workqueue_struct *md_misc_wq;
  77
  78#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
  79
  80/*
  81 * Default number of read corrections we'll attempt on an rdev
  82 * before ejecting it from the array. We divide the read error
  83 * count by 2 for every hour elapsed between read errors.
  84 */
  85#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
  86/*
  87 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  88 * is 1000 KB/sec, so the extra system load does not show up that much.
  89 * Increase it if you want to have more _guaranteed_ speed. Note that
  90 * the RAID driver will use the maximum available bandwidth if the IO
  91 * subsystem is idle. There is also an 'absolute maximum' reconstruction
  92 * speed limit - in case reconstruction slows down your system despite
  93 * idle IO detection.
  94 *
  95 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  96 * or /sys/block/mdX/md/sync_speed_{min,max}
  97 */
  98
  99static int sysctl_speed_limit_min = 1000;
 100static int sysctl_speed_limit_max = 200000;
 101static inline int speed_min(mddev_t *mddev)
 102{
 103	return mddev->sync_speed_min ?
 104		mddev->sync_speed_min : sysctl_speed_limit_min;
 105}
 106
 107static inline int speed_max(mddev_t *mddev)
 108{
 109	return mddev->sync_speed_max ?
 110		mddev->sync_speed_max : sysctl_speed_limit_max;
 111}
 112
 113static struct ctl_table_header *raid_table_header;
 114
 115static ctl_table raid_table[] = {
 116	{
 117		.procname	= "speed_limit_min",
 118		.data		= &sysctl_speed_limit_min,
 119		.maxlen		= sizeof(int),
 120		.mode		= S_IRUGO|S_IWUSR,
 121		.proc_handler	= proc_dointvec,
 122	},
 123	{
 124		.procname	= "speed_limit_max",
 125		.data		= &sysctl_speed_limit_max,
 126		.maxlen		= sizeof(int),
 127		.mode		= S_IRUGO|S_IWUSR,
 128		.proc_handler	= proc_dointvec,
 129	},
 130	{ }
 131};
 132
 133static ctl_table raid_dir_table[] = {
 134	{
 135		.procname	= "raid",
 136		.maxlen		= 0,
 137		.mode		= S_IRUGO|S_IXUGO,
 138		.child		= raid_table,
 139	},
 140	{ }
 141};
 142
 143static ctl_table raid_root_table[] = {
 144	{
 145		.procname	= "dev",
 146		.maxlen		= 0,
 147		.mode		= 0555,
 148		.child		= raid_dir_table,
 149	},
 150	{  }
 151};
 152
 153static const struct block_device_operations md_fops;
 154
 155static int start_readonly;
 156
 157/* bio_clone_mddev
 158 * like bio_clone, but with a local bio set
 159 */
 160
 161static void mddev_bio_destructor(struct bio *bio)
 162{
 163	mddev_t *mddev, **mddevp;
 164
 165	mddevp = (void*)bio;
 166	mddev = mddevp[-1];
 167
 168	bio_free(bio, mddev->bio_set);
 169}
 170
 171struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
 172			    mddev_t *mddev)
 173{
 174	struct bio *b;
 175	mddev_t **mddevp;
 176
 177	if (!mddev || !mddev->bio_set)
 178		return bio_alloc(gfp_mask, nr_iovecs);
 179
 180	b = bio_alloc_bioset(gfp_mask, nr_iovecs,
 181			     mddev->bio_set);
 182	if (!b)
 183		return NULL;
 184	mddevp = (void*)b;
 185	mddevp[-1] = mddev;
 186	b->bi_destructor = mddev_bio_destructor;
 187	return b;
 188}
 189EXPORT_SYMBOL_GPL(bio_alloc_mddev);
 190
 191struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
 192			    mddev_t *mddev)
 193{
 194	struct bio *b;
 195	mddev_t **mddevp;
 196
 197	if (!mddev || !mddev->bio_set)
 198		return bio_clone(bio, gfp_mask);
 199
 200	b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs,
 201			     mddev->bio_set);
 202	if (!b)
 203		return NULL;
 204	mddevp = (void*)b;
 205	mddevp[-1] = mddev;
 206	b->bi_destructor = mddev_bio_destructor;
 207	__bio_clone(b, bio);
 208	if (bio_integrity(bio)) {
 209		int ret;
 210
 211		ret = bio_integrity_clone(b, bio, gfp_mask, mddev->bio_set);
 212
 213		if (ret < 0) {
 214			bio_put(b);
 215			return NULL;
 216		}
 217	}
 218
 219	return b;
 220}
 221EXPORT_SYMBOL_GPL(bio_clone_mddev);
 222
 223void md_trim_bio(struct bio *bio, int offset, int size)
 224{
 225	/* 'bio' is a cloned bio which we need to trim to match
 226	 * the given offset and size.
 227	 * This requires adjusting bi_sector, bi_size, and bi_io_vec
 228	 */
 229	int i;
 230	struct bio_vec *bvec;
 231	int sofar = 0;
 232
 233	size <<= 9;
 234	if (offset == 0 && size == bio->bi_size)
 235		return;
 236
 237	bio->bi_sector += offset;
 238	bio->bi_size = size;
 239	offset <<= 9;
 240	clear_bit(BIO_SEG_VALID, &bio->bi_flags);
 241
 242	while (bio->bi_idx < bio->bi_vcnt &&
 243	       bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
 244		/* remove this whole bio_vec */
 245		offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
 246		bio->bi_idx++;
 247	}
 248	if (bio->bi_idx < bio->bi_vcnt) {
 249		bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
 250		bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
 251	}
 252	/* avoid any complications with bi_idx being non-zero*/
 253	if (bio->bi_idx) {
 254		memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
 255			(bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
 256		bio->bi_vcnt -= bio->bi_idx;
 257		bio->bi_idx = 0;
 258	}
 259	/* Make sure vcnt and last bv are not too big */
 260	bio_for_each_segment(bvec, bio, i) {
 261		if (sofar + bvec->bv_len > size)
 262			bvec->bv_len = size - sofar;
 263		if (bvec->bv_len == 0) {
 264			bio->bi_vcnt = i;
 265			break;
 266		}
 267		sofar += bvec->bv_len;
 268	}
 269}
 270EXPORT_SYMBOL_GPL(md_trim_bio);
 271
 272/*
 273 * We have a system wide 'event count' that is incremented
 274 * on any 'interesting' event, and readers of /proc/mdstat
 275 * can use 'poll' or 'select' to find out when the event
 276 * count increases.
 277 *
 278 * Events are:
 279 *  start array, stop array, error, add device, remove device,
 280 *  start build, activate spare
 281 */
 282static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 283static atomic_t md_event_count;
 284void md_new_event(mddev_t *mddev)
 285{
 286	atomic_inc(&md_event_count);
 287	wake_up(&md_event_waiters);
 288}
 289EXPORT_SYMBOL_GPL(md_new_event);
 290
 291/* Alternate version that can be called from interrupts
 292 * when calling sysfs_notify isn't needed.
 293 */
 294static void md_new_event_inintr(mddev_t *mddev)
 295{
 296	atomic_inc(&md_event_count);
 297	wake_up(&md_event_waiters);
 298}
 299
 300/*
 301 * Enables to iterate over all existing md arrays
 302 * all_mddevs_lock protects this list.
 303 */
 304static LIST_HEAD(all_mddevs);
 305static DEFINE_SPINLOCK(all_mddevs_lock);
 306
 307
 308/*
 309 * iterates through all used mddevs in the system.
 310 * We take care to grab the all_mddevs_lock whenever navigating
 311 * the list, and to always hold a refcount when unlocked.
 312 * Any code which breaks out of this loop while own
 313 * a reference to the current mddev and must mddev_put it.
 314 */
 315#define for_each_mddev(mddev,tmp)					\
 316									\
 317	for (({ spin_lock(&all_mddevs_lock); 				\
 318		tmp = all_mddevs.next;					\
 319		mddev = NULL;});					\
 320	     ({ if (tmp != &all_mddevs)					\
 321			mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 322		spin_unlock(&all_mddevs_lock);				\
 323		if (mddev) mddev_put(mddev);				\
 324		mddev = list_entry(tmp, mddev_t, all_mddevs);		\
 325		tmp != &all_mddevs;});					\
 326	     ({ spin_lock(&all_mddevs_lock);				\
 327		tmp = tmp->next;})					\
 328		)
 329
 330
 331/* Rather than calling directly into the personality make_request function,
 332 * IO requests come here first so that we can check if the device is
 333 * being suspended pending a reconfiguration.
 334 * We hold a refcount over the call to ->make_request.  By the time that
 335 * call has finished, the bio has been linked into some internal structure
 336 * and so is visible to ->quiesce(), so we don't need the refcount any more.
 337 */
 338static int md_make_request(struct request_queue *q, struct bio *bio)
 339{
 340	const int rw = bio_data_dir(bio);
 341	mddev_t *mddev = q->queuedata;
 342	int rv;
 343	int cpu;
 344	unsigned int sectors;
 345
 346	if (mddev == NULL || mddev->pers == NULL
 347	    || !mddev->ready) {
 348		bio_io_error(bio);
 349		return 0;
 350	}
 351	smp_rmb(); /* Ensure implications of  'active' are visible */
 352	rcu_read_lock();
 353	if (mddev->suspended) {
 354		DEFINE_WAIT(__wait);
 355		for (;;) {
 356			prepare_to_wait(&mddev->sb_wait, &__wait,
 357					TASK_UNINTERRUPTIBLE);
 358			if (!mddev->suspended)
 359				break;
 360			rcu_read_unlock();
 361			schedule();
 362			rcu_read_lock();
 363		}
 364		finish_wait(&mddev->sb_wait, &__wait);
 365	}
 366	atomic_inc(&mddev->active_io);
 367	rcu_read_unlock();
 368
 369	/*
 370	 * save the sectors now since our bio can
 371	 * go away inside make_request
 372	 */
 373	sectors = bio_sectors(bio);
 374	rv = mddev->pers->make_request(mddev, bio);
 375
 376	cpu = part_stat_lock();
 377	part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
 378	part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
 379	part_stat_unlock();
 380
 381	if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
 382		wake_up(&mddev->sb_wait);
 383
 384	return rv;
 385}
 386
 387/* mddev_suspend makes sure no new requests are submitted
 388 * to the device, and that any requests that have been submitted
 389 * are completely handled.
 390 * Once ->stop is called and completes, the module will be completely
 391 * unused.
 392 */
 393void mddev_suspend(mddev_t *mddev)
 394{
 395	BUG_ON(mddev->suspended);
 396	mddev->suspended = 1;
 397	synchronize_rcu();
 398	wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
 399	mddev->pers->quiesce(mddev, 1);
 
 
 400}
 401EXPORT_SYMBOL_GPL(mddev_suspend);
 402
 403void mddev_resume(mddev_t *mddev)
 404{
 405	mddev->suspended = 0;
 406	wake_up(&mddev->sb_wait);
 407	mddev->pers->quiesce(mddev, 0);
 408
 
 409	md_wakeup_thread(mddev->thread);
 410	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
 411}
 412EXPORT_SYMBOL_GPL(mddev_resume);
 413
 414int mddev_congested(mddev_t *mddev, int bits)
 415{
 416	return mddev->suspended;
 417}
 418EXPORT_SYMBOL(mddev_congested);
 419
 420/*
 421 * Generic flush handling for md
 422 */
 423
 424static void md_end_flush(struct bio *bio, int err)
 425{
 426	mdk_rdev_t *rdev = bio->bi_private;
 427	mddev_t *mddev = rdev->mddev;
 428
 429	rdev_dec_pending(rdev, mddev);
 430
 431	if (atomic_dec_and_test(&mddev->flush_pending)) {
 432		/* The pre-request flush has finished */
 433		queue_work(md_wq, &mddev->flush_work);
 434	}
 435	bio_put(bio);
 436}
 437
 438static void md_submit_flush_data(struct work_struct *ws);
 439
 440static void submit_flushes(struct work_struct *ws)
 441{
 442	mddev_t *mddev = container_of(ws, mddev_t, flush_work);
 443	mdk_rdev_t *rdev;
 444
 445	INIT_WORK(&mddev->flush_work, md_submit_flush_data);
 446	atomic_set(&mddev->flush_pending, 1);
 447	rcu_read_lock();
 448	list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
 449		if (rdev->raid_disk >= 0 &&
 450		    !test_bit(Faulty, &rdev->flags)) {
 451			/* Take two references, one is dropped
 452			 * when request finishes, one after
 453			 * we reclaim rcu_read_lock
 454			 */
 455			struct bio *bi;
 456			atomic_inc(&rdev->nr_pending);
 457			atomic_inc(&rdev->nr_pending);
 458			rcu_read_unlock();
 459			bi = bio_alloc_mddev(GFP_KERNEL, 0, mddev);
 460			bi->bi_end_io = md_end_flush;
 461			bi->bi_private = rdev;
 462			bi->bi_bdev = rdev->bdev;
 463			atomic_inc(&mddev->flush_pending);
 464			submit_bio(WRITE_FLUSH, bi);
 465			rcu_read_lock();
 466			rdev_dec_pending(rdev, mddev);
 467		}
 468	rcu_read_unlock();
 469	if (atomic_dec_and_test(&mddev->flush_pending))
 470		queue_work(md_wq, &mddev->flush_work);
 471}
 472
 473static void md_submit_flush_data(struct work_struct *ws)
 474{
 475	mddev_t *mddev = container_of(ws, mddev_t, flush_work);
 476	struct bio *bio = mddev->flush_bio;
 477
 478	if (bio->bi_size == 0)
 479		/* an empty barrier - all done */
 480		bio_endio(bio, 0);
 481	else {
 482		bio->bi_rw &= ~REQ_FLUSH;
 483		if (mddev->pers->make_request(mddev, bio))
 484			generic_make_request(bio);
 485	}
 486
 487	mddev->flush_bio = NULL;
 488	wake_up(&mddev->sb_wait);
 489}
 490
 491void md_flush_request(mddev_t *mddev, struct bio *bio)
 492{
 493	spin_lock_irq(&mddev->write_lock);
 494	wait_event_lock_irq(mddev->sb_wait,
 495			    !mddev->flush_bio,
 496			    mddev->write_lock, /*nothing*/);
 497	mddev->flush_bio = bio;
 498	spin_unlock_irq(&mddev->write_lock);
 499
 500	INIT_WORK(&mddev->flush_work, submit_flushes);
 501	queue_work(md_wq, &mddev->flush_work);
 502}
 503EXPORT_SYMBOL(md_flush_request);
 504
 505/* Support for plugging.
 506 * This mirrors the plugging support in request_queue, but does not
 507 * require having a whole queue or request structures.
 508 * We allocate an md_plug_cb for each md device and each thread it gets
 509 * plugged on.  This links tot the private plug_handle structure in the
 510 * personality data where we keep a count of the number of outstanding
 511 * plugs so other code can see if a plug is active.
 512 */
 513struct md_plug_cb {
 514	struct blk_plug_cb cb;
 515	mddev_t *mddev;
 516};
 517
 518static void plugger_unplug(struct blk_plug_cb *cb)
 519{
 520	struct md_plug_cb *mdcb = container_of(cb, struct md_plug_cb, cb);
 521	if (atomic_dec_and_test(&mdcb->mddev->plug_cnt))
 522		md_wakeup_thread(mdcb->mddev->thread);
 523	kfree(mdcb);
 524}
 525
 526/* Check that an unplug wakeup will come shortly.
 527 * If not, wakeup the md thread immediately
 528 */
 529int mddev_check_plugged(mddev_t *mddev)
 530{
 531	struct blk_plug *plug = current->plug;
 532	struct md_plug_cb *mdcb;
 533
 534	if (!plug)
 535		return 0;
 536
 537	list_for_each_entry(mdcb, &plug->cb_list, cb.list) {
 538		if (mdcb->cb.callback == plugger_unplug &&
 539		    mdcb->mddev == mddev) {
 540			/* Already on the list, move to top */
 541			if (mdcb != list_first_entry(&plug->cb_list,
 542						    struct md_plug_cb,
 543						    cb.list))
 544				list_move(&mdcb->cb.list, &plug->cb_list);
 545			return 1;
 546		}
 547	}
 548	/* Not currently on the callback list */
 549	mdcb = kmalloc(sizeof(*mdcb), GFP_ATOMIC);
 550	if (!mdcb)
 551		return 0;
 552
 553	mdcb->mddev = mddev;
 554	mdcb->cb.callback = plugger_unplug;
 555	atomic_inc(&mddev->plug_cnt);
 556	list_add(&mdcb->cb.list, &plug->cb_list);
 557	return 1;
 558}
 559EXPORT_SYMBOL_GPL(mddev_check_plugged);
 560
 561static inline mddev_t *mddev_get(mddev_t *mddev)
 562{
 563	atomic_inc(&mddev->active);
 564	return mddev;
 565}
 566
 567static void mddev_delayed_delete(struct work_struct *ws);
 568
 569static void mddev_put(mddev_t *mddev)
 570{
 571	struct bio_set *bs = NULL;
 572
 573	if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 574		return;
 575	if (!mddev->raid_disks && list_empty(&mddev->disks) &&
 576	    mddev->ctime == 0 && !mddev->hold_active) {
 577		/* Array is not configured at all, and not held active,
 578		 * so destroy it */
 579		list_del(&mddev->all_mddevs);
 580		bs = mddev->bio_set;
 581		mddev->bio_set = NULL;
 582		if (mddev->gendisk) {
 583			/* We did a probe so need to clean up.  Call
 584			 * queue_work inside the spinlock so that
 585			 * flush_workqueue() after mddev_find will
 586			 * succeed in waiting for the work to be done.
 587			 */
 588			INIT_WORK(&mddev->del_work, mddev_delayed_delete);
 589			queue_work(md_misc_wq, &mddev->del_work);
 590		} else
 591			kfree(mddev);
 592	}
 593	spin_unlock(&all_mddevs_lock);
 594	if (bs)
 595		bioset_free(bs);
 596}
 597
 598void mddev_init(mddev_t *mddev)
 599{
 600	mutex_init(&mddev->open_mutex);
 601	mutex_init(&mddev->reconfig_mutex);
 602	mutex_init(&mddev->bitmap_info.mutex);
 603	INIT_LIST_HEAD(&mddev->disks);
 604	INIT_LIST_HEAD(&mddev->all_mddevs);
 605	init_timer(&mddev->safemode_timer);
 606	atomic_set(&mddev->active, 1);
 607	atomic_set(&mddev->openers, 0);
 608	atomic_set(&mddev->active_io, 0);
 609	atomic_set(&mddev->plug_cnt, 0);
 610	spin_lock_init(&mddev->write_lock);
 611	atomic_set(&mddev->flush_pending, 0);
 612	init_waitqueue_head(&mddev->sb_wait);
 613	init_waitqueue_head(&mddev->recovery_wait);
 614	mddev->reshape_position = MaxSector;
 
 615	mddev->resync_min = 0;
 616	mddev->resync_max = MaxSector;
 617	mddev->level = LEVEL_NONE;
 618}
 619EXPORT_SYMBOL_GPL(mddev_init);
 620
 621static mddev_t * mddev_find(dev_t unit)
 622{
 623	mddev_t *mddev, *new = NULL;
 624
 625	if (unit && MAJOR(unit) != MD_MAJOR)
 626		unit &= ~((1<<MdpMinorShift)-1);
 627
 628 retry:
 629	spin_lock(&all_mddevs_lock);
 630
 631	if (unit) {
 632		list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 633			if (mddev->unit == unit) {
 634				mddev_get(mddev);
 635				spin_unlock(&all_mddevs_lock);
 636				kfree(new);
 637				return mddev;
 638			}
 639
 640		if (new) {
 641			list_add(&new->all_mddevs, &all_mddevs);
 642			spin_unlock(&all_mddevs_lock);
 643			new->hold_active = UNTIL_IOCTL;
 644			return new;
 645		}
 646	} else if (new) {
 647		/* find an unused unit number */
 648		static int next_minor = 512;
 649		int start = next_minor;
 650		int is_free = 0;
 651		int dev = 0;
 652		while (!is_free) {
 653			dev = MKDEV(MD_MAJOR, next_minor);
 654			next_minor++;
 655			if (next_minor > MINORMASK)
 656				next_minor = 0;
 657			if (next_minor == start) {
 658				/* Oh dear, all in use. */
 659				spin_unlock(&all_mddevs_lock);
 660				kfree(new);
 661				return NULL;
 662			}
 663				
 664			is_free = 1;
 665			list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 666				if (mddev->unit == dev) {
 667					is_free = 0;
 668					break;
 669				}
 670		}
 671		new->unit = dev;
 672		new->md_minor = MINOR(dev);
 673		new->hold_active = UNTIL_STOP;
 674		list_add(&new->all_mddevs, &all_mddevs);
 675		spin_unlock(&all_mddevs_lock);
 676		return new;
 677	}
 678	spin_unlock(&all_mddevs_lock);
 679
 680	new = kzalloc(sizeof(*new), GFP_KERNEL);
 681	if (!new)
 682		return NULL;
 683
 684	new->unit = unit;
 685	if (MAJOR(unit) == MD_MAJOR)
 686		new->md_minor = MINOR(unit);
 687	else
 688		new->md_minor = MINOR(unit) >> MdpMinorShift;
 689
 690	mddev_init(new);
 691
 692	goto retry;
 693}
 694
 695static inline int mddev_lock(mddev_t * mddev)
 696{
 697	return mutex_lock_interruptible(&mddev->reconfig_mutex);
 698}
 699
 700static inline int mddev_is_locked(mddev_t *mddev)
 701{
 702	return mutex_is_locked(&mddev->reconfig_mutex);
 703}
 704
 705static inline int mddev_trylock(mddev_t * mddev)
 706{
 707	return mutex_trylock(&mddev->reconfig_mutex);
 708}
 709
 710static struct attribute_group md_redundancy_group;
 711
 712static void mddev_unlock(mddev_t * mddev)
 713{
 714	if (mddev->to_remove) {
 715		/* These cannot be removed under reconfig_mutex as
 716		 * an access to the files will try to take reconfig_mutex
 717		 * while holding the file unremovable, which leads to
 718		 * a deadlock.
 719		 * So hold set sysfs_active while the remove in happeing,
 720		 * and anything else which might set ->to_remove or my
 721		 * otherwise change the sysfs namespace will fail with
 722		 * -EBUSY if sysfs_active is still set.
 723		 * We set sysfs_active under reconfig_mutex and elsewhere
 724		 * test it under the same mutex to ensure its correct value
 725		 * is seen.
 726		 */
 727		struct attribute_group *to_remove = mddev->to_remove;
 728		mddev->to_remove = NULL;
 729		mddev->sysfs_active = 1;
 730		mutex_unlock(&mddev->reconfig_mutex);
 731
 732		if (mddev->kobj.sd) {
 733			if (to_remove != &md_redundancy_group)
 734				sysfs_remove_group(&mddev->kobj, to_remove);
 735			if (mddev->pers == NULL ||
 736			    mddev->pers->sync_request == NULL) {
 737				sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
 738				if (mddev->sysfs_action)
 739					sysfs_put(mddev->sysfs_action);
 740				mddev->sysfs_action = NULL;
 741			}
 742		}
 743		mddev->sysfs_active = 0;
 744	} else
 745		mutex_unlock(&mddev->reconfig_mutex);
 746
 747	/* was we've dropped the mutex we need a spinlock to
 748	 * make sur the thread doesn't disappear
 749	 */
 750	spin_lock(&pers_lock);
 751	md_wakeup_thread(mddev->thread);
 752	spin_unlock(&pers_lock);
 753}
 754
 755static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 756{
 757	mdk_rdev_t *rdev;
 758
 759	list_for_each_entry(rdev, &mddev->disks, same_set)
 760		if (rdev->desc_nr == nr)
 761			return rdev;
 762
 763	return NULL;
 764}
 765
 766static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 767{
 768	mdk_rdev_t *rdev;
 769
 770	list_for_each_entry(rdev, &mddev->disks, same_set)
 771		if (rdev->bdev->bd_dev == dev)
 772			return rdev;
 773
 774	return NULL;
 775}
 776
 777static struct mdk_personality *find_pers(int level, char *clevel)
 778{
 779	struct mdk_personality *pers;
 780	list_for_each_entry(pers, &pers_list, list) {
 781		if (level != LEVEL_NONE && pers->level == level)
 782			return pers;
 783		if (strcmp(pers->name, clevel)==0)
 784			return pers;
 785	}
 786	return NULL;
 787}
 788
 789/* return the offset of the super block in 512byte sectors */
 790static inline sector_t calc_dev_sboffset(mdk_rdev_t *rdev)
 791{
 792	sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
 793	return MD_NEW_SIZE_SECTORS(num_sectors);
 794}
 795
 796static int alloc_disk_sb(mdk_rdev_t * rdev)
 797{
 798	if (rdev->sb_page)
 799		MD_BUG();
 800
 801	rdev->sb_page = alloc_page(GFP_KERNEL);
 802	if (!rdev->sb_page) {
 803		printk(KERN_ALERT "md: out of memory.\n");
 804		return -ENOMEM;
 805	}
 806
 807	return 0;
 808}
 809
 810static void free_disk_sb(mdk_rdev_t * rdev)
 811{
 812	if (rdev->sb_page) {
 813		put_page(rdev->sb_page);
 814		rdev->sb_loaded = 0;
 815		rdev->sb_page = NULL;
 816		rdev->sb_start = 0;
 817		rdev->sectors = 0;
 818	}
 819	if (rdev->bb_page) {
 820		put_page(rdev->bb_page);
 821		rdev->bb_page = NULL;
 822	}
 
 
 823}
 824
 825
 826static void super_written(struct bio *bio, int error)
 827{
 828	mdk_rdev_t *rdev = bio->bi_private;
 829	mddev_t *mddev = rdev->mddev;
 830
 831	if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
 832		printk("md: super_written gets error=%d, uptodate=%d\n",
 833		       error, test_bit(BIO_UPTODATE, &bio->bi_flags));
 834		WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
 835		md_error(mddev, rdev);
 836	}
 837
 838	if (atomic_dec_and_test(&mddev->pending_writes))
 839		wake_up(&mddev->sb_wait);
 840	bio_put(bio);
 841}
 842
 843void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 844		   sector_t sector, int size, struct page *page)
 845{
 846	/* write first size bytes of page to sector of rdev
 847	 * Increment mddev->pending_writes before returning
 848	 * and decrement it on completion, waking up sb_wait
 849	 * if zero is reached.
 850	 * If an error occurred, call md_error
 851	 */
 852	struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
 853
 854	bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
 855	bio->bi_sector = sector;
 856	bio_add_page(bio, page, size, 0);
 857	bio->bi_private = rdev;
 858	bio->bi_end_io = super_written;
 859
 860	atomic_inc(&mddev->pending_writes);
 861	submit_bio(WRITE_FLUSH_FUA, bio);
 862}
 863
 864void md_super_wait(mddev_t *mddev)
 865{
 866	/* wait for all superblock writes that were scheduled to complete */
 867	DEFINE_WAIT(wq);
 868	for(;;) {
 869		prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 870		if (atomic_read(&mddev->pending_writes)==0)
 871			break;
 872		schedule();
 873	}
 874	finish_wait(&mddev->sb_wait, &wq);
 875}
 876
 877static void bi_complete(struct bio *bio, int error)
 878{
 879	complete((struct completion*)bio->bi_private);
 880}
 881
 882int sync_page_io(mdk_rdev_t *rdev, sector_t sector, int size,
 883		 struct page *page, int rw, bool metadata_op)
 884{
 885	struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
 886	struct completion event;
 887	int ret;
 888
 889	rw |= REQ_SYNC;
 890
 891	bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
 892		rdev->meta_bdev : rdev->bdev;
 893	if (metadata_op)
 894		bio->bi_sector = sector + rdev->sb_start;
 
 
 
 
 895	else
 896		bio->bi_sector = sector + rdev->data_offset;
 897	bio_add_page(bio, page, size, 0);
 898	init_completion(&event);
 899	bio->bi_private = &event;
 900	bio->bi_end_io = bi_complete;
 901	submit_bio(rw, bio);
 902	wait_for_completion(&event);
 903
 904	ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 905	bio_put(bio);
 906	return ret;
 907}
 908EXPORT_SYMBOL_GPL(sync_page_io);
 909
 910static int read_disk_sb(mdk_rdev_t * rdev, int size)
 911{
 912	char b[BDEVNAME_SIZE];
 913	if (!rdev->sb_page) {
 914		MD_BUG();
 915		return -EINVAL;
 916	}
 917	if (rdev->sb_loaded)
 918		return 0;
 919
 920
 921	if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
 922		goto fail;
 923	rdev->sb_loaded = 1;
 924	return 0;
 925
 926fail:
 927	printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 928		bdevname(rdev->bdev,b));
 929	return -EINVAL;
 930}
 931
 932static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 933{
 934	return 	sb1->set_uuid0 == sb2->set_uuid0 &&
 935		sb1->set_uuid1 == sb2->set_uuid1 &&
 936		sb1->set_uuid2 == sb2->set_uuid2 &&
 937		sb1->set_uuid3 == sb2->set_uuid3;
 938}
 939
 940static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 941{
 942	int ret;
 943	mdp_super_t *tmp1, *tmp2;
 944
 945	tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 946	tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 947
 948	if (!tmp1 || !tmp2) {
 949		ret = 0;
 950		printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
 951		goto abort;
 952	}
 953
 954	*tmp1 = *sb1;
 955	*tmp2 = *sb2;
 956
 957	/*
 958	 * nr_disks is not constant
 959	 */
 960	tmp1->nr_disks = 0;
 961	tmp2->nr_disks = 0;
 962
 963	ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
 964abort:
 965	kfree(tmp1);
 966	kfree(tmp2);
 967	return ret;
 968}
 969
 970
 971static u32 md_csum_fold(u32 csum)
 972{
 973	csum = (csum & 0xffff) + (csum >> 16);
 974	return (csum & 0xffff) + (csum >> 16);
 975}
 976
 977static unsigned int calc_sb_csum(mdp_super_t * sb)
 978{
 979	u64 newcsum = 0;
 980	u32 *sb32 = (u32*)sb;
 981	int i;
 982	unsigned int disk_csum, csum;
 983
 984	disk_csum = sb->sb_csum;
 985	sb->sb_csum = 0;
 986
 987	for (i = 0; i < MD_SB_BYTES/4 ; i++)
 988		newcsum += sb32[i];
 989	csum = (newcsum & 0xffffffff) + (newcsum>>32);
 990
 991
 992#ifdef CONFIG_ALPHA
 993	/* This used to use csum_partial, which was wrong for several
 994	 * reasons including that different results are returned on
 995	 * different architectures.  It isn't critical that we get exactly
 996	 * the same return value as before (we always csum_fold before
 997	 * testing, and that removes any differences).  However as we
 998	 * know that csum_partial always returned a 16bit value on
 999	 * alphas, do a fold to maximise conformity to previous behaviour.
1000	 */
1001	sb->sb_csum = md_csum_fold(disk_csum);
1002#else
1003	sb->sb_csum = disk_csum;
1004#endif
1005	return csum;
1006}
1007
1008
1009/*
1010 * Handle superblock details.
1011 * We want to be able to handle multiple superblock formats
1012 * so we have a common interface to them all, and an array of
1013 * different handlers.
1014 * We rely on user-space to write the initial superblock, and support
1015 * reading and updating of superblocks.
1016 * Interface methods are:
1017 *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
1018 *      loads and validates a superblock on dev.
1019 *      if refdev != NULL, compare superblocks on both devices
1020 *    Return:
1021 *      0 - dev has a superblock that is compatible with refdev
1022 *      1 - dev has a superblock that is compatible and newer than refdev
1023 *          so dev should be used as the refdev in future
1024 *     -EINVAL superblock incompatible or invalid
1025 *     -othererror e.g. -EIO
1026 *
1027 *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
1028 *      Verify that dev is acceptable into mddev.
1029 *       The first time, mddev->raid_disks will be 0, and data from
1030 *       dev should be merged in.  Subsequent calls check that dev
1031 *       is new enough.  Return 0 or -EINVAL
1032 *
1033 *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
1034 *     Update the superblock for rdev with data in mddev
1035 *     This does not write to disc.
1036 *
1037 */
1038
1039struct super_type  {
1040	char		    *name;
1041	struct module	    *owner;
1042	int		    (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev,
 
1043					  int minor_version);
1044	int		    (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
1045	void		    (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
1046	unsigned long long  (*rdev_size_change)(mdk_rdev_t *rdev,
 
 
1047						sector_t num_sectors);
 
 
1048};
1049
1050/*
1051 * Check that the given mddev has no bitmap.
1052 *
1053 * This function is called from the run method of all personalities that do not
1054 * support bitmaps. It prints an error message and returns non-zero if mddev
1055 * has a bitmap. Otherwise, it returns 0.
1056 *
1057 */
1058int md_check_no_bitmap(mddev_t *mddev)
1059{
1060	if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
1061		return 0;
1062	printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1063		mdname(mddev), mddev->pers->name);
1064	return 1;
1065}
1066EXPORT_SYMBOL(md_check_no_bitmap);
1067
1068/*
1069 * load_super for 0.90.0 
1070 */
1071static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1072{
1073	char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1074	mdp_super_t *sb;
1075	int ret;
1076
1077	/*
1078	 * Calculate the position of the superblock (512byte sectors),
1079	 * it's at the end of the disk.
1080	 *
1081	 * It also happens to be a multiple of 4Kb.
1082	 */
1083	rdev->sb_start = calc_dev_sboffset(rdev);
1084
1085	ret = read_disk_sb(rdev, MD_SB_BYTES);
1086	if (ret) return ret;
1087
1088	ret = -EINVAL;
1089
1090	bdevname(rdev->bdev, b);
1091	sb = page_address(rdev->sb_page);
1092
1093	if (sb->md_magic != MD_SB_MAGIC) {
1094		printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1095		       b);
1096		goto abort;
1097	}
1098
1099	if (sb->major_version != 0 ||
1100	    sb->minor_version < 90 ||
1101	    sb->minor_version > 91) {
1102		printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1103			sb->major_version, sb->minor_version,
1104			b);
1105		goto abort;
1106	}
1107
1108	if (sb->raid_disks <= 0)
1109		goto abort;
1110
1111	if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
1112		printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1113			b);
1114		goto abort;
1115	}
1116
1117	rdev->preferred_minor = sb->md_minor;
1118	rdev->data_offset = 0;
 
1119	rdev->sb_size = MD_SB_BYTES;
1120	rdev->badblocks.shift = -1;
1121
1122	if (sb->level == LEVEL_MULTIPATH)
1123		rdev->desc_nr = -1;
1124	else
1125		rdev->desc_nr = sb->this_disk.number;
1126
1127	if (!refdev) {
1128		ret = 1;
1129	} else {
1130		__u64 ev1, ev2;
1131		mdp_super_t *refsb = page_address(refdev->sb_page);
1132		if (!uuid_equal(refsb, sb)) {
1133			printk(KERN_WARNING "md: %s has different UUID to %s\n",
1134				b, bdevname(refdev->bdev,b2));
1135			goto abort;
1136		}
1137		if (!sb_equal(refsb, sb)) {
1138			printk(KERN_WARNING "md: %s has same UUID"
1139			       " but different superblock to %s\n",
1140			       b, bdevname(refdev->bdev, b2));
1141			goto abort;
1142		}
1143		ev1 = md_event(sb);
1144		ev2 = md_event(refsb);
1145		if (ev1 > ev2)
1146			ret = 1;
1147		else 
1148			ret = 0;
1149	}
1150	rdev->sectors = rdev->sb_start;
1151	/* Limit to 4TB as metadata cannot record more than that */
1152	if (rdev->sectors >= (2ULL << 32))
 
 
 
1153		rdev->sectors = (2ULL << 32) - 2;
1154
1155	if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
1156		/* "this cannot possibly happen" ... */
1157		ret = -EINVAL;
1158
1159 abort:
1160	return ret;
1161}
1162
1163/*
1164 * validate_super for 0.90.0
1165 */
1166static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1167{
1168	mdp_disk_t *desc;
1169	mdp_super_t *sb = page_address(rdev->sb_page);
1170	__u64 ev1 = md_event(sb);
1171
1172	rdev->raid_disk = -1;
1173	clear_bit(Faulty, &rdev->flags);
1174	clear_bit(In_sync, &rdev->flags);
1175	clear_bit(WriteMostly, &rdev->flags);
1176
1177	if (mddev->raid_disks == 0) {
1178		mddev->major_version = 0;
1179		mddev->minor_version = sb->minor_version;
1180		mddev->patch_version = sb->patch_version;
1181		mddev->external = 0;
1182		mddev->chunk_sectors = sb->chunk_size >> 9;
1183		mddev->ctime = sb->ctime;
1184		mddev->utime = sb->utime;
1185		mddev->level = sb->level;
1186		mddev->clevel[0] = 0;
1187		mddev->layout = sb->layout;
1188		mddev->raid_disks = sb->raid_disks;
1189		mddev->dev_sectors = ((sector_t)sb->size) * 2;
1190		mddev->events = ev1;
1191		mddev->bitmap_info.offset = 0;
 
 
1192		mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
 
 
1193
1194		if (mddev->minor_version >= 91) {
1195			mddev->reshape_position = sb->reshape_position;
1196			mddev->delta_disks = sb->delta_disks;
1197			mddev->new_level = sb->new_level;
1198			mddev->new_layout = sb->new_layout;
1199			mddev->new_chunk_sectors = sb->new_chunk >> 9;
 
 
1200		} else {
1201			mddev->reshape_position = MaxSector;
1202			mddev->delta_disks = 0;
1203			mddev->new_level = mddev->level;
1204			mddev->new_layout = mddev->layout;
1205			mddev->new_chunk_sectors = mddev->chunk_sectors;
1206		}
1207
1208		if (sb->state & (1<<MD_SB_CLEAN))
1209			mddev->recovery_cp = MaxSector;
1210		else {
1211			if (sb->events_hi == sb->cp_events_hi && 
1212				sb->events_lo == sb->cp_events_lo) {
1213				mddev->recovery_cp = sb->recovery_cp;
1214			} else
1215				mddev->recovery_cp = 0;
1216		}
1217
1218		memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1219		memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1220		memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1221		memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1222
1223		mddev->max_disks = MD_SB_DISKS;
1224
1225		if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
1226		    mddev->bitmap_info.file == NULL)
1227			mddev->bitmap_info.offset =
1228				mddev->bitmap_info.default_offset;
 
 
 
1229
1230	} else if (mddev->pers == NULL) {
1231		/* Insist on good event counter while assembling, except
1232		 * for spares (which don't need an event count) */
1233		++ev1;
1234		if (sb->disks[rdev->desc_nr].state & (
1235			    (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1236			if (ev1 < mddev->events) 
1237				return -EINVAL;
1238	} else if (mddev->bitmap) {
1239		/* if adding to array with a bitmap, then we can accept an
1240		 * older device ... but not too old.
1241		 */
1242		if (ev1 < mddev->bitmap->events_cleared)
1243			return 0;
1244	} else {
1245		if (ev1 < mddev->events)
1246			/* just a hot-add of a new device, leave raid_disk at -1 */
1247			return 0;
1248	}
1249
1250	if (mddev->level != LEVEL_MULTIPATH) {
1251		desc = sb->disks + rdev->desc_nr;
1252
1253		if (desc->state & (1<<MD_DISK_FAULTY))
1254			set_bit(Faulty, &rdev->flags);
1255		else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1256			    desc->raid_disk < mddev->raid_disks */) {
1257			set_bit(In_sync, &rdev->flags);
1258			rdev->raid_disk = desc->raid_disk;
1259		} else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1260			/* active but not in sync implies recovery up to
1261			 * reshape position.  We don't know exactly where
1262			 * that is, so set to zero for now */
1263			if (mddev->minor_version >= 91) {
1264				rdev->recovery_offset = 0;
1265				rdev->raid_disk = desc->raid_disk;
1266			}
1267		}
1268		if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1269			set_bit(WriteMostly, &rdev->flags);
1270	} else /* MULTIPATH are always insync */
1271		set_bit(In_sync, &rdev->flags);
1272	return 0;
1273}
1274
1275/*
1276 * sync_super for 0.90.0
1277 */
1278static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1279{
1280	mdp_super_t *sb;
1281	mdk_rdev_t *rdev2;
1282	int next_spare = mddev->raid_disks;
1283
1284
1285	/* make rdev->sb match mddev data..
1286	 *
1287	 * 1/ zero out disks
1288	 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1289	 * 3/ any empty disks < next_spare become removed
1290	 *
1291	 * disks[0] gets initialised to REMOVED because
1292	 * we cannot be sure from other fields if it has
1293	 * been initialised or not.
1294	 */
1295	int i;
1296	int active=0, working=0,failed=0,spare=0,nr_disks=0;
1297
1298	rdev->sb_size = MD_SB_BYTES;
1299
1300	sb = page_address(rdev->sb_page);
1301
1302	memset(sb, 0, sizeof(*sb));
1303
1304	sb->md_magic = MD_SB_MAGIC;
1305	sb->major_version = mddev->major_version;
1306	sb->patch_version = mddev->patch_version;
1307	sb->gvalid_words  = 0; /* ignored */
1308	memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1309	memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1310	memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1311	memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1312
1313	sb->ctime = mddev->ctime;
1314	sb->level = mddev->level;
1315	sb->size = mddev->dev_sectors / 2;
1316	sb->raid_disks = mddev->raid_disks;
1317	sb->md_minor = mddev->md_minor;
1318	sb->not_persistent = 0;
1319	sb->utime = mddev->utime;
1320	sb->state = 0;
1321	sb->events_hi = (mddev->events>>32);
1322	sb->events_lo = (u32)mddev->events;
1323
1324	if (mddev->reshape_position == MaxSector)
1325		sb->minor_version = 90;
1326	else {
1327		sb->minor_version = 91;
1328		sb->reshape_position = mddev->reshape_position;
1329		sb->new_level = mddev->new_level;
1330		sb->delta_disks = mddev->delta_disks;
1331		sb->new_layout = mddev->new_layout;
1332		sb->new_chunk = mddev->new_chunk_sectors << 9;
1333	}
1334	mddev->minor_version = sb->minor_version;
1335	if (mddev->in_sync)
1336	{
1337		sb->recovery_cp = mddev->recovery_cp;
1338		sb->cp_events_hi = (mddev->events>>32);
1339		sb->cp_events_lo = (u32)mddev->events;
1340		if (mddev->recovery_cp == MaxSector)
1341			sb->state = (1<< MD_SB_CLEAN);
1342	} else
1343		sb->recovery_cp = 0;
1344
1345	sb->layout = mddev->layout;
1346	sb->chunk_size = mddev->chunk_sectors << 9;
1347
1348	if (mddev->bitmap && mddev->bitmap_info.file == NULL)
1349		sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1350
1351	sb->disks[0].state = (1<<MD_DISK_REMOVED);
1352	list_for_each_entry(rdev2, &mddev->disks, same_set) {
1353		mdp_disk_t *d;
1354		int desc_nr;
1355		int is_active = test_bit(In_sync, &rdev2->flags);
1356
1357		if (rdev2->raid_disk >= 0 &&
1358		    sb->minor_version >= 91)
1359			/* we have nowhere to store the recovery_offset,
1360			 * but if it is not below the reshape_position,
1361			 * we can piggy-back on that.
1362			 */
1363			is_active = 1;
1364		if (rdev2->raid_disk < 0 ||
1365		    test_bit(Faulty, &rdev2->flags))
1366			is_active = 0;
1367		if (is_active)
1368			desc_nr = rdev2->raid_disk;
1369		else
1370			desc_nr = next_spare++;
1371		rdev2->desc_nr = desc_nr;
1372		d = &sb->disks[rdev2->desc_nr];
1373		nr_disks++;
1374		d->number = rdev2->desc_nr;
1375		d->major = MAJOR(rdev2->bdev->bd_dev);
1376		d->minor = MINOR(rdev2->bdev->bd_dev);
1377		if (is_active)
1378			d->raid_disk = rdev2->raid_disk;
1379		else
1380			d->raid_disk = rdev2->desc_nr; /* compatibility */
1381		if (test_bit(Faulty, &rdev2->flags))
1382			d->state = (1<<MD_DISK_FAULTY);
1383		else if (is_active) {
1384			d->state = (1<<MD_DISK_ACTIVE);
1385			if (test_bit(In_sync, &rdev2->flags))
1386				d->state |= (1<<MD_DISK_SYNC);
1387			active++;
1388			working++;
1389		} else {
1390			d->state = 0;
1391			spare++;
1392			working++;
1393		}
1394		if (test_bit(WriteMostly, &rdev2->flags))
1395			d->state |= (1<<MD_DISK_WRITEMOSTLY);
1396	}
1397	/* now set the "removed" and "faulty" bits on any missing devices */
1398	for (i=0 ; i < mddev->raid_disks ; i++) {
1399		mdp_disk_t *d = &sb->disks[i];
1400		if (d->state == 0 && d->number == 0) {
1401			d->number = i;
1402			d->raid_disk = i;
1403			d->state = (1<<MD_DISK_REMOVED);
1404			d->state |= (1<<MD_DISK_FAULTY);
1405			failed++;
1406		}
1407	}
1408	sb->nr_disks = nr_disks;
1409	sb->active_disks = active;
1410	sb->working_disks = working;
1411	sb->failed_disks = failed;
1412	sb->spare_disks = spare;
1413
1414	sb->this_disk = sb->disks[rdev->desc_nr];
1415	sb->sb_csum = calc_sb_csum(sb);
1416}
1417
1418/*
1419 * rdev_size_change for 0.90.0
1420 */
1421static unsigned long long
1422super_90_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1423{
1424	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1425		return 0; /* component must fit device */
1426	if (rdev->mddev->bitmap_info.offset)
1427		return 0; /* can't move bitmap */
1428	rdev->sb_start = calc_dev_sboffset(rdev);
1429	if (!num_sectors || num_sectors > rdev->sb_start)
1430		num_sectors = rdev->sb_start;
1431	/* Limit to 4TB as metadata cannot record more than that.
1432	 * 4TB == 2^32 KB, or 2*2^32 sectors.
1433	 */
1434	if (num_sectors >= (2ULL << 32))
1435		num_sectors = (2ULL << 32) - 2;
1436	md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1437		       rdev->sb_page);
1438	md_super_wait(rdev->mddev);
1439	return num_sectors;
1440}
1441
 
 
 
 
 
 
1442
1443/*
1444 * version 1 superblock
1445 */
1446
1447static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
1448{
1449	__le32 disk_csum;
1450	u32 csum;
1451	unsigned long long newcsum;
1452	int size = 256 + le32_to_cpu(sb->max_dev)*2;
1453	__le32 *isuper = (__le32*)sb;
1454	int i;
1455
1456	disk_csum = sb->sb_csum;
1457	sb->sb_csum = 0;
1458	newcsum = 0;
1459	for (i=0; size>=4; size -= 4 )
1460		newcsum += le32_to_cpu(*isuper++);
1461
1462	if (size == 2)
1463		newcsum += le16_to_cpu(*(__le16*) isuper);
1464
1465	csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1466	sb->sb_csum = disk_csum;
1467	return cpu_to_le32(csum);
1468}
1469
1470static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1471			    int acknowledged);
1472static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
1473{
1474	struct mdp_superblock_1 *sb;
1475	int ret;
1476	sector_t sb_start;
 
1477	char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1478	int bmask;
1479
1480	/*
1481	 * Calculate the position of the superblock in 512byte sectors.
1482	 * It is always aligned to a 4K boundary and
1483	 * depeding on minor_version, it can be:
1484	 * 0: At least 8K, but less than 12K, from end of device
1485	 * 1: At start of device
1486	 * 2: 4K from start of device.
1487	 */
1488	switch(minor_version) {
1489	case 0:
1490		sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
1491		sb_start -= 8*2;
1492		sb_start &= ~(sector_t)(4*2-1);
1493		break;
1494	case 1:
1495		sb_start = 0;
1496		break;
1497	case 2:
1498		sb_start = 8;
1499		break;
1500	default:
1501		return -EINVAL;
1502	}
1503	rdev->sb_start = sb_start;
1504
1505	/* superblock is rarely larger than 1K, but it can be larger,
1506	 * and it is safe to read 4k, so we do that
1507	 */
1508	ret = read_disk_sb(rdev, 4096);
1509	if (ret) return ret;
1510
1511
1512	sb = page_address(rdev->sb_page);
1513
1514	if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1515	    sb->major_version != cpu_to_le32(1) ||
1516	    le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1517	    le64_to_cpu(sb->super_offset) != rdev->sb_start ||
1518	    (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1519		return -EINVAL;
1520
1521	if (calc_sb_1_csum(sb) != sb->sb_csum) {
1522		printk("md: invalid superblock checksum on %s\n",
1523			bdevname(rdev->bdev,b));
1524		return -EINVAL;
1525	}
1526	if (le64_to_cpu(sb->data_size) < 10) {
1527		printk("md: data_size too small on %s\n",
1528		       bdevname(rdev->bdev,b));
1529		return -EINVAL;
1530	}
 
 
 
 
 
1531
1532	rdev->preferred_minor = 0xffff;
1533	rdev->data_offset = le64_to_cpu(sb->data_offset);
 
 
 
 
1534	atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1535
1536	rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1537	bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1538	if (rdev->sb_size & bmask)
1539		rdev->sb_size = (rdev->sb_size | bmask) + 1;
1540
1541	if (minor_version
1542	    && rdev->data_offset < sb_start + (rdev->sb_size/512))
1543		return -EINVAL;
 
 
 
1544
1545	if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1546		rdev->desc_nr = -1;
1547	else
1548		rdev->desc_nr = le32_to_cpu(sb->dev_number);
1549
1550	if (!rdev->bb_page) {
1551		rdev->bb_page = alloc_page(GFP_KERNEL);
1552		if (!rdev->bb_page)
1553			return -ENOMEM;
1554	}
1555	if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1556	    rdev->badblocks.count == 0) {
1557		/* need to load the bad block list.
1558		 * Currently we limit it to one page.
1559		 */
1560		s32 offset;
1561		sector_t bb_sector;
1562		u64 *bbp;
1563		int i;
1564		int sectors = le16_to_cpu(sb->bblog_size);
1565		if (sectors > (PAGE_SIZE / 512))
1566			return -EINVAL;
1567		offset = le32_to_cpu(sb->bblog_offset);
1568		if (offset == 0)
1569			return -EINVAL;
1570		bb_sector = (long long)offset;
1571		if (!sync_page_io(rdev, bb_sector, sectors << 9,
1572				  rdev->bb_page, READ, true))
1573			return -EIO;
1574		bbp = (u64 *)page_address(rdev->bb_page);
1575		rdev->badblocks.shift = sb->bblog_shift;
1576		for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1577			u64 bb = le64_to_cpu(*bbp);
1578			int count = bb & (0x3ff);
1579			u64 sector = bb >> 10;
1580			sector <<= sb->bblog_shift;
1581			count <<= sb->bblog_shift;
1582			if (bb + 1 == 0)
1583				break;
1584			if (md_set_badblocks(&rdev->badblocks,
1585					     sector, count, 1) == 0)
1586				return -EINVAL;
1587		}
1588	} else if (sb->bblog_offset == 0)
1589		rdev->badblocks.shift = -1;
1590
1591	if (!refdev) {
1592		ret = 1;
1593	} else {
1594		__u64 ev1, ev2;
1595		struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
1596
1597		if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1598		    sb->level != refsb->level ||
1599		    sb->layout != refsb->layout ||
1600		    sb->chunksize != refsb->chunksize) {
1601			printk(KERN_WARNING "md: %s has strangely different"
1602				" superblock to %s\n",
1603				bdevname(rdev->bdev,b),
1604				bdevname(refdev->bdev,b2));
1605			return -EINVAL;
1606		}
1607		ev1 = le64_to_cpu(sb->events);
1608		ev2 = le64_to_cpu(refsb->events);
1609
1610		if (ev1 > ev2)
1611			ret = 1;
1612		else
1613			ret = 0;
1614	}
1615	if (minor_version)
1616		rdev->sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
1617			le64_to_cpu(sb->data_offset);
1618	else
1619		rdev->sectors = rdev->sb_start;
1620	if (rdev->sectors < le64_to_cpu(sb->data_size))
1621		return -EINVAL;
1622	rdev->sectors = le64_to_cpu(sb->data_size);
1623	if (le64_to_cpu(sb->size) > rdev->sectors)
1624		return -EINVAL;
1625	return ret;
1626}
1627
1628static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1629{
1630	struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
1631	__u64 ev1 = le64_to_cpu(sb->events);
1632
1633	rdev->raid_disk = -1;
1634	clear_bit(Faulty, &rdev->flags);
1635	clear_bit(In_sync, &rdev->flags);
1636	clear_bit(WriteMostly, &rdev->flags);
1637
1638	if (mddev->raid_disks == 0) {
1639		mddev->major_version = 1;
1640		mddev->patch_version = 0;
1641		mddev->external = 0;
1642		mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
1643		mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1644		mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1645		mddev->level = le32_to_cpu(sb->level);
1646		mddev->clevel[0] = 0;
1647		mddev->layout = le32_to_cpu(sb->layout);
1648		mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1649		mddev->dev_sectors = le64_to_cpu(sb->size);
1650		mddev->events = ev1;
1651		mddev->bitmap_info.offset = 0;
 
 
 
 
1652		mddev->bitmap_info.default_offset = 1024 >> 9;
1653		
 
 
1654		mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1655		memcpy(mddev->uuid, sb->set_uuid, 16);
1656
1657		mddev->max_disks =  (4096-256)/2;
1658
1659		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1660		    mddev->bitmap_info.file == NULL )
1661			mddev->bitmap_info.offset =
1662				(__s32)le32_to_cpu(sb->bitmap_offset);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1663
1664		if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1665			mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1666			mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1667			mddev->new_level = le32_to_cpu(sb->new_level);
1668			mddev->new_layout = le32_to_cpu(sb->new_layout);
1669			mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
 
 
 
 
 
1670		} else {
1671			mddev->reshape_position = MaxSector;
1672			mddev->delta_disks = 0;
1673			mddev->new_level = mddev->level;
1674			mddev->new_layout = mddev->layout;
1675			mddev->new_chunk_sectors = mddev->chunk_sectors;
1676		}
1677
1678	} else if (mddev->pers == NULL) {
1679		/* Insist of good event counter while assembling, except for
1680		 * spares (which don't need an event count) */
1681		++ev1;
1682		if (rdev->desc_nr >= 0 &&
1683		    rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1684		    le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1685			if (ev1 < mddev->events)
1686				return -EINVAL;
1687	} else if (mddev->bitmap) {
1688		/* If adding to array with a bitmap, then we can accept an
1689		 * older device, but not too old.
1690		 */
1691		if (ev1 < mddev->bitmap->events_cleared)
1692			return 0;
1693	} else {
1694		if (ev1 < mddev->events)
1695			/* just a hot-add of a new device, leave raid_disk at -1 */
1696			return 0;
1697	}
1698	if (mddev->level != LEVEL_MULTIPATH) {
1699		int role;
1700		if (rdev->desc_nr < 0 ||
1701		    rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1702			role = 0xffff;
1703			rdev->desc_nr = -1;
1704		} else
1705			role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1706		switch(role) {
1707		case 0xffff: /* spare */
1708			break;
1709		case 0xfffe: /* faulty */
1710			set_bit(Faulty, &rdev->flags);
1711			break;
1712		default:
1713			if ((le32_to_cpu(sb->feature_map) &
1714			     MD_FEATURE_RECOVERY_OFFSET))
1715				rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1716			else
1717				set_bit(In_sync, &rdev->flags);
1718			rdev->raid_disk = role;
1719			break;
1720		}
1721		if (sb->devflags & WriteMostly1)
1722			set_bit(WriteMostly, &rdev->flags);
 
 
1723	} else /* MULTIPATH are always insync */
1724		set_bit(In_sync, &rdev->flags);
1725
1726	return 0;
1727}
1728
1729static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1730{
1731	struct mdp_superblock_1 *sb;
1732	mdk_rdev_t *rdev2;
1733	int max_dev, i;
1734	/* make rdev->sb match mddev and rdev data. */
1735
1736	sb = page_address(rdev->sb_page);
1737
1738	sb->feature_map = 0;
1739	sb->pad0 = 0;
1740	sb->recovery_offset = cpu_to_le64(0);
1741	memset(sb->pad1, 0, sizeof(sb->pad1));
1742	memset(sb->pad3, 0, sizeof(sb->pad3));
1743
1744	sb->utime = cpu_to_le64((__u64)mddev->utime);
1745	sb->events = cpu_to_le64(mddev->events);
1746	if (mddev->in_sync)
1747		sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1748	else
1749		sb->resync_offset = cpu_to_le64(0);
1750
1751	sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
1752
1753	sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1754	sb->size = cpu_to_le64(mddev->dev_sectors);
1755	sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
1756	sb->level = cpu_to_le32(mddev->level);
1757	sb->layout = cpu_to_le32(mddev->layout);
1758
1759	if (test_bit(WriteMostly, &rdev->flags))
1760		sb->devflags |= WriteMostly1;
1761	else
1762		sb->devflags &= ~WriteMostly1;
 
 
1763
1764	if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1765		sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
1766		sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1767	}
1768
1769	if (rdev->raid_disk >= 0 &&
1770	    !test_bit(In_sync, &rdev->flags)) {
1771		sb->feature_map |=
1772			cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1773		sb->recovery_offset =
1774			cpu_to_le64(rdev->recovery_offset);
1775	}
 
 
 
1776
1777	if (mddev->reshape_position != MaxSector) {
1778		sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1779		sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1780		sb->new_layout = cpu_to_le32(mddev->new_layout);
1781		sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1782		sb->new_level = cpu_to_le32(mddev->new_level);
1783		sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
 
 
 
 
 
 
 
 
 
 
1784	}
1785
1786	if (rdev->badblocks.count == 0)
1787		/* Nothing to do for bad blocks*/ ;
1788	else if (sb->bblog_offset == 0)
1789		/* Cannot record bad blocks on this device */
1790		md_error(mddev, rdev);
1791	else {
1792		struct badblocks *bb = &rdev->badblocks;
1793		u64 *bbp = (u64 *)page_address(rdev->bb_page);
1794		u64 *p = bb->page;
1795		sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1796		if (bb->changed) {
1797			unsigned seq;
1798
1799retry:
1800			seq = read_seqbegin(&bb->lock);
1801
1802			memset(bbp, 0xff, PAGE_SIZE);
1803
1804			for (i = 0 ; i < bb->count ; i++) {
1805				u64 internal_bb = *p++;
1806				u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1807						| BB_LEN(internal_bb));
1808				*bbp++ = cpu_to_le64(store_bb);
1809			}
 
1810			if (read_seqretry(&bb->lock, seq))
1811				goto retry;
1812
1813			bb->sector = (rdev->sb_start +
1814				      (int)le32_to_cpu(sb->bblog_offset));
1815			bb->size = le16_to_cpu(sb->bblog_size);
1816			bb->changed = 0;
1817		}
1818	}
1819
1820	max_dev = 0;
1821	list_for_each_entry(rdev2, &mddev->disks, same_set)
1822		if (rdev2->desc_nr+1 > max_dev)
1823			max_dev = rdev2->desc_nr+1;
1824
1825	if (max_dev > le32_to_cpu(sb->max_dev)) {
1826		int bmask;
1827		sb->max_dev = cpu_to_le32(max_dev);
1828		rdev->sb_size = max_dev * 2 + 256;
1829		bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1830		if (rdev->sb_size & bmask)
1831			rdev->sb_size = (rdev->sb_size | bmask) + 1;
1832	} else
1833		max_dev = le32_to_cpu(sb->max_dev);
1834
1835	for (i=0; i<max_dev;i++)
1836		sb->dev_roles[i] = cpu_to_le16(0xfffe);
1837	
1838	list_for_each_entry(rdev2, &mddev->disks, same_set) {
1839		i = rdev2->desc_nr;
1840		if (test_bit(Faulty, &rdev2->flags))
1841			sb->dev_roles[i] = cpu_to_le16(0xfffe);
1842		else if (test_bit(In_sync, &rdev2->flags))
1843			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1844		else if (rdev2->raid_disk >= 0)
1845			sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1846		else
1847			sb->dev_roles[i] = cpu_to_le16(0xffff);
1848	}
1849
1850	sb->sb_csum = calc_sb_1_csum(sb);
1851}
1852
1853static unsigned long long
1854super_1_rdev_size_change(mdk_rdev_t *rdev, sector_t num_sectors)
1855{
1856	struct mdp_superblock_1 *sb;
1857	sector_t max_sectors;
1858	if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
1859		return 0; /* component must fit device */
 
 
1860	if (rdev->sb_start < rdev->data_offset) {
1861		/* minor versions 1 and 2; superblock before data */
1862		max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
1863		max_sectors -= rdev->data_offset;
1864		if (!num_sectors || num_sectors > max_sectors)
1865			num_sectors = max_sectors;
1866	} else if (rdev->mddev->bitmap_info.offset) {
1867		/* minor version 0 with bitmap we can't move */
1868		return 0;
1869	} else {
1870		/* minor version 0; superblock after data */
1871		sector_t sb_start;
1872		sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
1873		sb_start &= ~(sector_t)(4*2 - 1);
1874		max_sectors = rdev->sectors + sb_start - rdev->sb_start;
1875		if (!num_sectors || num_sectors > max_sectors)
1876			num_sectors = max_sectors;
1877		rdev->sb_start = sb_start;
1878	}
1879	sb = page_address(rdev->sb_page);
1880	sb->data_size = cpu_to_le64(num_sectors);
1881	sb->super_offset = rdev->sb_start;
1882	sb->sb_csum = calc_sb_1_csum(sb);
1883	md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
1884		       rdev->sb_page);
1885	md_super_wait(rdev->mddev);
1886	return num_sectors;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1887}
1888
1889static struct super_type super_types[] = {
1890	[0] = {
1891		.name	= "0.90.0",
1892		.owner	= THIS_MODULE,
1893		.load_super	    = super_90_load,
1894		.validate_super	    = super_90_validate,
1895		.sync_super	    = super_90_sync,
1896		.rdev_size_change   = super_90_rdev_size_change,
 
1897	},
1898	[1] = {
1899		.name	= "md-1",
1900		.owner	= THIS_MODULE,
1901		.load_super	    = super_1_load,
1902		.validate_super	    = super_1_validate,
1903		.sync_super	    = super_1_sync,
1904		.rdev_size_change   = super_1_rdev_size_change,
 
1905	},
1906};
1907
1908static void sync_super(mddev_t *mddev, mdk_rdev_t *rdev)
1909{
1910	if (mddev->sync_super) {
1911		mddev->sync_super(mddev, rdev);
1912		return;
1913	}
1914
1915	BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1916
1917	super_types[mddev->major_version].sync_super(mddev, rdev);
1918}
1919
1920static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1921{
1922	mdk_rdev_t *rdev, *rdev2;
1923
1924	rcu_read_lock();
1925	rdev_for_each_rcu(rdev, mddev1)
1926		rdev_for_each_rcu(rdev2, mddev2)
1927			if (rdev->bdev->bd_contains ==
1928			    rdev2->bdev->bd_contains) {
1929				rcu_read_unlock();
1930				return 1;
1931			}
1932	rcu_read_unlock();
1933	return 0;
1934}
1935
1936static LIST_HEAD(pending_raid_disks);
1937
1938/*
1939 * Try to register data integrity profile for an mddev
1940 *
1941 * This is called when an array is started and after a disk has been kicked
1942 * from the array. It only succeeds if all working and active component devices
1943 * are integrity capable with matching profiles.
1944 */
1945int md_integrity_register(mddev_t *mddev)
1946{
1947	mdk_rdev_t *rdev, *reference = NULL;
1948
1949	if (list_empty(&mddev->disks))
1950		return 0; /* nothing to do */
1951	if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1952		return 0; /* shouldn't register, or already is */
1953	list_for_each_entry(rdev, &mddev->disks, same_set) {
1954		/* skip spares and non-functional disks */
1955		if (test_bit(Faulty, &rdev->flags))
1956			continue;
1957		if (rdev->raid_disk < 0)
1958			continue;
1959		if (!reference) {
1960			/* Use the first rdev as the reference */
1961			reference = rdev;
1962			continue;
1963		}
1964		/* does this rdev's profile match the reference profile? */
1965		if (blk_integrity_compare(reference->bdev->bd_disk,
1966				rdev->bdev->bd_disk) < 0)
1967			return -EINVAL;
1968	}
1969	if (!reference || !bdev_get_integrity(reference->bdev))
1970		return 0;
1971	/*
1972	 * All component devices are integrity capable and have matching
1973	 * profiles, register the common profile for the md device.
1974	 */
1975	if (blk_integrity_register(mddev->gendisk,
1976			bdev_get_integrity(reference->bdev)) != 0) {
1977		printk(KERN_ERR "md: failed to register integrity for %s\n",
1978			mdname(mddev));
1979		return -EINVAL;
1980	}
1981	printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1982	if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1983		printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1984		       mdname(mddev));
1985		return -EINVAL;
1986	}
1987	return 0;
1988}
1989EXPORT_SYMBOL(md_integrity_register);
1990
1991/* Disable data integrity if non-capable/non-matching disk is being added */
1992void md_integrity_add_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
1993{
1994	struct blk_integrity *bi_rdev = bdev_get_integrity(rdev->bdev);
1995	struct blk_integrity *bi_mddev = blk_get_integrity(mddev->gendisk);
1996
1997	if (!bi_mddev) /* nothing to do */
1998		return;
1999	if (rdev->raid_disk < 0) /* skip spares */
2000		return;
2001	if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2002					     rdev->bdev->bd_disk) >= 0)
2003		return;
2004	printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2005	blk_integrity_unregister(mddev->gendisk);
2006}
2007EXPORT_SYMBOL(md_integrity_add_rdev);
2008
2009static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
2010{
2011	char b[BDEVNAME_SIZE];
2012	struct kobject *ko;
2013	char *s;
2014	int err;
2015
2016	if (rdev->mddev) {
2017		MD_BUG();
2018		return -EINVAL;
2019	}
2020
2021	/* prevent duplicates */
2022	if (find_rdev(mddev, rdev->bdev->bd_dev))
2023		return -EEXIST;
2024
2025	/* make sure rdev->sectors exceeds mddev->dev_sectors */
2026	if (rdev->sectors && (mddev->dev_sectors == 0 ||
2027			rdev->sectors < mddev->dev_sectors)) {
2028		if (mddev->pers) {
2029			/* Cannot change size, so fail
2030			 * If mddev->level <= 0, then we don't care
2031			 * about aligning sizes (e.g. linear)
2032			 */
2033			if (mddev->level > 0)
2034				return -ENOSPC;
2035		} else
2036			mddev->dev_sectors = rdev->sectors;
2037	}
2038
2039	/* Verify rdev->desc_nr is unique.
2040	 * If it is -1, assign a free number, else
2041	 * check number is not in use
2042	 */
2043	if (rdev->desc_nr < 0) {
2044		int choice = 0;
2045		if (mddev->pers) choice = mddev->raid_disks;
2046		while (find_rdev_nr(mddev, choice))
2047			choice++;
2048		rdev->desc_nr = choice;
2049	} else {
2050		if (find_rdev_nr(mddev, rdev->desc_nr))
2051			return -EBUSY;
2052	}
2053	if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2054		printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2055		       mdname(mddev), mddev->max_disks);
2056		return -EBUSY;
2057	}
2058	bdevname(rdev->bdev,b);
2059	while ( (s=strchr(b, '/')) != NULL)
2060		*s = '!';
2061
2062	rdev->mddev = mddev;
2063	printk(KERN_INFO "md: bind<%s>\n", b);
2064
2065	if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
2066		goto fail;
2067
2068	ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
2069	if (sysfs_create_link(&rdev->kobj, ko, "block"))
2070		/* failure here is OK */;
2071	rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
2072
2073	list_add_rcu(&rdev->same_set, &mddev->disks);
2074	bd_link_disk_holder(rdev->bdev, mddev->gendisk);
2075
2076	/* May as well allow recovery to be retried once */
2077	mddev->recovery_disabled++;
2078
2079	return 0;
2080
2081 fail:
2082	printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2083	       b, mdname(mddev));
2084	return err;
2085}
2086
2087static void md_delayed_delete(struct work_struct *ws)
2088{
2089	mdk_rdev_t *rdev = container_of(ws, mdk_rdev_t, del_work);
2090	kobject_del(&rdev->kobj);
2091	kobject_put(&rdev->kobj);
2092}
2093
2094static void unbind_rdev_from_array(mdk_rdev_t * rdev)
2095{
2096	char b[BDEVNAME_SIZE];
2097	if (!rdev->mddev) {
2098		MD_BUG();
2099		return;
2100	}
2101	bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
2102	list_del_rcu(&rdev->same_set);
2103	printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2104	rdev->mddev = NULL;
2105	sysfs_remove_link(&rdev->kobj, "block");
2106	sysfs_put(rdev->sysfs_state);
2107	rdev->sysfs_state = NULL;
2108	kfree(rdev->badblocks.page);
2109	rdev->badblocks.count = 0;
2110	rdev->badblocks.page = NULL;
2111	/* We need to delay this, otherwise we can deadlock when
2112	 * writing to 'remove' to "dev/state".  We also need
2113	 * to delay it due to rcu usage.
2114	 */
2115	synchronize_rcu();
2116	INIT_WORK(&rdev->del_work, md_delayed_delete);
2117	kobject_get(&rdev->kobj);
2118	queue_work(md_misc_wq, &rdev->del_work);
2119}
2120
2121/*
2122 * prevent the device from being mounted, repartitioned or
2123 * otherwise reused by a RAID array (or any other kernel
2124 * subsystem), by bd_claiming the device.
2125 */
2126static int lock_rdev(mdk_rdev_t *rdev, dev_t dev, int shared)
2127{
2128	int err = 0;
2129	struct block_device *bdev;
2130	char b[BDEVNAME_SIZE];
2131
2132	bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
2133				 shared ? (mdk_rdev_t *)lock_rdev : rdev);
2134	if (IS_ERR(bdev)) {
2135		printk(KERN_ERR "md: could not open %s.\n",
2136			__bdevname(dev, b));
2137		return PTR_ERR(bdev);
2138	}
2139	rdev->bdev = bdev;
2140	return err;
2141}
2142
2143static void unlock_rdev(mdk_rdev_t *rdev)
2144{
2145	struct block_device *bdev = rdev->bdev;
2146	rdev->bdev = NULL;
2147	if (!bdev)
2148		MD_BUG();
2149	blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
2150}
2151
2152void md_autodetect_dev(dev_t dev);
2153
2154static void export_rdev(mdk_rdev_t * rdev)
2155{
2156	char b[BDEVNAME_SIZE];
2157	printk(KERN_INFO "md: export_rdev(%s)\n",
2158		bdevname(rdev->bdev,b));
2159	if (rdev->mddev)
2160		MD_BUG();
2161	free_disk_sb(rdev);
2162#ifndef MODULE
2163	if (test_bit(AutoDetected, &rdev->flags))
2164		md_autodetect_dev(rdev->bdev->bd_dev);
2165#endif
2166	unlock_rdev(rdev);
2167	kobject_put(&rdev->kobj);
2168}
2169
2170static void kick_rdev_from_array(mdk_rdev_t * rdev)
2171{
2172	unbind_rdev_from_array(rdev);
2173	export_rdev(rdev);
2174}
2175
2176static void export_array(mddev_t *mddev)
2177{
2178	mdk_rdev_t *rdev, *tmp;
2179
2180	rdev_for_each(rdev, tmp, mddev) {
2181		if (!rdev->mddev) {
2182			MD_BUG();
2183			continue;
2184		}
2185		kick_rdev_from_array(rdev);
2186	}
2187	if (!list_empty(&mddev->disks))
2188		MD_BUG();
2189	mddev->raid_disks = 0;
2190	mddev->major_version = 0;
2191}
2192
2193static void print_desc(mdp_disk_t *desc)
2194{
2195	printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2196		desc->major,desc->minor,desc->raid_disk,desc->state);
2197}
2198
2199static void print_sb_90(mdp_super_t *sb)
2200{
2201	int i;
2202
2203	printk(KERN_INFO 
2204		"md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2205		sb->major_version, sb->minor_version, sb->patch_version,
2206		sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2207		sb->ctime);
2208	printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2209		sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2210		sb->md_minor, sb->layout, sb->chunk_size);
2211	printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
2212		" FD:%d SD:%d CSUM:%08x E:%08lx\n",
2213		sb->utime, sb->state, sb->active_disks, sb->working_disks,
2214		sb->failed_disks, sb->spare_disks,
2215		sb->sb_csum, (unsigned long)sb->events_lo);
2216
2217	printk(KERN_INFO);
2218	for (i = 0; i < MD_SB_DISKS; i++) {
2219		mdp_disk_t *desc;
2220
2221		desc = sb->disks + i;
2222		if (desc->number || desc->major || desc->minor ||
2223		    desc->raid_disk || (desc->state && (desc->state != 4))) {
2224			printk("     D %2d: ", i);
2225			print_desc(desc);
2226		}
2227	}
2228	printk(KERN_INFO "md:     THIS: ");
2229	print_desc(&sb->this_disk);
2230}
2231
2232static void print_sb_1(struct mdp_superblock_1 *sb)
2233{
2234	__u8 *uuid;
2235
2236	uuid = sb->set_uuid;
2237	printk(KERN_INFO
2238	       "md:  SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
2239	       "md:    Name: \"%s\" CT:%llu\n",
2240		le32_to_cpu(sb->major_version),
2241		le32_to_cpu(sb->feature_map),
2242		uuid,
2243		sb->set_name,
2244		(unsigned long long)le64_to_cpu(sb->ctime)
2245		       & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2246
2247	uuid = sb->device_uuid;
2248	printk(KERN_INFO
2249	       "md:       L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
2250			" RO:%llu\n"
2251	       "md:     Dev:%08x UUID: %pU\n"
2252	       "md:       (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2253	       "md:         (MaxDev:%u) \n",
2254		le32_to_cpu(sb->level),
2255		(unsigned long long)le64_to_cpu(sb->size),
2256		le32_to_cpu(sb->raid_disks),
2257		le32_to_cpu(sb->layout),
2258		le32_to_cpu(sb->chunksize),
2259		(unsigned long long)le64_to_cpu(sb->data_offset),
2260		(unsigned long long)le64_to_cpu(sb->data_size),
2261		(unsigned long long)le64_to_cpu(sb->super_offset),
2262		(unsigned long long)le64_to_cpu(sb->recovery_offset),
2263		le32_to_cpu(sb->dev_number),
2264		uuid,
2265		sb->devflags,
2266		(unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2267		(unsigned long long)le64_to_cpu(sb->events),
2268		(unsigned long long)le64_to_cpu(sb->resync_offset),
2269		le32_to_cpu(sb->sb_csum),
2270		le32_to_cpu(sb->max_dev)
2271		);
2272}
2273
2274static void print_rdev(mdk_rdev_t *rdev, int major_version)
2275{
2276	char b[BDEVNAME_SIZE];
2277	printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2278		bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
2279	        test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2280	        rdev->desc_nr);
2281	if (rdev->sb_loaded) {
2282		printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2283		switch (major_version) {
2284		case 0:
2285			print_sb_90(page_address(rdev->sb_page));
2286			break;
2287		case 1:
2288			print_sb_1(page_address(rdev->sb_page));
2289			break;
2290		}
2291	} else
2292		printk(KERN_INFO "md: no rdev superblock!\n");
2293}
2294
2295static void md_print_devices(void)
2296{
2297	struct list_head *tmp;
2298	mdk_rdev_t *rdev;
2299	mddev_t *mddev;
2300	char b[BDEVNAME_SIZE];
2301
2302	printk("\n");
2303	printk("md:	**********************************\n");
2304	printk("md:	* <COMPLETE RAID STATE PRINTOUT> *\n");
2305	printk("md:	**********************************\n");
2306	for_each_mddev(mddev, tmp) {
2307
2308		if (mddev->bitmap)
2309			bitmap_print_sb(mddev->bitmap);
2310		else
2311			printk("%s: ", mdname(mddev));
2312		list_for_each_entry(rdev, &mddev->disks, same_set)
2313			printk("<%s>", bdevname(rdev->bdev,b));
2314		printk("\n");
2315
2316		list_for_each_entry(rdev, &mddev->disks, same_set)
2317			print_rdev(rdev, mddev->major_version);
2318	}
2319	printk("md:	**********************************\n");
2320	printk("\n");
2321}
2322
2323
2324static void sync_sbs(mddev_t * mddev, int nospares)
2325{
2326	/* Update each superblock (in-memory image), but
2327	 * if we are allowed to, skip spares which already
2328	 * have the right event counter, or have one earlier
2329	 * (which would mean they aren't being marked as dirty
2330	 * with the rest of the array)
2331	 */
2332	mdk_rdev_t *rdev;
2333	list_for_each_entry(rdev, &mddev->disks, same_set) {
2334		if (rdev->sb_events == mddev->events ||
2335		    (nospares &&
2336		     rdev->raid_disk < 0 &&
2337		     rdev->sb_events+1 == mddev->events)) {
2338			/* Don't update this superblock */
2339			rdev->sb_loaded = 2;
2340		} else {
2341			sync_super(mddev, rdev);
2342			rdev->sb_loaded = 1;
2343		}
2344	}
2345}
2346
2347static void md_update_sb(mddev_t * mddev, int force_change)
2348{
2349	mdk_rdev_t *rdev;
2350	int sync_req;
2351	int nospares = 0;
2352	int any_badblocks_changed = 0;
2353
2354repeat:
2355	/* First make sure individual recovery_offsets are correct */
2356	list_for_each_entry(rdev, &mddev->disks, same_set) {
2357		if (rdev->raid_disk >= 0 &&
2358		    mddev->delta_disks >= 0 &&
2359		    !test_bit(In_sync, &rdev->flags) &&
2360		    mddev->curr_resync_completed > rdev->recovery_offset)
2361				rdev->recovery_offset = mddev->curr_resync_completed;
2362
2363	}	
2364	if (!mddev->persistent) {
2365		clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
2366		clear_bit(MD_CHANGE_DEVS, &mddev->flags);
2367		if (!mddev->external) {
2368			clear_bit(MD_CHANGE_PENDING, &mddev->flags);
2369			list_for_each_entry(rdev, &mddev->disks, same_set) {
2370				if (rdev->badblocks.changed) {
 
2371					md_ack_all_badblocks(&rdev->badblocks);
2372					md_error(mddev, rdev);
2373				}
2374				clear_bit(Blocked, &rdev->flags);
2375				clear_bit(BlockedBadBlocks, &rdev->flags);
2376				wake_up(&rdev->blocked_wait);
2377			}
2378		}
2379		wake_up(&mddev->sb_wait);
2380		return;
2381	}
2382
2383	spin_lock_irq(&mddev->write_lock);
2384
2385	mddev->utime = get_seconds();
2386
2387	if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2388		force_change = 1;
2389	if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2390		/* just a clean<-> dirty transition, possibly leave spares alone,
2391		 * though if events isn't the right even/odd, we will have to do
2392		 * spares after all
2393		 */
2394		nospares = 1;
2395	if (force_change)
2396		nospares = 0;
2397	if (mddev->degraded)
2398		/* If the array is degraded, then skipping spares is both
2399		 * dangerous and fairly pointless.
2400		 * Dangerous because a device that was removed from the array
2401		 * might have a event_count that still looks up-to-date,
2402		 * so it can be re-added without a resync.
2403		 * Pointless because if there are any spares to skip,
2404		 * then a recovery will happen and soon that array won't
2405		 * be degraded any more and the spare can go back to sleep then.
2406		 */
2407		nospares = 0;
2408
2409	sync_req = mddev->in_sync;
2410
2411	/* If this is just a dirty<->clean transition, and the array is clean
2412	 * and 'events' is odd, we can roll back to the previous clean state */
2413	if (nospares
2414	    && (mddev->in_sync && mddev->recovery_cp == MaxSector)
2415	    && mddev->can_decrease_events
2416	    && mddev->events != 1) {
2417		mddev->events--;
2418		mddev->can_decrease_events = 0;
2419	} else {
2420		/* otherwise we have to go forward and ... */
2421		mddev->events ++;
2422		mddev->can_decrease_events = nospares;
2423	}
2424
2425	if (!mddev->events) {
2426		/*
2427		 * oops, this 64-bit counter should never wrap.
2428		 * Either we are in around ~1 trillion A.C., assuming
2429		 * 1 reboot per second, or we have a bug:
2430		 */
2431		MD_BUG();
2432		mddev->events --;
2433	}
2434
2435	list_for_each_entry(rdev, &mddev->disks, same_set) {
2436		if (rdev->badblocks.changed)
2437			any_badblocks_changed++;
2438		if (test_bit(Faulty, &rdev->flags))
2439			set_bit(FaultRecorded, &rdev->flags);
2440	}
2441
2442	sync_sbs(mddev, nospares);
2443	spin_unlock_irq(&mddev->write_lock);
2444
2445	dprintk(KERN_INFO 
2446		"md: updating %s RAID superblock on device (in sync %d)\n",
2447		mdname(mddev),mddev->in_sync);
2448
2449	bitmap_update_sb(mddev->bitmap);
2450	list_for_each_entry(rdev, &mddev->disks, same_set) {
2451		char b[BDEVNAME_SIZE];
2452		dprintk(KERN_INFO "md: ");
2453		if (rdev->sb_loaded != 1)
2454			continue; /* no noise on spare devices */
2455		if (test_bit(Faulty, &rdev->flags))
2456			dprintk("(skipping faulty ");
2457
2458		dprintk("%s ", bdevname(rdev->bdev,b));
2459		if (!test_bit(Faulty, &rdev->flags)) {
2460			md_super_write(mddev,rdev,
2461				       rdev->sb_start, rdev->sb_size,
2462				       rdev->sb_page);
2463			dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
2464				bdevname(rdev->bdev,b),
2465				(unsigned long long)rdev->sb_start);
2466			rdev->sb_events = mddev->events;
2467			if (rdev->badblocks.size) {
2468				md_super_write(mddev, rdev,
2469					       rdev->badblocks.sector,
2470					       rdev->badblocks.size << 9,
2471					       rdev->bb_page);
2472				rdev->badblocks.size = 0;
2473			}
2474
2475		} else
2476			dprintk(")\n");
 
 
 
 
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	list_for_each_entry(rdev, &mddev->disks, same_set) {
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 */
2511static 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
2528struct rdev_sysfs_entry {
2529	struct attribute attr;
2530	ssize_t (*show)(mdk_rdev_t *, char *);
2531	ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
2532};
2533
2534static ssize_t
2535state_show(mdk_rdev_t *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		len += sprintf(page+len, "%sblocked", sep);
2556		sep = ",";
2557	}
2558	if (!test_bit(Faulty, &rdev->flags) &&
2559	    !test_bit(In_sync, &rdev->flags)) {
2560		len += sprintf(page+len, "%sspare", sep);
2561		sep = ",";
2562	}
2563	if (test_bit(WriteErrorSeen, &rdev->flags)) {
2564		len += sprintf(page+len, "%swrite_error", sep);
2565		sep = ",";
2566	}
 
 
 
 
 
 
 
 
 
2567	return len+sprintf(page+len, "\n");
2568}
2569
2570static ssize_t
2571state_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2572{
2573	/* can write
2574	 *  faulty  - simulates an error
2575	 *  remove  - disconnects the device
2576	 *  writemostly - sets write_mostly
2577	 *  -writemostly - clears write_mostly
2578	 *  blocked - sets the Blocked flags
2579	 *  -blocked - clears the Blocked and possibly simulates an error
2580	 *  insync - sets Insync providing device isn't active
2581	 *  write_error - sets WriteErrorSeen
2582	 *  -write_error - clears WriteErrorSeen
2583	 */
2584	int err = -EINVAL;
2585	if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2586		md_error(rdev->mddev, rdev);
2587		if (test_bit(Faulty, &rdev->flags))
2588			err = 0;
2589		else
2590			err = -EBUSY;
2591	} else if (cmd_match(buf, "remove")) {
2592		if (rdev->raid_disk >= 0)
2593			err = -EBUSY;
2594		else {
2595			mddev_t *mddev = rdev->mddev;
2596			kick_rdev_from_array(rdev);
2597			if (mddev->pers)
2598				md_update_sb(mddev, 1);
2599			md_new_event(mddev);
2600			err = 0;
2601		}
2602	} else if (cmd_match(buf, "writemostly")) {
2603		set_bit(WriteMostly, &rdev->flags);
2604		err = 0;
2605	} else if (cmd_match(buf, "-writemostly")) {
2606		clear_bit(WriteMostly, &rdev->flags);
2607		err = 0;
2608	} else if (cmd_match(buf, "blocked")) {
2609		set_bit(Blocked, &rdev->flags);
2610		err = 0;
2611	} else if (cmd_match(buf, "-blocked")) {
2612		if (!test_bit(Faulty, &rdev->flags) &&
2613		    rdev->badblocks.unacked_exist) {
2614			/* metadata handler doesn't understand badblocks,
2615			 * so we need to fail the device
2616			 */
2617			md_error(rdev->mddev, rdev);
2618		}
2619		clear_bit(Blocked, &rdev->flags);
2620		clear_bit(BlockedBadBlocks, &rdev->flags);
2621		wake_up(&rdev->blocked_wait);
2622		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2623		md_wakeup_thread(rdev->mddev->thread);
2624
2625		err = 0;
2626	} else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2627		set_bit(In_sync, &rdev->flags);
2628		err = 0;
2629	} else if (cmd_match(buf, "write_error")) {
2630		set_bit(WriteErrorSeen, &rdev->flags);
2631		err = 0;
2632	} else if (cmd_match(buf, "-write_error")) {
2633		clear_bit(WriteErrorSeen, &rdev->flags);
2634		err = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2635	}
2636	if (!err)
2637		sysfs_notify_dirent_safe(rdev->sysfs_state);
2638	return err ? err : len;
2639}
2640static struct rdev_sysfs_entry rdev_state =
2641__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
2642
2643static ssize_t
2644errors_show(mdk_rdev_t *rdev, char *page)
2645{
2646	return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2647}
2648
2649static ssize_t
2650errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2651{
2652	char *e;
2653	unsigned long n = simple_strtoul(buf, &e, 10);
2654	if (*buf && (*e == 0 || *e == '\n')) {
2655		atomic_set(&rdev->corrected_errors, n);
2656		return len;
2657	}
2658	return -EINVAL;
2659}
2660static struct rdev_sysfs_entry rdev_errors =
2661__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
2662
2663static ssize_t
2664slot_show(mdk_rdev_t *rdev, char *page)
2665{
2666	if (rdev->raid_disk < 0)
2667		return sprintf(page, "none\n");
2668	else
2669		return sprintf(page, "%d\n", rdev->raid_disk);
2670}
2671
2672static ssize_t
2673slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2674{
2675	char *e;
2676	int err;
2677	int slot = simple_strtoul(buf, &e, 10);
2678	if (strncmp(buf, "none", 4)==0)
2679		slot = -1;
2680	else if (e==buf || (*e && *e!= '\n'))
2681		return -EINVAL;
2682	if (rdev->mddev->pers && slot == -1) {
2683		/* Setting 'slot' on an active array requires also
2684		 * updating the 'rd%d' link, and communicating
2685		 * with the personality with ->hot_*_disk.
2686		 * For now we only support removing
2687		 * failed/spare devices.  This normally happens automatically,
2688		 * but not when the metadata is externally managed.
2689		 */
2690		if (rdev->raid_disk == -1)
2691			return -EEXIST;
2692		/* personality does all needed checks */
2693		if (rdev->mddev->pers->hot_remove_disk == NULL)
2694			return -EINVAL;
2695		err = rdev->mddev->pers->
2696			hot_remove_disk(rdev->mddev, rdev->raid_disk);
2697		if (err)
2698			return err;
2699		sysfs_unlink_rdev(rdev->mddev, rdev);
2700		rdev->raid_disk = -1;
2701		set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2702		md_wakeup_thread(rdev->mddev->thread);
2703	} else if (rdev->mddev->pers) {
2704		mdk_rdev_t *rdev2;
2705		/* Activating a spare .. or possibly reactivating
2706		 * if we ever get bitmaps working here.
2707		 */
2708
2709		if (rdev->raid_disk != -1)
2710			return -EBUSY;
2711
2712		if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2713			return -EBUSY;
2714
2715		if (rdev->mddev->pers->hot_add_disk == NULL)
2716			return -EINVAL;
2717
2718		list_for_each_entry(rdev2, &rdev->mddev->disks, same_set)
2719			if (rdev2->raid_disk == slot)
2720				return -EEXIST;
2721
2722		if (slot >= rdev->mddev->raid_disks &&
2723		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2724			return -ENOSPC;
2725
2726		rdev->raid_disk = slot;
2727		if (test_bit(In_sync, &rdev->flags))
2728			rdev->saved_raid_disk = slot;
2729		else
2730			rdev->saved_raid_disk = -1;
 
2731		err = rdev->mddev->pers->
2732			hot_add_disk(rdev->mddev, rdev);
2733		if (err) {
2734			rdev->raid_disk = -1;
2735			return err;
2736		} else
2737			sysfs_notify_dirent_safe(rdev->sysfs_state);
2738		if (sysfs_link_rdev(rdev->mddev, rdev))
2739			/* failure here is OK */;
2740		/* don't wakeup anyone, leave that to userspace. */
2741	} else {
2742		if (slot >= rdev->mddev->raid_disks &&
2743		    slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2744			return -ENOSPC;
2745		rdev->raid_disk = slot;
2746		/* assume it is working */
2747		clear_bit(Faulty, &rdev->flags);
2748		clear_bit(WriteMostly, &rdev->flags);
2749		set_bit(In_sync, &rdev->flags);
2750		sysfs_notify_dirent_safe(rdev->sysfs_state);
2751	}
2752	return len;
2753}
2754
2755
2756static struct rdev_sysfs_entry rdev_slot =
2757__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
2758
2759static ssize_t
2760offset_show(mdk_rdev_t *rdev, char *page)
2761{
2762	return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
2763}
2764
2765static ssize_t
2766offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2767{
2768	char *e;
2769	unsigned long long offset = simple_strtoull(buf, &e, 10);
2770	if (e==buf || (*e && *e != '\n'))
2771		return -EINVAL;
2772	if (rdev->mddev->pers && rdev->raid_disk >= 0)
2773		return -EBUSY;
2774	if (rdev->sectors && rdev->mddev->external)
2775		/* Must set offset before size, so overlap checks
2776		 * can be sane */
2777		return -EBUSY;
2778	rdev->data_offset = offset;
 
2779	return len;
2780}
2781
2782static struct rdev_sysfs_entry rdev_offset =
2783__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
2784
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2785static ssize_t
2786rdev_size_show(mdk_rdev_t *rdev, char *page)
2787{
2788	return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
2789}
2790
2791static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2792{
2793	/* check if two start/length pairs overlap */
2794	if (s1+l1 <= s2)
2795		return 0;
2796	if (s2+l2 <= s1)
2797		return 0;
2798	return 1;
2799}
2800
2801static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2802{
2803	unsigned long long blocks;
2804	sector_t new;
2805
2806	if (strict_strtoull(buf, 10, &blocks) < 0)
2807		return -EINVAL;
2808
2809	if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2810		return -EINVAL; /* sector conversion overflow */
2811
2812	new = blocks * 2;
2813	if (new != blocks * 2)
2814		return -EINVAL; /* unsigned long long to sector_t overflow */
2815
2816	*sectors = new;
2817	return 0;
2818}
2819
2820static ssize_t
2821rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2822{
2823	mddev_t *my_mddev = rdev->mddev;
2824	sector_t oldsectors = rdev->sectors;
2825	sector_t sectors;
2826
2827	if (strict_blocks_to_sectors(buf, &sectors) < 0)
2828		return -EINVAL;
 
 
2829	if (my_mddev->pers && rdev->raid_disk >= 0) {
2830		if (my_mddev->persistent) {
2831			sectors = super_types[my_mddev->major_version].
2832				rdev_size_change(rdev, sectors);
2833			if (!sectors)
2834				return -EBUSY;
2835		} else if (!sectors)
2836			sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
2837				rdev->data_offset;
2838	}
2839	if (sectors < my_mddev->dev_sectors)
2840		return -EINVAL; /* component must fit device */
2841
2842	rdev->sectors = sectors;
2843	if (sectors > oldsectors && my_mddev->external) {
2844		/* need to check that all other rdevs with the same ->bdev
2845		 * do not overlap.  We need to unlock the mddev to avoid
2846		 * a deadlock.  We have already changed rdev->sectors, and if
2847		 * we have to change it back, we will have the lock again.
2848		 */
2849		mddev_t *mddev;
2850		int overlap = 0;
2851		struct list_head *tmp;
2852
2853		mddev_unlock(my_mddev);
2854		for_each_mddev(mddev, tmp) {
2855			mdk_rdev_t *rdev2;
2856
2857			mddev_lock(mddev);
2858			list_for_each_entry(rdev2, &mddev->disks, same_set)
2859				if (rdev->bdev == rdev2->bdev &&
2860				    rdev != rdev2 &&
2861				    overlaps(rdev->data_offset, rdev->sectors,
2862					     rdev2->data_offset,
2863					     rdev2->sectors)) {
2864					overlap = 1;
2865					break;
2866				}
2867			mddev_unlock(mddev);
2868			if (overlap) {
2869				mddev_put(mddev);
2870				break;
2871			}
2872		}
2873		mddev_lock(my_mddev);
2874		if (overlap) {
2875			/* Someone else could have slipped in a size
2876			 * change here, but doing so is just silly.
2877			 * We put oldsectors back because we *know* it is
2878			 * safe, and trust userspace not to race with
2879			 * itself
2880			 */
2881			rdev->sectors = oldsectors;
2882			return -EBUSY;
2883		}
2884	}
2885	return len;
2886}
2887
2888static struct rdev_sysfs_entry rdev_size =
2889__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
2890
2891
2892static ssize_t recovery_start_show(mdk_rdev_t *rdev, char *page)
2893{
2894	unsigned long long recovery_start = rdev->recovery_offset;
2895
2896	if (test_bit(In_sync, &rdev->flags) ||
2897	    recovery_start == MaxSector)
2898		return sprintf(page, "none\n");
2899
2900	return sprintf(page, "%llu\n", recovery_start);
2901}
2902
2903static ssize_t recovery_start_store(mdk_rdev_t *rdev, const char *buf, size_t len)
2904{
2905	unsigned long long recovery_start;
2906
2907	if (cmd_match(buf, "none"))
2908		recovery_start = MaxSector;
2909	else if (strict_strtoull(buf, 10, &recovery_start))
2910		return -EINVAL;
2911
2912	if (rdev->mddev->pers &&
2913	    rdev->raid_disk >= 0)
2914		return -EBUSY;
2915
2916	rdev->recovery_offset = recovery_start;
2917	if (recovery_start == MaxSector)
2918		set_bit(In_sync, &rdev->flags);
2919	else
2920		clear_bit(In_sync, &rdev->flags);
2921	return len;
2922}
2923
2924static struct rdev_sysfs_entry rdev_recovery_start =
2925__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2926
2927
2928static ssize_t
2929badblocks_show(struct badblocks *bb, char *page, int unack);
2930static ssize_t
2931badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2932
2933static ssize_t bb_show(mdk_rdev_t *rdev, char *page)
2934{
2935	return badblocks_show(&rdev->badblocks, page, 0);
2936}
2937static ssize_t bb_store(mdk_rdev_t *rdev, const char *page, size_t len)
2938{
2939	int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2940	/* Maybe that ack was all we needed */
2941	if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2942		wake_up(&rdev->blocked_wait);
2943	return rv;
2944}
2945static struct rdev_sysfs_entry rdev_bad_blocks =
2946__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2947
2948
2949static ssize_t ubb_show(mdk_rdev_t *rdev, char *page)
2950{
2951	return badblocks_show(&rdev->badblocks, page, 1);
2952}
2953static ssize_t ubb_store(mdk_rdev_t *rdev, const char *page, size_t len)
2954{
2955	return badblocks_store(&rdev->badblocks, page, len, 1);
2956}
2957static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2958__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2959
2960static struct attribute *rdev_default_attrs[] = {
2961	&rdev_state.attr,
2962	&rdev_errors.attr,
2963	&rdev_slot.attr,
2964	&rdev_offset.attr,
 
2965	&rdev_size.attr,
2966	&rdev_recovery_start.attr,
2967	&rdev_bad_blocks.attr,
2968	&rdev_unack_bad_blocks.attr,
2969	NULL,
2970};
2971static ssize_t
2972rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2973{
2974	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2975	mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2976	mddev_t *mddev = rdev->mddev;
2977	ssize_t rv;
2978
2979	if (!entry->show)
2980		return -EIO;
2981
2982	rv = mddev ? mddev_lock(mddev) : -EBUSY;
2983	if (!rv) {
2984		if (rdev->mddev == NULL)
2985			rv = -EBUSY;
2986		else
2987			rv = entry->show(rdev, page);
2988		mddev_unlock(mddev);
2989	}
2990	return rv;
2991}
2992
2993static ssize_t
2994rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2995	      const char *page, size_t length)
2996{
2997	struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
2998	mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
2999	ssize_t rv;
3000	mddev_t *mddev = rdev->mddev;
3001
3002	if (!entry->store)
3003		return -EIO;
3004	if (!capable(CAP_SYS_ADMIN))
3005		return -EACCES;
3006	rv = mddev ? mddev_lock(mddev): -EBUSY;
3007	if (!rv) {
3008		if (rdev->mddev == NULL)
3009			rv = -EBUSY;
3010		else
3011			rv = entry->store(rdev, page, length);
3012		mddev_unlock(mddev);
3013	}
3014	return rv;
3015}
3016
3017static void rdev_free(struct kobject *ko)
3018{
3019	mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
3020	kfree(rdev);
3021}
3022static const struct sysfs_ops rdev_sysfs_ops = {
3023	.show		= rdev_attr_show,
3024	.store		= rdev_attr_store,
3025};
3026static struct kobj_type rdev_ktype = {
3027	.release	= rdev_free,
3028	.sysfs_ops	= &rdev_sysfs_ops,
3029	.default_attrs	= rdev_default_attrs,
3030};
3031
3032int md_rdev_init(mdk_rdev_t *rdev)
3033{
3034	rdev->desc_nr = -1;
3035	rdev->saved_raid_disk = -1;
3036	rdev->raid_disk = -1;
3037	rdev->flags = 0;
3038	rdev->data_offset = 0;
 
3039	rdev->sb_events = 0;
3040	rdev->last_read_error.tv_sec  = 0;
3041	rdev->last_read_error.tv_nsec = 0;
3042	rdev->sb_loaded = 0;
3043	rdev->bb_page = NULL;
3044	atomic_set(&rdev->nr_pending, 0);
3045	atomic_set(&rdev->read_errors, 0);
3046	atomic_set(&rdev->corrected_errors, 0);
3047
3048	INIT_LIST_HEAD(&rdev->same_set);
3049	init_waitqueue_head(&rdev->blocked_wait);
3050
3051	/* Add space to store bad block list.
3052	 * This reserves the space even on arrays where it cannot
3053	 * be used - I wonder if that matters
3054	 */
3055	rdev->badblocks.count = 0;
3056	rdev->badblocks.shift = 0;
3057	rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3058	seqlock_init(&rdev->badblocks.lock);
3059	if (rdev->badblocks.page == NULL)
3060		return -ENOMEM;
3061
3062	return 0;
3063}
3064EXPORT_SYMBOL_GPL(md_rdev_init);
3065/*
3066 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3067 *
3068 * mark the device faulty if:
3069 *
3070 *   - the device is nonexistent (zero size)
3071 *   - the device has no valid superblock
3072 *
3073 * a faulty rdev _never_ has rdev->sb set.
3074 */
3075static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
3076{
3077	char b[BDEVNAME_SIZE];
3078	int err;
3079	mdk_rdev_t *rdev;
3080	sector_t size;
3081
3082	rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
3083	if (!rdev) {
3084		printk(KERN_ERR "md: could not alloc mem for new device!\n");
3085		return ERR_PTR(-ENOMEM);
3086	}
3087
3088	err = md_rdev_init(rdev);
3089	if (err)
3090		goto abort_free;
3091	err = alloc_disk_sb(rdev);
3092	if (err)
3093		goto abort_free;
3094
3095	err = lock_rdev(rdev, newdev, super_format == -2);
3096	if (err)
3097		goto abort_free;
3098
3099	kobject_init(&rdev->kobj, &rdev_ktype);
3100
3101	size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
3102	if (!size) {
3103		printk(KERN_WARNING 
3104			"md: %s has zero or unknown size, marking faulty!\n",
3105			bdevname(rdev->bdev,b));
3106		err = -EINVAL;
3107		goto abort_free;
3108	}
3109
3110	if (super_format >= 0) {
3111		err = super_types[super_format].
3112			load_super(rdev, NULL, super_minor);
3113		if (err == -EINVAL) {
3114			printk(KERN_WARNING
3115				"md: %s does not have a valid v%d.%d "
3116			       "superblock, not importing!\n",
3117				bdevname(rdev->bdev,b),
3118			       super_format, super_minor);
3119			goto abort_free;
3120		}
3121		if (err < 0) {
3122			printk(KERN_WARNING 
3123				"md: could not read %s's sb, not importing!\n",
3124				bdevname(rdev->bdev,b));
3125			goto abort_free;
3126		}
3127	}
3128	if (super_format == -1)
3129		/* hot-add for 0.90, or non-persistent: so no badblocks */
3130		rdev->badblocks.shift = -1;
3131
3132	return rdev;
3133
3134abort_free:
3135	if (rdev->bdev)
3136		unlock_rdev(rdev);
3137	free_disk_sb(rdev);
3138	kfree(rdev->badblocks.page);
3139	kfree(rdev);
3140	return ERR_PTR(err);
3141}
3142
3143/*
3144 * Check a full RAID array for plausibility
3145 */
3146
3147
3148static void analyze_sbs(mddev_t * mddev)
3149{
3150	int i;
3151	mdk_rdev_t *rdev, *freshest, *tmp;
3152	char b[BDEVNAME_SIZE];
3153
3154	freshest = NULL;
3155	rdev_for_each(rdev, tmp, mddev)
3156		switch (super_types[mddev->major_version].
3157			load_super(rdev, freshest, mddev->minor_version)) {
3158		case 1:
3159			freshest = rdev;
3160			break;
3161		case 0:
3162			break;
3163		default:
3164			printk( KERN_ERR \
3165				"md: fatal superblock inconsistency in %s"
3166				" -- removing from array\n", 
3167				bdevname(rdev->bdev,b));
3168			kick_rdev_from_array(rdev);
3169		}
3170
3171
3172	super_types[mddev->major_version].
3173		validate_super(mddev, freshest);
3174
3175	i = 0;
3176	rdev_for_each(rdev, tmp, mddev) {
3177		if (mddev->max_disks &&
3178		    (rdev->desc_nr >= mddev->max_disks ||
3179		     i > mddev->max_disks)) {
3180			printk(KERN_WARNING
3181			       "md: %s: %s: only %d devices permitted\n",
3182			       mdname(mddev), bdevname(rdev->bdev, b),
3183			       mddev->max_disks);
3184			kick_rdev_from_array(rdev);
3185			continue;
3186		}
3187		if (rdev != freshest)
3188			if (super_types[mddev->major_version].
3189			    validate_super(mddev, rdev)) {
3190				printk(KERN_WARNING "md: kicking non-fresh %s"
3191					" from array!\n",
3192					bdevname(rdev->bdev,b));
3193				kick_rdev_from_array(rdev);
3194				continue;
3195			}
3196		if (mddev->level == LEVEL_MULTIPATH) {
3197			rdev->desc_nr = i++;
3198			rdev->raid_disk = rdev->desc_nr;
3199			set_bit(In_sync, &rdev->flags);
3200		} else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
3201			rdev->raid_disk = -1;
3202			clear_bit(In_sync, &rdev->flags);
3203		}
3204	}
3205}
3206
3207/* Read a fixed-point number.
3208 * Numbers in sysfs attributes should be in "standard" units where
3209 * possible, so time should be in seconds.
3210 * However we internally use a a much smaller unit such as 
3211 * milliseconds or jiffies.
3212 * This function takes a decimal number with a possible fractional
3213 * component, and produces an integer which is the result of
3214 * multiplying that number by 10^'scale'.
3215 * all without any floating-point arithmetic.
3216 */
3217int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3218{
3219	unsigned long result = 0;
3220	long decimals = -1;
3221	while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3222		if (*cp == '.')
3223			decimals = 0;
3224		else if (decimals < scale) {
3225			unsigned int value;
3226			value = *cp - '0';
3227			result = result * 10 + value;
3228			if (decimals >= 0)
3229				decimals++;
3230		}
3231		cp++;
3232	}
3233	if (*cp == '\n')
3234		cp++;
3235	if (*cp)
3236		return -EINVAL;
3237	if (decimals < 0)
3238		decimals = 0;
3239	while (decimals < scale) {
3240		result *= 10;
3241		decimals ++;
3242	}
3243	*res = result;
3244	return 0;
3245}
3246
3247
3248static void md_safemode_timeout(unsigned long data);
3249
3250static ssize_t
3251safe_delay_show(mddev_t *mddev, char *page)
3252{
3253	int msec = (mddev->safemode_delay*1000)/HZ;
3254	return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3255}
3256static ssize_t
3257safe_delay_store(mddev_t *mddev, const char *cbuf, size_t len)
3258{
3259	unsigned long msec;
3260
3261	if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
3262		return -EINVAL;
3263	if (msec == 0)
3264		mddev->safemode_delay = 0;
3265	else {
3266		unsigned long old_delay = mddev->safemode_delay;
3267		mddev->safemode_delay = (msec*HZ)/1000;
3268		if (mddev->safemode_delay == 0)
3269			mddev->safemode_delay = 1;
3270		if (mddev->safemode_delay < old_delay)
3271			md_safemode_timeout((unsigned long)mddev);
3272	}
3273	return len;
3274}
3275static struct md_sysfs_entry md_safe_delay =
3276__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
3277
3278static ssize_t
3279level_show(mddev_t *mddev, char *page)
3280{
3281	struct mdk_personality *p = mddev->pers;
3282	if (p)
3283		return sprintf(page, "%s\n", p->name);
3284	else if (mddev->clevel[0])
3285		return sprintf(page, "%s\n", mddev->clevel);
3286	else if (mddev->level != LEVEL_NONE)
3287		return sprintf(page, "%d\n", mddev->level);
3288	else
3289		return 0;
3290}
3291
3292static ssize_t
3293level_store(mddev_t *mddev, const char *buf, size_t len)
3294{
3295	char clevel[16];
3296	ssize_t rv = len;
3297	struct mdk_personality *pers;
3298	long level;
3299	void *priv;
3300	mdk_rdev_t *rdev;
3301
3302	if (mddev->pers == NULL) {
3303		if (len == 0)
3304			return 0;
3305		if (len >= sizeof(mddev->clevel))
3306			return -ENOSPC;
3307		strncpy(mddev->clevel, buf, len);
3308		if (mddev->clevel[len-1] == '\n')
3309			len--;
3310		mddev->clevel[len] = 0;
3311		mddev->level = LEVEL_NONE;
3312		return rv;
3313	}
3314
3315	/* request to change the personality.  Need to ensure:
3316	 *  - array is not engaged in resync/recovery/reshape
3317	 *  - old personality can be suspended
3318	 *  - new personality will access other array.
3319	 */
3320
3321	if (mddev->sync_thread ||
3322	    mddev->reshape_position != MaxSector ||
3323	    mddev->sysfs_active)
3324		return -EBUSY;
3325
3326	if (!mddev->pers->quiesce) {
3327		printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3328		       mdname(mddev), mddev->pers->name);
3329		return -EINVAL;
3330	}
3331
3332	/* Now find the new personality */
3333	if (len == 0 || len >= sizeof(clevel))
3334		return -EINVAL;
3335	strncpy(clevel, buf, len);
3336	if (clevel[len-1] == '\n')
3337		len--;
3338	clevel[len] = 0;
3339	if (strict_strtol(clevel, 10, &level))
3340		level = LEVEL_NONE;
3341
3342	if (request_module("md-%s", clevel) != 0)
3343		request_module("md-level-%s", clevel);
3344	spin_lock(&pers_lock);
3345	pers = find_pers(level, clevel);
3346	if (!pers || !try_module_get(pers->owner)) {
3347		spin_unlock(&pers_lock);
3348		printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
3349		return -EINVAL;
3350	}
3351	spin_unlock(&pers_lock);
3352
3353	if (pers == mddev->pers) {
3354		/* Nothing to do! */
3355		module_put(pers->owner);
3356		return rv;
3357	}
3358	if (!pers->takeover) {
3359		module_put(pers->owner);
3360		printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
3361		       mdname(mddev), clevel);
3362		return -EINVAL;
3363	}
3364
3365	list_for_each_entry(rdev, &mddev->disks, same_set)
3366		rdev->new_raid_disk = rdev->raid_disk;
3367
3368	/* ->takeover must set new_* and/or delta_disks
3369	 * if it succeeds, and may set them when it fails.
3370	 */
3371	priv = pers->takeover(mddev);
3372	if (IS_ERR(priv)) {
3373		mddev->new_level = mddev->level;
3374		mddev->new_layout = mddev->layout;
3375		mddev->new_chunk_sectors = mddev->chunk_sectors;
3376		mddev->raid_disks -= mddev->delta_disks;
3377		mddev->delta_disks = 0;
 
3378		module_put(pers->owner);
3379		printk(KERN_WARNING "md: %s: %s would not accept array\n",
3380		       mdname(mddev), clevel);
3381		return PTR_ERR(priv);
3382	}
3383
3384	/* Looks like we have a winner */
3385	mddev_suspend(mddev);
3386	mddev->pers->stop(mddev);
3387	
3388	if (mddev->pers->sync_request == NULL &&
3389	    pers->sync_request != NULL) {
3390		/* need to add the md_redundancy_group */
3391		if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3392			printk(KERN_WARNING
3393			       "md: cannot register extra attributes for %s\n",
3394			       mdname(mddev));
3395		mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
3396	}		
3397	if (mddev->pers->sync_request != NULL &&
3398	    pers->sync_request == NULL) {
3399		/* need to remove the md_redundancy_group */
3400		if (mddev->to_remove == NULL)
3401			mddev->to_remove = &md_redundancy_group;
3402	}
3403
3404	if (mddev->pers->sync_request == NULL &&
3405	    mddev->external) {
3406		/* We are converting from a no-redundancy array
3407		 * to a redundancy array and metadata is managed
3408		 * externally so we need to be sure that writes
3409		 * won't block due to a need to transition
3410		 *      clean->dirty
3411		 * until external management is started.
3412		 */
3413		mddev->in_sync = 0;
3414		mddev->safemode_delay = 0;
3415		mddev->safemode = 0;
3416	}
3417
3418	list_for_each_entry(rdev, &mddev->disks, same_set) {
3419		if (rdev->raid_disk < 0)
3420			continue;
3421		if (rdev->new_raid_disk >= mddev->raid_disks)
3422			rdev->new_raid_disk = -1;
3423		if (rdev->new_raid_disk == rdev->raid_disk)
3424			continue;
3425		sysfs_unlink_rdev(mddev, rdev);
3426	}
3427	list_for_each_entry(rdev, &mddev->disks, same_set) {
3428		if (rdev->raid_disk < 0)
3429			continue;
3430		if (rdev->new_raid_disk == rdev->raid_disk)
3431			continue;
3432		rdev->raid_disk = rdev->new_raid_disk;
3433		if (rdev->raid_disk < 0)
3434			clear_bit(In_sync, &rdev->flags);
3435		else {
3436			if (sysfs_link_rdev(mddev, rdev))
3437				printk(KERN_WARNING "md: cannot register rd%d"
3438				       " for %s after level change\n",
3439				       rdev->raid_disk, mdname(mddev));
3440		}
3441	}
3442
3443	module_put(mddev->pers->owner);
3444	mddev->pers = pers;
3445	mddev->private = priv;
3446	strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3447	mddev->level = mddev->new_level;
3448	mddev->layout = mddev->new_layout;
3449	mddev->chunk_sectors = mddev->new_chunk_sectors;
3450	mddev->delta_disks = 0;
 
3451	mddev->degraded = 0;
3452	if (mddev->pers->sync_request == NULL) {
3453		/* this is now an array without redundancy, so
3454		 * it must always be in_sync
3455		 */
3456		mddev->in_sync = 1;
3457		del_timer_sync(&mddev->safemode_timer);
3458	}
3459	pers->run(mddev);
3460	mddev_resume(mddev);
3461	set_bit(MD_CHANGE_DEVS, &mddev->flags);
3462	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3463	md_wakeup_thread(mddev->thread);
3464	sysfs_notify(&mddev->kobj, NULL, "level");
3465	md_new_event(mddev);
3466	return rv;
3467}
3468
3469static struct md_sysfs_entry md_level =
3470__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
3471
3472
3473static ssize_t
3474layout_show(mddev_t *mddev, char *page)
3475{
3476	/* just a number, not meaningful for all levels */
3477	if (mddev->reshape_position != MaxSector &&
3478	    mddev->layout != mddev->new_layout)
3479		return sprintf(page, "%d (%d)\n",
3480			       mddev->new_layout, mddev->layout);
3481	return sprintf(page, "%d\n", mddev->layout);
3482}
3483
3484static ssize_t
3485layout_store(mddev_t *mddev, const char *buf, size_t len)
3486{
3487	char *e;
3488	unsigned long n = simple_strtoul(buf, &e, 10);
3489
3490	if (!*buf || (*e && *e != '\n'))
3491		return -EINVAL;
3492
3493	if (mddev->pers) {
3494		int err;
3495		if (mddev->pers->check_reshape == NULL)
3496			return -EBUSY;
3497		mddev->new_layout = n;
3498		err = mddev->pers->check_reshape(mddev);
3499		if (err) {
3500			mddev->new_layout = mddev->layout;
3501			return err;
3502		}
3503	} else {
3504		mddev->new_layout = n;
3505		if (mddev->reshape_position == MaxSector)
3506			mddev->layout = n;
3507	}
3508	return len;
3509}
3510static struct md_sysfs_entry md_layout =
3511__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
3512
3513
3514static ssize_t
3515raid_disks_show(mddev_t *mddev, char *page)
3516{
3517	if (mddev->raid_disks == 0)
3518		return 0;
3519	if (mddev->reshape_position != MaxSector &&
3520	    mddev->delta_disks != 0)
3521		return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3522			       mddev->raid_disks - mddev->delta_disks);
3523	return sprintf(page, "%d\n", mddev->raid_disks);
3524}
3525
3526static int update_raid_disks(mddev_t *mddev, int raid_disks);
3527
3528static ssize_t
3529raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
3530{
3531	char *e;
3532	int rv = 0;
3533	unsigned long n = simple_strtoul(buf, &e, 10);
3534
3535	if (!*buf || (*e && *e != '\n'))
3536		return -EINVAL;
3537
3538	if (mddev->pers)
3539		rv = update_raid_disks(mddev, n);
3540	else if (mddev->reshape_position != MaxSector) {
 
3541		int olddisks = mddev->raid_disks - mddev->delta_disks;
 
 
 
 
 
 
 
 
 
3542		mddev->delta_disks = n - olddisks;
3543		mddev->raid_disks = n;
 
3544	} else
3545		mddev->raid_disks = n;
3546	return rv ? rv : len;
3547}
3548static struct md_sysfs_entry md_raid_disks =
3549__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
3550
3551static ssize_t
3552chunk_size_show(mddev_t *mddev, char *page)
3553{
3554	if (mddev->reshape_position != MaxSector &&
3555	    mddev->chunk_sectors != mddev->new_chunk_sectors)
3556		return sprintf(page, "%d (%d)\n",
3557			       mddev->new_chunk_sectors << 9,
3558			       mddev->chunk_sectors << 9);
3559	return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
3560}
3561
3562static ssize_t
3563chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
3564{
3565	char *e;
3566	unsigned long n = simple_strtoul(buf, &e, 10);
3567
3568	if (!*buf || (*e && *e != '\n'))
3569		return -EINVAL;
3570
3571	if (mddev->pers) {
3572		int err;
3573		if (mddev->pers->check_reshape == NULL)
3574			return -EBUSY;
3575		mddev->new_chunk_sectors = n >> 9;
3576		err = mddev->pers->check_reshape(mddev);
3577		if (err) {
3578			mddev->new_chunk_sectors = mddev->chunk_sectors;
3579			return err;
3580		}
3581	} else {
3582		mddev->new_chunk_sectors = n >> 9;
3583		if (mddev->reshape_position == MaxSector)
3584			mddev->chunk_sectors = n >> 9;
3585	}
3586	return len;
3587}
3588static struct md_sysfs_entry md_chunk_size =
3589__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
3590
3591static ssize_t
3592resync_start_show(mddev_t *mddev, char *page)
3593{
3594	if (mddev->recovery_cp == MaxSector)
3595		return sprintf(page, "none\n");
3596	return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3597}
3598
3599static ssize_t
3600resync_start_store(mddev_t *mddev, const char *buf, size_t len)
3601{
3602	char *e;
3603	unsigned long long n = simple_strtoull(buf, &e, 10);
3604
3605	if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
3606		return -EBUSY;
3607	if (cmd_match(buf, "none"))
3608		n = MaxSector;
3609	else if (!*buf || (*e && *e != '\n'))
3610		return -EINVAL;
3611
3612	mddev->recovery_cp = n;
3613	return len;
3614}
3615static struct md_sysfs_entry md_resync_start =
3616__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
3617
3618/*
3619 * The array state can be:
3620 *
3621 * clear
3622 *     No devices, no size, no level
3623 *     Equivalent to STOP_ARRAY ioctl
3624 * inactive
3625 *     May have some settings, but array is not active
3626 *        all IO results in error
3627 *     When written, doesn't tear down array, but just stops it
3628 * suspended (not supported yet)
3629 *     All IO requests will block. The array can be reconfigured.
3630 *     Writing this, if accepted, will block until array is quiescent
3631 * readonly
3632 *     no resync can happen.  no superblocks get written.
3633 *     write requests fail
3634 * read-auto
3635 *     like readonly, but behaves like 'clean' on a write request.
3636 *
3637 * clean - no pending writes, but otherwise active.
3638 *     When written to inactive array, starts without resync
3639 *     If a write request arrives then
3640 *       if metadata is known, mark 'dirty' and switch to 'active'.
3641 *       if not known, block and switch to write-pending
3642 *     If written to an active array that has pending writes, then fails.
3643 * active
3644 *     fully active: IO and resync can be happening.
3645 *     When written to inactive array, starts with resync
3646 *
3647 * write-pending
3648 *     clean, but writes are blocked waiting for 'active' to be written.
3649 *
3650 * active-idle
3651 *     like active, but no writes have been seen for a while (100msec).
3652 *
3653 */
3654enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3655		   write_pending, active_idle, bad_word};
3656static char *array_states[] = {
3657	"clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3658	"write-pending", "active-idle", NULL };
3659
3660static int match_word(const char *word, char **list)
3661{
3662	int n;
3663	for (n=0; list[n]; n++)
3664		if (cmd_match(word, list[n]))
3665			break;
3666	return n;
3667}
3668
3669static ssize_t
3670array_state_show(mddev_t *mddev, char *page)
3671{
3672	enum array_state st = inactive;
3673
3674	if (mddev->pers)
3675		switch(mddev->ro) {
3676		case 1:
3677			st = readonly;
3678			break;
3679		case 2:
3680			st = read_auto;
3681			break;
3682		case 0:
3683			if (mddev->in_sync)
3684				st = clean;
3685			else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
3686				st = write_pending;
3687			else if (mddev->safemode)
3688				st = active_idle;
3689			else
3690				st = active;
3691		}
3692	else {
3693		if (list_empty(&mddev->disks) &&
3694		    mddev->raid_disks == 0 &&
3695		    mddev->dev_sectors == 0)
3696			st = clear;
3697		else
3698			st = inactive;
3699	}
3700	return sprintf(page, "%s\n", array_states[st]);
3701}
3702
3703static int do_md_stop(mddev_t * mddev, int ro, int is_open);
3704static int md_set_readonly(mddev_t * mddev, int is_open);
3705static int do_md_run(mddev_t * mddev);
3706static int restart_array(mddev_t *mddev);
3707
3708static ssize_t
3709array_state_store(mddev_t *mddev, const char *buf, size_t len)
3710{
3711	int err = -EINVAL;
3712	enum array_state st = match_word(buf, array_states);
3713	switch(st) {
3714	case bad_word:
3715		break;
3716	case clear:
3717		/* stopping an active array */
3718		if (atomic_read(&mddev->openers) > 0)
3719			return -EBUSY;
3720		err = do_md_stop(mddev, 0, 0);
3721		break;
3722	case inactive:
3723		/* stopping an active array */
3724		if (mddev->pers) {
3725			if (atomic_read(&mddev->openers) > 0)
3726				return -EBUSY;
3727			err = do_md_stop(mddev, 2, 0);
3728		} else
3729			err = 0; /* already inactive */
3730		break;
3731	case suspended:
3732		break; /* not supported yet */
3733	case readonly:
3734		if (mddev->pers)
3735			err = md_set_readonly(mddev, 0);
3736		else {
3737			mddev->ro = 1;
3738			set_disk_ro(mddev->gendisk, 1);
3739			err = do_md_run(mddev);
3740		}
3741		break;
3742	case read_auto:
3743		if (mddev->pers) {
3744			if (mddev->ro == 0)
3745				err = md_set_readonly(mddev, 0);
3746			else if (mddev->ro == 1)
3747				err = restart_array(mddev);
3748			if (err == 0) {
3749				mddev->ro = 2;
3750				set_disk_ro(mddev->gendisk, 0);
3751			}
3752		} else {
3753			mddev->ro = 2;
3754			err = do_md_run(mddev);
3755		}
3756		break;
3757	case clean:
3758		if (mddev->pers) {
3759			restart_array(mddev);
3760			spin_lock_irq(&mddev->write_lock);
3761			if (atomic_read(&mddev->writes_pending) == 0) {
3762				if (mddev->in_sync == 0) {
3763					mddev->in_sync = 1;
3764					if (mddev->safemode == 1)
3765						mddev->safemode = 0;
3766					set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3767				}
3768				err = 0;
3769			} else
3770				err = -EBUSY;
3771			spin_unlock_irq(&mddev->write_lock);
3772		} else
3773			err = -EINVAL;
3774		break;
3775	case active:
3776		if (mddev->pers) {
3777			restart_array(mddev);
3778			clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3779			wake_up(&mddev->sb_wait);
3780			err = 0;
3781		} else {
3782			mddev->ro = 0;
3783			set_disk_ro(mddev->gendisk, 0);
3784			err = do_md_run(mddev);
3785		}
3786		break;
3787	case write_pending:
3788	case active_idle:
3789		/* these cannot be set */
3790		break;
3791	}
3792	if (err)
3793		return err;
3794	else {
 
 
3795		sysfs_notify_dirent_safe(mddev->sysfs_state);
3796		return len;
3797	}
3798}
3799static struct md_sysfs_entry md_array_state =
3800__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
3801
3802static ssize_t
3803max_corrected_read_errors_show(mddev_t *mddev, char *page) {
3804	return sprintf(page, "%d\n",
3805		       atomic_read(&mddev->max_corr_read_errors));
3806}
3807
3808static ssize_t
3809max_corrected_read_errors_store(mddev_t *mddev, const char *buf, size_t len)
3810{
3811	char *e;
3812	unsigned long n = simple_strtoul(buf, &e, 10);
3813
3814	if (*buf && (*e == 0 || *e == '\n')) {
3815		atomic_set(&mddev->max_corr_read_errors, n);
3816		return len;
3817	}
3818	return -EINVAL;
3819}
3820
3821static struct md_sysfs_entry max_corr_read_errors =
3822__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3823	max_corrected_read_errors_store);
3824
3825static ssize_t
3826null_show(mddev_t *mddev, char *page)
3827{
3828	return -EINVAL;
3829}
3830
3831static ssize_t
3832new_dev_store(mddev_t *mddev, const char *buf, size_t len)
3833{
3834	/* buf must be %d:%d\n? giving major and minor numbers */
3835	/* The new device is added to the array.
3836	 * If the array has a persistent superblock, we read the
3837	 * superblock to initialise info and check validity.
3838	 * Otherwise, only checking done is that in bind_rdev_to_array,
3839	 * which mainly checks size.
3840	 */
3841	char *e;
3842	int major = simple_strtoul(buf, &e, 10);
3843	int minor;
3844	dev_t dev;
3845	mdk_rdev_t *rdev;
3846	int err;
3847
3848	if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3849		return -EINVAL;
3850	minor = simple_strtoul(e+1, &e, 10);
3851	if (*e && *e != '\n')
3852		return -EINVAL;
3853	dev = MKDEV(major, minor);
3854	if (major != MAJOR(dev) ||
3855	    minor != MINOR(dev))
3856		return -EOVERFLOW;
3857
3858
3859	if (mddev->persistent) {
3860		rdev = md_import_device(dev, mddev->major_version,
3861					mddev->minor_version);
3862		if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
3863			mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3864						       mdk_rdev_t, same_set);
 
3865			err = super_types[mddev->major_version]
3866				.load_super(rdev, rdev0, mddev->minor_version);
3867			if (err < 0)
3868				goto out;
3869		}
3870	} else if (mddev->external)
3871		rdev = md_import_device(dev, -2, -1);
3872	else
3873		rdev = md_import_device(dev, -1, -1);
3874
3875	if (IS_ERR(rdev))
3876		return PTR_ERR(rdev);
3877	err = bind_rdev_to_array(rdev, mddev);
3878 out:
3879	if (err)
3880		export_rdev(rdev);
3881	return err ? err : len;
3882}
3883
3884static struct md_sysfs_entry md_new_device =
3885__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
3886
3887static ssize_t
3888bitmap_store(mddev_t *mddev, const char *buf, size_t len)
3889{
3890	char *end;
3891	unsigned long chunk, end_chunk;
3892
3893	if (!mddev->bitmap)
3894		goto out;
3895	/* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3896	while (*buf) {
3897		chunk = end_chunk = simple_strtoul(buf, &end, 0);
3898		if (buf == end) break;
3899		if (*end == '-') { /* range */
3900			buf = end + 1;
3901			end_chunk = simple_strtoul(buf, &end, 0);
3902			if (buf == end) break;
3903		}
3904		if (*end && !isspace(*end)) break;
3905		bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
3906		buf = skip_spaces(end);
3907	}
3908	bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3909out:
3910	return len;
3911}
3912
3913static struct md_sysfs_entry md_bitmap =
3914__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3915
3916static ssize_t
3917size_show(mddev_t *mddev, char *page)
3918{
3919	return sprintf(page, "%llu\n",
3920		(unsigned long long)mddev->dev_sectors / 2);
3921}
3922
3923static int update_size(mddev_t *mddev, sector_t num_sectors);
3924
3925static ssize_t
3926size_store(mddev_t *mddev, const char *buf, size_t len)
3927{
3928	/* If array is inactive, we can reduce the component size, but
3929	 * not increase it (except from 0).
3930	 * If array is active, we can try an on-line resize
3931	 */
3932	sector_t sectors;
3933	int err = strict_blocks_to_sectors(buf, &sectors);
3934
3935	if (err < 0)
3936		return err;
3937	if (mddev->pers) {
3938		err = update_size(mddev, sectors);
3939		md_update_sb(mddev, 1);
3940	} else {
3941		if (mddev->dev_sectors == 0 ||
3942		    mddev->dev_sectors > sectors)
3943			mddev->dev_sectors = sectors;
3944		else
3945			err = -ENOSPC;
3946	}
3947	return err ? err : len;
3948}
3949
3950static struct md_sysfs_entry md_size =
3951__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
3952
3953
3954/* Metdata version.
3955 * This is one of
3956 *   'none' for arrays with no metadata (good luck...)
3957 *   'external' for arrays with externally managed metadata,
3958 * or N.M for internally known formats
3959 */
3960static ssize_t
3961metadata_show(mddev_t *mddev, char *page)
3962{
3963	if (mddev->persistent)
3964		return sprintf(page, "%d.%d\n",
3965			       mddev->major_version, mddev->minor_version);
3966	else if (mddev->external)
3967		return sprintf(page, "external:%s\n", mddev->metadata_type);
3968	else
3969		return sprintf(page, "none\n");
3970}
3971
3972static ssize_t
3973metadata_store(mddev_t *mddev, const char *buf, size_t len)
3974{
3975	int major, minor;
3976	char *e;
3977	/* Changing the details of 'external' metadata is
3978	 * always permitted.  Otherwise there must be
3979	 * no devices attached to the array.
3980	 */
3981	if (mddev->external && strncmp(buf, "external:", 9) == 0)
3982		;
3983	else if (!list_empty(&mddev->disks))
3984		return -EBUSY;
3985
3986	if (cmd_match(buf, "none")) {
3987		mddev->persistent = 0;
3988		mddev->external = 0;
3989		mddev->major_version = 0;
3990		mddev->minor_version = 90;
3991		return len;
3992	}
3993	if (strncmp(buf, "external:", 9) == 0) {
3994		size_t namelen = len-9;
3995		if (namelen >= sizeof(mddev->metadata_type))
3996			namelen = sizeof(mddev->metadata_type)-1;
3997		strncpy(mddev->metadata_type, buf+9, namelen);
3998		mddev->metadata_type[namelen] = 0;
3999		if (namelen && mddev->metadata_type[namelen-1] == '\n')
4000			mddev->metadata_type[--namelen] = 0;
4001		mddev->persistent = 0;
4002		mddev->external = 1;
4003		mddev->major_version = 0;
4004		mddev->minor_version = 90;
4005		return len;
4006	}
4007	major = simple_strtoul(buf, &e, 10);
4008	if (e==buf || *e != '.')
4009		return -EINVAL;
4010	buf = e+1;
4011	minor = simple_strtoul(buf, &e, 10);
4012	if (e==buf || (*e && *e != '\n') )
4013		return -EINVAL;
4014	if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
4015		return -ENOENT;
4016	mddev->major_version = major;
4017	mddev->minor_version = minor;
4018	mddev->persistent = 1;
4019	mddev->external = 0;
4020	return len;
4021}
4022
4023static struct md_sysfs_entry md_metadata =
4024__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
4025
4026static ssize_t
4027action_show(mddev_t *mddev, char *page)
4028{
4029	char *type = "idle";
4030	if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4031		type = "frozen";
4032	else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4033	    (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
4034		if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4035			type = "reshape";
4036		else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4037			if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4038				type = "resync";
4039			else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4040				type = "check";
4041			else
4042				type = "repair";
4043		} else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
4044			type = "recover";
4045	}
4046	return sprintf(page, "%s\n", type);
4047}
4048
4049static void reap_sync_thread(mddev_t *mddev);
4050
4051static ssize_t
4052action_store(mddev_t *mddev, const char *page, size_t len)
4053{
4054	if (!mddev->pers || !mddev->pers->sync_request)
4055		return -EINVAL;
4056
4057	if (cmd_match(page, "frozen"))
4058		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4059	else
4060		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4061
4062	if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
4063		if (mddev->sync_thread) {
4064			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4065			reap_sync_thread(mddev);
4066		}
4067	} else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4068		   test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
4069		return -EBUSY;
4070	else if (cmd_match(page, "resync"))
4071		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4072	else if (cmd_match(page, "recover")) {
4073		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4074		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4075	} else if (cmd_match(page, "reshape")) {
4076		int err;
4077		if (mddev->pers->start_reshape == NULL)
4078			return -EINVAL;
4079		err = mddev->pers->start_reshape(mddev);
4080		if (err)
4081			return err;
4082		sysfs_notify(&mddev->kobj, NULL, "degraded");
4083	} else {
4084		if (cmd_match(page, "check"))
4085			set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4086		else if (!cmd_match(page, "repair"))
4087			return -EINVAL;
4088		set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4089		set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4090	}
4091	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4092	md_wakeup_thread(mddev->thread);
4093	sysfs_notify_dirent_safe(mddev->sysfs_action);
4094	return len;
4095}
4096
4097static ssize_t
4098mismatch_cnt_show(mddev_t *mddev, char *page)
4099{
4100	return sprintf(page, "%llu\n",
4101		       (unsigned long long) mddev->resync_mismatches);
4102}
4103
4104static struct md_sysfs_entry md_scan_mode =
4105__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
4106
4107
4108static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
4109
4110static ssize_t
4111sync_min_show(mddev_t *mddev, char *page)
4112{
4113	return sprintf(page, "%d (%s)\n", speed_min(mddev),
4114		       mddev->sync_speed_min ? "local": "system");
4115}
4116
4117static ssize_t
4118sync_min_store(mddev_t *mddev, const char *buf, size_t len)
4119{
4120	int min;
4121	char *e;
4122	if (strncmp(buf, "system", 6)==0) {
4123		mddev->sync_speed_min = 0;
4124		return len;
4125	}
4126	min = simple_strtoul(buf, &e, 10);
4127	if (buf == e || (*e && *e != '\n') || min <= 0)
4128		return -EINVAL;
4129	mddev->sync_speed_min = min;
4130	return len;
4131}
4132
4133static struct md_sysfs_entry md_sync_min =
4134__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4135
4136static ssize_t
4137sync_max_show(mddev_t *mddev, char *page)
4138{
4139	return sprintf(page, "%d (%s)\n", speed_max(mddev),
4140		       mddev->sync_speed_max ? "local": "system");
4141}
4142
4143static ssize_t
4144sync_max_store(mddev_t *mddev, const char *buf, size_t len)
4145{
4146	int max;
4147	char *e;
4148	if (strncmp(buf, "system", 6)==0) {
4149		mddev->sync_speed_max = 0;
4150		return len;
4151	}
4152	max = simple_strtoul(buf, &e, 10);
4153	if (buf == e || (*e && *e != '\n') || max <= 0)
4154		return -EINVAL;
4155	mddev->sync_speed_max = max;
4156	return len;
4157}
4158
4159static struct md_sysfs_entry md_sync_max =
4160__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4161
4162static ssize_t
4163degraded_show(mddev_t *mddev, char *page)
4164{
4165	return sprintf(page, "%d\n", mddev->degraded);
4166}
4167static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
4168
4169static ssize_t
4170sync_force_parallel_show(mddev_t *mddev, char *page)
4171{
4172	return sprintf(page, "%d\n", mddev->parallel_resync);
4173}
4174
4175static ssize_t
4176sync_force_parallel_store(mddev_t *mddev, const char *buf, size_t len)
4177{
4178	long n;
4179
4180	if (strict_strtol(buf, 10, &n))
4181		return -EINVAL;
4182
4183	if (n != 0 && n != 1)
4184		return -EINVAL;
4185
4186	mddev->parallel_resync = n;
4187
4188	if (mddev->sync_thread)
4189		wake_up(&resync_wait);
4190
4191	return len;
4192}
4193
4194/* force parallel resync, even with shared block devices */
4195static struct md_sysfs_entry md_sync_force_parallel =
4196__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4197       sync_force_parallel_show, sync_force_parallel_store);
4198
4199static ssize_t
4200sync_speed_show(mddev_t *mddev, char *page)
4201{
4202	unsigned long resync, dt, db;
4203	if (mddev->curr_resync == 0)
4204		return sprintf(page, "none\n");
4205	resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4206	dt = (jiffies - mddev->resync_mark) / HZ;
4207	if (!dt) dt++;
4208	db = resync - mddev->resync_mark_cnt;
4209	return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
4210}
4211
4212static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
4213
4214static ssize_t
4215sync_completed_show(mddev_t *mddev, char *page)
4216{
4217	unsigned long long max_sectors, resync;
4218
4219	if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4220		return sprintf(page, "none\n");
4221
4222	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
 
4223		max_sectors = mddev->resync_max_sectors;
4224	else
4225		max_sectors = mddev->dev_sectors;
4226
4227	resync = mddev->curr_resync_completed;
4228	return sprintf(page, "%llu / %llu\n", resync, max_sectors);
4229}
4230
4231static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
4232
4233static ssize_t
4234min_sync_show(mddev_t *mddev, char *page)
4235{
4236	return sprintf(page, "%llu\n",
4237		       (unsigned long long)mddev->resync_min);
4238}
4239static ssize_t
4240min_sync_store(mddev_t *mddev, const char *buf, size_t len)
4241{
4242	unsigned long long min;
4243	if (strict_strtoull(buf, 10, &min))
4244		return -EINVAL;
4245	if (min > mddev->resync_max)
4246		return -EINVAL;
4247	if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4248		return -EBUSY;
4249
4250	/* Must be a multiple of chunk_size */
4251	if (mddev->chunk_sectors) {
4252		sector_t temp = min;
4253		if (sector_div(temp, mddev->chunk_sectors))
4254			return -EINVAL;
4255	}
4256	mddev->resync_min = min;
4257
4258	return len;
4259}
4260
4261static struct md_sysfs_entry md_min_sync =
4262__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4263
4264static ssize_t
4265max_sync_show(mddev_t *mddev, char *page)
4266{
4267	if (mddev->resync_max == MaxSector)
4268		return sprintf(page, "max\n");
4269	else
4270		return sprintf(page, "%llu\n",
4271			       (unsigned long long)mddev->resync_max);
4272}
4273static ssize_t
4274max_sync_store(mddev_t *mddev, const char *buf, size_t len)
4275{
4276	if (strncmp(buf, "max", 3) == 0)
4277		mddev->resync_max = MaxSector;
4278	else {
4279		unsigned long long max;
4280		if (strict_strtoull(buf, 10, &max))
4281			return -EINVAL;
4282		if (max < mddev->resync_min)
4283			return -EINVAL;
4284		if (max < mddev->resync_max &&
4285		    mddev->ro == 0 &&
4286		    test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4287			return -EBUSY;
4288
4289		/* Must be a multiple of chunk_size */
4290		if (mddev->chunk_sectors) {
4291			sector_t temp = max;
4292			if (sector_div(temp, mddev->chunk_sectors))
4293				return -EINVAL;
4294		}
4295		mddev->resync_max = max;
4296	}
4297	wake_up(&mddev->recovery_wait);
4298	return len;
4299}
4300
4301static struct md_sysfs_entry md_max_sync =
4302__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4303
4304static ssize_t
4305suspend_lo_show(mddev_t *mddev, char *page)
4306{
4307	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4308}
4309
4310static ssize_t
4311suspend_lo_store(mddev_t *mddev, const char *buf, size_t len)
4312{
4313	char *e;
4314	unsigned long long new = simple_strtoull(buf, &e, 10);
4315	unsigned long long old = mddev->suspend_lo;
4316
4317	if (mddev->pers == NULL || 
4318	    mddev->pers->quiesce == NULL)
4319		return -EINVAL;
4320	if (buf == e || (*e && *e != '\n'))
4321		return -EINVAL;
4322
4323	mddev->suspend_lo = new;
4324	if (new >= old)
4325		/* Shrinking suspended region */
4326		mddev->pers->quiesce(mddev, 2);
4327	else {
4328		/* Expanding suspended region - need to wait */
4329		mddev->pers->quiesce(mddev, 1);
4330		mddev->pers->quiesce(mddev, 0);
4331	}
4332	return len;
4333}
4334static struct md_sysfs_entry md_suspend_lo =
4335__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4336
4337
4338static ssize_t
4339suspend_hi_show(mddev_t *mddev, char *page)
4340{
4341	return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4342}
4343
4344static ssize_t
4345suspend_hi_store(mddev_t *mddev, const char *buf, size_t len)
4346{
4347	char *e;
4348	unsigned long long new = simple_strtoull(buf, &e, 10);
4349	unsigned long long old = mddev->suspend_hi;
4350
4351	if (mddev->pers == NULL ||
4352	    mddev->pers->quiesce == NULL)
4353		return -EINVAL;
4354	if (buf == e || (*e && *e != '\n'))
4355		return -EINVAL;
4356
4357	mddev->suspend_hi = new;
4358	if (new <= old)
4359		/* Shrinking suspended region */
4360		mddev->pers->quiesce(mddev, 2);
4361	else {
4362		/* Expanding suspended region - need to wait */
4363		mddev->pers->quiesce(mddev, 1);
4364		mddev->pers->quiesce(mddev, 0);
4365	}
4366	return len;
4367}
4368static struct md_sysfs_entry md_suspend_hi =
4369__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4370
4371static ssize_t
4372reshape_position_show(mddev_t *mddev, char *page)
4373{
4374	if (mddev->reshape_position != MaxSector)
4375		return sprintf(page, "%llu\n",
4376			       (unsigned long long)mddev->reshape_position);
4377	strcpy(page, "none\n");
4378	return 5;
4379}
4380
4381static ssize_t
4382reshape_position_store(mddev_t *mddev, const char *buf, size_t len)
4383{
 
4384	char *e;
4385	unsigned long long new = simple_strtoull(buf, &e, 10);
4386	if (mddev->pers)
4387		return -EBUSY;
4388	if (buf == e || (*e && *e != '\n'))
4389		return -EINVAL;
4390	mddev->reshape_position = new;
4391	mddev->delta_disks = 0;
 
4392	mddev->new_level = mddev->level;
4393	mddev->new_layout = mddev->layout;
4394	mddev->new_chunk_sectors = mddev->chunk_sectors;
 
 
4395	return len;
4396}
4397
4398static struct md_sysfs_entry md_reshape_position =
4399__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4400       reshape_position_store);
4401
4402static ssize_t
4403array_size_show(mddev_t *mddev, char *page)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4404{
4405	if (mddev->external_size)
4406		return sprintf(page, "%llu\n",
4407			       (unsigned long long)mddev->array_sectors/2);
4408	else
4409		return sprintf(page, "default\n");
4410}
4411
4412static ssize_t
4413array_size_store(mddev_t *mddev, const char *buf, size_t len)
4414{
4415	sector_t sectors;
4416
4417	if (strncmp(buf, "default", 7) == 0) {
4418		if (mddev->pers)
4419			sectors = mddev->pers->size(mddev, 0, 0);
4420		else
4421			sectors = mddev->array_sectors;
4422
4423		mddev->external_size = 0;
4424	} else {
4425		if (strict_blocks_to_sectors(buf, &sectors) < 0)
4426			return -EINVAL;
4427		if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4428			return -E2BIG;
4429
4430		mddev->external_size = 1;
4431	}
4432
4433	mddev->array_sectors = sectors;
4434	if (mddev->pers) {
4435		set_capacity(mddev->gendisk, mddev->array_sectors);
4436		revalidate_disk(mddev->gendisk);
4437	}
4438	return len;
4439}
4440
4441static struct md_sysfs_entry md_array_size =
4442__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4443       array_size_store);
4444
4445static struct attribute *md_default_attrs[] = {
4446	&md_level.attr,
4447	&md_layout.attr,
4448	&md_raid_disks.attr,
4449	&md_chunk_size.attr,
4450	&md_size.attr,
4451	&md_resync_start.attr,
4452	&md_metadata.attr,
4453	&md_new_device.attr,
4454	&md_safe_delay.attr,
4455	&md_array_state.attr,
4456	&md_reshape_position.attr,
 
4457	&md_array_size.attr,
4458	&max_corr_read_errors.attr,
4459	NULL,
4460};
4461
4462static struct attribute *md_redundancy_attrs[] = {
4463	&md_scan_mode.attr,
4464	&md_mismatches.attr,
4465	&md_sync_min.attr,
4466	&md_sync_max.attr,
4467	&md_sync_speed.attr,
4468	&md_sync_force_parallel.attr,
4469	&md_sync_completed.attr,
4470	&md_min_sync.attr,
4471	&md_max_sync.attr,
4472	&md_suspend_lo.attr,
4473	&md_suspend_hi.attr,
4474	&md_bitmap.attr,
4475	&md_degraded.attr,
4476	NULL,
4477};
4478static struct attribute_group md_redundancy_group = {
4479	.name = NULL,
4480	.attrs = md_redundancy_attrs,
4481};
4482
4483
4484static ssize_t
4485md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4486{
4487	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4488	mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4489	ssize_t rv;
4490
4491	if (!entry->show)
4492		return -EIO;
 
 
 
 
 
 
 
 
4493	rv = mddev_lock(mddev);
4494	if (!rv) {
4495		rv = entry->show(mddev, page);
4496		mddev_unlock(mddev);
4497	}
 
4498	return rv;
4499}
4500
4501static ssize_t
4502md_attr_store(struct kobject *kobj, struct attribute *attr,
4503	      const char *page, size_t length)
4504{
4505	struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
4506	mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
4507	ssize_t rv;
4508
4509	if (!entry->store)
4510		return -EIO;
4511	if (!capable(CAP_SYS_ADMIN))
4512		return -EACCES;
 
 
 
 
 
 
 
4513	rv = mddev_lock(mddev);
4514	if (mddev->hold_active == UNTIL_IOCTL)
4515		mddev->hold_active = 0;
4516	if (!rv) {
4517		rv = entry->store(mddev, page, length);
4518		mddev_unlock(mddev);
4519	}
 
4520	return rv;
4521}
4522
4523static void md_free(struct kobject *ko)
4524{
4525	mddev_t *mddev = container_of(ko, mddev_t, kobj);
4526
4527	if (mddev->sysfs_state)
4528		sysfs_put(mddev->sysfs_state);
4529
4530	if (mddev->gendisk) {
4531		del_gendisk(mddev->gendisk);
4532		put_disk(mddev->gendisk);
4533	}
4534	if (mddev->queue)
4535		blk_cleanup_queue(mddev->queue);
4536
4537	kfree(mddev);
4538}
4539
4540static const struct sysfs_ops md_sysfs_ops = {
4541	.show	= md_attr_show,
4542	.store	= md_attr_store,
4543};
4544static struct kobj_type md_ktype = {
4545	.release	= md_free,
4546	.sysfs_ops	= &md_sysfs_ops,
4547	.default_attrs	= md_default_attrs,
4548};
4549
4550int mdp_major = 0;
4551
4552static void mddev_delayed_delete(struct work_struct *ws)
4553{
4554	mddev_t *mddev = container_of(ws, mddev_t, del_work);
4555
4556	sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
4557	kobject_del(&mddev->kobj);
4558	kobject_put(&mddev->kobj);
4559}
4560
4561static int md_alloc(dev_t dev, char *name)
4562{
4563	static DEFINE_MUTEX(disks_mutex);
4564	mddev_t *mddev = mddev_find(dev);
4565	struct gendisk *disk;
4566	int partitioned;
4567	int shift;
4568	int unit;
4569	int error;
4570
4571	if (!mddev)
4572		return -ENODEV;
4573
4574	partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4575	shift = partitioned ? MdpMinorShift : 0;
4576	unit = MINOR(mddev->unit) >> shift;
4577
4578	/* wait for any previous instance of this device to be
4579	 * completely removed (mddev_delayed_delete).
4580	 */
4581	flush_workqueue(md_misc_wq);
4582
4583	mutex_lock(&disks_mutex);
4584	error = -EEXIST;
4585	if (mddev->gendisk)
4586		goto abort;
4587
4588	if (name) {
4589		/* Need to ensure that 'name' is not a duplicate.
4590		 */
4591		mddev_t *mddev2;
4592		spin_lock(&all_mddevs_lock);
4593
4594		list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4595			if (mddev2->gendisk &&
4596			    strcmp(mddev2->gendisk->disk_name, name) == 0) {
4597				spin_unlock(&all_mddevs_lock);
4598				goto abort;
4599			}
4600		spin_unlock(&all_mddevs_lock);
4601	}
4602
4603	error = -ENOMEM;
4604	mddev->queue = blk_alloc_queue(GFP_KERNEL);
4605	if (!mddev->queue)
4606		goto abort;
4607	mddev->queue->queuedata = mddev;
4608
4609	blk_queue_make_request(mddev->queue, md_make_request);
 
4610
4611	disk = alloc_disk(1 << shift);
4612	if (!disk) {
4613		blk_cleanup_queue(mddev->queue);
4614		mddev->queue = NULL;
4615		goto abort;
4616	}
4617	disk->major = MAJOR(mddev->unit);
4618	disk->first_minor = unit << shift;
4619	if (name)
4620		strcpy(disk->disk_name, name);
4621	else if (partitioned)
4622		sprintf(disk->disk_name, "md_d%d", unit);
4623	else
4624		sprintf(disk->disk_name, "md%d", unit);
4625	disk->fops = &md_fops;
4626	disk->private_data = mddev;
4627	disk->queue = mddev->queue;
4628	blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
4629	/* Allow extended partitions.  This makes the
4630	 * 'mdp' device redundant, but we can't really
4631	 * remove it now.
4632	 */
4633	disk->flags |= GENHD_FL_EXT_DEVT;
4634	mddev->gendisk = disk;
4635	/* As soon as we call add_disk(), another thread could get
4636	 * through to md_open, so make sure it doesn't get too far
4637	 */
4638	mutex_lock(&mddev->open_mutex);
4639	add_disk(disk);
4640
4641	error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4642				     &disk_to_dev(disk)->kobj, "%s", "md");
4643	if (error) {
4644		/* This isn't possible, but as kobject_init_and_add is marked
4645		 * __must_check, we must do something with the result
4646		 */
4647		printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4648		       disk->disk_name);
4649		error = 0;
4650	}
4651	if (mddev->kobj.sd &&
4652	    sysfs_create_group(&mddev->kobj, &md_bitmap_group))
4653		printk(KERN_DEBUG "pointless warning\n");
4654	mutex_unlock(&mddev->open_mutex);
4655 abort:
4656	mutex_unlock(&disks_mutex);
4657	if (!error && mddev->kobj.sd) {
4658		kobject_uevent(&mddev->kobj, KOBJ_ADD);
4659		mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
4660	}
4661	mddev_put(mddev);
4662	return error;
4663}
4664
4665static struct kobject *md_probe(dev_t dev, int *part, void *data)
4666{
4667	md_alloc(dev, NULL);
4668	return NULL;
4669}
4670
4671static int add_named_array(const char *val, struct kernel_param *kp)
4672{
4673	/* val must be "md_*" where * is not all digits.
4674	 * We allocate an array with a large free minor number, and
4675	 * set the name to val.  val must not already be an active name.
4676	 */
4677	int len = strlen(val);
4678	char buf[DISK_NAME_LEN];
4679
4680	while (len && val[len-1] == '\n')
4681		len--;
4682	if (len >= DISK_NAME_LEN)
4683		return -E2BIG;
4684	strlcpy(buf, val, len+1);
4685	if (strncmp(buf, "md_", 3) != 0)
4686		return -EINVAL;
4687	return md_alloc(0, buf);
4688}
4689
4690static void md_safemode_timeout(unsigned long data)
4691{
4692	mddev_t *mddev = (mddev_t *) data;
4693
4694	if (!atomic_read(&mddev->writes_pending)) {
4695		mddev->safemode = 1;
4696		if (mddev->external)
4697			sysfs_notify_dirent_safe(mddev->sysfs_state);
4698	}
4699	md_wakeup_thread(mddev->thread);
4700}
4701
4702static int start_dirty_degraded;
4703
4704int md_run(mddev_t *mddev)
4705{
4706	int err;
4707	mdk_rdev_t *rdev;
4708	struct mdk_personality *pers;
4709
4710	if (list_empty(&mddev->disks))
4711		/* cannot run an array with no devices.. */
4712		return -EINVAL;
4713
4714	if (mddev->pers)
4715		return -EBUSY;
4716	/* Cannot run until previous stop completes properly */
4717	if (mddev->sysfs_active)
4718		return -EBUSY;
4719
4720	/*
4721	 * Analyze all RAID superblock(s)
4722	 */
4723	if (!mddev->raid_disks) {
4724		if (!mddev->persistent)
4725			return -EINVAL;
4726		analyze_sbs(mddev);
4727	}
4728
4729	if (mddev->level != LEVEL_NONE)
4730		request_module("md-level-%d", mddev->level);
4731	else if (mddev->clevel[0])
4732		request_module("md-%s", mddev->clevel);
4733
4734	/*
4735	 * Drop all container device buffers, from now on
4736	 * the only valid external interface is through the md
4737	 * device.
4738	 */
4739	list_for_each_entry(rdev, &mddev->disks, same_set) {
4740		if (test_bit(Faulty, &rdev->flags))
4741			continue;
4742		sync_blockdev(rdev->bdev);
4743		invalidate_bdev(rdev->bdev);
4744
4745		/* perform some consistency tests on the device.
4746		 * We don't want the data to overlap the metadata,
4747		 * Internal Bitmap issues have been handled elsewhere.
4748		 */
4749		if (rdev->meta_bdev) {
4750			/* Nothing to check */;
4751		} else if (rdev->data_offset < rdev->sb_start) {
4752			if (mddev->dev_sectors &&
4753			    rdev->data_offset + mddev->dev_sectors
4754			    > rdev->sb_start) {
4755				printk("md: %s: data overlaps metadata\n",
4756				       mdname(mddev));
4757				return -EINVAL;
4758			}
4759		} else {
4760			if (rdev->sb_start + rdev->sb_size/512
4761			    > rdev->data_offset) {
4762				printk("md: %s: metadata overlaps data\n",
4763				       mdname(mddev));
4764				return -EINVAL;
4765			}
4766		}
4767		sysfs_notify_dirent_safe(rdev->sysfs_state);
4768	}
4769
4770	if (mddev->bio_set == NULL)
4771		mddev->bio_set = bioset_create(BIO_POOL_SIZE,
4772					       sizeof(mddev_t *));
4773
4774	spin_lock(&pers_lock);
4775	pers = find_pers(mddev->level, mddev->clevel);
4776	if (!pers || !try_module_get(pers->owner)) {
4777		spin_unlock(&pers_lock);
4778		if (mddev->level != LEVEL_NONE)
4779			printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
4780			       mddev->level);
4781		else
4782			printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
4783			       mddev->clevel);
4784		return -EINVAL;
4785	}
4786	mddev->pers = pers;
4787	spin_unlock(&pers_lock);
4788	if (mddev->level != pers->level) {
4789		mddev->level = pers->level;
4790		mddev->new_level = pers->level;
4791	}
4792	strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
4793
4794	if (mddev->reshape_position != MaxSector &&
4795	    pers->start_reshape == NULL) {
4796		/* This personality cannot handle reshaping... */
4797		mddev->pers = NULL;
4798		module_put(pers->owner);
4799		return -EINVAL;
4800	}
4801
4802	if (pers->sync_request) {
4803		/* Warn if this is a potentially silly
4804		 * configuration.
4805		 */
4806		char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
4807		mdk_rdev_t *rdev2;
4808		int warned = 0;
4809
4810		list_for_each_entry(rdev, &mddev->disks, same_set)
4811			list_for_each_entry(rdev2, &mddev->disks, same_set) {
4812				if (rdev < rdev2 &&
4813				    rdev->bdev->bd_contains ==
4814				    rdev2->bdev->bd_contains) {
4815					printk(KERN_WARNING
4816					       "%s: WARNING: %s appears to be"
4817					       " on the same physical disk as"
4818					       " %s.\n",
4819					       mdname(mddev),
4820					       bdevname(rdev->bdev,b),
4821					       bdevname(rdev2->bdev,b2));
4822					warned = 1;
4823				}
4824			}
4825
4826		if (warned)
4827			printk(KERN_WARNING
4828			       "True protection against single-disk"
4829			       " failure might be compromised.\n");
4830	}
4831
4832	mddev->recovery = 0;
4833	/* may be over-ridden by personality */
4834	mddev->resync_max_sectors = mddev->dev_sectors;
4835
4836	mddev->ok_start_degraded = start_dirty_degraded;
4837
4838	if (start_readonly && mddev->ro == 0)
4839		mddev->ro = 2; /* read-only, but switch on first write */
4840
4841	err = mddev->pers->run(mddev);
4842	if (err)
4843		printk(KERN_ERR "md: pers->run() failed ...\n");
4844	else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
4845		WARN_ONCE(!mddev->external_size, "%s: default size too small,"
4846			  " but 'external_size' not in effect?\n", __func__);
4847		printk(KERN_ERR
4848		       "md: invalid array_size %llu > default size %llu\n",
4849		       (unsigned long long)mddev->array_sectors / 2,
4850		       (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
4851		err = -EINVAL;
4852		mddev->pers->stop(mddev);
4853	}
4854	if (err == 0 && mddev->pers->sync_request) {
 
4855		err = bitmap_create(mddev);
4856		if (err) {
4857			printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
4858			       mdname(mddev), err);
4859			mddev->pers->stop(mddev);
4860		}
4861	}
4862	if (err) {
4863		module_put(mddev->pers->owner);
4864		mddev->pers = NULL;
4865		bitmap_destroy(mddev);
4866		return err;
4867	}
4868	if (mddev->pers->sync_request) {
4869		if (mddev->kobj.sd &&
4870		    sysfs_create_group(&mddev->kobj, &md_redundancy_group))
4871			printk(KERN_WARNING
4872			       "md: cannot register extra attributes for %s\n",
4873			       mdname(mddev));
4874		mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
4875	} else if (mddev->ro == 2) /* auto-readonly not meaningful */
4876		mddev->ro = 0;
4877
4878 	atomic_set(&mddev->writes_pending,0);
4879	atomic_set(&mddev->max_corr_read_errors,
4880		   MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
4881	mddev->safemode = 0;
4882	mddev->safemode_timer.function = md_safemode_timeout;
4883	mddev->safemode_timer.data = (unsigned long) mddev;
4884	mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
4885	mddev->in_sync = 1;
4886	smp_wmb();
4887	mddev->ready = 1;
4888	list_for_each_entry(rdev, &mddev->disks, same_set)
4889		if (rdev->raid_disk >= 0)
4890			if (sysfs_link_rdev(mddev, rdev))
4891				/* failure here is OK */;
4892	
4893	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4894	
4895	if (mddev->flags)
4896		md_update_sb(mddev, 0);
4897
4898	md_new_event(mddev);
4899	sysfs_notify_dirent_safe(mddev->sysfs_state);
4900	sysfs_notify_dirent_safe(mddev->sysfs_action);
4901	sysfs_notify(&mddev->kobj, NULL, "degraded");
4902	return 0;
4903}
4904EXPORT_SYMBOL_GPL(md_run);
4905
4906static int do_md_run(mddev_t *mddev)
4907{
4908	int err;
4909
4910	err = md_run(mddev);
4911	if (err)
4912		goto out;
4913	err = bitmap_load(mddev);
4914	if (err) {
4915		bitmap_destroy(mddev);
4916		goto out;
4917	}
4918
4919	md_wakeup_thread(mddev->thread);
4920	md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
4921
4922	set_capacity(mddev->gendisk, mddev->array_sectors);
4923	revalidate_disk(mddev->gendisk);
4924	mddev->changed = 1;
4925	kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
4926out:
4927	return err;
4928}
4929
4930static int restart_array(mddev_t *mddev)
4931{
4932	struct gendisk *disk = mddev->gendisk;
4933
4934	/* Complain if it has no devices */
4935	if (list_empty(&mddev->disks))
4936		return -ENXIO;
4937	if (!mddev->pers)
4938		return -EINVAL;
4939	if (!mddev->ro)
4940		return -EBUSY;
4941	mddev->safemode = 0;
4942	mddev->ro = 0;
4943	set_disk_ro(disk, 0);
4944	printk(KERN_INFO "md: %s switched to read-write mode.\n",
4945		mdname(mddev));
4946	/* Kick recovery or resync if necessary */
4947	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4948	md_wakeup_thread(mddev->thread);
4949	md_wakeup_thread(mddev->sync_thread);
4950	sysfs_notify_dirent_safe(mddev->sysfs_state);
4951	return 0;
4952}
4953
4954/* similar to deny_write_access, but accounts for our holding a reference
4955 * to the file ourselves */
4956static int deny_bitmap_write_access(struct file * file)
4957{
4958	struct inode *inode = file->f_mapping->host;
4959
4960	spin_lock(&inode->i_lock);
4961	if (atomic_read(&inode->i_writecount) > 1) {
4962		spin_unlock(&inode->i_lock);
4963		return -ETXTBSY;
4964	}
4965	atomic_set(&inode->i_writecount, -1);
4966	spin_unlock(&inode->i_lock);
4967
4968	return 0;
4969}
4970
4971void restore_bitmap_write_access(struct file *file)
4972{
4973	struct inode *inode = file->f_mapping->host;
4974
4975	spin_lock(&inode->i_lock);
4976	atomic_set(&inode->i_writecount, 1);
4977	spin_unlock(&inode->i_lock);
4978}
4979
4980static void md_clean(mddev_t *mddev)
4981{
4982	mddev->array_sectors = 0;
4983	mddev->external_size = 0;
4984	mddev->dev_sectors = 0;
4985	mddev->raid_disks = 0;
4986	mddev->recovery_cp = 0;
4987	mddev->resync_min = 0;
4988	mddev->resync_max = MaxSector;
4989	mddev->reshape_position = MaxSector;
4990	mddev->external = 0;
4991	mddev->persistent = 0;
4992	mddev->level = LEVEL_NONE;
4993	mddev->clevel[0] = 0;
4994	mddev->flags = 0;
4995	mddev->ro = 0;
4996	mddev->metadata_type[0] = 0;
4997	mddev->chunk_sectors = 0;
4998	mddev->ctime = mddev->utime = 0;
4999	mddev->layout = 0;
5000	mddev->max_disks = 0;
5001	mddev->events = 0;
5002	mddev->can_decrease_events = 0;
5003	mddev->delta_disks = 0;
 
5004	mddev->new_level = LEVEL_NONE;
5005	mddev->new_layout = 0;
5006	mddev->new_chunk_sectors = 0;
5007	mddev->curr_resync = 0;
5008	mddev->resync_mismatches = 0;
5009	mddev->suspend_lo = mddev->suspend_hi = 0;
5010	mddev->sync_speed_min = mddev->sync_speed_max = 0;
5011	mddev->recovery = 0;
5012	mddev->in_sync = 0;
5013	mddev->changed = 0;
5014	mddev->degraded = 0;
5015	mddev->safemode = 0;
 
5016	mddev->bitmap_info.offset = 0;
5017	mddev->bitmap_info.default_offset = 0;
 
5018	mddev->bitmap_info.chunksize = 0;
5019	mddev->bitmap_info.daemon_sleep = 0;
5020	mddev->bitmap_info.max_write_behind = 0;
5021}
5022
5023static void __md_stop_writes(mddev_t *mddev)
5024{
5025	if (mddev->sync_thread) {
5026		set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5027		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
5028		reap_sync_thread(mddev);
5029	}
5030
5031	del_timer_sync(&mddev->safemode_timer);
5032
5033	bitmap_flush(mddev);
5034	md_super_wait(mddev);
5035
5036	if (!mddev->in_sync || mddev->flags) {
5037		/* mark array as shutdown cleanly */
5038		mddev->in_sync = 1;
5039		md_update_sb(mddev, 1);
5040	}
5041}
5042
5043void md_stop_writes(mddev_t *mddev)
5044{
5045	mddev_lock(mddev);
5046	__md_stop_writes(mddev);
5047	mddev_unlock(mddev);
5048}
5049EXPORT_SYMBOL_GPL(md_stop_writes);
5050
5051void md_stop(mddev_t *mddev)
5052{
5053	mddev->ready = 0;
5054	mddev->pers->stop(mddev);
5055	if (mddev->pers->sync_request && mddev->to_remove == NULL)
5056		mddev->to_remove = &md_redundancy_group;
5057	module_put(mddev->pers->owner);
5058	mddev->pers = NULL;
5059	clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5060}
5061EXPORT_SYMBOL_GPL(md_stop);
5062
5063static int md_set_readonly(mddev_t *mddev, int is_open)
5064{
5065	int err = 0;
5066	mutex_lock(&mddev->open_mutex);
5067	if (atomic_read(&mddev->openers) > is_open) {
5068		printk("md: %s still in use.\n",mdname(mddev));
5069		err = -EBUSY;
5070		goto out;
5071	}
 
 
5072	if (mddev->pers) {
5073		__md_stop_writes(mddev);
5074
5075		err  = -ENXIO;
5076		if (mddev->ro==1)
5077			goto out;
5078		mddev->ro = 1;
5079		set_disk_ro(mddev->gendisk, 1);
5080		clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5081		sysfs_notify_dirent_safe(mddev->sysfs_state);
5082		err = 0;	
5083	}
5084out:
5085	mutex_unlock(&mddev->open_mutex);
5086	return err;
5087}
5088
5089/* mode:
5090 *   0 - completely stop and dis-assemble array
5091 *   2 - stop but do not disassemble array
5092 */
5093static int do_md_stop(mddev_t * mddev, int mode, int is_open)
 
5094{
5095	struct gendisk *disk = mddev->gendisk;
5096	mdk_rdev_t *rdev;
5097
5098	mutex_lock(&mddev->open_mutex);
5099	if (atomic_read(&mddev->openers) > is_open ||
5100	    mddev->sysfs_active) {
5101		printk("md: %s still in use.\n",mdname(mddev));
5102		mutex_unlock(&mddev->open_mutex);
5103		return -EBUSY;
5104	}
 
 
 
 
 
 
 
5105
5106	if (mddev->pers) {
5107		if (mddev->ro)
5108			set_disk_ro(disk, 0);
5109
5110		__md_stop_writes(mddev);
5111		md_stop(mddev);
5112		mddev->queue->merge_bvec_fn = NULL;
5113		mddev->queue->backing_dev_info.congested_fn = NULL;
5114
5115		/* tell userspace to handle 'inactive' */
5116		sysfs_notify_dirent_safe(mddev->sysfs_state);
5117
5118		list_for_each_entry(rdev, &mddev->disks, same_set)
5119			if (rdev->raid_disk >= 0)
5120				sysfs_unlink_rdev(mddev, rdev);
5121
5122		set_capacity(disk, 0);
5123		mutex_unlock(&mddev->open_mutex);
5124		mddev->changed = 1;
5125		revalidate_disk(disk);
5126
5127		if (mddev->ro)
5128			mddev->ro = 0;
5129	} else
5130		mutex_unlock(&mddev->open_mutex);
5131	/*
5132	 * Free resources if final stop
5133	 */
5134	if (mode == 0) {
5135		printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5136
5137		bitmap_destroy(mddev);
5138		if (mddev->bitmap_info.file) {
5139			restore_bitmap_write_access(mddev->bitmap_info.file);
5140			fput(mddev->bitmap_info.file);
5141			mddev->bitmap_info.file = NULL;
5142		}
5143		mddev->bitmap_info.offset = 0;
5144
5145		export_array(mddev);
5146
5147		md_clean(mddev);
5148		kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5149		if (mddev->hold_active == UNTIL_STOP)
5150			mddev->hold_active = 0;
5151	}
5152	blk_integrity_unregister(disk);
5153	md_new_event(mddev);
5154	sysfs_notify_dirent_safe(mddev->sysfs_state);
5155	return 0;
5156}
5157
5158#ifndef MODULE
5159static void autorun_array(mddev_t *mddev)
5160{
5161	mdk_rdev_t *rdev;
5162	int err;
5163
5164	if (list_empty(&mddev->disks))
5165		return;
5166
5167	printk(KERN_INFO "md: running: ");
5168
5169	list_for_each_entry(rdev, &mddev->disks, same_set) {
5170		char b[BDEVNAME_SIZE];
5171		printk("<%s>", bdevname(rdev->bdev,b));
5172	}
5173	printk("\n");
5174
5175	err = do_md_run(mddev);
5176	if (err) {
5177		printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
5178		do_md_stop(mddev, 0, 0);
5179	}
5180}
5181
5182/*
5183 * lets try to run arrays based on all disks that have arrived
5184 * until now. (those are in pending_raid_disks)
5185 *
5186 * the method: pick the first pending disk, collect all disks with
5187 * the same UUID, remove all from the pending list and put them into
5188 * the 'same_array' list. Then order this list based on superblock
5189 * update time (freshest comes first), kick out 'old' disks and
5190 * compare superblocks. If everything's fine then run it.
5191 *
5192 * If "unit" is allocated, then bump its reference count
5193 */
5194static void autorun_devices(int part)
5195{
5196	mdk_rdev_t *rdev0, *rdev, *tmp;
5197	mddev_t *mddev;
5198	char b[BDEVNAME_SIZE];
5199
5200	printk(KERN_INFO "md: autorun ...\n");
5201	while (!list_empty(&pending_raid_disks)) {
5202		int unit;
5203		dev_t dev;
5204		LIST_HEAD(candidates);
5205		rdev0 = list_entry(pending_raid_disks.next,
5206					 mdk_rdev_t, same_set);
5207
5208		printk(KERN_INFO "md: considering %s ...\n",
5209			bdevname(rdev0->bdev,b));
5210		INIT_LIST_HEAD(&candidates);
5211		rdev_for_each_list(rdev, tmp, &pending_raid_disks)
5212			if (super_90_load(rdev, rdev0, 0) >= 0) {
5213				printk(KERN_INFO "md:  adding %s ...\n",
5214					bdevname(rdev->bdev,b));
5215				list_move(&rdev->same_set, &candidates);
5216			}
5217		/*
5218		 * now we have a set of devices, with all of them having
5219		 * mostly sane superblocks. It's time to allocate the
5220		 * mddev.
5221		 */
5222		if (part) {
5223			dev = MKDEV(mdp_major,
5224				    rdev0->preferred_minor << MdpMinorShift);
5225			unit = MINOR(dev) >> MdpMinorShift;
5226		} else {
5227			dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5228			unit = MINOR(dev);
5229		}
5230		if (rdev0->preferred_minor != unit) {
5231			printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5232			       bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5233			break;
5234		}
5235
5236		md_probe(dev, NULL, NULL);
5237		mddev = mddev_find(dev);
5238		if (!mddev || !mddev->gendisk) {
5239			if (mddev)
5240				mddev_put(mddev);
5241			printk(KERN_ERR
5242				"md: cannot allocate memory for md drive.\n");
5243			break;
5244		}
5245		if (mddev_lock(mddev)) 
5246			printk(KERN_WARNING "md: %s locked, cannot run\n",
5247			       mdname(mddev));
5248		else if (mddev->raid_disks || mddev->major_version
5249			 || !list_empty(&mddev->disks)) {
5250			printk(KERN_WARNING 
5251				"md: %s already running, cannot run %s\n",
5252				mdname(mddev), bdevname(rdev0->bdev,b));
5253			mddev_unlock(mddev);
5254		} else {
5255			printk(KERN_INFO "md: created %s\n", mdname(mddev));
5256			mddev->persistent = 1;
5257			rdev_for_each_list(rdev, tmp, &candidates) {
5258				list_del_init(&rdev->same_set);
5259				if (bind_rdev_to_array(rdev, mddev))
5260					export_rdev(rdev);
5261			}
5262			autorun_array(mddev);
5263			mddev_unlock(mddev);
5264		}
5265		/* on success, candidates will be empty, on error
5266		 * it won't...
5267		 */
5268		rdev_for_each_list(rdev, tmp, &candidates) {
5269			list_del_init(&rdev->same_set);
5270			export_rdev(rdev);
5271		}
5272		mddev_put(mddev);
5273	}
5274	printk(KERN_INFO "md: ... autorun DONE.\n");
5275}
5276#endif /* !MODULE */
5277
5278static int get_version(void __user * arg)
5279{
5280	mdu_version_t ver;
5281
5282	ver.major = MD_MAJOR_VERSION;
5283	ver.minor = MD_MINOR_VERSION;
5284	ver.patchlevel = MD_PATCHLEVEL_VERSION;
5285
5286	if (copy_to_user(arg, &ver, sizeof(ver)))
5287		return -EFAULT;
5288
5289	return 0;
5290}
5291
5292static int get_array_info(mddev_t * mddev, void __user * arg)
5293{
5294	mdu_array_info_t info;
5295	int nr,working,insync,failed,spare;
5296	mdk_rdev_t *rdev;
5297
5298	nr=working=insync=failed=spare=0;
5299	list_for_each_entry(rdev, &mddev->disks, same_set) {
5300		nr++;
5301		if (test_bit(Faulty, &rdev->flags))
5302			failed++;
5303		else {
5304			working++;
5305			if (test_bit(In_sync, &rdev->flags))
5306				insync++;	
5307			else
5308				spare++;
5309		}
5310	}
5311
5312	info.major_version = mddev->major_version;
5313	info.minor_version = mddev->minor_version;
5314	info.patch_version = MD_PATCHLEVEL_VERSION;
5315	info.ctime         = mddev->ctime;
5316	info.level         = mddev->level;
5317	info.size          = mddev->dev_sectors / 2;
5318	if (info.size != mddev->dev_sectors / 2) /* overflow */
5319		info.size = -1;
5320	info.nr_disks      = nr;
5321	info.raid_disks    = mddev->raid_disks;
5322	info.md_minor      = mddev->md_minor;
5323	info.not_persistent= !mddev->persistent;
5324
5325	info.utime         = mddev->utime;
5326	info.state         = 0;
5327	if (mddev->in_sync)
5328		info.state = (1<<MD_SB_CLEAN);
5329	if (mddev->bitmap && mddev->bitmap_info.offset)
5330		info.state = (1<<MD_SB_BITMAP_PRESENT);
5331	info.active_disks  = insync;
5332	info.working_disks = working;
5333	info.failed_disks  = failed;
5334	info.spare_disks   = spare;
5335
5336	info.layout        = mddev->layout;
5337	info.chunk_size    = mddev->chunk_sectors << 9;
5338
5339	if (copy_to_user(arg, &info, sizeof(info)))
5340		return -EFAULT;
5341
5342	return 0;
5343}
5344
5345static int get_bitmap_file(mddev_t * mddev, void __user * arg)
5346{
5347	mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5348	char *ptr, *buf = NULL;
5349	int err = -ENOMEM;
5350
5351	if (md_allow_write(mddev))
5352		file = kmalloc(sizeof(*file), GFP_NOIO);
5353	else
5354		file = kmalloc(sizeof(*file), GFP_KERNEL);
5355
5356	if (!file)
5357		goto out;
5358
5359	/* bitmap disabled, zero the first byte and copy out */
5360	if (!mddev->bitmap || !mddev->bitmap->file) {
5361		file->pathname[0] = '\0';
5362		goto copy_out;
5363	}
5364
5365	buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5366	if (!buf)
5367		goto out;
5368
5369	ptr = d_path(&mddev->bitmap->file->f_path, buf, sizeof(file->pathname));
 
5370	if (IS_ERR(ptr))
5371		goto out;
5372
5373	strcpy(file->pathname, ptr);
5374
5375copy_out:
5376	err = 0;
5377	if (copy_to_user(arg, file, sizeof(*file)))
5378		err = -EFAULT;
5379out:
5380	kfree(buf);
5381	kfree(file);
5382	return err;
5383}
5384
5385static int get_disk_info(mddev_t * mddev, void __user * arg)
5386{
5387	mdu_disk_info_t info;
5388	mdk_rdev_t *rdev;
5389
5390	if (copy_from_user(&info, arg, sizeof(info)))
5391		return -EFAULT;
5392
5393	rdev = find_rdev_nr(mddev, info.number);
5394	if (rdev) {
5395		info.major = MAJOR(rdev->bdev->bd_dev);
5396		info.minor = MINOR(rdev->bdev->bd_dev);
5397		info.raid_disk = rdev->raid_disk;
5398		info.state = 0;
5399		if (test_bit(Faulty, &rdev->flags))
5400			info.state |= (1<<MD_DISK_FAULTY);
5401		else if (test_bit(In_sync, &rdev->flags)) {
5402			info.state |= (1<<MD_DISK_ACTIVE);
5403			info.state |= (1<<MD_DISK_SYNC);
5404		}
5405		if (test_bit(WriteMostly, &rdev->flags))
5406			info.state |= (1<<MD_DISK_WRITEMOSTLY);
5407	} else {
5408		info.major = info.minor = 0;
5409		info.raid_disk = -1;
5410		info.state = (1<<MD_DISK_REMOVED);
5411	}
5412
5413	if (copy_to_user(arg, &info, sizeof(info)))
5414		return -EFAULT;
5415
5416	return 0;
5417}
5418
5419static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
5420{
5421	char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
5422	mdk_rdev_t *rdev;
5423	dev_t dev = MKDEV(info->major,info->minor);
5424
5425	if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5426		return -EOVERFLOW;
5427
5428	if (!mddev->raid_disks) {
5429		int err;
5430		/* expecting a device which has a superblock */
5431		rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5432		if (IS_ERR(rdev)) {
5433			printk(KERN_WARNING 
5434				"md: md_import_device returned %ld\n",
5435				PTR_ERR(rdev));
5436			return PTR_ERR(rdev);
5437		}
5438		if (!list_empty(&mddev->disks)) {
5439			mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
5440							mdk_rdev_t, same_set);
 
5441			err = super_types[mddev->major_version]
5442				.load_super(rdev, rdev0, mddev->minor_version);
5443			if (err < 0) {
5444				printk(KERN_WARNING 
5445					"md: %s has different UUID to %s\n",
5446					bdevname(rdev->bdev,b), 
5447					bdevname(rdev0->bdev,b2));
5448				export_rdev(rdev);
5449				return -EINVAL;
5450			}
5451		}
5452		err = bind_rdev_to_array(rdev, mddev);
5453		if (err)
5454			export_rdev(rdev);
5455		return err;
5456	}
5457
5458	/*
5459	 * add_new_disk can be used once the array is assembled
5460	 * to add "hot spares".  They must already have a superblock
5461	 * written
5462	 */
5463	if (mddev->pers) {
5464		int err;
5465		if (!mddev->pers->hot_add_disk) {
5466			printk(KERN_WARNING 
5467				"%s: personality does not support diskops!\n",
5468			       mdname(mddev));
5469			return -EINVAL;
5470		}
5471		if (mddev->persistent)
5472			rdev = md_import_device(dev, mddev->major_version,
5473						mddev->minor_version);
5474		else
5475			rdev = md_import_device(dev, -1, -1);
5476		if (IS_ERR(rdev)) {
5477			printk(KERN_WARNING 
5478				"md: md_import_device returned %ld\n",
5479				PTR_ERR(rdev));
5480			return PTR_ERR(rdev);
5481		}
5482		/* set saved_raid_disk if appropriate */
5483		if (!mddev->persistent) {
5484			if (info->state & (1<<MD_DISK_SYNC)  &&
5485			    info->raid_disk < mddev->raid_disks) {
5486				rdev->raid_disk = info->raid_disk;
5487				set_bit(In_sync, &rdev->flags);
5488			} else
5489				rdev->raid_disk = -1;
5490		} else
5491			super_types[mddev->major_version].
5492				validate_super(mddev, rdev);
5493		if ((info->state & (1<<MD_DISK_SYNC)) &&
5494		    (!test_bit(In_sync, &rdev->flags) ||
5495		     rdev->raid_disk != info->raid_disk)) {
5496			/* This was a hot-add request, but events doesn't
5497			 * match, so reject it.
5498			 */
5499			export_rdev(rdev);
5500			return -EINVAL;
5501		}
5502
5503		if (test_bit(In_sync, &rdev->flags))
5504			rdev->saved_raid_disk = rdev->raid_disk;
5505		else
5506			rdev->saved_raid_disk = -1;
5507
5508		clear_bit(In_sync, &rdev->flags); /* just to be sure */
5509		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5510			set_bit(WriteMostly, &rdev->flags);
5511		else
5512			clear_bit(WriteMostly, &rdev->flags);
5513
5514		rdev->raid_disk = -1;
5515		err = bind_rdev_to_array(rdev, mddev);
5516		if (!err && !mddev->pers->hot_remove_disk) {
5517			/* If there is hot_add_disk but no hot_remove_disk
5518			 * then added disks for geometry changes,
5519			 * and should be added immediately.
5520			 */
5521			super_types[mddev->major_version].
5522				validate_super(mddev, rdev);
5523			err = mddev->pers->hot_add_disk(mddev, rdev);
5524			if (err)
5525				unbind_rdev_from_array(rdev);
5526		}
5527		if (err)
5528			export_rdev(rdev);
5529		else
5530			sysfs_notify_dirent_safe(rdev->sysfs_state);
5531
5532		md_update_sb(mddev, 1);
5533		if (mddev->degraded)
5534			set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
5535		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5536		if (!err)
5537			md_new_event(mddev);
5538		md_wakeup_thread(mddev->thread);
5539		return err;
5540	}
5541
5542	/* otherwise, add_new_disk is only allowed
5543	 * for major_version==0 superblocks
5544	 */
5545	if (mddev->major_version != 0) {
5546		printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5547		       mdname(mddev));
5548		return -EINVAL;
5549	}
5550
5551	if (!(info->state & (1<<MD_DISK_FAULTY))) {
5552		int err;
5553		rdev = md_import_device(dev, -1, 0);
5554		if (IS_ERR(rdev)) {
5555			printk(KERN_WARNING 
5556				"md: error, md_import_device() returned %ld\n",
5557				PTR_ERR(rdev));
5558			return PTR_ERR(rdev);
5559		}
5560		rdev->desc_nr = info->number;
5561		if (info->raid_disk < mddev->raid_disks)
5562			rdev->raid_disk = info->raid_disk;
5563		else
5564			rdev->raid_disk = -1;
5565
5566		if (rdev->raid_disk < mddev->raid_disks)
5567			if (info->state & (1<<MD_DISK_SYNC))
5568				set_bit(In_sync, &rdev->flags);
5569
5570		if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5571			set_bit(WriteMostly, &rdev->flags);
5572
5573		if (!mddev->persistent) {
5574			printk(KERN_INFO "md: nonpersistent superblock ...\n");
5575			rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5576		} else
5577			rdev->sb_start = calc_dev_sboffset(rdev);
5578		rdev->sectors = rdev->sb_start;
5579
5580		err = bind_rdev_to_array(rdev, mddev);
5581		if (err) {
5582			export_rdev(rdev);
5583			return err;
5584		}
5585	}
5586
5587	return 0;
5588}
5589
5590static int hot_remove_disk(mddev_t * mddev, dev_t dev)
5591{
5592	char b[BDEVNAME_SIZE];
5593	mdk_rdev_t *rdev;
5594
5595	rdev = find_rdev(mddev, dev);
5596	if (!rdev)
5597		return -ENXIO;
5598
5599	if (rdev->raid_disk >= 0)
5600		goto busy;
5601
5602	kick_rdev_from_array(rdev);
5603	md_update_sb(mddev, 1);
5604	md_new_event(mddev);
5605
5606	return 0;
5607busy:
5608	printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
5609		bdevname(rdev->bdev,b), mdname(mddev));
5610	return -EBUSY;
5611}
5612
5613static int hot_add_disk(mddev_t * mddev, dev_t dev)
5614{
5615	char b[BDEVNAME_SIZE];
5616	int err;
5617	mdk_rdev_t *rdev;
5618
5619	if (!mddev->pers)
5620		return -ENODEV;
5621
5622	if (mddev->major_version != 0) {
5623		printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5624			" version-0 superblocks.\n",
5625			mdname(mddev));
5626		return -EINVAL;
5627	}
5628	if (!mddev->pers->hot_add_disk) {
5629		printk(KERN_WARNING 
5630			"%s: personality does not support diskops!\n",
5631			mdname(mddev));
5632		return -EINVAL;
5633	}
5634
5635	rdev = md_import_device(dev, -1, 0);
5636	if (IS_ERR(rdev)) {
5637		printk(KERN_WARNING 
5638			"md: error, md_import_device() returned %ld\n",
5639			PTR_ERR(rdev));
5640		return -EINVAL;
5641	}
5642
5643	if (mddev->persistent)
5644		rdev->sb_start = calc_dev_sboffset(rdev);
5645	else
5646		rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5647
5648	rdev->sectors = rdev->sb_start;
5649
5650	if (test_bit(Faulty, &rdev->flags)) {
5651		printk(KERN_WARNING 
5652			"md: can not hot-add faulty %s disk to %s!\n",
5653			bdevname(rdev->bdev,b), mdname(mddev));
5654		err = -EINVAL;
5655		goto abort_export;
5656	}
5657	clear_bit(In_sync, &rdev->flags);
5658	rdev->desc_nr = -1;
5659	rdev->saved_raid_disk = -1;
5660	err = bind_rdev_to_array(rdev, mddev);
5661	if (err)
5662		goto abort_export;
5663
5664	/*
5665	 * The rest should better be atomic, we can have disk failures
5666	 * noticed in interrupt contexts ...
5667	 */
5668
5669	rdev->raid_disk = -1;
5670
5671	md_update_sb(mddev, 1);
5672
5673	/*
5674	 * Kick recovery, maybe this spare has to be added to the
5675	 * array immediately.
5676	 */
5677	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5678	md_wakeup_thread(mddev->thread);
5679	md_new_event(mddev);
5680	return 0;
5681
5682abort_export:
5683	export_rdev(rdev);
5684	return err;
5685}
5686
5687static int set_bitmap_file(mddev_t *mddev, int fd)
5688{
5689	int err;
5690
5691	if (mddev->pers) {
5692		if (!mddev->pers->quiesce)
5693			return -EBUSY;
5694		if (mddev->recovery || mddev->sync_thread)
5695			return -EBUSY;
5696		/* we should be able to change the bitmap.. */
5697	}
5698
5699
5700	if (fd >= 0) {
5701		if (mddev->bitmap)
5702			return -EEXIST; /* cannot add when bitmap is present */
5703		mddev->bitmap_info.file = fget(fd);
5704
5705		if (mddev->bitmap_info.file == NULL) {
5706			printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5707			       mdname(mddev));
5708			return -EBADF;
5709		}
5710
5711		err = deny_bitmap_write_access(mddev->bitmap_info.file);
5712		if (err) {
5713			printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5714			       mdname(mddev));
5715			fput(mddev->bitmap_info.file);
5716			mddev->bitmap_info.file = NULL;
5717			return err;
5718		}
5719		mddev->bitmap_info.offset = 0; /* file overrides offset */
5720	} else if (mddev->bitmap == NULL)
5721		return -ENOENT; /* cannot remove what isn't there */
5722	err = 0;
5723	if (mddev->pers) {
5724		mddev->pers->quiesce(mddev, 1);
5725		if (fd >= 0) {
5726			err = bitmap_create(mddev);
5727			if (!err)
5728				err = bitmap_load(mddev);
5729		}
5730		if (fd < 0 || err) {
5731			bitmap_destroy(mddev);
5732			fd = -1; /* make sure to put the file */
5733		}
5734		mddev->pers->quiesce(mddev, 0);
5735	}
5736	if (fd < 0) {
5737		if (mddev->bitmap_info.file) {
5738			restore_bitmap_write_access(mddev->bitmap_info.file);
5739			fput(mddev->bitmap_info.file);
5740		}
5741		mddev->bitmap_info.file = NULL;
5742	}
5743
5744	return err;
5745}
5746
5747/*
5748 * set_array_info is used two different ways
5749 * The original usage is when creating a new array.
5750 * In this usage, raid_disks is > 0 and it together with
5751 *  level, size, not_persistent,layout,chunksize determine the
5752 *  shape of the array.
5753 *  This will always create an array with a type-0.90.0 superblock.
5754 * The newer usage is when assembling an array.
5755 *  In this case raid_disks will be 0, and the major_version field is
5756 *  use to determine which style super-blocks are to be found on the devices.
5757 *  The minor and patch _version numbers are also kept incase the
5758 *  super_block handler wishes to interpret them.
5759 */
5760static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
5761{
5762
5763	if (info->raid_disks == 0) {
5764		/* just setting version number for superblock loading */
5765		if (info->major_version < 0 ||
5766		    info->major_version >= ARRAY_SIZE(super_types) ||
5767		    super_types[info->major_version].name == NULL) {
5768			/* maybe try to auto-load a module? */
5769			printk(KERN_INFO 
5770				"md: superblock version %d not known\n",
5771				info->major_version);
5772			return -EINVAL;
5773		}
5774		mddev->major_version = info->major_version;
5775		mddev->minor_version = info->minor_version;
5776		mddev->patch_version = info->patch_version;
5777		mddev->persistent = !info->not_persistent;
5778		/* ensure mddev_put doesn't delete this now that there
5779		 * is some minimal configuration.
5780		 */
5781		mddev->ctime         = get_seconds();
5782		return 0;
5783	}
5784	mddev->major_version = MD_MAJOR_VERSION;
5785	mddev->minor_version = MD_MINOR_VERSION;
5786	mddev->patch_version = MD_PATCHLEVEL_VERSION;
5787	mddev->ctime         = get_seconds();
5788
5789	mddev->level         = info->level;
5790	mddev->clevel[0]     = 0;
5791	mddev->dev_sectors   = 2 * (sector_t)info->size;
5792	mddev->raid_disks    = info->raid_disks;
5793	/* don't set md_minor, it is determined by which /dev/md* was
5794	 * openned
5795	 */
5796	if (info->state & (1<<MD_SB_CLEAN))
5797		mddev->recovery_cp = MaxSector;
5798	else
5799		mddev->recovery_cp = 0;
5800	mddev->persistent    = ! info->not_persistent;
5801	mddev->external	     = 0;
5802
5803	mddev->layout        = info->layout;
5804	mddev->chunk_sectors = info->chunk_size >> 9;
5805
5806	mddev->max_disks     = MD_SB_DISKS;
5807
5808	if (mddev->persistent)
5809		mddev->flags         = 0;
5810	set_bit(MD_CHANGE_DEVS, &mddev->flags);
5811
5812	mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
 
5813	mddev->bitmap_info.offset = 0;
5814
5815	mddev->reshape_position = MaxSector;
5816
5817	/*
5818	 * Generate a 128 bit UUID
5819	 */
5820	get_random_bytes(mddev->uuid, 16);
5821
5822	mddev->new_level = mddev->level;
5823	mddev->new_chunk_sectors = mddev->chunk_sectors;
5824	mddev->new_layout = mddev->layout;
5825	mddev->delta_disks = 0;
 
5826
5827	return 0;
5828}
5829
5830void md_set_array_sectors(mddev_t *mddev, sector_t array_sectors)
5831{
5832	WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
5833
5834	if (mddev->external_size)
5835		return;
5836
5837	mddev->array_sectors = array_sectors;
5838}
5839EXPORT_SYMBOL(md_set_array_sectors);
5840
5841static int update_size(mddev_t *mddev, sector_t num_sectors)
5842{
5843	mdk_rdev_t *rdev;
5844	int rv;
5845	int fit = (num_sectors == 0);
5846
5847	if (mddev->pers->resize == NULL)
5848		return -EINVAL;
5849	/* The "num_sectors" is the number of sectors of each device that
5850	 * is used.  This can only make sense for arrays with redundancy.
5851	 * linear and raid0 always use whatever space is available. We can only
5852	 * consider changing this number if no resync or reconstruction is
5853	 * happening, and if the new size is acceptable. It must fit before the
5854	 * sb_start or, if that is <data_offset, it must fit before the size
5855	 * of each device.  If num_sectors is zero, we find the largest size
5856	 * that fits.
5857	 */
5858	if (mddev->sync_thread)
5859		return -EBUSY;
5860	if (mddev->bitmap)
5861		/* Sorry, cannot grow a bitmap yet, just remove it,
5862		 * grow, and re-add.
5863		 */
5864		return -EBUSY;
5865	list_for_each_entry(rdev, &mddev->disks, same_set) {
5866		sector_t avail = rdev->sectors;
5867
5868		if (fit && (num_sectors == 0 || num_sectors > avail))
5869			num_sectors = avail;
5870		if (avail < num_sectors)
5871			return -ENOSPC;
5872	}
5873	rv = mddev->pers->resize(mddev, num_sectors);
5874	if (!rv)
5875		revalidate_disk(mddev->gendisk);
5876	return rv;
5877}
5878
5879static int update_raid_disks(mddev_t *mddev, int raid_disks)
5880{
5881	int rv;
 
5882	/* change the number of raid disks */
5883	if (mddev->pers->check_reshape == NULL)
5884		return -EINVAL;
5885	if (raid_disks <= 0 ||
5886	    (mddev->max_disks && raid_disks >= mddev->max_disks))
5887		return -EINVAL;
5888	if (mddev->sync_thread || mddev->reshape_position != MaxSector)
5889		return -EBUSY;
 
 
 
 
 
 
 
 
 
 
5890	mddev->delta_disks = raid_disks - mddev->raid_disks;
 
 
 
 
5891
5892	rv = mddev->pers->check_reshape(mddev);
5893	if (rv < 0)
5894		mddev->delta_disks = 0;
 
 
5895	return rv;
5896}
5897
5898
5899/*
5900 * update_array_info is used to change the configuration of an
5901 * on-line array.
5902 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
5903 * fields in the info are checked against the array.
5904 * Any differences that cannot be handled will cause an error.
5905 * Normally, only one change can be managed at a time.
5906 */
5907static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
5908{
5909	int rv = 0;
5910	int cnt = 0;
5911	int state = 0;
5912
5913	/* calculate expected state,ignoring low bits */
5914	if (mddev->bitmap && mddev->bitmap_info.offset)
5915		state |= (1 << MD_SB_BITMAP_PRESENT);
5916
5917	if (mddev->major_version != info->major_version ||
5918	    mddev->minor_version != info->minor_version ||
5919/*	    mddev->patch_version != info->patch_version || */
5920	    mddev->ctime         != info->ctime         ||
5921	    mddev->level         != info->level         ||
5922/*	    mddev->layout        != info->layout        || */
5923	    !mddev->persistent	 != info->not_persistent||
5924	    mddev->chunk_sectors != info->chunk_size >> 9 ||
5925	    /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
5926	    ((state^info->state) & 0xfffffe00)
5927		)
5928		return -EINVAL;
5929	/* Check there is only one change */
5930	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5931		cnt++;
5932	if (mddev->raid_disks != info->raid_disks)
5933		cnt++;
5934	if (mddev->layout != info->layout)
5935		cnt++;
5936	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
5937		cnt++;
5938	if (cnt == 0)
5939		return 0;
5940	if (cnt > 1)
5941		return -EINVAL;
5942
5943	if (mddev->layout != info->layout) {
5944		/* Change layout
5945		 * we don't need to do anything at the md level, the
5946		 * personality will take care of it all.
5947		 */
5948		if (mddev->pers->check_reshape == NULL)
5949			return -EINVAL;
5950		else {
5951			mddev->new_layout = info->layout;
5952			rv = mddev->pers->check_reshape(mddev);
5953			if (rv)
5954				mddev->new_layout = mddev->layout;
5955			return rv;
5956		}
5957	}
5958	if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
5959		rv = update_size(mddev, (sector_t)info->size * 2);
5960
5961	if (mddev->raid_disks    != info->raid_disks)
5962		rv = update_raid_disks(mddev, info->raid_disks);
5963
5964	if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
5965		if (mddev->pers->quiesce == NULL)
5966			return -EINVAL;
5967		if (mddev->recovery || mddev->sync_thread)
5968			return -EBUSY;
5969		if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
5970			/* add the bitmap */
5971			if (mddev->bitmap)
5972				return -EEXIST;
5973			if (mddev->bitmap_info.default_offset == 0)
5974				return -EINVAL;
5975			mddev->bitmap_info.offset =
5976				mddev->bitmap_info.default_offset;
 
 
5977			mddev->pers->quiesce(mddev, 1);
5978			rv = bitmap_create(mddev);
5979			if (!rv)
5980				rv = bitmap_load(mddev);
5981			if (rv)
5982				bitmap_destroy(mddev);
5983			mddev->pers->quiesce(mddev, 0);
5984		} else {
5985			/* remove the bitmap */
5986			if (!mddev->bitmap)
5987				return -ENOENT;
5988			if (mddev->bitmap->file)
5989				return -EINVAL;
5990			mddev->pers->quiesce(mddev, 1);
5991			bitmap_destroy(mddev);
5992			mddev->pers->quiesce(mddev, 0);
5993			mddev->bitmap_info.offset = 0;
5994		}
5995	}
5996	md_update_sb(mddev, 1);
5997	return rv;
5998}
5999
6000static int set_disk_faulty(mddev_t *mddev, dev_t dev)
6001{
6002	mdk_rdev_t *rdev;
6003
6004	if (mddev->pers == NULL)
6005		return -ENODEV;
6006
6007	rdev = find_rdev(mddev, dev);
6008	if (!rdev)
6009		return -ENODEV;
6010
6011	md_error(mddev, rdev);
6012	if (!test_bit(Faulty, &rdev->flags))
6013		return -EBUSY;
6014	return 0;
6015}
6016
6017/*
6018 * We have a problem here : there is no easy way to give a CHS
6019 * virtual geometry. We currently pretend that we have a 2 heads
6020 * 4 sectors (with a BIG number of cylinders...). This drives
6021 * dosfs just mad... ;-)
6022 */
6023static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6024{
6025	mddev_t *mddev = bdev->bd_disk->private_data;
6026
6027	geo->heads = 2;
6028	geo->sectors = 4;
6029	geo->cylinders = mddev->array_sectors / 8;
6030	return 0;
6031}
6032
6033static int md_ioctl(struct block_device *bdev, fmode_t mode,
6034			unsigned int cmd, unsigned long arg)
6035{
6036	int err = 0;
6037	void __user *argp = (void __user *)arg;
6038	mddev_t *mddev = NULL;
6039	int ro;
6040
6041	if (!capable(CAP_SYS_ADMIN))
6042		return -EACCES;
 
 
 
 
 
 
 
6043
6044	/*
6045	 * Commands dealing with the RAID driver but not any
6046	 * particular array:
6047	 */
6048	switch (cmd)
6049	{
6050		case RAID_VERSION:
6051			err = get_version(argp);
6052			goto done;
6053
6054		case PRINT_RAID_DEBUG:
6055			err = 0;
6056			md_print_devices();
6057			goto done;
6058
6059#ifndef MODULE
6060		case RAID_AUTORUN:
6061			err = 0;
6062			autostart_arrays(arg);
6063			goto done;
6064#endif
6065		default:;
6066	}
6067
6068	/*
6069	 * Commands creating/starting a new array:
6070	 */
6071
6072	mddev = bdev->bd_disk->private_data;
6073
6074	if (!mddev) {
6075		BUG();
6076		goto abort;
6077	}
6078
6079	err = mddev_lock(mddev);
6080	if (err) {
6081		printk(KERN_INFO 
6082			"md: ioctl lock interrupted, reason %d, cmd %d\n",
6083			err, cmd);
6084		goto abort;
6085	}
6086
6087	switch (cmd)
6088	{
6089		case SET_ARRAY_INFO:
6090			{
6091				mdu_array_info_t info;
6092				if (!arg)
6093					memset(&info, 0, sizeof(info));
6094				else if (copy_from_user(&info, argp, sizeof(info))) {
6095					err = -EFAULT;
6096					goto abort_unlock;
6097				}
6098				if (mddev->pers) {
6099					err = update_array_info(mddev, &info);
6100					if (err) {
6101						printk(KERN_WARNING "md: couldn't update"
6102						       " array info. %d\n", err);
6103						goto abort_unlock;
6104					}
6105					goto done_unlock;
6106				}
6107				if (!list_empty(&mddev->disks)) {
6108					printk(KERN_WARNING
6109					       "md: array %s already has disks!\n",
6110					       mdname(mddev));
6111					err = -EBUSY;
6112					goto abort_unlock;
6113				}
6114				if (mddev->raid_disks) {
6115					printk(KERN_WARNING
6116					       "md: array %s already initialised!\n",
6117					       mdname(mddev));
6118					err = -EBUSY;
6119					goto abort_unlock;
6120				}
6121				err = set_array_info(mddev, &info);
6122				if (err) {
6123					printk(KERN_WARNING "md: couldn't set"
6124					       " array info. %d\n", err);
6125					goto abort_unlock;
6126				}
6127			}
6128			goto done_unlock;
6129
6130		default:;
6131	}
6132
6133	/*
6134	 * Commands querying/configuring an existing array:
6135	 */
6136	/* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
6137	 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
6138	if ((!mddev->raid_disks && !mddev->external)
6139	    && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6140	    && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6141	    && cmd != GET_BITMAP_FILE) {
6142		err = -ENODEV;
6143		goto abort_unlock;
6144	}
6145
6146	/*
6147	 * Commands even a read-only array can execute:
6148	 */
6149	switch (cmd)
6150	{
6151		case GET_ARRAY_INFO:
6152			err = get_array_info(mddev, argp);
6153			goto done_unlock;
6154
6155		case GET_BITMAP_FILE:
6156			err = get_bitmap_file(mddev, argp);
6157			goto done_unlock;
6158
6159		case GET_DISK_INFO:
6160			err = get_disk_info(mddev, argp);
6161			goto done_unlock;
6162
6163		case RESTART_ARRAY_RW:
6164			err = restart_array(mddev);
6165			goto done_unlock;
6166
6167		case STOP_ARRAY:
6168			err = do_md_stop(mddev, 0, 1);
6169			goto done_unlock;
6170
6171		case STOP_ARRAY_RO:
6172			err = md_set_readonly(mddev, 1);
6173			goto done_unlock;
6174
6175		case BLKROSET:
6176			if (get_user(ro, (int __user *)(arg))) {
6177				err = -EFAULT;
6178				goto done_unlock;
6179			}
6180			err = -EINVAL;
6181
6182			/* if the bdev is going readonly the value of mddev->ro
6183			 * does not matter, no writes are coming
6184			 */
6185			if (ro)
6186				goto done_unlock;
6187
6188			/* are we are already prepared for writes? */
6189			if (mddev->ro != 1)
6190				goto done_unlock;
6191
6192			/* transitioning to readauto need only happen for
6193			 * arrays that call md_write_start
6194			 */
6195			if (mddev->pers) {
6196				err = restart_array(mddev);
6197				if (err == 0) {
6198					mddev->ro = 2;
6199					set_disk_ro(mddev->gendisk, 0);
6200				}
6201			}
6202			goto done_unlock;
6203	}
6204
6205	/*
6206	 * The remaining ioctls are changing the state of the
6207	 * superblock, so we do not allow them on read-only arrays.
6208	 * However non-MD ioctls (e.g. get-size) will still come through
6209	 * here and hit the 'default' below, so only disallow
6210	 * 'md' ioctls, and switch to rw mode if started auto-readonly.
6211	 */
6212	if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
6213		if (mddev->ro == 2) {
6214			mddev->ro = 0;
6215			sysfs_notify_dirent_safe(mddev->sysfs_state);
6216			set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6217			md_wakeup_thread(mddev->thread);
6218		} else {
6219			err = -EROFS;
6220			goto abort_unlock;
6221		}
6222	}
6223
6224	switch (cmd)
6225	{
6226		case ADD_NEW_DISK:
6227		{
6228			mdu_disk_info_t info;
6229			if (copy_from_user(&info, argp, sizeof(info)))
6230				err = -EFAULT;
6231			else
6232				err = add_new_disk(mddev, &info);
6233			goto done_unlock;
6234		}
6235
6236		case HOT_REMOVE_DISK:
6237			err = hot_remove_disk(mddev, new_decode_dev(arg));
6238			goto done_unlock;
6239
6240		case HOT_ADD_DISK:
6241			err = hot_add_disk(mddev, new_decode_dev(arg));
6242			goto done_unlock;
6243
6244		case SET_DISK_FAULTY:
6245			err = set_disk_faulty(mddev, new_decode_dev(arg));
6246			goto done_unlock;
6247
6248		case RUN_ARRAY:
6249			err = do_md_run(mddev);
6250			goto done_unlock;
6251
6252		case SET_BITMAP_FILE:
6253			err = set_bitmap_file(mddev, (int)arg);
6254			goto done_unlock;
6255
6256		default:
6257			err = -EINVAL;
6258			goto abort_unlock;
6259	}
6260
6261done_unlock:
6262abort_unlock:
6263	if (mddev->hold_active == UNTIL_IOCTL &&
6264	    err != -EINVAL)
6265		mddev->hold_active = 0;
6266	mddev_unlock(mddev);
6267
6268	return err;
6269done:
6270	if (err)
6271		MD_BUG();
6272abort:
6273	return err;
6274}
6275#ifdef CONFIG_COMPAT
6276static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6277		    unsigned int cmd, unsigned long arg)
6278{
6279	switch (cmd) {
6280	case HOT_REMOVE_DISK:
6281	case HOT_ADD_DISK:
6282	case SET_DISK_FAULTY:
6283	case SET_BITMAP_FILE:
6284		/* These take in integer arg, do not convert */
6285		break;
6286	default:
6287		arg = (unsigned long)compat_ptr(arg);
6288		break;
6289	}
6290
6291	return md_ioctl(bdev, mode, cmd, arg);
6292}
6293#endif /* CONFIG_COMPAT */
6294
6295static int md_open(struct block_device *bdev, fmode_t mode)
6296{
6297	/*
6298	 * Succeed if we can lock the mddev, which confirms that
6299	 * it isn't being stopped right now.
6300	 */
6301	mddev_t *mddev = mddev_find(bdev->bd_dev);
6302	int err;
6303
 
 
 
6304	if (mddev->gendisk != bdev->bd_disk) {
6305		/* we are racing with mddev_put which is discarding this
6306		 * bd_disk.
6307		 */
6308		mddev_put(mddev);
6309		/* Wait until bdev->bd_disk is definitely gone */
6310		flush_workqueue(md_misc_wq);
6311		/* Then retry the open from the top */
6312		return -ERESTARTSYS;
6313	}
6314	BUG_ON(mddev != bdev->bd_disk->private_data);
6315
6316	if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
6317		goto out;
6318
6319	err = 0;
6320	atomic_inc(&mddev->openers);
6321	mutex_unlock(&mddev->open_mutex);
6322
6323	check_disk_change(bdev);
6324 out:
6325	return err;
6326}
6327
6328static int md_release(struct gendisk *disk, fmode_t mode)
6329{
6330 	mddev_t *mddev = disk->private_data;
6331
6332	BUG_ON(!mddev);
6333	atomic_dec(&mddev->openers);
6334	mddev_put(mddev);
6335
6336	return 0;
6337}
6338
6339static int md_media_changed(struct gendisk *disk)
6340{
6341	mddev_t *mddev = disk->private_data;
6342
6343	return mddev->changed;
6344}
6345
6346static int md_revalidate(struct gendisk *disk)
6347{
6348	mddev_t *mddev = disk->private_data;
6349
6350	mddev->changed = 0;
6351	return 0;
6352}
6353static const struct block_device_operations md_fops =
6354{
6355	.owner		= THIS_MODULE,
6356	.open		= md_open,
6357	.release	= md_release,
6358	.ioctl		= md_ioctl,
6359#ifdef CONFIG_COMPAT
6360	.compat_ioctl	= md_compat_ioctl,
6361#endif
6362	.getgeo		= md_getgeo,
6363	.media_changed  = md_media_changed,
6364	.revalidate_disk= md_revalidate,
6365};
6366
6367static int md_thread(void * arg)
6368{
6369	mdk_thread_t *thread = arg;
6370
6371	/*
6372	 * md_thread is a 'system-thread', it's priority should be very
6373	 * high. We avoid resource deadlocks individually in each
6374	 * raid personality. (RAID5 does preallocation) We also use RR and
6375	 * the very same RT priority as kswapd, thus we will never get
6376	 * into a priority inversion deadlock.
6377	 *
6378	 * we definitely have to have equal or higher priority than
6379	 * bdflush, otherwise bdflush will deadlock if there are too
6380	 * many dirty RAID5 blocks.
6381	 */
6382
6383	allow_signal(SIGKILL);
6384	while (!kthread_should_stop()) {
6385
6386		/* We need to wait INTERRUPTIBLE so that
6387		 * we don't add to the load-average.
6388		 * That means we need to be sure no signals are
6389		 * pending
6390		 */
6391		if (signal_pending(current))
6392			flush_signals(current);
6393
6394		wait_event_interruptible_timeout
6395			(thread->wqueue,
6396			 test_bit(THREAD_WAKEUP, &thread->flags)
6397			 || kthread_should_stop(),
6398			 thread->timeout);
6399
6400		clear_bit(THREAD_WAKEUP, &thread->flags);
6401		if (!kthread_should_stop())
6402			thread->run(thread->mddev);
6403	}
6404
6405	return 0;
6406}
6407
6408void md_wakeup_thread(mdk_thread_t *thread)
6409{
6410	if (thread) {
6411		dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
6412		set_bit(THREAD_WAKEUP, &thread->flags);
6413		wake_up(&thread->wqueue);
6414	}
6415}
6416
6417mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
6418				 const char *name)
6419{
6420	mdk_thread_t *thread;
6421
6422	thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
6423	if (!thread)
6424		return NULL;
6425
6426	init_waitqueue_head(&thread->wqueue);
6427
6428	thread->run = run;
6429	thread->mddev = mddev;
6430	thread->timeout = MAX_SCHEDULE_TIMEOUT;
6431	thread->tsk = kthread_run(md_thread, thread,
6432				  "%s_%s",
6433				  mdname(thread->mddev),
6434				  name ?: mddev->pers->name);
6435	if (IS_ERR(thread->tsk)) {
6436		kfree(thread);
6437		return NULL;
6438	}
6439	return thread;
6440}
6441
6442void md_unregister_thread(mdk_thread_t **threadp)
6443{
6444	mdk_thread_t *thread = *threadp;
6445	if (!thread)
6446		return;
6447	dprintk("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
6448	/* Locking ensures that mddev_unlock does not wake_up a
6449	 * non-existent thread
6450	 */
6451	spin_lock(&pers_lock);
6452	*threadp = NULL;
6453	spin_unlock(&pers_lock);
6454
6455	kthread_stop(thread->tsk);
6456	kfree(thread);
6457}
6458
6459void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
6460{
6461	if (!mddev) {
6462		MD_BUG();
6463		return;
6464	}
6465
6466	if (!rdev || test_bit(Faulty, &rdev->flags))
6467		return;
6468
6469	if (!mddev->pers || !mddev->pers->error_handler)
6470		return;
6471	mddev->pers->error_handler(mddev,rdev);
6472	if (mddev->degraded)
6473		set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
6474	sysfs_notify_dirent_safe(rdev->sysfs_state);
6475	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6476	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6477	md_wakeup_thread(mddev->thread);
6478	if (mddev->event_work.func)
6479		queue_work(md_misc_wq, &mddev->event_work);
6480	md_new_event_inintr(mddev);
6481}
6482
6483/* seq_file implementation /proc/mdstat */
6484
6485static void status_unused(struct seq_file *seq)
6486{
6487	int i = 0;
6488	mdk_rdev_t *rdev;
6489
6490	seq_printf(seq, "unused devices: ");
6491
6492	list_for_each_entry(rdev, &pending_raid_disks, same_set) {
6493		char b[BDEVNAME_SIZE];
6494		i++;
6495		seq_printf(seq, "%s ",
6496			      bdevname(rdev->bdev,b));
6497	}
6498	if (!i)
6499		seq_printf(seq, "<none>");
6500
6501	seq_printf(seq, "\n");
6502}
6503
6504
6505static void status_resync(struct seq_file *seq, mddev_t * mddev)
6506{
6507	sector_t max_sectors, resync, res;
6508	unsigned long dt, db;
6509	sector_t rt;
6510	int scale;
6511	unsigned int per_milli;
6512
6513	resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
6514
6515	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
 
6516		max_sectors = mddev->resync_max_sectors;
6517	else
6518		max_sectors = mddev->dev_sectors;
6519
6520	/*
6521	 * Should not happen.
6522	 */
6523	if (!max_sectors) {
6524		MD_BUG();
6525		return;
6526	}
6527	/* Pick 'scale' such that (resync>>scale)*1000 will fit
6528	 * in a sector_t, and (max_sectors>>scale) will fit in a
6529	 * u32, as those are the requirements for sector_div.
6530	 * Thus 'scale' must be at least 10
6531	 */
6532	scale = 10;
6533	if (sizeof(sector_t) > sizeof(unsigned long)) {
6534		while ( max_sectors/2 > (1ULL<<(scale+32)))
6535			scale++;
6536	}
6537	res = (resync>>scale)*1000;
6538	sector_div(res, (u32)((max_sectors>>scale)+1));
6539
6540	per_milli = res;
6541	{
6542		int i, x = per_milli/50, y = 20-x;
6543		seq_printf(seq, "[");
6544		for (i = 0; i < x; i++)
6545			seq_printf(seq, "=");
6546		seq_printf(seq, ">");
6547		for (i = 0; i < y; i++)
6548			seq_printf(seq, ".");
6549		seq_printf(seq, "] ");
6550	}
6551	seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
6552		   (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6553		    "reshape" :
6554		    (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6555		     "check" :
6556		     (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6557		      "resync" : "recovery"))),
6558		   per_milli/10, per_milli % 10,
6559		   (unsigned long long) resync/2,
6560		   (unsigned long long) max_sectors/2);
6561
6562	/*
6563	 * dt: time from mark until now
6564	 * db: blocks written from mark until now
6565	 * rt: remaining time
6566	 *
6567	 * rt is a sector_t, so could be 32bit or 64bit.
6568	 * So we divide before multiply in case it is 32bit and close
6569	 * to the limit.
6570	 * We scale the divisor (db) by 32 to avoid losing precision
6571	 * near the end of resync when the number of remaining sectors
6572	 * is close to 'db'.
6573	 * We then divide rt by 32 after multiplying by db to compensate.
6574	 * The '+1' avoids division by zero if db is very small.
6575	 */
6576	dt = ((jiffies - mddev->resync_mark) / HZ);
6577	if (!dt) dt++;
6578	db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6579		- mddev->resync_mark_cnt;
6580
6581	rt = max_sectors - resync;    /* number of remaining sectors */
6582	sector_div(rt, db/32+1);
6583	rt *= dt;
6584	rt >>= 5;
6585
6586	seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6587		   ((unsigned long)rt % 60)/6);
6588
6589	seq_printf(seq, " speed=%ldK/sec", db/2/dt);
6590}
6591
6592static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6593{
6594	struct list_head *tmp;
6595	loff_t l = *pos;
6596	mddev_t *mddev;
6597
6598	if (l >= 0x10000)
6599		return NULL;
6600	if (!l--)
6601		/* header */
6602		return (void*)1;
6603
6604	spin_lock(&all_mddevs_lock);
6605	list_for_each(tmp,&all_mddevs)
6606		if (!l--) {
6607			mddev = list_entry(tmp, mddev_t, all_mddevs);
6608			mddev_get(mddev);
6609			spin_unlock(&all_mddevs_lock);
6610			return mddev;
6611		}
6612	spin_unlock(&all_mddevs_lock);
6613	if (!l--)
6614		return (void*)2;/* tail */
6615	return NULL;
6616}
6617
6618static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6619{
6620	struct list_head *tmp;
6621	mddev_t *next_mddev, *mddev = v;
6622	
6623	++*pos;
6624	if (v == (void*)2)
6625		return NULL;
6626
6627	spin_lock(&all_mddevs_lock);
6628	if (v == (void*)1)
6629		tmp = all_mddevs.next;
6630	else
6631		tmp = mddev->all_mddevs.next;
6632	if (tmp != &all_mddevs)
6633		next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
6634	else {
6635		next_mddev = (void*)2;
6636		*pos = 0x10000;
6637	}		
6638	spin_unlock(&all_mddevs_lock);
6639
6640	if (v != (void*)1)
6641		mddev_put(mddev);
6642	return next_mddev;
6643
6644}
6645
6646static void md_seq_stop(struct seq_file *seq, void *v)
6647{
6648	mddev_t *mddev = v;
6649
6650	if (mddev && v != (void*)1 && v != (void*)2)
6651		mddev_put(mddev);
6652}
6653
6654static int md_seq_show(struct seq_file *seq, void *v)
6655{
6656	mddev_t *mddev = v;
6657	sector_t sectors;
6658	mdk_rdev_t *rdev;
6659	struct bitmap *bitmap;
6660
6661	if (v == (void*)1) {
6662		struct mdk_personality *pers;
6663		seq_printf(seq, "Personalities : ");
6664		spin_lock(&pers_lock);
6665		list_for_each_entry(pers, &pers_list, list)
6666			seq_printf(seq, "[%s] ", pers->name);
6667
6668		spin_unlock(&pers_lock);
6669		seq_printf(seq, "\n");
6670		seq->poll_event = atomic_read(&md_event_count);
6671		return 0;
6672	}
6673	if (v == (void*)2) {
6674		status_unused(seq);
6675		return 0;
6676	}
6677
6678	if (mddev_lock(mddev) < 0)
6679		return -EINTR;
6680
6681	if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6682		seq_printf(seq, "%s : %sactive", mdname(mddev),
6683						mddev->pers ? "" : "in");
6684		if (mddev->pers) {
6685			if (mddev->ro==1)
6686				seq_printf(seq, " (read-only)");
6687			if (mddev->ro==2)
6688				seq_printf(seq, " (auto-read-only)");
6689			seq_printf(seq, " %s", mddev->pers->name);
6690		}
6691
6692		sectors = 0;
6693		list_for_each_entry(rdev, &mddev->disks, same_set) {
6694			char b[BDEVNAME_SIZE];
6695			seq_printf(seq, " %s[%d]",
6696				bdevname(rdev->bdev,b), rdev->desc_nr);
6697			if (test_bit(WriteMostly, &rdev->flags))
6698				seq_printf(seq, "(W)");
6699			if (test_bit(Faulty, &rdev->flags)) {
6700				seq_printf(seq, "(F)");
6701				continue;
6702			} else if (rdev->raid_disk < 0)
 
6703				seq_printf(seq, "(S)"); /* spare */
 
 
6704			sectors += rdev->sectors;
6705		}
6706
6707		if (!list_empty(&mddev->disks)) {
6708			if (mddev->pers)
6709				seq_printf(seq, "\n      %llu blocks",
6710					   (unsigned long long)
6711					   mddev->array_sectors / 2);
6712			else
6713				seq_printf(seq, "\n      %llu blocks",
6714					   (unsigned long long)sectors / 2);
6715		}
6716		if (mddev->persistent) {
6717			if (mddev->major_version != 0 ||
6718			    mddev->minor_version != 90) {
6719				seq_printf(seq," super %d.%d",
6720					   mddev->major_version,
6721					   mddev->minor_version);
6722			}
6723		} else if (mddev->external)
6724			seq_printf(seq, " super external:%s",
6725				   mddev->metadata_type);
6726		else
6727			seq_printf(seq, " super non-persistent");
6728
6729		if (mddev->pers) {
6730			mddev->pers->status(seq, mddev);
6731	 		seq_printf(seq, "\n      ");
6732			if (mddev->pers->sync_request) {
6733				if (mddev->curr_resync > 2) {
6734					status_resync(seq, mddev);
6735					seq_printf(seq, "\n      ");
6736				} else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
6737					seq_printf(seq, "\tresync=DELAYED\n      ");
6738				else if (mddev->recovery_cp < MaxSector)
6739					seq_printf(seq, "\tresync=PENDING\n      ");
6740			}
6741		} else
6742			seq_printf(seq, "\n       ");
6743
6744		if ((bitmap = mddev->bitmap)) {
6745			unsigned long chunk_kb;
6746			unsigned long flags;
6747			spin_lock_irqsave(&bitmap->lock, flags);
6748			chunk_kb = mddev->bitmap_info.chunksize >> 10;
6749			seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
6750				"%lu%s chunk",
6751				bitmap->pages - bitmap->missing_pages,
6752				bitmap->pages,
6753				(bitmap->pages - bitmap->missing_pages)
6754					<< (PAGE_SHIFT - 10),
6755				chunk_kb ? chunk_kb : mddev->bitmap_info.chunksize,
6756				chunk_kb ? "KB" : "B");
6757			if (bitmap->file) {
6758				seq_printf(seq, ", file: ");
6759				seq_path(seq, &bitmap->file->f_path, " \t\n");
6760			}
6761
6762			seq_printf(seq, "\n");
6763			spin_unlock_irqrestore(&bitmap->lock, flags);
6764		}
6765
6766		seq_printf(seq, "\n");
6767	}
6768	mddev_unlock(mddev);
6769	
6770	return 0;
6771}
6772
6773static const struct seq_operations md_seq_ops = {
6774	.start  = md_seq_start,
6775	.next   = md_seq_next,
6776	.stop   = md_seq_stop,
6777	.show   = md_seq_show,
6778};
6779
6780static int md_seq_open(struct inode *inode, struct file *file)
6781{
6782	struct seq_file *seq;
6783	int error;
6784
6785	error = seq_open(file, &md_seq_ops);
6786	if (error)
6787		return error;
6788
6789	seq = file->private_data;
6790	seq->poll_event = atomic_read(&md_event_count);
6791	return error;
6792}
6793
6794static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
6795{
6796	struct seq_file *seq = filp->private_data;
6797	int mask;
6798
6799	poll_wait(filp, &md_event_waiters, wait);
6800
6801	/* always allow read */
6802	mask = POLLIN | POLLRDNORM;
6803
6804	if (seq->poll_event != atomic_read(&md_event_count))
6805		mask |= POLLERR | POLLPRI;
6806	return mask;
6807}
6808
6809static const struct file_operations md_seq_fops = {
6810	.owner		= THIS_MODULE,
6811	.open           = md_seq_open,
6812	.read           = seq_read,
6813	.llseek         = seq_lseek,
6814	.release	= seq_release_private,
6815	.poll		= mdstat_poll,
6816};
6817
6818int register_md_personality(struct mdk_personality *p)
6819{
6820	spin_lock(&pers_lock);
6821	list_add_tail(&p->list, &pers_list);
6822	printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
6823	spin_unlock(&pers_lock);
6824	return 0;
6825}
6826
6827int unregister_md_personality(struct mdk_personality *p)
6828{
6829	printk(KERN_INFO "md: %s personality unregistered\n", p->name);
6830	spin_lock(&pers_lock);
6831	list_del_init(&p->list);
6832	spin_unlock(&pers_lock);
6833	return 0;
6834}
6835
6836static int is_mddev_idle(mddev_t *mddev, int init)
6837{
6838	mdk_rdev_t * rdev;
6839	int idle;
6840	int curr_events;
6841
6842	idle = 1;
6843	rcu_read_lock();
6844	rdev_for_each_rcu(rdev, mddev) {
6845		struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
6846		curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
6847			      (int)part_stat_read(&disk->part0, sectors[1]) -
6848			      atomic_read(&disk->sync_io);
6849		/* sync IO will cause sync_io to increase before the disk_stats
6850		 * as sync_io is counted when a request starts, and
6851		 * disk_stats is counted when it completes.
6852		 * So resync activity will cause curr_events to be smaller than
6853		 * when there was no such activity.
6854		 * non-sync IO will cause disk_stat to increase without
6855		 * increasing sync_io so curr_events will (eventually)
6856		 * be larger than it was before.  Once it becomes
6857		 * substantially larger, the test below will cause
6858		 * the array to appear non-idle, and resync will slow
6859		 * down.
6860		 * If there is a lot of outstanding resync activity when
6861		 * we set last_event to curr_events, then all that activity
6862		 * completing might cause the array to appear non-idle
6863		 * and resync will be slowed down even though there might
6864		 * not have been non-resync activity.  This will only
6865		 * happen once though.  'last_events' will soon reflect
6866		 * the state where there is little or no outstanding
6867		 * resync requests, and further resync activity will
6868		 * always make curr_events less than last_events.
6869		 *
6870		 */
6871		if (init || curr_events - rdev->last_events > 64) {
6872			rdev->last_events = curr_events;
6873			idle = 0;
6874		}
6875	}
6876	rcu_read_unlock();
6877	return idle;
6878}
6879
6880void md_done_sync(mddev_t *mddev, int blocks, int ok)
6881{
6882	/* another "blocks" (512byte) blocks have been synced */
6883	atomic_sub(blocks, &mddev->recovery_active);
6884	wake_up(&mddev->recovery_wait);
6885	if (!ok) {
6886		set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6887		md_wakeup_thread(mddev->thread);
6888		// stop recovery, signal do_sync ....
6889	}
6890}
6891
6892
6893/* md_write_start(mddev, bi)
6894 * If we need to update some array metadata (e.g. 'active' flag
6895 * in superblock) before writing, schedule a superblock update
6896 * and wait for it to complete.
6897 */
6898void md_write_start(mddev_t *mddev, struct bio *bi)
6899{
6900	int did_change = 0;
6901	if (bio_data_dir(bi) != WRITE)
6902		return;
6903
6904	BUG_ON(mddev->ro == 1);
6905	if (mddev->ro == 2) {
6906		/* need to switch to read/write */
6907		mddev->ro = 0;
6908		set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6909		md_wakeup_thread(mddev->thread);
6910		md_wakeup_thread(mddev->sync_thread);
6911		did_change = 1;
6912	}
6913	atomic_inc(&mddev->writes_pending);
6914	if (mddev->safemode == 1)
6915		mddev->safemode = 0;
6916	if (mddev->in_sync) {
6917		spin_lock_irq(&mddev->write_lock);
6918		if (mddev->in_sync) {
6919			mddev->in_sync = 0;
6920			set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6921			set_bit(MD_CHANGE_PENDING, &mddev->flags);
6922			md_wakeup_thread(mddev->thread);
6923			did_change = 1;
6924		}
6925		spin_unlock_irq(&mddev->write_lock);
6926	}
6927	if (did_change)
6928		sysfs_notify_dirent_safe(mddev->sysfs_state);
6929	wait_event(mddev->sb_wait,
6930		   !test_bit(MD_CHANGE_PENDING, &mddev->flags));
6931}
6932
6933void md_write_end(mddev_t *mddev)
6934{
6935	if (atomic_dec_and_test(&mddev->writes_pending)) {
6936		if (mddev->safemode == 2)
6937			md_wakeup_thread(mddev->thread);
6938		else if (mddev->safemode_delay)
6939			mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
6940	}
6941}
6942
6943/* md_allow_write(mddev)
6944 * Calling this ensures that the array is marked 'active' so that writes
6945 * may proceed without blocking.  It is important to call this before
6946 * attempting a GFP_KERNEL allocation while holding the mddev lock.
6947 * Must be called with mddev_lock held.
6948 *
6949 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
6950 * is dropped, so return -EAGAIN after notifying userspace.
6951 */
6952int md_allow_write(mddev_t *mddev)
6953{
6954	if (!mddev->pers)
6955		return 0;
6956	if (mddev->ro)
6957		return 0;
6958	if (!mddev->pers->sync_request)
6959		return 0;
6960
6961	spin_lock_irq(&mddev->write_lock);
6962	if (mddev->in_sync) {
6963		mddev->in_sync = 0;
6964		set_bit(MD_CHANGE_CLEAN, &mddev->flags);
6965		set_bit(MD_CHANGE_PENDING, &mddev->flags);
6966		if (mddev->safemode_delay &&
6967		    mddev->safemode == 0)
6968			mddev->safemode = 1;
6969		spin_unlock_irq(&mddev->write_lock);
6970		md_update_sb(mddev, 0);
6971		sysfs_notify_dirent_safe(mddev->sysfs_state);
6972	} else
6973		spin_unlock_irq(&mddev->write_lock);
6974
6975	if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
6976		return -EAGAIN;
6977	else
6978		return 0;
6979}
6980EXPORT_SYMBOL_GPL(md_allow_write);
6981
6982#define SYNC_MARKS	10
6983#define	SYNC_MARK_STEP	(3*HZ)
6984void md_do_sync(mddev_t *mddev)
6985{
6986	mddev_t *mddev2;
6987	unsigned int currspeed = 0,
6988		 window;
6989	sector_t max_sectors,j, io_sectors;
6990	unsigned long mark[SYNC_MARKS];
6991	sector_t mark_cnt[SYNC_MARKS];
6992	int last_mark,m;
6993	struct list_head *tmp;
6994	sector_t last_check;
6995	int skipped = 0;
6996	mdk_rdev_t *rdev;
6997	char *desc;
 
6998
6999	/* just incase thread restarts... */
7000	if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7001		return;
7002	if (mddev->ro) /* never try to sync a read-only array */
7003		return;
7004
7005	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7006		if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7007			desc = "data-check";
7008		else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7009			desc = "requested-resync";
7010		else
7011			desc = "resync";
7012	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7013		desc = "reshape";
7014	else
7015		desc = "recovery";
7016
7017	/* we overload curr_resync somewhat here.
7018	 * 0 == not engaged in resync at all
7019	 * 2 == checking that there is no conflict with another sync
7020	 * 1 == like 2, but have yielded to allow conflicting resync to
7021	 *		commense
7022	 * other == active in resync - this many blocks
7023	 *
7024	 * Before starting a resync we must have set curr_resync to
7025	 * 2, and then checked that every "conflicting" array has curr_resync
7026	 * less than ours.  When we find one that is the same or higher
7027	 * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
7028	 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7029	 * This will mean we have to start checking from the beginning again.
7030	 *
7031	 */
7032
7033	do {
7034		mddev->curr_resync = 2;
7035
7036	try_again:
7037		if (kthread_should_stop())
7038			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7039
7040		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7041			goto skip;
7042		for_each_mddev(mddev2, tmp) {
7043			if (mddev2 == mddev)
7044				continue;
7045			if (!mddev->parallel_resync
7046			&&  mddev2->curr_resync
7047			&&  match_mddev_units(mddev, mddev2)) {
7048				DEFINE_WAIT(wq);
7049				if (mddev < mddev2 && mddev->curr_resync == 2) {
7050					/* arbitrarily yield */
7051					mddev->curr_resync = 1;
7052					wake_up(&resync_wait);
7053				}
7054				if (mddev > mddev2 && mddev->curr_resync == 1)
7055					/* no need to wait here, we can wait the next
7056					 * time 'round when curr_resync == 2
7057					 */
7058					continue;
7059				/* We need to wait 'interruptible' so as not to
7060				 * contribute to the load average, and not to
7061				 * be caught by 'softlockup'
7062				 */
7063				prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
7064				if (!kthread_should_stop() &&
7065				    mddev2->curr_resync >= mddev->curr_resync) {
7066					printk(KERN_INFO "md: delaying %s of %s"
7067					       " until %s has finished (they"
7068					       " share one or more physical units)\n",
7069					       desc, mdname(mddev), mdname(mddev2));
7070					mddev_put(mddev2);
7071					if (signal_pending(current))
7072						flush_signals(current);
7073					schedule();
7074					finish_wait(&resync_wait, &wq);
7075					goto try_again;
7076				}
7077				finish_wait(&resync_wait, &wq);
7078			}
7079		}
7080	} while (mddev->curr_resync < 2);
7081
7082	j = 0;
7083	if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7084		/* resync follows the size requested by the personality,
7085		 * which defaults to physical size, but can be virtual size
7086		 */
7087		max_sectors = mddev->resync_max_sectors;
7088		mddev->resync_mismatches = 0;
7089		/* we don't use the checkpoint if there's a bitmap */
7090		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7091			j = mddev->resync_min;
7092		else if (!mddev->bitmap)
7093			j = mddev->recovery_cp;
7094
7095	} else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7096		max_sectors = mddev->dev_sectors;
7097	else {
7098		/* recovery follows the physical size of devices */
7099		max_sectors = mddev->dev_sectors;
7100		j = MaxSector;
7101		rcu_read_lock();
7102		list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
7103			if (rdev->raid_disk >= 0 &&
7104			    !test_bit(Faulty, &rdev->flags) &&
7105			    !test_bit(In_sync, &rdev->flags) &&
7106			    rdev->recovery_offset < j)
7107				j = rdev->recovery_offset;
7108		rcu_read_unlock();
7109	}
7110
7111	printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7112	printk(KERN_INFO "md: minimum _guaranteed_  speed:"
7113		" %d KB/sec/disk.\n", speed_min(mddev));
7114	printk(KERN_INFO "md: using maximum available idle IO bandwidth "
7115	       "(but not more than %d KB/sec) for %s.\n",
7116	       speed_max(mddev), desc);
7117
7118	is_mddev_idle(mddev, 1); /* this initializes IO event counters */
7119
7120	io_sectors = 0;
7121	for (m = 0; m < SYNC_MARKS; m++) {
7122		mark[m] = jiffies;
7123		mark_cnt[m] = io_sectors;
7124	}
7125	last_mark = 0;
7126	mddev->resync_mark = mark[last_mark];
7127	mddev->resync_mark_cnt = mark_cnt[last_mark];
7128
7129	/*
7130	 * Tune reconstruction:
7131	 */
7132	window = 32*(PAGE_SIZE/512);
7133	printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7134		window/2, (unsigned long long)max_sectors/2);
7135
7136	atomic_set(&mddev->recovery_active, 0);
7137	last_check = 0;
7138
7139	if (j>2) {
7140		printk(KERN_INFO 
7141		       "md: resuming %s of %s from checkpoint.\n",
7142		       desc, mdname(mddev));
7143		mddev->curr_resync = j;
7144	}
7145	mddev->curr_resync_completed = j;
7146
 
7147	while (j < max_sectors) {
7148		sector_t sectors;
7149
7150		skipped = 0;
7151
7152		if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7153		    ((mddev->curr_resync > mddev->curr_resync_completed &&
7154		      (mddev->curr_resync - mddev->curr_resync_completed)
7155		      > (max_sectors >> 4)) ||
7156		     (j - mddev->curr_resync_completed)*2
7157		     >= mddev->resync_max - mddev->curr_resync_completed
7158			    )) {
7159			/* time to update curr_resync_completed */
7160			wait_event(mddev->recovery_wait,
7161				   atomic_read(&mddev->recovery_active) == 0);
7162			mddev->curr_resync_completed = j;
7163			set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7164			sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7165		}
7166
7167		while (j >= mddev->resync_max && !kthread_should_stop()) {
7168			/* As this condition is controlled by user-space,
7169			 * we can block indefinitely, so use '_interruptible'
7170			 * to avoid triggering warnings.
7171			 */
7172			flush_signals(current); /* just in case */
7173			wait_event_interruptible(mddev->recovery_wait,
7174						 mddev->resync_max > j
7175						 || kthread_should_stop());
7176		}
7177
7178		if (kthread_should_stop())
7179			goto interrupted;
7180
7181		sectors = mddev->pers->sync_request(mddev, j, &skipped,
7182						  currspeed < speed_min(mddev));
7183		if (sectors == 0) {
7184			set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7185			goto out;
7186		}
7187
7188		if (!skipped) { /* actual IO requested */
7189			io_sectors += sectors;
7190			atomic_add(sectors, &mddev->recovery_active);
7191		}
7192
7193		if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7194			break;
7195
7196		j += sectors;
7197		if (j>1) mddev->curr_resync = j;
7198		mddev->curr_mark_cnt = io_sectors;
7199		if (last_check == 0)
7200			/* this is the earliest that rebuild will be
7201			 * visible in /proc/mdstat
7202			 */
7203			md_new_event(mddev);
7204
7205		if (last_check + window > io_sectors || j == max_sectors)
7206			continue;
7207
7208		last_check = io_sectors;
7209	repeat:
7210		if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7211			/* step marks */
7212			int next = (last_mark+1) % SYNC_MARKS;
7213
7214			mddev->resync_mark = mark[next];
7215			mddev->resync_mark_cnt = mark_cnt[next];
7216			mark[next] = jiffies;
7217			mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
7218			last_mark = next;
7219		}
7220
7221
7222		if (kthread_should_stop())
7223			goto interrupted;
7224
7225
7226		/*
7227		 * this loop exits only if either when we are slower than
7228		 * the 'hard' speed limit, or the system was IO-idle for
7229		 * a jiffy.
7230		 * the system might be non-idle CPU-wise, but we only care
7231		 * about not overloading the IO subsystem. (things like an
7232		 * e2fsck being done on the RAID array should execute fast)
7233		 */
7234		cond_resched();
7235
7236		currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7237			/((jiffies-mddev->resync_mark)/HZ +1) +1;
7238
7239		if (currspeed > speed_min(mddev)) {
7240			if ((currspeed > speed_max(mddev)) ||
7241					!is_mddev_idle(mddev, 0)) {
7242				msleep(500);
7243				goto repeat;
7244			}
7245		}
7246	}
7247	printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
7248	/*
7249	 * this also signals 'finished resyncing' to md_stop
7250	 */
7251 out:
 
7252	wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7253
7254	/* tell personality that we are finished */
7255	mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
7256
7257	if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
7258	    mddev->curr_resync > 2) {
7259		if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7260			if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7261				if (mddev->curr_resync >= mddev->recovery_cp) {
7262					printk(KERN_INFO
7263					       "md: checkpointing %s of %s.\n",
7264					       desc, mdname(mddev));
7265					mddev->recovery_cp = mddev->curr_resync;
 
7266				}
7267			} else
7268				mddev->recovery_cp = MaxSector;
7269		} else {
7270			if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7271				mddev->curr_resync = MaxSector;
7272			rcu_read_lock();
7273			list_for_each_entry_rcu(rdev, &mddev->disks, same_set)
7274				if (rdev->raid_disk >= 0 &&
7275				    mddev->delta_disks >= 0 &&
7276				    !test_bit(Faulty, &rdev->flags) &&
7277				    !test_bit(In_sync, &rdev->flags) &&
7278				    rdev->recovery_offset < mddev->curr_resync)
7279					rdev->recovery_offset = mddev->curr_resync;
7280			rcu_read_unlock();
7281		}
7282	}
 
7283	set_bit(MD_CHANGE_DEVS, &mddev->flags);
7284
7285 skip:
7286	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7287		/* We completed so min/max setting can be forgotten if used. */
7288		if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7289			mddev->resync_min = 0;
7290		mddev->resync_max = MaxSector;
7291	} else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7292		mddev->resync_min = mddev->curr_resync_completed;
7293	mddev->curr_resync = 0;
7294	wake_up(&resync_wait);
7295	set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7296	md_wakeup_thread(mddev->thread);
7297	return;
7298
7299 interrupted:
7300	/*
7301	 * got a signal, exit.
7302	 */
7303	printk(KERN_INFO
7304	       "md: md_do_sync() got signal ... exiting\n");
7305	set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7306	goto out;
7307
7308}
7309EXPORT_SYMBOL_GPL(md_do_sync);
7310
7311static int remove_and_add_spares(mddev_t *mddev)
7312{
7313	mdk_rdev_t *rdev;
7314	int spares = 0;
 
7315
7316	mddev->curr_resync_completed = 0;
7317
7318	list_for_each_entry(rdev, &mddev->disks, same_set)
7319		if (rdev->raid_disk >= 0 &&
7320		    !test_bit(Blocked, &rdev->flags) &&
7321		    (test_bit(Faulty, &rdev->flags) ||
7322		     ! test_bit(In_sync, &rdev->flags)) &&
7323		    atomic_read(&rdev->nr_pending)==0) {
7324			if (mddev->pers->hot_remove_disk(
7325				    mddev, rdev->raid_disk)==0) {
7326				sysfs_unlink_rdev(mddev, rdev);
7327				rdev->raid_disk = -1;
 
7328			}
7329		}
 
 
 
7330
7331	if (mddev->degraded) {
7332		list_for_each_entry(rdev, &mddev->disks, same_set) {
7333			if (rdev->raid_disk >= 0 &&
7334			    !test_bit(In_sync, &rdev->flags) &&
7335			    !test_bit(Faulty, &rdev->flags))
 
 
 
 
 
 
 
 
7336				spares++;
7337			if (rdev->raid_disk < 0
7338			    && !test_bit(Faulty, &rdev->flags)) {
7339				rdev->recovery_offset = 0;
7340				if (mddev->pers->
7341				    hot_add_disk(mddev, rdev) == 0) {
7342					if (sysfs_link_rdev(mddev, rdev))
7343						/* failure here is OK */;
7344					spares++;
7345					md_new_event(mddev);
7346					set_bit(MD_CHANGE_DEVS, &mddev->flags);
7347				} else
7348					break;
7349			}
7350		}
7351	}
 
 
7352	return spares;
7353}
7354
7355static void reap_sync_thread(mddev_t *mddev)
7356{
7357	mdk_rdev_t *rdev;
7358
7359	/* resync has finished, collect result */
7360	md_unregister_thread(&mddev->sync_thread);
7361	if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7362	    !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7363		/* success...*/
7364		/* activate any spares */
7365		if (mddev->pers->spare_active(mddev))
7366			sysfs_notify(&mddev->kobj, NULL,
7367				     "degraded");
 
 
7368	}
7369	if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7370	    mddev->pers->finish_reshape)
7371		mddev->pers->finish_reshape(mddev);
7372	md_update_sb(mddev, 1);
7373
7374	/* if array is no-longer degraded, then any saved_raid_disk
7375	 * information must be scrapped
7376	 */
7377	if (!mddev->degraded)
7378		list_for_each_entry(rdev, &mddev->disks, same_set)
 
 
 
 
7379			rdev->saved_raid_disk = -1;
7380
 
7381	clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7382	clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7383	clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7384	clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7385	clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7386	/* flag recovery needed just to double check */
7387	set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7388	sysfs_notify_dirent_safe(mddev->sysfs_action);
7389	md_new_event(mddev);
7390	if (mddev->event_work.func)
7391		queue_work(md_misc_wq, &mddev->event_work);
7392}
7393
7394/*
7395 * This routine is regularly called by all per-raid-array threads to
7396 * deal with generic issues like resync and super-block update.
7397 * Raid personalities that don't have a thread (linear/raid0) do not
7398 * need this as they never do any recovery or update the superblock.
7399 *
7400 * It does not do any resync itself, but rather "forks" off other threads
7401 * to do that as needed.
7402 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7403 * "->recovery" and create a thread at ->sync_thread.
7404 * When the thread finishes it sets MD_RECOVERY_DONE
7405 * and wakeups up this thread which will reap the thread and finish up.
7406 * This thread also removes any faulty devices (with nr_pending == 0).
7407 *
7408 * The overall approach is:
7409 *  1/ if the superblock needs updating, update it.
7410 *  2/ If a recovery thread is running, don't do anything else.
7411 *  3/ If recovery has finished, clean up, possibly marking spares active.
7412 *  4/ If there are any faulty devices, remove them.
7413 *  5/ If array is degraded, try to add spares devices
7414 *  6/ If array has spares or is not in-sync, start a resync thread.
7415 */
7416void md_check_recovery(mddev_t *mddev)
7417{
7418	if (mddev->suspended)
7419		return;
7420
7421	if (mddev->bitmap)
7422		bitmap_daemon_work(mddev);
7423
7424	if (signal_pending(current)) {
7425		if (mddev->pers->sync_request && !mddev->external) {
7426			printk(KERN_INFO "md: %s in immediate safe mode\n",
7427			       mdname(mddev));
7428			mddev->safemode = 2;
7429		}
7430		flush_signals(current);
7431	}
7432
7433	if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7434		return;
7435	if ( ! (
7436		(mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
7437		test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7438		test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
7439		(mddev->external == 0 && mddev->safemode == 1) ||
7440		(mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7441		 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
7442		))
7443		return;
7444
7445	if (mddev_trylock(mddev)) {
7446		int spares = 0;
7447
7448		if (mddev->ro) {
7449			/* Only thing we do on a ro array is remove
7450			 * failed devices.
7451			 */
7452			mdk_rdev_t *rdev;
7453			list_for_each_entry(rdev, &mddev->disks, same_set)
7454				if (rdev->raid_disk >= 0 &&
7455				    !test_bit(Blocked, &rdev->flags) &&
7456				    test_bit(Faulty, &rdev->flags) &&
7457				    atomic_read(&rdev->nr_pending)==0) {
7458					if (mddev->pers->hot_remove_disk(
7459						    mddev, rdev->raid_disk)==0) {
7460						sysfs_unlink_rdev(mddev, rdev);
7461						rdev->raid_disk = -1;
7462					}
7463				}
7464			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7465			goto unlock;
7466		}
7467
7468		if (!mddev->external) {
7469			int did_change = 0;
7470			spin_lock_irq(&mddev->write_lock);
7471			if (mddev->safemode &&
7472			    !atomic_read(&mddev->writes_pending) &&
7473			    !mddev->in_sync &&
7474			    mddev->recovery_cp == MaxSector) {
7475				mddev->in_sync = 1;
7476				did_change = 1;
7477				set_bit(MD_CHANGE_CLEAN, &mddev->flags);
7478			}
7479			if (mddev->safemode == 1)
7480				mddev->safemode = 0;
7481			spin_unlock_irq(&mddev->write_lock);
7482			if (did_change)
7483				sysfs_notify_dirent_safe(mddev->sysfs_state);
7484		}
7485
7486		if (mddev->flags)
7487			md_update_sb(mddev, 0);
7488
7489		if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7490		    !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7491			/* resync/recovery still happening */
7492			clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7493			goto unlock;
7494		}
7495		if (mddev->sync_thread) {
7496			reap_sync_thread(mddev);
7497			goto unlock;
7498		}
7499		/* Set RUNNING before clearing NEEDED to avoid
7500		 * any transients in the value of "sync_action".
7501		 */
7502		set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7503		clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7504		/* Clear some bits that don't mean anything, but
7505		 * might be left set
7506		 */
7507		clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7508		clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
7509
7510		if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
 
7511			goto unlock;
7512		/* no recovery is running.
7513		 * remove any failed drives, then
7514		 * add spares if possible.
7515		 * Spare are also removed and re-added, to allow
7516		 * the personality to fail the re-add.
7517		 */
7518
7519		if (mddev->reshape_position != MaxSector) {
7520			if (mddev->pers->check_reshape == NULL ||
7521			    mddev->pers->check_reshape(mddev) != 0)
7522				/* Cannot proceed */
7523				goto unlock;
7524			set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7525			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7526		} else if ((spares = remove_and_add_spares(mddev))) {
7527			clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7528			clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7529			clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7530			set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7531		} else if (mddev->recovery_cp < MaxSector) {
7532			set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7533			clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
7534		} else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7535			/* nothing to be done ... */
7536			goto unlock;
7537
7538		if (mddev->pers->sync_request) {
7539			if (spares && mddev->bitmap && ! mddev->bitmap->file) {
7540				/* We are adding a device or devices to an array
7541				 * which has the bitmap stored on all devices.
7542				 * So make sure all bitmap pages get written
7543				 */
7544				bitmap_write_all(mddev->bitmap);
7545			}
7546			mddev->sync_thread = md_register_thread(md_do_sync,
7547								mddev,
7548								"resync");
7549			if (!mddev->sync_thread) {
7550				printk(KERN_ERR "%s: could not start resync"
7551					" thread...\n", 
7552					mdname(mddev));
7553				/* leave the spares where they are, it shouldn't hurt */
7554				clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7555				clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7556				clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7557				clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7558				clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7559			} else
7560				md_wakeup_thread(mddev->sync_thread);
7561			sysfs_notify_dirent_safe(mddev->sysfs_action);
7562			md_new_event(mddev);
7563		}
7564	unlock:
7565		if (!mddev->sync_thread) {
7566			clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7567			if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7568					       &mddev->recovery))
7569				if (mddev->sysfs_action)
7570					sysfs_notify_dirent_safe(mddev->sysfs_action);
7571		}
7572		mddev_unlock(mddev);
7573	}
7574}
7575
7576void md_wait_for_blocked_rdev(mdk_rdev_t *rdev, mddev_t *mddev)
7577{
7578	sysfs_notify_dirent_safe(rdev->sysfs_state);
7579	wait_event_timeout(rdev->blocked_wait,
7580			   !test_bit(Blocked, &rdev->flags) &&
7581			   !test_bit(BlockedBadBlocks, &rdev->flags),
7582			   msecs_to_jiffies(5000));
7583	rdev_dec_pending(rdev, mddev);
7584}
7585EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7586
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7587
7588/* Bad block management.
7589 * We can record which blocks on each device are 'bad' and so just
7590 * fail those blocks, or that stripe, rather than the whole device.
7591 * Entries in the bad-block table are 64bits wide.  This comprises:
7592 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7593 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7594 *  A 'shift' can be set so that larger blocks are tracked and
7595 *  consequently larger devices can be covered.
7596 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7597 *
7598 * Locking of the bad-block table uses a seqlock so md_is_badblock
7599 * might need to retry if it is very unlucky.
7600 * We will sometimes want to check for bad blocks in a bi_end_io function,
7601 * so we use the write_seqlock_irq variant.
7602 *
7603 * When looking for a bad block we specify a range and want to
7604 * know if any block in the range is bad.  So we binary-search
7605 * to the last range that starts at-or-before the given endpoint,
7606 * (or "before the sector after the target range")
7607 * then see if it ends after the given start.
7608 * We return
7609 *  0 if there are no known bad blocks in the range
7610 *  1 if there are known bad block which are all acknowledged
7611 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7612 * plus the start/length of the first bad section we overlap.
7613 */
7614int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7615		   sector_t *first_bad, int *bad_sectors)
7616{
7617	int hi;
7618	int lo = 0;
7619	u64 *p = bb->page;
7620	int rv = 0;
7621	sector_t target = s + sectors;
7622	unsigned seq;
7623
7624	if (bb->shift > 0) {
7625		/* round the start down, and the end up */
7626		s >>= bb->shift;
7627		target += (1<<bb->shift) - 1;
7628		target >>= bb->shift;
7629		sectors = target - s;
7630	}
7631	/* 'target' is now the first block after the bad range */
7632
7633retry:
7634	seq = read_seqbegin(&bb->lock);
7635
7636	hi = bb->count;
7637
7638	/* Binary search between lo and hi for 'target'
7639	 * i.e. for the last range that starts before 'target'
7640	 */
7641	/* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7642	 * are known not to be the last range before target.
7643	 * VARIANT: hi-lo is the number of possible
7644	 * ranges, and decreases until it reaches 1
7645	 */
7646	while (hi - lo > 1) {
7647		int mid = (lo + hi) / 2;
7648		sector_t a = BB_OFFSET(p[mid]);
7649		if (a < target)
7650			/* This could still be the one, earlier ranges
7651			 * could not. */
7652			lo = mid;
7653		else
7654			/* This and later ranges are definitely out. */
7655			hi = mid;
7656	}
7657	/* 'lo' might be the last that started before target, but 'hi' isn't */
7658	if (hi > lo) {
7659		/* need to check all range that end after 's' to see if
7660		 * any are unacknowledged.
7661		 */
7662		while (lo >= 0 &&
7663		       BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7664			if (BB_OFFSET(p[lo]) < target) {
7665				/* starts before the end, and finishes after
7666				 * the start, so they must overlap
7667				 */
7668				if (rv != -1 && BB_ACK(p[lo]))
7669					rv = 1;
7670				else
7671					rv = -1;
7672				*first_bad = BB_OFFSET(p[lo]);
7673				*bad_sectors = BB_LEN(p[lo]);
7674			}
7675			lo--;
7676		}
7677	}
7678
7679	if (read_seqretry(&bb->lock, seq))
7680		goto retry;
7681
7682	return rv;
7683}
7684EXPORT_SYMBOL_GPL(md_is_badblock);
7685
7686/*
7687 * Add a range of bad blocks to the table.
7688 * This might extend the table, or might contract it
7689 * if two adjacent ranges can be merged.
7690 * We binary-search to find the 'insertion' point, then
7691 * decide how best to handle it.
7692 */
7693static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7694			    int acknowledged)
7695{
7696	u64 *p;
7697	int lo, hi;
7698	int rv = 1;
7699
7700	if (bb->shift < 0)
7701		/* badblocks are disabled */
7702		return 0;
7703
7704	if (bb->shift) {
7705		/* round the start down, and the end up */
7706		sector_t next = s + sectors;
7707		s >>= bb->shift;
7708		next += (1<<bb->shift) - 1;
7709		next >>= bb->shift;
7710		sectors = next - s;
7711	}
7712
7713	write_seqlock_irq(&bb->lock);
7714
7715	p = bb->page;
7716	lo = 0;
7717	hi = bb->count;
7718	/* Find the last range that starts at-or-before 's' */
7719	while (hi - lo > 1) {
7720		int mid = (lo + hi) / 2;
7721		sector_t a = BB_OFFSET(p[mid]);
7722		if (a <= s)
7723			lo = mid;
7724		else
7725			hi = mid;
7726	}
7727	if (hi > lo && BB_OFFSET(p[lo]) > s)
7728		hi = lo;
7729
7730	if (hi > lo) {
7731		/* we found a range that might merge with the start
7732		 * of our new range
7733		 */
7734		sector_t a = BB_OFFSET(p[lo]);
7735		sector_t e = a + BB_LEN(p[lo]);
7736		int ack = BB_ACK(p[lo]);
7737		if (e >= s) {
7738			/* Yes, we can merge with a previous range */
7739			if (s == a && s + sectors >= e)
7740				/* new range covers old */
7741				ack = acknowledged;
7742			else
7743				ack = ack && acknowledged;
7744
7745			if (e < s + sectors)
7746				e = s + sectors;
7747			if (e - a <= BB_MAX_LEN) {
7748				p[lo] = BB_MAKE(a, e-a, ack);
7749				s = e;
7750			} else {
7751				/* does not all fit in one range,
7752				 * make p[lo] maximal
7753				 */
7754				if (BB_LEN(p[lo]) != BB_MAX_LEN)
7755					p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
7756				s = a + BB_MAX_LEN;
7757			}
7758			sectors = e - s;
7759		}
7760	}
7761	if (sectors && hi < bb->count) {
7762		/* 'hi' points to the first range that starts after 's'.
7763		 * Maybe we can merge with the start of that range */
7764		sector_t a = BB_OFFSET(p[hi]);
7765		sector_t e = a + BB_LEN(p[hi]);
7766		int ack = BB_ACK(p[hi]);
7767		if (a <= s + sectors) {
7768			/* merging is possible */
7769			if (e <= s + sectors) {
7770				/* full overlap */
7771				e = s + sectors;
7772				ack = acknowledged;
7773			} else
7774				ack = ack && acknowledged;
7775
7776			a = s;
7777			if (e - a <= BB_MAX_LEN) {
7778				p[hi] = BB_MAKE(a, e-a, ack);
7779				s = e;
7780			} else {
7781				p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
7782				s = a + BB_MAX_LEN;
7783			}
7784			sectors = e - s;
7785			lo = hi;
7786			hi++;
7787		}
7788	}
7789	if (sectors == 0 && hi < bb->count) {
7790		/* we might be able to combine lo and hi */
7791		/* Note: 's' is at the end of 'lo' */
7792		sector_t a = BB_OFFSET(p[hi]);
7793		int lolen = BB_LEN(p[lo]);
7794		int hilen = BB_LEN(p[hi]);
7795		int newlen = lolen + hilen - (s - a);
7796		if (s >= a && newlen < BB_MAX_LEN) {
7797			/* yes, we can combine them */
7798			int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
7799			p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
7800			memmove(p + hi, p + hi + 1,
7801				(bb->count - hi - 1) * 8);
7802			bb->count--;
7803		}
7804	}
7805	while (sectors) {
7806		/* didn't merge (it all).
7807		 * Need to add a range just before 'hi' */
7808		if (bb->count >= MD_MAX_BADBLOCKS) {
7809			/* No room for more */
7810			rv = 0;
7811			break;
7812		} else {
7813			int this_sectors = sectors;
7814			memmove(p + hi + 1, p + hi,
7815				(bb->count - hi) * 8);
7816			bb->count++;
7817
7818			if (this_sectors > BB_MAX_LEN)
7819				this_sectors = BB_MAX_LEN;
7820			p[hi] = BB_MAKE(s, this_sectors, acknowledged);
7821			sectors -= this_sectors;
7822			s += this_sectors;
7823		}
7824	}
7825
7826	bb->changed = 1;
7827	if (!acknowledged)
7828		bb->unacked_exist = 1;
7829	write_sequnlock_irq(&bb->lock);
7830
7831	return rv;
7832}
7833
7834int rdev_set_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors,
7835		       int acknowledged)
7836{
7837	int rv = md_set_badblocks(&rdev->badblocks,
7838				  s + rdev->data_offset, sectors, acknowledged);
 
 
 
 
 
7839	if (rv) {
7840		/* Make sure they get written out promptly */
 
7841		set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
7842		md_wakeup_thread(rdev->mddev->thread);
7843	}
7844	return rv;
7845}
7846EXPORT_SYMBOL_GPL(rdev_set_badblocks);
7847
7848/*
7849 * Remove a range of bad blocks from the table.
7850 * This may involve extending the table if we spilt a region,
7851 * but it must not fail.  So if the table becomes full, we just
7852 * drop the remove request.
7853 */
7854static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
7855{
7856	u64 *p;
7857	int lo, hi;
7858	sector_t target = s + sectors;
7859	int rv = 0;
7860
7861	if (bb->shift > 0) {
7862		/* When clearing we round the start up and the end down.
7863		 * This should not matter as the shift should align with
7864		 * the block size and no rounding should ever be needed.
7865		 * However it is better the think a block is bad when it
7866		 * isn't than to think a block is not bad when it is.
7867		 */
7868		s += (1<<bb->shift) - 1;
7869		s >>= bb->shift;
7870		target >>= bb->shift;
7871		sectors = target - s;
7872	}
7873
7874	write_seqlock_irq(&bb->lock);
7875
7876	p = bb->page;
7877	lo = 0;
7878	hi = bb->count;
7879	/* Find the last range that starts before 'target' */
7880	while (hi - lo > 1) {
7881		int mid = (lo + hi) / 2;
7882		sector_t a = BB_OFFSET(p[mid]);
7883		if (a < target)
7884			lo = mid;
7885		else
7886			hi = mid;
7887	}
7888	if (hi > lo) {
7889		/* p[lo] is the last range that could overlap the
7890		 * current range.  Earlier ranges could also overlap,
7891		 * but only this one can overlap the end of the range.
7892		 */
7893		if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
7894			/* Partial overlap, leave the tail of this range */
7895			int ack = BB_ACK(p[lo]);
7896			sector_t a = BB_OFFSET(p[lo]);
7897			sector_t end = a + BB_LEN(p[lo]);
7898
7899			if (a < s) {
7900				/* we need to split this range */
7901				if (bb->count >= MD_MAX_BADBLOCKS) {
7902					rv = 0;
7903					goto out;
7904				}
7905				memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
7906				bb->count++;
7907				p[lo] = BB_MAKE(a, s-a, ack);
7908				lo++;
7909			}
7910			p[lo] = BB_MAKE(target, end - target, ack);
7911			/* there is no longer an overlap */
7912			hi = lo;
7913			lo--;
7914		}
7915		while (lo >= 0 &&
7916		       BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7917			/* This range does overlap */
7918			if (BB_OFFSET(p[lo]) < s) {
7919				/* Keep the early parts of this range. */
7920				int ack = BB_ACK(p[lo]);
7921				sector_t start = BB_OFFSET(p[lo]);
7922				p[lo] = BB_MAKE(start, s - start, ack);
7923				/* now low doesn't overlap, so.. */
7924				break;
7925			}
7926			lo--;
7927		}
7928		/* 'lo' is strictly before, 'hi' is strictly after,
7929		 * anything between needs to be discarded
7930		 */
7931		if (hi - lo > 1) {
7932			memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
7933			bb->count -= (hi - lo - 1);
7934		}
7935	}
7936
7937	bb->changed = 1;
7938out:
7939	write_sequnlock_irq(&bb->lock);
7940	return rv;
7941}
7942
7943int rdev_clear_badblocks(mdk_rdev_t *rdev, sector_t s, int sectors)
 
7944{
 
 
 
 
7945	return md_clear_badblocks(&rdev->badblocks,
7946				  s + rdev->data_offset,
7947				  sectors);
7948}
7949EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
7950
7951/*
7952 * Acknowledge all bad blocks in a list.
7953 * This only succeeds if ->changed is clear.  It is used by
7954 * in-kernel metadata updates
7955 */
7956void md_ack_all_badblocks(struct badblocks *bb)
7957{
7958	if (bb->page == NULL || bb->changed)
7959		/* no point even trying */
7960		return;
7961	write_seqlock_irq(&bb->lock);
7962
7963	if (bb->changed == 0) {
7964		u64 *p = bb->page;
7965		int i;
7966		for (i = 0; i < bb->count ; i++) {
7967			if (!BB_ACK(p[i])) {
7968				sector_t start = BB_OFFSET(p[i]);
7969				int len = BB_LEN(p[i]);
7970				p[i] = BB_MAKE(start, len, 1);
7971			}
7972		}
7973		bb->unacked_exist = 0;
7974	}
7975	write_sequnlock_irq(&bb->lock);
7976}
7977EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
7978
7979/* sysfs access to bad-blocks list.
7980 * We present two files.
7981 * 'bad-blocks' lists sector numbers and lengths of ranges that
7982 *    are recorded as bad.  The list is truncated to fit within
7983 *    the one-page limit of sysfs.
7984 *    Writing "sector length" to this file adds an acknowledged
7985 *    bad block list.
7986 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
7987 *    been acknowledged.  Writing to this file adds bad blocks
7988 *    without acknowledging them.  This is largely for testing.
7989 */
7990
7991static ssize_t
7992badblocks_show(struct badblocks *bb, char *page, int unack)
7993{
7994	size_t len;
7995	int i;
7996	u64 *p = bb->page;
7997	unsigned seq;
7998
7999	if (bb->shift < 0)
8000		return 0;
8001
8002retry:
8003	seq = read_seqbegin(&bb->lock);
8004
8005	len = 0;
8006	i = 0;
8007
8008	while (len < PAGE_SIZE && i < bb->count) {
8009		sector_t s = BB_OFFSET(p[i]);
8010		unsigned int length = BB_LEN(p[i]);
8011		int ack = BB_ACK(p[i]);
8012		i++;
8013
8014		if (unack && ack)
8015			continue;
8016
8017		len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8018				(unsigned long long)s << bb->shift,
8019				length << bb->shift);
8020	}
8021	if (unack && len == 0)
8022		bb->unacked_exist = 0;
8023
8024	if (read_seqretry(&bb->lock, seq))
8025		goto retry;
8026
8027	return len;
8028}
8029
8030#define DO_DEBUG 1
8031
8032static ssize_t
8033badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8034{
8035	unsigned long long sector;
8036	int length;
8037	char newline;
8038#ifdef DO_DEBUG
8039	/* Allow clearing via sysfs *only* for testing/debugging.
8040	 * Normally only a successful write may clear a badblock
8041	 */
8042	int clear = 0;
8043	if (page[0] == '-') {
8044		clear = 1;
8045		page++;
8046	}
8047#endif /* DO_DEBUG */
8048
8049	switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8050	case 3:
8051		if (newline != '\n')
8052			return -EINVAL;
8053	case 2:
8054		if (length <= 0)
8055			return -EINVAL;
8056		break;
8057	default:
8058		return -EINVAL;
8059	}
8060
8061#ifdef DO_DEBUG
8062	if (clear) {
8063		md_clear_badblocks(bb, sector, length);
8064		return len;
8065	}
8066#endif /* DO_DEBUG */
8067	if (md_set_badblocks(bb, sector, length, !unack))
8068		return len;
8069	else
8070		return -ENOSPC;
8071}
8072
8073static int md_notify_reboot(struct notifier_block *this,
8074			    unsigned long code, void *x)
8075{
8076	struct list_head *tmp;
8077	mddev_t *mddev;
 
8078
8079	if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
8080
8081		printk(KERN_INFO "md: stopping all md devices.\n");
8082
8083		for_each_mddev(mddev, tmp)
8084			if (mddev_trylock(mddev)) {
8085				/* Force a switch to readonly even array
8086				 * appears to still be in use.  Hence
8087				 * the '100'.
8088				 */
8089				md_set_readonly(mddev, 100);
8090				mddev_unlock(mddev);
8091			}
8092		/*
8093		 * certain more exotic SCSI devices are known to be
8094		 * volatile wrt too early system reboots. While the
8095		 * right place to handle this issue is the given
8096		 * driver, we do want to have a safe RAID driver ...
8097		 */
8098		mdelay(1000*1);
8099	}
 
 
 
 
 
 
 
 
 
8100	return NOTIFY_DONE;
8101}
8102
8103static struct notifier_block md_notifier = {
8104	.notifier_call	= md_notify_reboot,
8105	.next		= NULL,
8106	.priority	= INT_MAX, /* before any real devices */
8107};
8108
8109static void md_geninit(void)
8110{
8111	dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
8112
8113	proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
8114}
8115
8116static int __init md_init(void)
8117{
8118	int ret = -ENOMEM;
8119
8120	md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
8121	if (!md_wq)
8122		goto err_wq;
8123
8124	md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8125	if (!md_misc_wq)
8126		goto err_misc_wq;
8127
8128	if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8129		goto err_md;
8130
8131	if ((ret = register_blkdev(0, "mdp")) < 0)
8132		goto err_mdp;
8133	mdp_major = ret;
8134
8135	blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
8136			    md_probe, NULL, NULL);
8137	blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
8138			    md_probe, NULL, NULL);
8139
8140	register_reboot_notifier(&md_notifier);
8141	raid_table_header = register_sysctl_table(raid_root_table);
8142
8143	md_geninit();
8144	return 0;
8145
8146err_mdp:
8147	unregister_blkdev(MD_MAJOR, "md");
8148err_md:
8149	destroy_workqueue(md_misc_wq);
8150err_misc_wq:
8151	destroy_workqueue(md_wq);
8152err_wq:
8153	return ret;
8154}
8155
8156#ifndef MODULE
8157
8158/*
8159 * Searches all registered partitions for autorun RAID arrays
8160 * at boot time.
8161 */
8162
8163static LIST_HEAD(all_detected_devices);
8164struct detected_devices_node {
8165	struct list_head list;
8166	dev_t dev;
8167};
8168
8169void md_autodetect_dev(dev_t dev)
8170{
8171	struct detected_devices_node *node_detected_dev;
8172
8173	node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8174	if (node_detected_dev) {
8175		node_detected_dev->dev = dev;
8176		list_add_tail(&node_detected_dev->list, &all_detected_devices);
8177	} else {
8178		printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8179			", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8180	}
8181}
8182
8183
8184static void autostart_arrays(int part)
8185{
8186	mdk_rdev_t *rdev;
8187	struct detected_devices_node *node_detected_dev;
8188	dev_t dev;
8189	int i_scanned, i_passed;
8190
8191	i_scanned = 0;
8192	i_passed = 0;
8193
8194	printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8195
8196	while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8197		i_scanned++;
8198		node_detected_dev = list_entry(all_detected_devices.next,
8199					struct detected_devices_node, list);
8200		list_del(&node_detected_dev->list);
8201		dev = node_detected_dev->dev;
8202		kfree(node_detected_dev);
8203		rdev = md_import_device(dev,0, 90);
8204		if (IS_ERR(rdev))
8205			continue;
8206
8207		if (test_bit(Faulty, &rdev->flags)) {
8208			MD_BUG();
8209			continue;
8210		}
8211		set_bit(AutoDetected, &rdev->flags);
8212		list_add(&rdev->same_set, &pending_raid_disks);
8213		i_passed++;
8214	}
8215
8216	printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8217						i_scanned, i_passed);
8218
8219	autorun_devices(part);
8220}
8221
8222#endif /* !MODULE */
8223
8224static __exit void md_exit(void)
8225{
8226	mddev_t *mddev;
8227	struct list_head *tmp;
8228
8229	blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
8230	blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
8231
8232	unregister_blkdev(MD_MAJOR,"md");
8233	unregister_blkdev(mdp_major, "mdp");
8234	unregister_reboot_notifier(&md_notifier);
8235	unregister_sysctl_table(raid_table_header);
8236	remove_proc_entry("mdstat", NULL);
8237	for_each_mddev(mddev, tmp) {
8238		export_array(mddev);
8239		mddev->hold_active = 0;
8240	}
8241	destroy_workqueue(md_misc_wq);
8242	destroy_workqueue(md_wq);
8243}
8244
8245subsys_initcall(md_init);
8246module_exit(md_exit)
8247
8248static int get_ro(char *buffer, struct kernel_param *kp)
8249{
8250	return sprintf(buffer, "%d", start_readonly);
8251}
8252static int set_ro(const char *val, struct kernel_param *kp)
8253{
8254	char *e;
8255	int num = simple_strtoul(val, &e, 10);
8256	if (*val && (*e == '\0' || *e == '\n')) {
8257		start_readonly = num;
8258		return 0;
8259	}
8260	return -EINVAL;
8261}
8262
8263module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8264module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
8265
8266module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
8267
8268EXPORT_SYMBOL(register_md_personality);
8269EXPORT_SYMBOL(unregister_md_personality);
8270EXPORT_SYMBOL(md_error);
8271EXPORT_SYMBOL(md_done_sync);
8272EXPORT_SYMBOL(md_write_start);
8273EXPORT_SYMBOL(md_write_end);
8274EXPORT_SYMBOL(md_register_thread);
8275EXPORT_SYMBOL(md_unregister_thread);
8276EXPORT_SYMBOL(md_wakeup_thread);
8277EXPORT_SYMBOL(md_check_recovery);
8278MODULE_LICENSE("GPL");
8279MODULE_DESCRIPTION("MD RAID framework");
8280MODULE_ALIAS("md");
8281MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);