blob: 7564c44b8045a1fee38cf31daf28c8de467c3c09 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
NeilBrown32a76272005-06-21 17:17:14 -070022 - persistent bitmap code
23 Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
24
Linus Torvalds1da177e2005-04-16 15:20:36 -070025 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
NeilBrowna6fb0932005-09-09 16:23:56 -070035#include <linux/kthread.h>
NeilBrownbff61972009-03-31 14:33:13 +110036#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/sysctl.h>
NeilBrownbff61972009-03-31 14:33:13 +110038#include <linux/seq_file.h>
Al Viroff01bb42011-09-16 02:31:11 -040039#include <linux/fs.h>
NeilBrownd7603b72006-01-06 00:20:30 -080040#include <linux/poll.h>
NeilBrown16f17b32006-06-26 00:27:37 -070041#include <linux/ctype.h>
André Goddard Rosae7d28602009-12-14 18:01:06 -080042#include <linux/string.h>
NeilBrownfb4d8c72008-10-13 11:55:12 +110043#include <linux/hdreg.h>
44#include <linux/proc_fs.h>
45#include <linux/random.h>
Paul Gortmaker056075c2011-07-03 13:58:33 -040046#include <linux/module.h>
NeilBrownfb4d8c72008-10-13 11:55:12 +110047#include <linux/reboot.h>
NeilBrown32a76272005-06-21 17:17:14 -070048#include <linux/file.h>
Arnd Bergmannaa98aa32009-12-14 12:50:05 +110049#include <linux/compat.h>
Stephen Rothwell25570722008-10-15 09:09:21 +110050#include <linux/delay.h>
NeilBrownbff61972009-03-31 14:33:13 +110051#include <linux/raid/md_p.h>
52#include <linux/raid/md_u.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090053#include <linux/slab.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110054#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110055#include "bitmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Linus Torvalds1da177e2005-04-16 15:20:36 -070057#ifndef MODULE
NeilBrownd710e132008-10-13 11:55:12 +110058static void autostart_arrays(int part);
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#endif
60
NeilBrown01f96c02011-09-21 15:30:20 +100061/* pers_list is a list of registered personalities protected
62 * by pers_lock.
63 * pers_lock does extra service to protect accesses to
64 * mddev->thread when the mutex cannot be held.
65 */
NeilBrown2604b702006-01-06 00:20:36 -080066static LIST_HEAD(pers_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067static DEFINE_SPINLOCK(pers_lock);
68
Adrian Bunk5e563412006-06-26 00:27:42 -070069static void md_print_devices(void);
70
Bernd Schubert90b08712008-05-23 13:04:38 -070071static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
Tejun Heoe804ac72010-10-15 15:36:08 +020072static struct workqueue_struct *md_wq;
73static struct workqueue_struct *md_misc_wq;
Bernd Schubert90b08712008-05-23 13:04:38 -070074
Adrian Bunk5e563412006-06-26 00:27:42 -070075#define MD_BUG(x...) { printk("md: bug in file %s, line %d\n", __FILE__, __LINE__); md_print_devices(); }
76
Linus Torvalds1da177e2005-04-16 15:20:36 -070077/*
Robert Becker1e509152009-12-14 12:49:58 +110078 * Default number of read corrections we'll attempt on an rdev
79 * before ejecting it from the array. We divide the read error
80 * count by 2 for every hour elapsed between read errors.
81 */
82#define MD_DEFAULT_MAX_CORRECTED_READ_ERRORS 20
83/*
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
85 * is 1000 KB/sec, so the extra system load does not show up that much.
86 * Increase it if you want to have more _guaranteed_ speed. Note that
Adrian Bunk338cec32005-09-10 00:26:54 -070087 * the RAID driver will use the maximum available bandwidth if the IO
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * subsystem is idle. There is also an 'absolute maximum' reconstruction
89 * speed limit - in case reconstruction slows down your system despite
90 * idle IO detection.
91 *
92 * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
NeilBrown88202a02006-01-06 00:21:36 -080093 * or /sys/block/mdX/md/sync_speed_{min,max}
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 */
95
96static int sysctl_speed_limit_min = 1000;
97static int sysctl_speed_limit_max = 200000;
NeilBrownfd01b882011-10-11 16:47:53 +110098static inline int speed_min(struct mddev *mddev)
NeilBrown88202a02006-01-06 00:21:36 -080099{
100 return mddev->sync_speed_min ?
101 mddev->sync_speed_min : sysctl_speed_limit_min;
102}
103
NeilBrownfd01b882011-10-11 16:47:53 +1100104static inline int speed_max(struct mddev *mddev)
NeilBrown88202a02006-01-06 00:21:36 -0800105{
106 return mddev->sync_speed_max ?
107 mddev->sync_speed_max : sysctl_speed_limit_max;
108}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109
110static struct ctl_table_header *raid_table_header;
111
112static ctl_table raid_table[] = {
113 {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 .procname = "speed_limit_min",
115 .data = &sysctl_speed_limit_min,
116 .maxlen = sizeof(int),
NeilBrown80ca3a42006-07-10 04:44:18 -0700117 .mode = S_IRUGO|S_IWUSR,
Eric W. Biederman6d456112009-11-16 03:11:48 -0800118 .proc_handler = proc_dointvec,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 },
120 {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 .procname = "speed_limit_max",
122 .data = &sysctl_speed_limit_max,
123 .maxlen = sizeof(int),
NeilBrown80ca3a42006-07-10 04:44:18 -0700124 .mode = S_IRUGO|S_IWUSR,
Eric W. Biederman6d456112009-11-16 03:11:48 -0800125 .proc_handler = proc_dointvec,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 },
Eric W. Biederman894d2492009-11-05 14:34:02 -0800127 { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128};
129
130static ctl_table raid_dir_table[] = {
131 {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 .procname = "raid",
133 .maxlen = 0,
NeilBrown80ca3a42006-07-10 04:44:18 -0700134 .mode = S_IRUGO|S_IXUGO,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 .child = raid_table,
136 },
Eric W. Biederman894d2492009-11-05 14:34:02 -0800137 { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138};
139
140static ctl_table raid_root_table[] = {
141 {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 .procname = "dev",
143 .maxlen = 0,
144 .mode = 0555,
145 .child = raid_dir_table,
146 },
Eric W. Biederman894d2492009-11-05 14:34:02 -0800147 { }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148};
149
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -0700150static const struct block_device_operations md_fops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
NeilBrownf91de922005-11-08 21:39:36 -0800152static int start_readonly;
153
NeilBrowna167f662010-10-26 18:31:13 +1100154/* bio_clone_mddev
155 * like bio_clone, but with a local bio set
156 */
157
NeilBrowna167f662010-10-26 18:31:13 +1100158struct bio *bio_alloc_mddev(gfp_t gfp_mask, int nr_iovecs,
NeilBrownfd01b882011-10-11 16:47:53 +1100159 struct mddev *mddev)
NeilBrowna167f662010-10-26 18:31:13 +1100160{
161 struct bio *b;
NeilBrowna167f662010-10-26 18:31:13 +1100162
163 if (!mddev || !mddev->bio_set)
164 return bio_alloc(gfp_mask, nr_iovecs);
165
Kent Overstreet395c72a2012-09-06 15:34:55 -0700166 b = bio_alloc_bioset(gfp_mask, nr_iovecs, mddev->bio_set);
NeilBrowna167f662010-10-26 18:31:13 +1100167 if (!b)
168 return NULL;
NeilBrowna167f662010-10-26 18:31:13 +1100169 return b;
170}
171EXPORT_SYMBOL_GPL(bio_alloc_mddev);
172
173struct bio *bio_clone_mddev(struct bio *bio, gfp_t gfp_mask,
NeilBrownfd01b882011-10-11 16:47:53 +1100174 struct mddev *mddev)
NeilBrowna167f662010-10-26 18:31:13 +1100175{
NeilBrowna167f662010-10-26 18:31:13 +1100176 if (!mddev || !mddev->bio_set)
177 return bio_clone(bio, gfp_mask);
178
Kent Overstreetbf800ef2012-09-06 15:35:02 -0700179 return bio_clone_bioset(bio, gfp_mask, mddev->bio_set);
NeilBrowna167f662010-10-26 18:31:13 +1100180}
181EXPORT_SYMBOL_GPL(bio_clone_mddev);
182
NeilBrownd2eb35a2011-07-28 11:31:48 +1000183void md_trim_bio(struct bio *bio, int offset, int size)
184{
185 /* 'bio' is a cloned bio which we need to trim to match
186 * the given offset and size.
187 * This requires adjusting bi_sector, bi_size, and bi_io_vec
188 */
189 int i;
190 struct bio_vec *bvec;
191 int sofar = 0;
192
193 size <<= 9;
194 if (offset == 0 && size == bio->bi_size)
195 return;
196
197 bio->bi_sector += offset;
198 bio->bi_size = size;
199 offset <<= 9;
200 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
201
202 while (bio->bi_idx < bio->bi_vcnt &&
203 bio->bi_io_vec[bio->bi_idx].bv_len <= offset) {
204 /* remove this whole bio_vec */
205 offset -= bio->bi_io_vec[bio->bi_idx].bv_len;
206 bio->bi_idx++;
207 }
208 if (bio->bi_idx < bio->bi_vcnt) {
209 bio->bi_io_vec[bio->bi_idx].bv_offset += offset;
210 bio->bi_io_vec[bio->bi_idx].bv_len -= offset;
211 }
212 /* avoid any complications with bi_idx being non-zero*/
213 if (bio->bi_idx) {
214 memmove(bio->bi_io_vec, bio->bi_io_vec+bio->bi_idx,
215 (bio->bi_vcnt - bio->bi_idx) * sizeof(struct bio_vec));
216 bio->bi_vcnt -= bio->bi_idx;
217 bio->bi_idx = 0;
218 }
219 /* Make sure vcnt and last bv are not too big */
220 bio_for_each_segment(bvec, bio, i) {
221 if (sofar + bvec->bv_len > size)
222 bvec->bv_len = size - sofar;
223 if (bvec->bv_len == 0) {
224 bio->bi_vcnt = i;
225 break;
226 }
227 sofar += bvec->bv_len;
228 }
229}
230EXPORT_SYMBOL_GPL(md_trim_bio);
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232/*
NeilBrownd7603b72006-01-06 00:20:30 -0800233 * We have a system wide 'event count' that is incremented
234 * on any 'interesting' event, and readers of /proc/mdstat
235 * can use 'poll' or 'select' to find out when the event
236 * count increases.
237 *
238 * Events are:
239 * start array, stop array, error, add device, remove device,
240 * start build, activate spare
241 */
NeilBrown2989ddb2006-01-06 00:20:43 -0800242static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
NeilBrownd7603b72006-01-06 00:20:30 -0800243static atomic_t md_event_count;
NeilBrownfd01b882011-10-11 16:47:53 +1100244void md_new_event(struct mddev *mddev)
NeilBrownd7603b72006-01-06 00:20:30 -0800245{
246 atomic_inc(&md_event_count);
247 wake_up(&md_event_waiters);
248}
NeilBrown29269552006-03-27 01:18:10 -0800249EXPORT_SYMBOL_GPL(md_new_event);
NeilBrownd7603b72006-01-06 00:20:30 -0800250
NeilBrownc331eb02006-05-30 21:27:13 -0700251/* Alternate version that can be called from interrupts
252 * when calling sysfs_notify isn't needed.
253 */
NeilBrownfd01b882011-10-11 16:47:53 +1100254static void md_new_event_inintr(struct mddev *mddev)
NeilBrownc331eb02006-05-30 21:27:13 -0700255{
256 atomic_inc(&md_event_count);
257 wake_up(&md_event_waiters);
258}
259
NeilBrownd7603b72006-01-06 00:20:30 -0800260/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 * Enables to iterate over all existing md arrays
262 * all_mddevs_lock protects this list.
263 */
264static LIST_HEAD(all_mddevs);
265static DEFINE_SPINLOCK(all_mddevs_lock);
266
267
268/*
269 * iterates through all used mddevs in the system.
270 * We take care to grab the all_mddevs_lock whenever navigating
271 * the list, and to always hold a refcount when unlocked.
272 * Any code which breaks out of this loop while own
273 * a reference to the current mddev and must mddev_put it.
274 */
NeilBrownfd01b882011-10-11 16:47:53 +1100275#define for_each_mddev(_mddev,_tmp) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 \
277 for (({ spin_lock(&all_mddevs_lock); \
NeilBrownfd01b882011-10-11 16:47:53 +1100278 _tmp = all_mddevs.next; \
279 _mddev = NULL;}); \
280 ({ if (_tmp != &all_mddevs) \
281 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 spin_unlock(&all_mddevs_lock); \
NeilBrownfd01b882011-10-11 16:47:53 +1100283 if (_mddev) mddev_put(_mddev); \
284 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
285 _tmp != &all_mddevs;}); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 ({ spin_lock(&all_mddevs_lock); \
NeilBrownfd01b882011-10-11 16:47:53 +1100287 _tmp = _tmp->next;}) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 )
289
290
NeilBrown409c57f2009-03-31 14:39:39 +1100291/* Rather than calling directly into the personality make_request function,
292 * IO requests come here first so that we can check if the device is
293 * being suspended pending a reconfiguration.
294 * We hold a refcount over the call to ->make_request. By the time that
295 * call has finished, the bio has been linked into some internal structure
296 * and so is visible to ->quiesce(), so we don't need the refcount any more.
297 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200298static void md_make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
NeilBrown49077322010-03-25 16:20:56 +1100300 const int rw = bio_data_dir(bio);
NeilBrownfd01b882011-10-11 16:47:53 +1100301 struct mddev *mddev = q->queuedata;
NeilBrown49077322010-03-25 16:20:56 +1100302 int cpu;
Chris Masone91ece52011-02-07 19:21:48 -0500303 unsigned int sectors;
NeilBrown49077322010-03-25 16:20:56 +1100304
NeilBrown0ca69882011-01-14 09:14:33 +1100305 if (mddev == NULL || mddev->pers == NULL
306 || !mddev->ready) {
NeilBrown409c57f2009-03-31 14:39:39 +1100307 bio_io_error(bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200308 return;
NeilBrown409c57f2009-03-31 14:39:39 +1100309 }
NeilBrown0ca69882011-01-14 09:14:33 +1100310 smp_rmb(); /* Ensure implications of 'active' are visible */
NeilBrown409c57f2009-03-31 14:39:39 +1100311 rcu_read_lock();
Tejun Heoe9c74692010-09-03 11:56:18 +0200312 if (mddev->suspended) {
NeilBrown409c57f2009-03-31 14:39:39 +1100313 DEFINE_WAIT(__wait);
314 for (;;) {
315 prepare_to_wait(&mddev->sb_wait, &__wait,
316 TASK_UNINTERRUPTIBLE);
Tejun Heoe9c74692010-09-03 11:56:18 +0200317 if (!mddev->suspended)
NeilBrown409c57f2009-03-31 14:39:39 +1100318 break;
319 rcu_read_unlock();
320 schedule();
321 rcu_read_lock();
322 }
323 finish_wait(&mddev->sb_wait, &__wait);
324 }
325 atomic_inc(&mddev->active_io);
326 rcu_read_unlock();
NeilBrown49077322010-03-25 16:20:56 +1100327
Chris Masone91ece52011-02-07 19:21:48 -0500328 /*
329 * save the sectors now since our bio can
330 * go away inside make_request
331 */
332 sectors = bio_sectors(bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200333 mddev->pers->make_request(mddev, bio);
NeilBrown49077322010-03-25 16:20:56 +1100334
335 cpu = part_stat_lock();
336 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
Chris Masone91ece52011-02-07 19:21:48 -0500337 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
NeilBrown49077322010-03-25 16:20:56 +1100338 part_stat_unlock();
339
NeilBrown409c57f2009-03-31 14:39:39 +1100340 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
341 wake_up(&mddev->sb_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342}
343
NeilBrown9e35b992010-04-06 14:23:02 +1000344/* mddev_suspend makes sure no new requests are submitted
345 * to the device, and that any requests that have been submitted
346 * are completely handled.
347 * Once ->stop is called and completes, the module will be completely
348 * unused.
349 */
NeilBrownfd01b882011-10-11 16:47:53 +1100350void mddev_suspend(struct mddev *mddev)
NeilBrown409c57f2009-03-31 14:39:39 +1100351{
352 BUG_ON(mddev->suspended);
353 mddev->suspended = 1;
354 synchronize_rcu();
355 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
356 mddev->pers->quiesce(mddev, 1);
Jonathan Brassow0d9f4f12012-05-16 04:06:14 -0500357
358 del_timer_sync(&mddev->safemode_timer);
NeilBrown409c57f2009-03-31 14:39:39 +1100359}
NeilBrown390ee602010-06-01 19:37:27 +1000360EXPORT_SYMBOL_GPL(mddev_suspend);
NeilBrown409c57f2009-03-31 14:39:39 +1100361
NeilBrownfd01b882011-10-11 16:47:53 +1100362void mddev_resume(struct mddev *mddev)
NeilBrown409c57f2009-03-31 14:39:39 +1100363{
364 mddev->suspended = 0;
365 wake_up(&mddev->sb_wait);
366 mddev->pers->quiesce(mddev, 0);
Jonathan Brassow0fd018a2011-06-07 17:49:36 -0500367
Jonathan Brassow47525e52012-05-22 13:55:29 +1000368 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
Jonathan Brassow0fd018a2011-06-07 17:49:36 -0500369 md_wakeup_thread(mddev->thread);
370 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
NeilBrown409c57f2009-03-31 14:39:39 +1100371}
NeilBrown390ee602010-06-01 19:37:27 +1000372EXPORT_SYMBOL_GPL(mddev_resume);
NeilBrown409c57f2009-03-31 14:39:39 +1100373
NeilBrownfd01b882011-10-11 16:47:53 +1100374int mddev_congested(struct mddev *mddev, int bits)
NeilBrown3fa841d2009-09-23 18:10:29 +1000375{
376 return mddev->suspended;
377}
378EXPORT_SYMBOL(mddev_congested);
379
NeilBrowna2826aa2009-12-14 12:49:49 +1100380/*
Tejun Heoe9c74692010-09-03 11:56:18 +0200381 * Generic flush handling for md
NeilBrowna2826aa2009-12-14 12:49:49 +1100382 */
383
Tejun Heoe9c74692010-09-03 11:56:18 +0200384static void md_end_flush(struct bio *bio, int err)
NeilBrowna2826aa2009-12-14 12:49:49 +1100385{
NeilBrown3cb03002011-10-11 16:45:26 +1100386 struct md_rdev *rdev = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +1100387 struct mddev *mddev = rdev->mddev;
NeilBrowna2826aa2009-12-14 12:49:49 +1100388
389 rdev_dec_pending(rdev, mddev);
390
391 if (atomic_dec_and_test(&mddev->flush_pending)) {
Tejun Heoe9c74692010-09-03 11:56:18 +0200392 /* The pre-request flush has finished */
Tejun Heoe804ac72010-10-15 15:36:08 +0200393 queue_work(md_wq, &mddev->flush_work);
NeilBrowna2826aa2009-12-14 12:49:49 +1100394 }
395 bio_put(bio);
396}
397
NeilBrowna7a07e62010-12-09 16:04:25 +1100398static void md_submit_flush_data(struct work_struct *ws);
399
NeilBrowna035fc32010-12-09 16:17:51 +1100400static void submit_flushes(struct work_struct *ws)
NeilBrowna2826aa2009-12-14 12:49:49 +1100401{
NeilBrownfd01b882011-10-11 16:47:53 +1100402 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
NeilBrown3cb03002011-10-11 16:45:26 +1100403 struct md_rdev *rdev;
NeilBrowna2826aa2009-12-14 12:49:49 +1100404
NeilBrowna7a07e62010-12-09 16:04:25 +1100405 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
406 atomic_set(&mddev->flush_pending, 1);
NeilBrowna2826aa2009-12-14 12:49:49 +1100407 rcu_read_lock();
NeilBrowndafb20f2012-03-19 12:46:39 +1100408 rdev_for_each_rcu(rdev, mddev)
NeilBrowna2826aa2009-12-14 12:49:49 +1100409 if (rdev->raid_disk >= 0 &&
410 !test_bit(Faulty, &rdev->flags)) {
411 /* Take two references, one is dropped
412 * when request finishes, one after
413 * we reclaim rcu_read_lock
414 */
415 struct bio *bi;
416 atomic_inc(&rdev->nr_pending);
417 atomic_inc(&rdev->nr_pending);
418 rcu_read_unlock();
Shaohua Lib5e1b8c2012-05-21 09:26:59 +1000419 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
Tejun Heoe9c74692010-09-03 11:56:18 +0200420 bi->bi_end_io = md_end_flush;
NeilBrowna2826aa2009-12-14 12:49:49 +1100421 bi->bi_private = rdev;
422 bi->bi_bdev = rdev->bdev;
423 atomic_inc(&mddev->flush_pending);
Tejun Heoe9c74692010-09-03 11:56:18 +0200424 submit_bio(WRITE_FLUSH, bi);
NeilBrowna2826aa2009-12-14 12:49:49 +1100425 rcu_read_lock();
426 rdev_dec_pending(rdev, mddev);
427 }
428 rcu_read_unlock();
NeilBrowna7a07e62010-12-09 16:04:25 +1100429 if (atomic_dec_and_test(&mddev->flush_pending))
430 queue_work(md_wq, &mddev->flush_work);
NeilBrowna2826aa2009-12-14 12:49:49 +1100431}
432
Tejun Heoe9c74692010-09-03 11:56:18 +0200433static void md_submit_flush_data(struct work_struct *ws)
NeilBrowna2826aa2009-12-14 12:49:49 +1100434{
NeilBrownfd01b882011-10-11 16:47:53 +1100435 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
Tejun Heoe9c74692010-09-03 11:56:18 +0200436 struct bio *bio = mddev->flush_bio;
NeilBrowna2826aa2009-12-14 12:49:49 +1100437
Tejun Heoe9c74692010-09-03 11:56:18 +0200438 if (bio->bi_size == 0)
NeilBrowna2826aa2009-12-14 12:49:49 +1100439 /* an empty barrier - all done */
440 bio_endio(bio, 0);
441 else {
Tejun Heoe9c74692010-09-03 11:56:18 +0200442 bio->bi_rw &= ~REQ_FLUSH;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200443 mddev->pers->make_request(mddev, bio);
NeilBrowna2826aa2009-12-14 12:49:49 +1100444 }
NeilBrown2b74e122010-12-09 15:59:01 +1100445
446 mddev->flush_bio = NULL;
447 wake_up(&mddev->sb_wait);
NeilBrowna2826aa2009-12-14 12:49:49 +1100448}
449
NeilBrownfd01b882011-10-11 16:47:53 +1100450void md_flush_request(struct mddev *mddev, struct bio *bio)
NeilBrowna2826aa2009-12-14 12:49:49 +1100451{
452 spin_lock_irq(&mddev->write_lock);
453 wait_event_lock_irq(mddev->sb_wait,
Tejun Heoe9c74692010-09-03 11:56:18 +0200454 !mddev->flush_bio,
NeilBrowna2826aa2009-12-14 12:49:49 +1100455 mddev->write_lock, /*nothing*/);
Tejun Heoe9c74692010-09-03 11:56:18 +0200456 mddev->flush_bio = bio;
NeilBrowna2826aa2009-12-14 12:49:49 +1100457 spin_unlock_irq(&mddev->write_lock);
458
NeilBrowna035fc32010-12-09 16:17:51 +1100459 INIT_WORK(&mddev->flush_work, submit_flushes);
460 queue_work(md_wq, &mddev->flush_work);
NeilBrowna2826aa2009-12-14 12:49:49 +1100461}
Tejun Heoe9c74692010-09-03 11:56:18 +0200462EXPORT_SYMBOL(md_flush_request);
NeilBrown409c57f2009-03-31 14:39:39 +1100463
NeilBrown74018dc2012-07-31 09:08:15 +0200464void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
NeilBrown97658cd2011-04-18 18:25:42 +1000465{
NeilBrown9cbb1752012-07-31 09:08:14 +0200466 struct mddev *mddev = cb->data;
467 md_wakeup_thread(mddev->thread);
468 kfree(cb);
NeilBrown97658cd2011-04-18 18:25:42 +1000469}
NeilBrown9cbb1752012-07-31 09:08:14 +0200470EXPORT_SYMBOL(md_unplug);
NeilBrown2ac87402010-06-01 19:37:29 +1000471
NeilBrownfd01b882011-10-11 16:47:53 +1100472static inline struct mddev *mddev_get(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473{
474 atomic_inc(&mddev->active);
475 return mddev;
476}
477
Dan Williams5fd3a172009-03-04 00:57:25 -0700478static void mddev_delayed_delete(struct work_struct *ws);
NeilBrownd3374822009-01-09 08:31:10 +1100479
NeilBrownfd01b882011-10-11 16:47:53 +1100480static void mddev_put(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
NeilBrowna167f662010-10-26 18:31:13 +1100482 struct bio_set *bs = NULL;
483
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
485 return;
NeilBrownd3374822009-01-09 08:31:10 +1100486 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
NeilBrowncbd19982009-12-30 12:08:49 +1100487 mddev->ctime == 0 && !mddev->hold_active) {
488 /* Array is not configured at all, and not held active,
489 * so destroy it */
NeilBrownaf8a2432011-12-08 15:49:46 +1100490 list_del_init(&mddev->all_mddevs);
NeilBrowna167f662010-10-26 18:31:13 +1100491 bs = mddev->bio_set;
492 mddev->bio_set = NULL;
NeilBrownd3374822009-01-09 08:31:10 +1100493 if (mddev->gendisk) {
Tejun Heoe804ac72010-10-15 15:36:08 +0200494 /* We did a probe so need to clean up. Call
495 * queue_work inside the spinlock so that
496 * flush_workqueue() after mddev_find will
497 * succeed in waiting for the work to be done.
NeilBrownd3374822009-01-09 08:31:10 +1100498 */
499 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
Tejun Heoe804ac72010-10-15 15:36:08 +0200500 queue_work(md_misc_wq, &mddev->del_work);
NeilBrownd3374822009-01-09 08:31:10 +1100501 } else
502 kfree(mddev);
503 }
504 spin_unlock(&all_mddevs_lock);
NeilBrowna167f662010-10-26 18:31:13 +1100505 if (bs)
506 bioset_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
508
NeilBrownfd01b882011-10-11 16:47:53 +1100509void mddev_init(struct mddev *mddev)
NeilBrownfafd7fb2010-04-01 15:55:30 +1100510{
511 mutex_init(&mddev->open_mutex);
512 mutex_init(&mddev->reconfig_mutex);
513 mutex_init(&mddev->bitmap_info.mutex);
514 INIT_LIST_HEAD(&mddev->disks);
515 INIT_LIST_HEAD(&mddev->all_mddevs);
516 init_timer(&mddev->safemode_timer);
517 atomic_set(&mddev->active, 1);
518 atomic_set(&mddev->openers, 0);
519 atomic_set(&mddev->active_io, 0);
520 spin_lock_init(&mddev->write_lock);
521 atomic_set(&mddev->flush_pending, 0);
522 init_waitqueue_head(&mddev->sb_wait);
523 init_waitqueue_head(&mddev->recovery_wait);
524 mddev->reshape_position = MaxSector;
NeilBrown2c810cd2012-05-21 09:27:00 +1000525 mddev->reshape_backwards = 0;
NeilBrownfafd7fb2010-04-01 15:55:30 +1100526 mddev->resync_min = 0;
527 mddev->resync_max = MaxSector;
528 mddev->level = LEVEL_NONE;
529}
NeilBrown390ee602010-06-01 19:37:27 +1000530EXPORT_SYMBOL_GPL(mddev_init);
NeilBrownfafd7fb2010-04-01 15:55:30 +1100531
NeilBrownfd01b882011-10-11 16:47:53 +1100532static struct mddev * mddev_find(dev_t unit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533{
NeilBrownfd01b882011-10-11 16:47:53 +1100534 struct mddev *mddev, *new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
NeilBrown8f5f02c2011-02-16 13:58:51 +1100536 if (unit && MAJOR(unit) != MD_MAJOR)
537 unit &= ~((1<<MdpMinorShift)-1);
538
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 retry:
540 spin_lock(&all_mddevs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700541
NeilBrownefeb53c2009-01-09 08:31:10 +1100542 if (unit) {
543 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
544 if (mddev->unit == unit) {
545 mddev_get(mddev);
546 spin_unlock(&all_mddevs_lock);
547 kfree(new);
548 return mddev;
549 }
550
551 if (new) {
552 list_add(&new->all_mddevs, &all_mddevs);
553 spin_unlock(&all_mddevs_lock);
554 new->hold_active = UNTIL_IOCTL;
555 return new;
556 }
557 } else if (new) {
558 /* find an unused unit number */
559 static int next_minor = 512;
560 int start = next_minor;
561 int is_free = 0;
562 int dev = 0;
563 while (!is_free) {
564 dev = MKDEV(MD_MAJOR, next_minor);
565 next_minor++;
566 if (next_minor > MINORMASK)
567 next_minor = 0;
568 if (next_minor == start) {
569 /* Oh dear, all in use. */
570 spin_unlock(&all_mddevs_lock);
571 kfree(new);
572 return NULL;
573 }
574
575 is_free = 1;
576 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
577 if (mddev->unit == dev) {
578 is_free = 0;
579 break;
580 }
581 }
582 new->unit = dev;
583 new->md_minor = MINOR(dev);
584 new->hold_active = UNTIL_STOP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 list_add(&new->all_mddevs, &all_mddevs);
586 spin_unlock(&all_mddevs_lock);
587 return new;
588 }
589 spin_unlock(&all_mddevs_lock);
590
NeilBrown9ffae0c2006-01-06 00:20:32 -0800591 new = kzalloc(sizeof(*new), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 if (!new)
593 return NULL;
594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 new->unit = unit;
596 if (MAJOR(unit) == MD_MAJOR)
597 new->md_minor = MINOR(unit);
598 else
599 new->md_minor = MINOR(unit) >> MdpMinorShift;
600
NeilBrownfafd7fb2010-04-01 15:55:30 +1100601 mddev_init(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 goto retry;
604}
605
NeilBrownfd01b882011-10-11 16:47:53 +1100606static inline int mddev_lock(struct mddev * mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607{
NeilBrowndf5b89b2006-03-27 01:18:20 -0800608 return mutex_lock_interruptible(&mddev->reconfig_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609}
610
NeilBrownfd01b882011-10-11 16:47:53 +1100611static inline int mddev_is_locked(struct mddev *mddev)
Dan Williamsb522adc2009-03-31 15:00:31 +1100612{
613 return mutex_is_locked(&mddev->reconfig_mutex);
614}
615
NeilBrownfd01b882011-10-11 16:47:53 +1100616static inline int mddev_trylock(struct mddev * mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
NeilBrowndf5b89b2006-03-27 01:18:20 -0800618 return mutex_trylock(&mddev->reconfig_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619}
620
NeilBrownb6eb1272010-04-15 10:13:47 +1000621static struct attribute_group md_redundancy_group;
622
NeilBrownfd01b882011-10-11 16:47:53 +1100623static void mddev_unlock(struct mddev * mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624{
NeilBrowna64c8762010-04-14 17:15:37 +1000625 if (mddev->to_remove) {
NeilBrownb6eb1272010-04-15 10:13:47 +1000626 /* These cannot be removed under reconfig_mutex as
627 * an access to the files will try to take reconfig_mutex
628 * while holding the file unremovable, which leads to
629 * a deadlock.
NeilBrownbb4f1e92010-08-08 21:18:03 +1000630 * So hold set sysfs_active while the remove in happeing,
631 * and anything else which might set ->to_remove or my
632 * otherwise change the sysfs namespace will fail with
633 * -EBUSY if sysfs_active is still set.
634 * We set sysfs_active under reconfig_mutex and elsewhere
635 * test it under the same mutex to ensure its correct value
636 * is seen.
NeilBrownb6eb1272010-04-15 10:13:47 +1000637 */
NeilBrowna64c8762010-04-14 17:15:37 +1000638 struct attribute_group *to_remove = mddev->to_remove;
639 mddev->to_remove = NULL;
NeilBrownbb4f1e92010-08-08 21:18:03 +1000640 mddev->sysfs_active = 1;
NeilBrownb6eb1272010-04-15 10:13:47 +1000641 mutex_unlock(&mddev->reconfig_mutex);
642
NeilBrown00bcb4a2010-06-01 19:37:23 +1000643 if (mddev->kobj.sd) {
644 if (to_remove != &md_redundancy_group)
645 sysfs_remove_group(&mddev->kobj, to_remove);
646 if (mddev->pers == NULL ||
647 mddev->pers->sync_request == NULL) {
648 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
649 if (mddev->sysfs_action)
650 sysfs_put(mddev->sysfs_action);
651 mddev->sysfs_action = NULL;
652 }
NeilBrowna64c8762010-04-14 17:15:37 +1000653 }
NeilBrownbb4f1e92010-08-08 21:18:03 +1000654 mddev->sysfs_active = 0;
NeilBrownb6eb1272010-04-15 10:13:47 +1000655 } else
656 mutex_unlock(&mddev->reconfig_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
Chris Dunlop751e67c2011-10-19 16:48:26 +1100658 /* As we've dropped the mutex we need a spinlock to
659 * make sure the thread doesn't disappear
NeilBrown01f96c02011-09-21 15:30:20 +1000660 */
661 spin_lock(&pers_lock);
NeilBrown005eca52005-08-22 13:11:08 -0700662 md_wakeup_thread(mddev->thread);
NeilBrown01f96c02011-09-21 15:30:20 +1000663 spin_unlock(&pers_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664}
665
NeilBrownfd01b882011-10-11 16:47:53 +1100666static struct md_rdev * find_rdev_nr(struct mddev *mddev, int nr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
NeilBrown3cb03002011-10-11 16:45:26 +1100668 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669
NeilBrowndafb20f2012-03-19 12:46:39 +1100670 rdev_for_each(rdev, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700671 if (rdev->desc_nr == nr)
672 return rdev;
Cheng Renquan159ec1f2009-01-09 08:31:08 +1100673
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 return NULL;
675}
676
NeilBrown1ca69c42012-10-11 13:37:33 +1100677static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
678{
679 struct md_rdev *rdev;
680
681 rdev_for_each_rcu(rdev, mddev)
682 if (rdev->desc_nr == nr)
683 return rdev;
684
685 return NULL;
686}
687
688static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689{
NeilBrown3cb03002011-10-11 16:45:26 +1100690 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691
NeilBrowndafb20f2012-03-19 12:46:39 +1100692 rdev_for_each(rdev, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 if (rdev->bdev->bd_dev == dev)
694 return rdev;
Cheng Renquan159ec1f2009-01-09 08:31:08 +1100695
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 return NULL;
697}
698
NeilBrown1ca69c42012-10-11 13:37:33 +1100699static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
700{
701 struct md_rdev *rdev;
702
703 rdev_for_each_rcu(rdev, mddev)
704 if (rdev->bdev->bd_dev == dev)
705 return rdev;
706
707 return NULL;
708}
709
NeilBrown84fc4b52011-10-11 16:49:58 +1100710static struct md_personality *find_pers(int level, char *clevel)
NeilBrown2604b702006-01-06 00:20:36 -0800711{
NeilBrown84fc4b52011-10-11 16:49:58 +1100712 struct md_personality *pers;
NeilBrownd9d166c2006-01-06 00:20:51 -0800713 list_for_each_entry(pers, &pers_list, list) {
714 if (level != LEVEL_NONE && pers->level == level)
NeilBrown2604b702006-01-06 00:20:36 -0800715 return pers;
NeilBrownd9d166c2006-01-06 00:20:51 -0800716 if (strcmp(pers->name, clevel)==0)
717 return pers;
718 }
NeilBrown2604b702006-01-06 00:20:36 -0800719 return NULL;
720}
721
Andre Nollb73df2d2008-07-11 22:02:23 +1000722/* return the offset of the super block in 512byte sectors */
NeilBrown3cb03002011-10-11 16:45:26 +1100723static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724{
Jonathan Brassow57b2caa2011-01-14 09:14:33 +1100725 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
Andre Nollb73df2d2008-07-11 22:02:23 +1000726 return MD_NEW_SIZE_SECTORS(num_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727}
728
NeilBrown3cb03002011-10-11 16:45:26 +1100729static int alloc_disk_sb(struct md_rdev * rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730{
731 if (rdev->sb_page)
732 MD_BUG();
733
734 rdev->sb_page = alloc_page(GFP_KERNEL);
735 if (!rdev->sb_page) {
736 printk(KERN_ALERT "md: out of memory.\n");
Andre Nollebc24332008-07-11 22:02:20 +1000737 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738 }
739
740 return 0;
741}
742
NeilBrown545c8792012-05-22 13:54:30 +1000743void md_rdev_clear(struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744{
745 if (rdev->sb_page) {
NeilBrown2d1f3b52006-01-06 00:20:31 -0800746 put_page(rdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 rdev->sb_loaded = 0;
748 rdev->sb_page = NULL;
Andre Noll0f420352008-07-11 22:02:23 +1000749 rdev->sb_start = 0;
Andre Nolldd8ac332009-03-31 14:33:13 +1100750 rdev->sectors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700751 }
NeilBrown2699b672011-07-28 11:31:47 +1000752 if (rdev->bb_page) {
753 put_page(rdev->bb_page);
754 rdev->bb_page = NULL;
755 }
NeilBrown4fa2f322012-05-22 13:55:01 +1000756 kfree(rdev->badblocks.page);
757 rdev->badblocks.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758}
NeilBrown545c8792012-05-22 13:54:30 +1000759EXPORT_SYMBOL_GPL(md_rdev_clear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760
NeilBrown6712ecf2007-09-27 12:47:43 +0200761static void super_written(struct bio *bio, int error)
NeilBrown7bfa19f2005-06-21 17:17:28 -0700762{
NeilBrown3cb03002011-10-11 16:45:26 +1100763 struct md_rdev *rdev = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +1100764 struct mddev *mddev = rdev->mddev;
NeilBrown7bfa19f2005-06-21 17:17:28 -0700765
NeilBrown3a0f5bb2006-10-03 01:16:03 -0700766 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
767 printk("md: super_written gets error=%d, uptodate=%d\n",
768 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
769 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
NeilBrowna9701a32005-11-08 21:39:34 -0800770 md_error(mddev, rdev);
NeilBrown3a0f5bb2006-10-03 01:16:03 -0700771 }
NeilBrown7bfa19f2005-06-21 17:17:28 -0700772
NeilBrowna9701a32005-11-08 21:39:34 -0800773 if (atomic_dec_and_test(&mddev->pending_writes))
774 wake_up(&mddev->sb_wait);
Neil Brownf8b58ed2005-06-27 22:29:34 -0700775 bio_put(bio);
NeilBrown7bfa19f2005-06-21 17:17:28 -0700776}
777
NeilBrownfd01b882011-10-11 16:47:53 +1100778void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
NeilBrown7bfa19f2005-06-21 17:17:28 -0700779 sector_t sector, int size, struct page *page)
780{
781 /* write first size bytes of page to sector of rdev
782 * Increment mddev->pending_writes before returning
783 * and decrement it on completion, waking up sb_wait
784 * if zero is reached.
785 * If an error occurred, call md_error
786 */
NeilBrowna167f662010-10-26 18:31:13 +1100787 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
NeilBrown7bfa19f2005-06-21 17:17:28 -0700788
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100789 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
NeilBrown7bfa19f2005-06-21 17:17:28 -0700790 bio->bi_sector = sector;
791 bio_add_page(bio, page, size, 0);
792 bio->bi_private = rdev;
793 bio->bi_end_io = super_written;
NeilBrowna9701a32005-11-08 21:39:34 -0800794
NeilBrown7bfa19f2005-06-21 17:17:28 -0700795 atomic_inc(&mddev->pending_writes);
Namhyung Kima5bf4df2011-08-25 14:43:34 +1000796 submit_bio(WRITE_FLUSH_FUA, bio);
NeilBrowna9701a32005-11-08 21:39:34 -0800797}
798
NeilBrownfd01b882011-10-11 16:47:53 +1100799void md_super_wait(struct mddev *mddev)
NeilBrowna9701a32005-11-08 21:39:34 -0800800{
Tejun Heoe9c74692010-09-03 11:56:18 +0200801 /* wait for all superblock writes that were scheduled to complete */
NeilBrowna9701a32005-11-08 21:39:34 -0800802 DEFINE_WAIT(wq);
803 for(;;) {
804 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
805 if (atomic_read(&mddev->pending_writes)==0)
806 break;
NeilBrowna9701a32005-11-08 21:39:34 -0800807 schedule();
808 }
809 finish_wait(&mddev->sb_wait, &wq);
NeilBrown7bfa19f2005-06-21 17:17:28 -0700810}
811
NeilBrown6712ecf2007-09-27 12:47:43 +0200812static void bi_complete(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814 complete((struct completion*)bio->bi_private);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815}
816
NeilBrown3cb03002011-10-11 16:45:26 +1100817int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100818 struct page *page, int rw, bool metadata_op)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819{
NeilBrowna167f662010-10-26 18:31:13 +1100820 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821 struct completion event;
822 int ret;
823
Jens Axboe721a9602011-03-09 11:56:30 +0100824 rw |= REQ_SYNC;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100826 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
827 rdev->meta_bdev : rdev->bdev;
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100828 if (metadata_op)
829 bio->bi_sector = sector + rdev->sb_start;
NeilBrown1fdd6fc92012-05-21 09:28:32 +1000830 else if (rdev->mddev->reshape_position != MaxSector &&
831 (rdev->mddev->reshape_backwards ==
832 (sector >= rdev->mddev->reshape_position)))
833 bio->bi_sector = sector + rdev->new_data_offset;
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100834 else
835 bio->bi_sector = sector + rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 bio_add_page(bio, page, size, 0);
837 init_completion(&event);
838 bio->bi_private = &event;
839 bio->bi_end_io = bi_complete;
840 submit_bio(rw, bio);
841 wait_for_completion(&event);
842
843 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
844 bio_put(bio);
845 return ret;
846}
NeilBrowna8745db2006-01-06 00:20:34 -0800847EXPORT_SYMBOL_GPL(sync_page_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
NeilBrown3cb03002011-10-11 16:45:26 +1100849static int read_disk_sb(struct md_rdev * rdev, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850{
851 char b[BDEVNAME_SIZE];
852 if (!rdev->sb_page) {
853 MD_BUG();
854 return -EINVAL;
855 }
856 if (rdev->sb_loaded)
857 return 0;
858
859
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100860 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 goto fail;
862 rdev->sb_loaded = 1;
863 return 0;
864
865fail:
866 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
867 bdevname(rdev->bdev,b));
868 return -EINVAL;
869}
870
871static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
872{
Andre Noll05710462008-07-11 22:02:20 +1000873 return sb1->set_uuid0 == sb2->set_uuid0 &&
874 sb1->set_uuid1 == sb2->set_uuid1 &&
875 sb1->set_uuid2 == sb2->set_uuid2 &&
876 sb1->set_uuid3 == sb2->set_uuid3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877}
878
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
880{
881 int ret;
882 mdp_super_t *tmp1, *tmp2;
883
884 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
885 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
886
887 if (!tmp1 || !tmp2) {
888 ret = 0;
Andre Noll35020f12008-03-23 15:10:33 +0100889 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 goto abort;
891 }
892
893 *tmp1 = *sb1;
894 *tmp2 = *sb2;
895
896 /*
897 * nr_disks is not constant
898 */
899 tmp1->nr_disks = 0;
900 tmp2->nr_disks = 0;
901
Andre Nollce0c8e02008-07-11 22:02:20 +1000902 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903abort:
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700904 kfree(tmp1);
905 kfree(tmp2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700906 return ret;
907}
908
NeilBrown4d167f02007-05-09 02:35:37 -0700909
910static u32 md_csum_fold(u32 csum)
911{
912 csum = (csum & 0xffff) + (csum >> 16);
913 return (csum & 0xffff) + (csum >> 16);
914}
915
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916static unsigned int calc_sb_csum(mdp_super_t * sb)
917{
NeilBrown4d167f02007-05-09 02:35:37 -0700918 u64 newcsum = 0;
919 u32 *sb32 = (u32*)sb;
920 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921 unsigned int disk_csum, csum;
922
923 disk_csum = sb->sb_csum;
924 sb->sb_csum = 0;
NeilBrown4d167f02007-05-09 02:35:37 -0700925
926 for (i = 0; i < MD_SB_BYTES/4 ; i++)
927 newcsum += sb32[i];
928 csum = (newcsum & 0xffffffff) + (newcsum>>32);
929
930
931#ifdef CONFIG_ALPHA
932 /* This used to use csum_partial, which was wrong for several
933 * reasons including that different results are returned on
934 * different architectures. It isn't critical that we get exactly
935 * the same return value as before (we always csum_fold before
936 * testing, and that removes any differences). However as we
937 * know that csum_partial always returned a 16bit value on
938 * alphas, do a fold to maximise conformity to previous behaviour.
939 */
940 sb->sb_csum = md_csum_fold(disk_csum);
941#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 sb->sb_csum = disk_csum;
NeilBrown4d167f02007-05-09 02:35:37 -0700943#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 return csum;
945}
946
947
948/*
949 * Handle superblock details.
950 * We want to be able to handle multiple superblock formats
951 * so we have a common interface to them all, and an array of
952 * different handlers.
953 * We rely on user-space to write the initial superblock, and support
954 * reading and updating of superblocks.
955 * Interface methods are:
NeilBrown3cb03002011-10-11 16:45:26 +1100956 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 * loads and validates a superblock on dev.
958 * if refdev != NULL, compare superblocks on both devices
959 * Return:
960 * 0 - dev has a superblock that is compatible with refdev
961 * 1 - dev has a superblock that is compatible and newer than refdev
962 * so dev should be used as the refdev in future
963 * -EINVAL superblock incompatible or invalid
964 * -othererror e.g. -EIO
965 *
NeilBrownfd01b882011-10-11 16:47:53 +1100966 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 * Verify that dev is acceptable into mddev.
968 * The first time, mddev->raid_disks will be 0, and data from
969 * dev should be merged in. Subsequent calls check that dev
970 * is new enough. Return 0 or -EINVAL
971 *
NeilBrownfd01b882011-10-11 16:47:53 +1100972 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973 * Update the superblock for rdev with data in mddev
974 * This does not write to disc.
975 *
976 */
977
978struct super_type {
Chris Webb0cd17fe2008-06-28 08:31:46 +1000979 char *name;
980 struct module *owner;
NeilBrownc6563a82012-05-21 09:27:00 +1000981 int (*load_super)(struct md_rdev *rdev,
982 struct md_rdev *refdev,
Chris Webb0cd17fe2008-06-28 08:31:46 +1000983 int minor_version);
NeilBrownc6563a82012-05-21 09:27:00 +1000984 int (*validate_super)(struct mddev *mddev,
985 struct md_rdev *rdev);
986 void (*sync_super)(struct mddev *mddev,
987 struct md_rdev *rdev);
NeilBrown3cb03002011-10-11 16:45:26 +1100988 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
Andre Noll15f4a5f2008-07-21 14:42:12 +1000989 sector_t num_sectors);
NeilBrownc6563a82012-05-21 09:27:00 +1000990 int (*allow_new_offset)(struct md_rdev *rdev,
991 unsigned long long new_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992};
993
994/*
Andre Noll0894cc32009-06-18 08:49:23 +1000995 * Check that the given mddev has no bitmap.
996 *
997 * This function is called from the run method of all personalities that do not
998 * support bitmaps. It prints an error message and returns non-zero if mddev
999 * has a bitmap. Otherwise, it returns 0.
1000 *
1001 */
NeilBrownfd01b882011-10-11 16:47:53 +11001002int md_check_no_bitmap(struct mddev *mddev)
Andre Noll0894cc32009-06-18 08:49:23 +10001003{
NeilBrownc3d97142009-12-14 12:49:52 +11001004 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
Andre Noll0894cc32009-06-18 08:49:23 +10001005 return 0;
1006 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
1007 mdname(mddev), mddev->pers->name);
1008 return 1;
1009}
1010EXPORT_SYMBOL(md_check_no_bitmap);
1011
1012/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 * load_super for 0.90.0
1014 */
NeilBrown3cb03002011-10-11 16:45:26 +11001015static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016{
1017 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1018 mdp_super_t *sb;
1019 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020
1021 /*
Andre Noll0f420352008-07-11 22:02:23 +10001022 * Calculate the position of the superblock (512byte sectors),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023 * it's at the end of the disk.
1024 *
1025 * It also happens to be a multiple of 4Kb.
1026 */
Jonathan Brassow57b2caa2011-01-14 09:14:33 +11001027 rdev->sb_start = calc_dev_sboffset(rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
NeilBrown0002b272005-09-09 16:23:53 -07001029 ret = read_disk_sb(rdev, MD_SB_BYTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 if (ret) return ret;
1031
1032 ret = -EINVAL;
1033
1034 bdevname(rdev->bdev, b);
Namhyung Kim65a06f02011-07-27 11:00:36 +10001035 sb = page_address(rdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036
1037 if (sb->md_magic != MD_SB_MAGIC) {
1038 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
1039 b);
1040 goto abort;
1041 }
1042
1043 if (sb->major_version != 0 ||
NeilBrownf6705572006-03-27 01:18:11 -08001044 sb->minor_version < 90 ||
1045 sb->minor_version > 91) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
1047 sb->major_version, sb->minor_version,
1048 b);
1049 goto abort;
1050 }
1051
1052 if (sb->raid_disks <= 0)
1053 goto abort;
1054
NeilBrown4d167f02007-05-09 02:35:37 -07001055 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
1057 b);
1058 goto abort;
1059 }
1060
1061 rdev->preferred_minor = sb->md_minor;
1062 rdev->data_offset = 0;
NeilBrownc6563a82012-05-21 09:27:00 +10001063 rdev->new_data_offset = 0;
NeilBrown0002b272005-09-09 16:23:53 -07001064 rdev->sb_size = MD_SB_BYTES;
NeilBrown9f2f3832011-07-28 11:31:47 +10001065 rdev->badblocks.shift = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
1067 if (sb->level == LEVEL_MULTIPATH)
1068 rdev->desc_nr = -1;
1069 else
1070 rdev->desc_nr = sb->this_disk.number;
1071
Harvey Harrison9a7b2b02008-04-28 02:15:49 -07001072 if (!refdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 ret = 1;
Harvey Harrison9a7b2b02008-04-28 02:15:49 -07001074 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075 __u64 ev1, ev2;
Namhyung Kim65a06f02011-07-27 11:00:36 +10001076 mdp_super_t *refsb = page_address(refdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 if (!uuid_equal(refsb, sb)) {
1078 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1079 b, bdevname(refdev->bdev,b2));
1080 goto abort;
1081 }
1082 if (!sb_equal(refsb, sb)) {
1083 printk(KERN_WARNING "md: %s has same UUID"
1084 " but different superblock to %s\n",
1085 b, bdevname(refdev->bdev, b2));
1086 goto abort;
1087 }
1088 ev1 = md_event(sb);
1089 ev2 = md_event(refsb);
1090 if (ev1 > ev2)
1091 ret = 1;
1092 else
1093 ret = 0;
1094 }
NeilBrown8190e752009-06-18 08:48:58 +10001095 rdev->sectors = rdev->sb_start;
NeilBrown667a5312012-08-16 16:46:12 +10001096 /* Limit to 4TB as metadata cannot record more than that.
1097 * (not needed for Linear and RAID0 as metadata doesn't
1098 * record this size)
1099 */
1100 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
NeilBrown27a7b262011-09-10 17:21:28 +10001101 rdev->sectors = (2ULL << 32) - 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
NeilBrown27a7b262011-09-10 17:21:28 +10001103 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
NeilBrown2bf071b2006-01-06 00:20:55 -08001104 /* "this cannot possibly happen" ... */
1105 ret = -EINVAL;
1106
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107 abort:
1108 return ret;
1109}
1110
1111/*
1112 * validate_super for 0.90.0
1113 */
NeilBrownfd01b882011-10-11 16:47:53 +11001114static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115{
1116 mdp_disk_t *desc;
Namhyung Kim65a06f02011-07-27 11:00:36 +10001117 mdp_super_t *sb = page_address(rdev->sb_page);
NeilBrown07d84d102006-06-26 00:27:56 -07001118 __u64 ev1 = md_event(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
NeilBrown41158c72005-06-21 17:17:25 -07001120 rdev->raid_disk = -1;
NeilBrownc5d79ad2008-02-06 01:39:54 -08001121 clear_bit(Faulty, &rdev->flags);
1122 clear_bit(In_sync, &rdev->flags);
1123 clear_bit(WriteMostly, &rdev->flags);
NeilBrownc5d79ad2008-02-06 01:39:54 -08001124
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 if (mddev->raid_disks == 0) {
1126 mddev->major_version = 0;
1127 mddev->minor_version = sb->minor_version;
1128 mddev->patch_version = sb->patch_version;
NeilBrowne6910632008-02-06 01:39:51 -08001129 mddev->external = 0;
Andre Noll9d8f0362009-06-18 08:45:01 +10001130 mddev->chunk_sectors = sb->chunk_size >> 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 mddev->ctime = sb->ctime;
1132 mddev->utime = sb->utime;
1133 mddev->level = sb->level;
NeilBrownd9d166c2006-01-06 00:20:51 -08001134 mddev->clevel[0] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135 mddev->layout = sb->layout;
1136 mddev->raid_disks = sb->raid_disks;
NeilBrown27a7b262011-09-10 17:21:28 +10001137 mddev->dev_sectors = ((sector_t)sb->size) * 2;
NeilBrown07d84d102006-06-26 00:27:56 -07001138 mddev->events = ev1;
NeilBrownc3d97142009-12-14 12:49:52 +11001139 mddev->bitmap_info.offset = 0;
NeilBrown6409bb02012-05-22 13:55:07 +10001140 mddev->bitmap_info.space = 0;
1141 /* bitmap can use 60 K after the 4K superblocks */
NeilBrownc3d97142009-12-14 12:49:52 +11001142 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
NeilBrown6409bb02012-05-22 13:55:07 +10001143 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
NeilBrown2c810cd2012-05-21 09:27:00 +10001144 mddev->reshape_backwards = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145
NeilBrownf6705572006-03-27 01:18:11 -08001146 if (mddev->minor_version >= 91) {
1147 mddev->reshape_position = sb->reshape_position;
1148 mddev->delta_disks = sb->delta_disks;
1149 mddev->new_level = sb->new_level;
1150 mddev->new_layout = sb->new_layout;
Andre Noll664e7c42009-06-18 08:45:27 +10001151 mddev->new_chunk_sectors = sb->new_chunk >> 9;
NeilBrown2c810cd2012-05-21 09:27:00 +10001152 if (mddev->delta_disks < 0)
1153 mddev->reshape_backwards = 1;
NeilBrownf6705572006-03-27 01:18:11 -08001154 } else {
1155 mddev->reshape_position = MaxSector;
1156 mddev->delta_disks = 0;
1157 mddev->new_level = mddev->level;
1158 mddev->new_layout = mddev->layout;
Andre Noll664e7c42009-06-18 08:45:27 +10001159 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08001160 }
1161
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 if (sb->state & (1<<MD_SB_CLEAN))
1163 mddev->recovery_cp = MaxSector;
1164 else {
1165 if (sb->events_hi == sb->cp_events_hi &&
1166 sb->events_lo == sb->cp_events_lo) {
1167 mddev->recovery_cp = sb->recovery_cp;
1168 } else
1169 mddev->recovery_cp = 0;
1170 }
1171
1172 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1173 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1174 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1175 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1176
1177 mddev->max_disks = MD_SB_DISKS;
NeilBrowna654b9d82005-06-21 17:17:27 -07001178
1179 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
NeilBrown6409bb02012-05-22 13:55:07 +10001180 mddev->bitmap_info.file == NULL) {
NeilBrownc3d97142009-12-14 12:49:52 +11001181 mddev->bitmap_info.offset =
1182 mddev->bitmap_info.default_offset;
NeilBrown6409bb02012-05-22 13:55:07 +10001183 mddev->bitmap_info.space =
1184 mddev->bitmap_info.space;
1185 }
NeilBrowna654b9d82005-06-21 17:17:27 -07001186
NeilBrown41158c72005-06-21 17:17:25 -07001187 } else if (mddev->pers == NULL) {
NeilBrownbe6800a2010-05-18 10:17:09 +10001188 /* Insist on good event counter while assembling, except
1189 * for spares (which don't need an event count) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 ++ev1;
NeilBrownbe6800a2010-05-18 10:17:09 +10001191 if (sb->disks[rdev->desc_nr].state & (
1192 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
1193 if (ev1 < mddev->events)
1194 return -EINVAL;
NeilBrown41158c72005-06-21 17:17:25 -07001195 } else if (mddev->bitmap) {
1196 /* if adding to array with a bitmap, then we can accept an
1197 * older device ... but not too old.
1198 */
NeilBrown41158c72005-06-21 17:17:25 -07001199 if (ev1 < mddev->bitmap->events_cleared)
1200 return 0;
NeilBrown07d84d102006-06-26 00:27:56 -07001201 } else {
1202 if (ev1 < mddev->events)
1203 /* just a hot-add of a new device, leave raid_disk at -1 */
1204 return 0;
1205 }
NeilBrown41158c72005-06-21 17:17:25 -07001206
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 if (mddev->level != LEVEL_MULTIPATH) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 desc = sb->disks + rdev->desc_nr;
1209
1210 if (desc->state & (1<<MD_DISK_FAULTY))
NeilBrownb2d444d2005-11-08 21:39:31 -08001211 set_bit(Faulty, &rdev->flags);
NeilBrown7c7546c2006-06-26 00:27:41 -07001212 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1213 desc->raid_disk < mddev->raid_disks */) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001214 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 rdev->raid_disk = desc->raid_disk;
NeilBrown0261cd9f2009-11-13 17:40:48 +11001216 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1217 /* active but not in sync implies recovery up to
1218 * reshape position. We don't know exactly where
1219 * that is, so set to zero for now */
1220 if (mddev->minor_version >= 91) {
1221 rdev->recovery_offset = 0;
1222 rdev->raid_disk = desc->raid_disk;
1223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -07001225 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1226 set_bit(WriteMostly, &rdev->flags);
NeilBrown41158c72005-06-21 17:17:25 -07001227 } else /* MULTIPATH are always insync */
NeilBrownb2d444d2005-11-08 21:39:31 -08001228 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 return 0;
1230}
1231
1232/*
1233 * sync_super for 0.90.0
1234 */
NeilBrownfd01b882011-10-11 16:47:53 +11001235static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236{
1237 mdp_super_t *sb;
NeilBrown3cb03002011-10-11 16:45:26 +11001238 struct md_rdev *rdev2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 int next_spare = mddev->raid_disks;
NeilBrown19133a42005-11-08 21:39:35 -08001240
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241
1242 /* make rdev->sb match mddev data..
1243 *
1244 * 1/ zero out disks
1245 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1246 * 3/ any empty disks < next_spare become removed
1247 *
1248 * disks[0] gets initialised to REMOVED because
1249 * we cannot be sure from other fields if it has
1250 * been initialised or not.
1251 */
1252 int i;
1253 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1254
NeilBrown61181562005-09-09 16:24:02 -07001255 rdev->sb_size = MD_SB_BYTES;
1256
Namhyung Kim65a06f02011-07-27 11:00:36 +10001257 sb = page_address(rdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258
1259 memset(sb, 0, sizeof(*sb));
1260
1261 sb->md_magic = MD_SB_MAGIC;
1262 sb->major_version = mddev->major_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 sb->patch_version = mddev->patch_version;
1264 sb->gvalid_words = 0; /* ignored */
1265 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1266 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1267 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1268 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1269
1270 sb->ctime = mddev->ctime;
1271 sb->level = mddev->level;
Andre Noll58c0fed2009-03-31 14:33:13 +11001272 sb->size = mddev->dev_sectors / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 sb->raid_disks = mddev->raid_disks;
1274 sb->md_minor = mddev->md_minor;
NeilBrowne6910632008-02-06 01:39:51 -08001275 sb->not_persistent = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 sb->utime = mddev->utime;
1277 sb->state = 0;
1278 sb->events_hi = (mddev->events>>32);
1279 sb->events_lo = (u32)mddev->events;
1280
NeilBrownf6705572006-03-27 01:18:11 -08001281 if (mddev->reshape_position == MaxSector)
1282 sb->minor_version = 90;
1283 else {
1284 sb->minor_version = 91;
1285 sb->reshape_position = mddev->reshape_position;
1286 sb->new_level = mddev->new_level;
1287 sb->delta_disks = mddev->delta_disks;
1288 sb->new_layout = mddev->new_layout;
Andre Noll664e7c42009-06-18 08:45:27 +10001289 sb->new_chunk = mddev->new_chunk_sectors << 9;
NeilBrownf6705572006-03-27 01:18:11 -08001290 }
1291 mddev->minor_version = sb->minor_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292 if (mddev->in_sync)
1293 {
1294 sb->recovery_cp = mddev->recovery_cp;
1295 sb->cp_events_hi = (mddev->events>>32);
1296 sb->cp_events_lo = (u32)mddev->events;
1297 if (mddev->recovery_cp == MaxSector)
1298 sb->state = (1<< MD_SB_CLEAN);
1299 } else
1300 sb->recovery_cp = 0;
1301
1302 sb->layout = mddev->layout;
Andre Noll9d8f0362009-06-18 08:45:01 +10001303 sb->chunk_size = mddev->chunk_sectors << 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
NeilBrownc3d97142009-12-14 12:49:52 +11001305 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
NeilBrowna654b9d82005-06-21 17:17:27 -07001306 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1307
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 sb->disks[0].state = (1<<MD_DISK_REMOVED);
NeilBrowndafb20f2012-03-19 12:46:39 +11001309 rdev_for_each(rdev2, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310 mdp_disk_t *d;
NeilBrown86e6ffd2005-11-08 21:39:24 -08001311 int desc_nr;
NeilBrown0261cd9f2009-11-13 17:40:48 +11001312 int is_active = test_bit(In_sync, &rdev2->flags);
1313
1314 if (rdev2->raid_disk >= 0 &&
1315 sb->minor_version >= 91)
1316 /* we have nowhere to store the recovery_offset,
1317 * but if it is not below the reshape_position,
1318 * we can piggy-back on that.
1319 */
1320 is_active = 1;
1321 if (rdev2->raid_disk < 0 ||
1322 test_bit(Faulty, &rdev2->flags))
1323 is_active = 0;
1324 if (is_active)
NeilBrown86e6ffd2005-11-08 21:39:24 -08001325 desc_nr = rdev2->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 else
NeilBrown86e6ffd2005-11-08 21:39:24 -08001327 desc_nr = next_spare++;
NeilBrown19133a42005-11-08 21:39:35 -08001328 rdev2->desc_nr = desc_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 d = &sb->disks[rdev2->desc_nr];
1330 nr_disks++;
1331 d->number = rdev2->desc_nr;
1332 d->major = MAJOR(rdev2->bdev->bd_dev);
1333 d->minor = MINOR(rdev2->bdev->bd_dev);
NeilBrown0261cd9f2009-11-13 17:40:48 +11001334 if (is_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 d->raid_disk = rdev2->raid_disk;
1336 else
1337 d->raid_disk = rdev2->desc_nr; /* compatibility */
NeilBrown1be78922006-03-27 01:18:03 -08001338 if (test_bit(Faulty, &rdev2->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 d->state = (1<<MD_DISK_FAULTY);
NeilBrown0261cd9f2009-11-13 17:40:48 +11001340 else if (is_active) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 d->state = (1<<MD_DISK_ACTIVE);
NeilBrown0261cd9f2009-11-13 17:40:48 +11001342 if (test_bit(In_sync, &rdev2->flags))
1343 d->state |= (1<<MD_DISK_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 active++;
1345 working++;
1346 } else {
1347 d->state = 0;
1348 spare++;
1349 working++;
1350 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -07001351 if (test_bit(WriteMostly, &rdev2->flags))
1352 d->state |= (1<<MD_DISK_WRITEMOSTLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 /* now set the "removed" and "faulty" bits on any missing devices */
1355 for (i=0 ; i < mddev->raid_disks ; i++) {
1356 mdp_disk_t *d = &sb->disks[i];
1357 if (d->state == 0 && d->number == 0) {
1358 d->number = i;
1359 d->raid_disk = i;
1360 d->state = (1<<MD_DISK_REMOVED);
1361 d->state |= (1<<MD_DISK_FAULTY);
1362 failed++;
1363 }
1364 }
1365 sb->nr_disks = nr_disks;
1366 sb->active_disks = active;
1367 sb->working_disks = working;
1368 sb->failed_disks = failed;
1369 sb->spare_disks = spare;
1370
1371 sb->this_disk = sb->disks[rdev->desc_nr];
1372 sb->sb_csum = calc_sb_csum(sb);
1373}
1374
1375/*
Chris Webb0cd17fe2008-06-28 08:31:46 +10001376 * rdev_size_change for 0.90.0
1377 */
1378static unsigned long long
NeilBrown3cb03002011-10-11 16:45:26 +11001379super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
Chris Webb0cd17fe2008-06-28 08:31:46 +10001380{
Andre Noll58c0fed2009-03-31 14:33:13 +11001381 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
Chris Webb0cd17fe2008-06-28 08:31:46 +10001382 return 0; /* component must fit device */
NeilBrownc3d97142009-12-14 12:49:52 +11001383 if (rdev->mddev->bitmap_info.offset)
Chris Webb0cd17fe2008-06-28 08:31:46 +10001384 return 0; /* can't move bitmap */
Jonathan Brassow57b2caa2011-01-14 09:14:33 +11001385 rdev->sb_start = calc_dev_sboffset(rdev);
Andre Noll15f4a5f2008-07-21 14:42:12 +10001386 if (!num_sectors || num_sectors > rdev->sb_start)
1387 num_sectors = rdev->sb_start;
NeilBrown27a7b262011-09-10 17:21:28 +10001388 /* Limit to 4TB as metadata cannot record more than that.
1389 * 4TB == 2^32 KB, or 2*2^32 sectors.
1390 */
NeilBrown667a5312012-08-16 16:46:12 +10001391 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
NeilBrown27a7b262011-09-10 17:21:28 +10001392 num_sectors = (2ULL << 32) - 2;
Andre Noll0f420352008-07-11 22:02:23 +10001393 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
Chris Webb0cd17fe2008-06-28 08:31:46 +10001394 rdev->sb_page);
1395 md_super_wait(rdev->mddev);
Justin Maggardc26a44e2010-11-24 16:36:17 +11001396 return num_sectors;
Chris Webb0cd17fe2008-06-28 08:31:46 +10001397}
1398
NeilBrownc6563a82012-05-21 09:27:00 +10001399static int
1400super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1401{
1402 /* non-zero offset changes not possible with v0.90 */
1403 return new_offset == 0;
1404}
Chris Webb0cd17fe2008-06-28 08:31:46 +10001405
1406/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 * version 1 superblock
1408 */
1409
NeilBrown1c05b4b2006-10-21 10:24:08 -07001410static __le32 calc_sb_1_csum(struct mdp_superblock_1 * sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411{
NeilBrown1c05b4b2006-10-21 10:24:08 -07001412 __le32 disk_csum;
1413 u32 csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 unsigned long long newcsum;
1415 int size = 256 + le32_to_cpu(sb->max_dev)*2;
NeilBrown1c05b4b2006-10-21 10:24:08 -07001416 __le32 *isuper = (__le32*)sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 int i;
1418
1419 disk_csum = sb->sb_csum;
1420 sb->sb_csum = 0;
1421 newcsum = 0;
1422 for (i=0; size>=4; size -= 4 )
1423 newcsum += le32_to_cpu(*isuper++);
1424
1425 if (size == 2)
NeilBrown1c05b4b2006-10-21 10:24:08 -07001426 newcsum += le16_to_cpu(*(__le16*) isuper);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
1428 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1429 sb->sb_csum = disk_csum;
1430 return cpu_to_le32(csum);
1431}
1432
NeilBrown2699b672011-07-28 11:31:47 +10001433static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1434 int acknowledged);
NeilBrown3cb03002011-10-11 16:45:26 +11001435static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436{
1437 struct mdp_superblock_1 *sb;
1438 int ret;
Andre Noll0f420352008-07-11 22:02:23 +10001439 sector_t sb_start;
NeilBrownc6563a82012-05-21 09:27:00 +10001440 sector_t sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
NeilBrown0002b272005-09-09 16:23:53 -07001442 int bmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
1444 /*
Andre Noll0f420352008-07-11 22:02:23 +10001445 * Calculate the position of the superblock in 512byte sectors.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 * It is always aligned to a 4K boundary and
1447 * depeding on minor_version, it can be:
1448 * 0: At least 8K, but less than 12K, from end of device
1449 * 1: At start of device
1450 * 2: 4K from start of device.
1451 */
1452 switch(minor_version) {
1453 case 0:
Mike Snitzer77304d22010-11-08 14:39:12 +01001454 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
Andre Noll0f420352008-07-11 22:02:23 +10001455 sb_start -= 8*2;
1456 sb_start &= ~(sector_t)(4*2-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 break;
1458 case 1:
Andre Noll0f420352008-07-11 22:02:23 +10001459 sb_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460 break;
1461 case 2:
Andre Noll0f420352008-07-11 22:02:23 +10001462 sb_start = 8;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 break;
1464 default:
1465 return -EINVAL;
1466 }
Andre Noll0f420352008-07-11 22:02:23 +10001467 rdev->sb_start = sb_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
NeilBrown0002b272005-09-09 16:23:53 -07001469 /* superblock is rarely larger than 1K, but it can be larger,
1470 * and it is safe to read 4k, so we do that
1471 */
1472 ret = read_disk_sb(rdev, 4096);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 if (ret) return ret;
1474
1475
Namhyung Kim65a06f02011-07-27 11:00:36 +10001476 sb = page_address(rdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477
1478 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1479 sb->major_version != cpu_to_le32(1) ||
1480 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
Andre Noll0f420352008-07-11 22:02:23 +10001481 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
NeilBrown71c08052005-09-09 16:23:51 -07001482 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 return -EINVAL;
1484
1485 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1486 printk("md: invalid superblock checksum on %s\n",
1487 bdevname(rdev->bdev,b));
1488 return -EINVAL;
1489 }
1490 if (le64_to_cpu(sb->data_size) < 10) {
1491 printk("md: data_size too small on %s\n",
1492 bdevname(rdev->bdev,b));
1493 return -EINVAL;
1494 }
NeilBrownc6563a82012-05-21 09:27:00 +10001495 if (sb->pad0 ||
1496 sb->pad3[0] ||
1497 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1498 /* Some padding is non-zero, might be a new feature */
1499 return -EINVAL;
NeilBrowne11e93f2007-05-09 02:35:36 -07001500
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 rdev->preferred_minor = 0xffff;
1502 rdev->data_offset = le64_to_cpu(sb->data_offset);
NeilBrownc6563a82012-05-21 09:27:00 +10001503 rdev->new_data_offset = rdev->data_offset;
1504 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1505 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1506 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
NeilBrown4dbcdc72006-01-06 00:20:52 -08001507 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508
NeilBrown0002b272005-09-09 16:23:53 -07001509 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
Martin K. Petersene1defc42009-05-22 17:17:49 -04001510 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
NeilBrown0002b272005-09-09 16:23:53 -07001511 if (rdev->sb_size & bmask)
NeilBrowna1801f82008-03-04 14:29:31 -08001512 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1513
1514 if (minor_version
Andre Noll0f420352008-07-11 22:02:23 +10001515 && rdev->data_offset < sb_start + (rdev->sb_size/512))
NeilBrowna1801f82008-03-04 14:29:31 -08001516 return -EINVAL;
NeilBrownc6563a82012-05-21 09:27:00 +10001517 if (minor_version
1518 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1519 return -EINVAL;
NeilBrown0002b272005-09-09 16:23:53 -07001520
NeilBrown31b65a02006-07-10 04:44:14 -07001521 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1522 rdev->desc_nr = -1;
1523 else
1524 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1525
NeilBrown2699b672011-07-28 11:31:47 +10001526 if (!rdev->bb_page) {
1527 rdev->bb_page = alloc_page(GFP_KERNEL);
1528 if (!rdev->bb_page)
1529 return -ENOMEM;
1530 }
1531 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1532 rdev->badblocks.count == 0) {
1533 /* need to load the bad block list.
1534 * Currently we limit it to one page.
1535 */
1536 s32 offset;
1537 sector_t bb_sector;
1538 u64 *bbp;
1539 int i;
1540 int sectors = le16_to_cpu(sb->bblog_size);
1541 if (sectors > (PAGE_SIZE / 512))
1542 return -EINVAL;
1543 offset = le32_to_cpu(sb->bblog_offset);
1544 if (offset == 0)
1545 return -EINVAL;
1546 bb_sector = (long long)offset;
1547 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1548 rdev->bb_page, READ, true))
1549 return -EIO;
1550 bbp = (u64 *)page_address(rdev->bb_page);
1551 rdev->badblocks.shift = sb->bblog_shift;
1552 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1553 u64 bb = le64_to_cpu(*bbp);
1554 int count = bb & (0x3ff);
1555 u64 sector = bb >> 10;
1556 sector <<= sb->bblog_shift;
1557 count <<= sb->bblog_shift;
1558 if (bb + 1 == 0)
1559 break;
1560 if (md_set_badblocks(&rdev->badblocks,
1561 sector, count, 1) == 0)
1562 return -EINVAL;
1563 }
1564 } else if (sb->bblog_offset == 0)
1565 rdev->badblocks.shift = -1;
1566
Harvey Harrison9a7b2b02008-04-28 02:15:49 -07001567 if (!refdev) {
NeilBrown8ed75462006-02-03 03:03:41 -08001568 ret = 1;
Harvey Harrison9a7b2b02008-04-28 02:15:49 -07001569 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 __u64 ev1, ev2;
Namhyung Kim65a06f02011-07-27 11:00:36 +10001571 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572
1573 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1574 sb->level != refsb->level ||
1575 sb->layout != refsb->layout ||
1576 sb->chunksize != refsb->chunksize) {
1577 printk(KERN_WARNING "md: %s has strangely different"
1578 " superblock to %s\n",
1579 bdevname(rdev->bdev,b),
1580 bdevname(refdev->bdev,b2));
1581 return -EINVAL;
1582 }
1583 ev1 = le64_to_cpu(sb->events);
1584 ev2 = le64_to_cpu(refsb->events);
1585
1586 if (ev1 > ev2)
NeilBrown8ed75462006-02-03 03:03:41 -08001587 ret = 1;
1588 else
1589 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 }
NeilBrownc6563a82012-05-21 09:27:00 +10001591 if (minor_version) {
1592 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1593 sectors -= rdev->data_offset;
1594 } else
1595 sectors = rdev->sb_start;
1596 if (sectors < le64_to_cpu(sb->data_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597 return -EINVAL;
Andre Nolldd8ac332009-03-31 14:33:13 +11001598 rdev->sectors = le64_to_cpu(sb->data_size);
NeilBrown8ed75462006-02-03 03:03:41 -08001599 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600}
1601
NeilBrownfd01b882011-10-11 16:47:53 +11001602static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603{
Namhyung Kim65a06f02011-07-27 11:00:36 +10001604 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
NeilBrown07d84d102006-06-26 00:27:56 -07001605 __u64 ev1 = le64_to_cpu(sb->events);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606
NeilBrown41158c72005-06-21 17:17:25 -07001607 rdev->raid_disk = -1;
NeilBrownc5d79ad2008-02-06 01:39:54 -08001608 clear_bit(Faulty, &rdev->flags);
1609 clear_bit(In_sync, &rdev->flags);
1610 clear_bit(WriteMostly, &rdev->flags);
NeilBrownc5d79ad2008-02-06 01:39:54 -08001611
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612 if (mddev->raid_disks == 0) {
1613 mddev->major_version = 1;
1614 mddev->patch_version = 0;
NeilBrowne6910632008-02-06 01:39:51 -08001615 mddev->external = 0;
Andre Noll9d8f0362009-06-18 08:45:01 +10001616 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1618 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1619 mddev->level = le32_to_cpu(sb->level);
NeilBrownd9d166c2006-01-06 00:20:51 -08001620 mddev->clevel[0] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621 mddev->layout = le32_to_cpu(sb->layout);
1622 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
Andre Noll58c0fed2009-03-31 14:33:13 +11001623 mddev->dev_sectors = le64_to_cpu(sb->size);
NeilBrown07d84d102006-06-26 00:27:56 -07001624 mddev->events = ev1;
NeilBrownc3d97142009-12-14 12:49:52 +11001625 mddev->bitmap_info.offset = 0;
NeilBrown6409bb02012-05-22 13:55:07 +10001626 mddev->bitmap_info.space = 0;
1627 /* Default location for bitmap is 1K after superblock
1628 * using 3K - total of 4K
1629 */
NeilBrownc3d97142009-12-14 12:49:52 +11001630 mddev->bitmap_info.default_offset = 1024 >> 9;
NeilBrown6409bb02012-05-22 13:55:07 +10001631 mddev->bitmap_info.default_space = (4096-1024) >> 9;
NeilBrown2c810cd2012-05-21 09:27:00 +10001632 mddev->reshape_backwards = 0;
1633
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1635 memcpy(mddev->uuid, sb->set_uuid, 16);
1636
1637 mddev->max_disks = (4096-256)/2;
NeilBrowna654b9d82005-06-21 17:17:27 -07001638
NeilBrown71c08052005-09-09 16:23:51 -07001639 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
NeilBrown6409bb02012-05-22 13:55:07 +10001640 mddev->bitmap_info.file == NULL) {
NeilBrownc3d97142009-12-14 12:49:52 +11001641 mddev->bitmap_info.offset =
1642 (__s32)le32_to_cpu(sb->bitmap_offset);
NeilBrown6409bb02012-05-22 13:55:07 +10001643 /* Metadata doesn't record how much space is available.
1644 * For 1.0, we assume we can use up to the superblock
1645 * if before, else to 4K beyond superblock.
1646 * For others, assume no change is possible.
1647 */
1648 if (mddev->minor_version > 0)
1649 mddev->bitmap_info.space = 0;
1650 else if (mddev->bitmap_info.offset > 0)
1651 mddev->bitmap_info.space =
1652 8 - mddev->bitmap_info.offset;
1653 else
1654 mddev->bitmap_info.space =
1655 -mddev->bitmap_info.offset;
1656 }
NeilBrowne11e93f2007-05-09 02:35:36 -07001657
NeilBrownf6705572006-03-27 01:18:11 -08001658 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1659 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1660 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1661 mddev->new_level = le32_to_cpu(sb->new_level);
1662 mddev->new_layout = le32_to_cpu(sb->new_layout);
Andre Noll664e7c42009-06-18 08:45:27 +10001663 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
NeilBrown2c810cd2012-05-21 09:27:00 +10001664 if (mddev->delta_disks < 0 ||
1665 (mddev->delta_disks == 0 &&
1666 (le32_to_cpu(sb->feature_map)
1667 & MD_FEATURE_RESHAPE_BACKWARDS)))
1668 mddev->reshape_backwards = 1;
NeilBrownf6705572006-03-27 01:18:11 -08001669 } else {
1670 mddev->reshape_position = MaxSector;
1671 mddev->delta_disks = 0;
1672 mddev->new_level = mddev->level;
1673 mddev->new_layout = mddev->layout;
Andre Noll664e7c42009-06-18 08:45:27 +10001674 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08001675 }
1676
NeilBrown41158c72005-06-21 17:17:25 -07001677 } else if (mddev->pers == NULL) {
NeilBrownbe6800a2010-05-18 10:17:09 +10001678 /* Insist of good event counter while assembling, except for
1679 * spares (which don't need an event count) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 ++ev1;
NeilBrownbe6800a2010-05-18 10:17:09 +10001681 if (rdev->desc_nr >= 0 &&
1682 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1683 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1684 if (ev1 < mddev->events)
1685 return -EINVAL;
NeilBrown41158c72005-06-21 17:17:25 -07001686 } else if (mddev->bitmap) {
1687 /* If adding to array with a bitmap, then we can accept an
1688 * older device, but not too old.
1689 */
NeilBrown41158c72005-06-21 17:17:25 -07001690 if (ev1 < mddev->bitmap->events_cleared)
1691 return 0;
NeilBrown07d84d102006-06-26 00:27:56 -07001692 } else {
1693 if (ev1 < mddev->events)
1694 /* just a hot-add of a new device, leave raid_disk at -1 */
1695 return 0;
1696 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 if (mddev->level != LEVEL_MULTIPATH) {
1698 int role;
NeilBrown3673f302009-08-03 10:59:56 +10001699 if (rdev->desc_nr < 0 ||
1700 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1701 role = 0xffff;
1702 rdev->desc_nr = -1;
1703 } else
1704 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705 switch(role) {
1706 case 0xffff: /* spare */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 break;
1708 case 0xfffe: /* faulty */
NeilBrownb2d444d2005-11-08 21:39:31 -08001709 set_bit(Faulty, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 break;
1711 default:
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001712 if ((le32_to_cpu(sb->feature_map) &
1713 MD_FEATURE_RECOVERY_OFFSET))
1714 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
1715 else
1716 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 rdev->raid_disk = role;
1718 break;
1719 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -07001720 if (sb->devflags & WriteMostly1)
1721 set_bit(WriteMostly, &rdev->flags);
NeilBrown2d78f8c2011-12-23 10:17:51 +11001722 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1723 set_bit(Replacement, &rdev->flags);
NeilBrown41158c72005-06-21 17:17:25 -07001724 } else /* MULTIPATH are always insync */
NeilBrownb2d444d2005-11-08 21:39:31 -08001725 set_bit(In_sync, &rdev->flags);
NeilBrown41158c72005-06-21 17:17:25 -07001726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 return 0;
1728}
1729
NeilBrownfd01b882011-10-11 16:47:53 +11001730static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731{
1732 struct mdp_superblock_1 *sb;
NeilBrown3cb03002011-10-11 16:45:26 +11001733 struct md_rdev *rdev2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 int max_dev, i;
1735 /* make rdev->sb match mddev and rdev data. */
1736
Namhyung Kim65a06f02011-07-27 11:00:36 +10001737 sb = page_address(rdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001738
1739 sb->feature_map = 0;
1740 sb->pad0 = 0;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001741 sb->recovery_offset = cpu_to_le64(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 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
NeilBrown1c05b4b2006-10-21 10:24:08 -07001751 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
NeilBrown4dbcdc72006-01-06 00:20:52 -08001752
NeilBrownf0ca3402006-02-02 14:28:04 -08001753 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
Andre Noll58c0fed2009-03-31 14:33:13 +11001754 sb->size = cpu_to_le64(mddev->dev_sectors);
Andre Noll9d8f0362009-06-18 08:45:01 +10001755 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
NeilBrown62e1e382009-05-26 09:40:59 +10001756 sb->level = cpu_to_le32(mddev->level);
1757 sb->layout = cpu_to_le32(mddev->layout);
NeilBrownf0ca3402006-02-02 14:28:04 -08001758
NeilBrownaeb9b212011-08-25 14:43:08 +10001759 if (test_bit(WriteMostly, &rdev->flags))
1760 sb->devflags |= WriteMostly1;
1761 else
1762 sb->devflags &= ~WriteMostly1;
NeilBrownc6563a82012-05-21 09:27:00 +10001763 sb->data_offset = cpu_to_le64(rdev->data_offset);
1764 sb->data_size = cpu_to_le64(rdev->sectors);
NeilBrownaeb9b212011-08-25 14:43:08 +10001765
NeilBrownc3d97142009-12-14 12:49:52 +11001766 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1767 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
NeilBrown71c08052005-09-09 16:23:51 -07001768 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
NeilBrowna654b9d82005-06-21 17:17:27 -07001769 }
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001770
1771 if (rdev->raid_disk >= 0 &&
NeilBrown97e4f422009-03-31 14:33:13 +11001772 !test_bit(In_sync, &rdev->flags)) {
NeilBrown93be75f2009-12-14 12:50:06 +11001773 sb->feature_map |=
1774 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1775 sb->recovery_offset =
1776 cpu_to_le64(rdev->recovery_offset);
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001777 }
NeilBrown2d78f8c2011-12-23 10:17:51 +11001778 if (test_bit(Replacement, &rdev->flags))
1779 sb->feature_map |=
1780 cpu_to_le32(MD_FEATURE_REPLACEMENT);
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001781
NeilBrownf6705572006-03-27 01:18:11 -08001782 if (mddev->reshape_position != MaxSector) {
1783 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1784 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1785 sb->new_layout = cpu_to_le32(mddev->new_layout);
1786 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1787 sb->new_level = cpu_to_le32(mddev->new_level);
Andre Noll664e7c42009-06-18 08:45:27 +10001788 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
NeilBrown2c810cd2012-05-21 09:27:00 +10001789 if (mddev->delta_disks == 0 &&
1790 mddev->reshape_backwards)
1791 sb->feature_map
1792 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
NeilBrownc6563a82012-05-21 09:27:00 +10001793 if (rdev->new_data_offset != rdev->data_offset) {
1794 sb->feature_map
1795 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1796 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1797 - rdev->data_offset));
1798 }
NeilBrownf6705572006-03-27 01:18:11 -08001799 }
NeilBrowna654b9d82005-06-21 17:17:27 -07001800
NeilBrown2699b672011-07-28 11:31:47 +10001801 if (rdev->badblocks.count == 0)
1802 /* Nothing to do for bad blocks*/ ;
1803 else if (sb->bblog_offset == 0)
1804 /* Cannot record bad blocks on this device */
1805 md_error(mddev, rdev);
1806 else {
1807 struct badblocks *bb = &rdev->badblocks;
1808 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1809 u64 *p = bb->page;
1810 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1811 if (bb->changed) {
1812 unsigned seq;
1813
1814retry:
1815 seq = read_seqbegin(&bb->lock);
1816
1817 memset(bbp, 0xff, PAGE_SIZE);
1818
1819 for (i = 0 ; i < bb->count ; i++) {
1820 u64 internal_bb = *p++;
1821 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1822 | BB_LEN(internal_bb));
1823 *bbp++ = cpu_to_le64(store_bb);
1824 }
NeilBrownd0962932012-03-19 12:46:41 +11001825 bb->changed = 0;
NeilBrown2699b672011-07-28 11:31:47 +10001826 if (read_seqretry(&bb->lock, seq))
1827 goto retry;
1828
1829 bb->sector = (rdev->sb_start +
1830 (int)le32_to_cpu(sb->bblog_offset));
1831 bb->size = le16_to_cpu(sb->bblog_size);
NeilBrown2699b672011-07-28 11:31:47 +10001832 }
1833 }
1834
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 max_dev = 0;
NeilBrowndafb20f2012-03-19 12:46:39 +11001836 rdev_for_each(rdev2, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 if (rdev2->desc_nr+1 > max_dev)
1838 max_dev = rdev2->desc_nr+1;
NeilBrowna778b732007-05-23 13:58:10 -07001839
NeilBrown70471da2009-08-03 10:59:57 +10001840 if (max_dev > le32_to_cpu(sb->max_dev)) {
1841 int bmask;
NeilBrowna778b732007-05-23 13:58:10 -07001842 sb->max_dev = cpu_to_le32(max_dev);
NeilBrown70471da2009-08-03 10:59:57 +10001843 rdev->sb_size = max_dev * 2 + 256;
1844 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1845 if (rdev->sb_size & bmask)
1846 rdev->sb_size = (rdev->sb_size | bmask) + 1;
NeilBrownddcf3522010-09-08 16:48:17 +10001847 } else
1848 max_dev = le32_to_cpu(sb->max_dev);
1849
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 for (i=0; i<max_dev;i++)
1851 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1852
NeilBrowndafb20f2012-03-19 12:46:39 +11001853 rdev_for_each(rdev2, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 i = rdev2->desc_nr;
NeilBrownb2d444d2005-11-08 21:39:31 -08001855 if (test_bit(Faulty, &rdev2->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 sb->dev_roles[i] = cpu_to_le16(0xfffe);
NeilBrownb2d444d2005-11-08 21:39:31 -08001857 else if (test_bit(In_sync, &rdev2->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
NeilBrown93be75f2009-12-14 12:50:06 +11001859 else if (rdev2->raid_disk >= 0)
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001860 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 else
1862 sb->dev_roles[i] = cpu_to_le16(0xffff);
1863 }
1864
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 sb->sb_csum = calc_sb_1_csum(sb);
1866}
1867
Chris Webb0cd17fe2008-06-28 08:31:46 +10001868static unsigned long long
NeilBrown3cb03002011-10-11 16:45:26 +11001869super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
Chris Webb0cd17fe2008-06-28 08:31:46 +10001870{
1871 struct mdp_superblock_1 *sb;
Andre Noll15f4a5f2008-07-21 14:42:12 +10001872 sector_t max_sectors;
Andre Noll58c0fed2009-03-31 14:33:13 +11001873 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
Chris Webb0cd17fe2008-06-28 08:31:46 +10001874 return 0; /* component must fit device */
NeilBrownc6563a82012-05-21 09:27:00 +10001875 if (rdev->data_offset != rdev->new_data_offset)
1876 return 0; /* too confusing */
Andre Noll0f420352008-07-11 22:02:23 +10001877 if (rdev->sb_start < rdev->data_offset) {
Chris Webb0cd17fe2008-06-28 08:31:46 +10001878 /* minor versions 1 and 2; superblock before data */
Mike Snitzer77304d22010-11-08 14:39:12 +01001879 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
Andre Noll15f4a5f2008-07-21 14:42:12 +10001880 max_sectors -= rdev->data_offset;
1881 if (!num_sectors || num_sectors > max_sectors)
1882 num_sectors = max_sectors;
NeilBrownc3d97142009-12-14 12:49:52 +11001883 } else if (rdev->mddev->bitmap_info.offset) {
Chris Webb0cd17fe2008-06-28 08:31:46 +10001884 /* minor version 0 with bitmap we can't move */
1885 return 0;
1886 } else {
1887 /* minor version 0; superblock after data */
Andre Noll0f420352008-07-11 22:02:23 +10001888 sector_t sb_start;
Mike Snitzer77304d22010-11-08 14:39:12 +01001889 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
Andre Noll0f420352008-07-11 22:02:23 +10001890 sb_start &= ~(sector_t)(4*2 - 1);
Andre Nolldd8ac332009-03-31 14:33:13 +11001891 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
Andre Noll15f4a5f2008-07-21 14:42:12 +10001892 if (!num_sectors || num_sectors > max_sectors)
1893 num_sectors = max_sectors;
Andre Noll0f420352008-07-11 22:02:23 +10001894 rdev->sb_start = sb_start;
Chris Webb0cd17fe2008-06-28 08:31:46 +10001895 }
Namhyung Kim65a06f02011-07-27 11:00:36 +10001896 sb = page_address(rdev->sb_page);
Andre Noll15f4a5f2008-07-21 14:42:12 +10001897 sb->data_size = cpu_to_le64(num_sectors);
Andre Noll0f420352008-07-11 22:02:23 +10001898 sb->super_offset = rdev->sb_start;
Chris Webb0cd17fe2008-06-28 08:31:46 +10001899 sb->sb_csum = calc_sb_1_csum(sb);
Andre Noll0f420352008-07-11 22:02:23 +10001900 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
Chris Webb0cd17fe2008-06-28 08:31:46 +10001901 rdev->sb_page);
1902 md_super_wait(rdev->mddev);
Justin Maggardc26a44e2010-11-24 16:36:17 +11001903 return num_sectors;
NeilBrownc6563a82012-05-21 09:27:00 +10001904
1905}
1906
1907static int
1908super_1_allow_new_offset(struct md_rdev *rdev,
1909 unsigned long long new_offset)
1910{
1911 /* All necessary checks on new >= old have been done */
1912 struct bitmap *bitmap;
1913 if (new_offset >= rdev->data_offset)
1914 return 1;
1915
1916 /* with 1.0 metadata, there is no metadata to tread on
1917 * so we can always move back */
1918 if (rdev->mddev->minor_version == 0)
1919 return 1;
1920
1921 /* otherwise we must be sure not to step on
1922 * any metadata, so stay:
1923 * 36K beyond start of superblock
1924 * beyond end of badblocks
1925 * beyond write-intent bitmap
1926 */
1927 if (rdev->sb_start + (32+4)*2 > new_offset)
1928 return 0;
1929 bitmap = rdev->mddev->bitmap;
1930 if (bitmap && !rdev->mddev->bitmap_info.file &&
1931 rdev->sb_start + rdev->mddev->bitmap_info.offset +
NeilBrown1ec885c2012-05-22 13:55:10 +10001932 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
NeilBrownc6563a82012-05-21 09:27:00 +10001933 return 0;
1934 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1935 return 0;
1936
1937 return 1;
Chris Webb0cd17fe2008-06-28 08:31:46 +10001938}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939
Adrian Bunk75c96f82005-05-05 16:16:09 -07001940static struct super_type super_types[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 [0] = {
1942 .name = "0.90.0",
1943 .owner = THIS_MODULE,
Chris Webb0cd17fe2008-06-28 08:31:46 +10001944 .load_super = super_90_load,
1945 .validate_super = super_90_validate,
1946 .sync_super = super_90_sync,
1947 .rdev_size_change = super_90_rdev_size_change,
NeilBrownc6563a82012-05-21 09:27:00 +10001948 .allow_new_offset = super_90_allow_new_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 },
1950 [1] = {
1951 .name = "md-1",
1952 .owner = THIS_MODULE,
Chris Webb0cd17fe2008-06-28 08:31:46 +10001953 .load_super = super_1_load,
1954 .validate_super = super_1_validate,
1955 .sync_super = super_1_sync,
1956 .rdev_size_change = super_1_rdev_size_change,
NeilBrownc6563a82012-05-21 09:27:00 +10001957 .allow_new_offset = super_1_allow_new_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 },
1959};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960
NeilBrownfd01b882011-10-11 16:47:53 +11001961static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
Jonathan Brassow076f9682011-06-07 17:51:30 -05001962{
1963 if (mddev->sync_super) {
1964 mddev->sync_super(mddev, rdev);
1965 return;
1966 }
1967
1968 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1969
1970 super_types[mddev->major_version].sync_super(mddev, rdev);
1971}
1972
NeilBrownfd01b882011-10-11 16:47:53 +11001973static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974{
NeilBrown3cb03002011-10-11 16:45:26 +11001975 struct md_rdev *rdev, *rdev2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
NeilBrown4b809912008-07-21 17:05:25 +10001977 rcu_read_lock();
1978 rdev_for_each_rcu(rdev, mddev1)
1979 rdev_for_each_rcu(rdev2, mddev2)
NeilBrown7dd5e7c32007-02-28 20:11:35 -08001980 if (rdev->bdev->bd_contains ==
NeilBrown4b809912008-07-21 17:05:25 +10001981 rdev2->bdev->bd_contains) {
1982 rcu_read_unlock();
NeilBrown7dd5e7c32007-02-28 20:11:35 -08001983 return 1;
NeilBrown4b809912008-07-21 17:05:25 +10001984 }
1985 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 return 0;
1987}
1988
1989static LIST_HEAD(pending_raid_disks);
1990
Andre Nollac5e7112009-08-03 10:59:47 +10001991/*
1992 * Try to register data integrity profile for an mddev
1993 *
1994 * This is called when an array is started and after a disk has been kicked
1995 * from the array. It only succeeds if all working and active component devices
1996 * are integrity capable with matching profiles.
1997 */
NeilBrownfd01b882011-10-11 16:47:53 +11001998int md_integrity_register(struct mddev *mddev)
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11001999{
NeilBrown3cb03002011-10-11 16:45:26 +11002000 struct md_rdev *rdev, *reference = NULL;
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11002001
Andre Nollac5e7112009-08-03 10:59:47 +10002002 if (list_empty(&mddev->disks))
2003 return 0; /* nothing to do */
Jonathan Brassow629acb62011-06-08 15:10:08 +10002004 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
2005 return 0; /* shouldn't register, or already is */
NeilBrowndafb20f2012-03-19 12:46:39 +11002006 rdev_for_each(rdev, mddev) {
Andre Nollac5e7112009-08-03 10:59:47 +10002007 /* skip spares and non-functional disks */
2008 if (test_bit(Faulty, &rdev->flags))
2009 continue;
2010 if (rdev->raid_disk < 0)
2011 continue;
Andre Nollac5e7112009-08-03 10:59:47 +10002012 if (!reference) {
2013 /* Use the first rdev as the reference */
2014 reference = rdev;
2015 continue;
2016 }
2017 /* does this rdev's profile match the reference profile? */
2018 if (blk_integrity_compare(reference->bdev->bd_disk,
2019 rdev->bdev->bd_disk) < 0)
2020 return -EINVAL;
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11002021 }
Martin K. Petersen89078d52011-03-28 20:09:12 -04002022 if (!reference || !bdev_get_integrity(reference->bdev))
2023 return 0;
Andre Nollac5e7112009-08-03 10:59:47 +10002024 /*
2025 * All component devices are integrity capable and have matching
2026 * profiles, register the common profile for the md device.
2027 */
2028 if (blk_integrity_register(mddev->gendisk,
2029 bdev_get_integrity(reference->bdev)) != 0) {
2030 printk(KERN_ERR "md: failed to register integrity for %s\n",
2031 mdname(mddev));
2032 return -EINVAL;
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11002033 }
Martin K. Petersena91a2782011-03-17 11:11:05 +01002034 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
2035 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
2036 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
2037 mdname(mddev));
2038 return -EINVAL;
2039 }
Andre Nollac5e7112009-08-03 10:59:47 +10002040 return 0;
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11002041}
Andre Nollac5e7112009-08-03 10:59:47 +10002042EXPORT_SYMBOL(md_integrity_register);
2043
2044/* Disable data integrity if non-capable/non-matching disk is being added */
NeilBrownfd01b882011-10-11 16:47:53 +11002045void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
Andre Nollac5e7112009-08-03 10:59:47 +10002046{
Jonathan Brassow2863b9e2012-10-11 13:38:58 +11002047 struct blk_integrity *bi_rdev;
2048 struct blk_integrity *bi_mddev;
2049
2050 if (!mddev->gendisk)
2051 return;
2052
2053 bi_rdev = bdev_get_integrity(rdev->bdev);
2054 bi_mddev = blk_get_integrity(mddev->gendisk);
Andre Nollac5e7112009-08-03 10:59:47 +10002055
2056 if (!bi_mddev) /* nothing to do */
2057 return;
2058 if (rdev->raid_disk < 0) /* skip spares */
2059 return;
2060 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2061 rdev->bdev->bd_disk) >= 0)
2062 return;
2063 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2064 blk_integrity_unregister(mddev->gendisk);
2065}
2066EXPORT_SYMBOL(md_integrity_add_rdev);
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11002067
NeilBrownfd01b882011-10-11 16:47:53 +11002068static int bind_rdev_to_array(struct md_rdev * rdev, struct mddev * mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069{
NeilBrown7dd5e7c32007-02-28 20:11:35 -08002070 char b[BDEVNAME_SIZE];
NeilBrownf637b9f2005-11-08 21:39:37 -08002071 struct kobject *ko;
Neil Brown1edf80d2006-01-12 01:05:23 -08002072 char *s;
NeilBrown5e55e2f2007-03-26 21:32:14 -08002073 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
2075 if (rdev->mddev) {
2076 MD_BUG();
2077 return -EINVAL;
2078 }
Dan Williams11e2ede2008-04-30 00:52:32 -07002079
2080 /* prevent duplicates */
2081 if (find_rdev(mddev, rdev->bdev->bd_dev))
2082 return -EEXIST;
2083
Andre Nolldd8ac332009-03-31 14:33:13 +11002084 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2085 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2086 rdev->sectors < mddev->dev_sectors)) {
NeilBrowna778b732007-05-23 13:58:10 -07002087 if (mddev->pers) {
2088 /* Cannot change size, so fail
2089 * If mddev->level <= 0, then we don't care
2090 * about aligning sizes (e.g. linear)
2091 */
2092 if (mddev->level > 0)
2093 return -ENOSPC;
2094 } else
Andre Nolldd8ac332009-03-31 14:33:13 +11002095 mddev->dev_sectors = rdev->sectors;
NeilBrown2bf071b2006-01-06 00:20:55 -08002096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097
2098 /* Verify rdev->desc_nr is unique.
2099 * If it is -1, assign a free number, else
2100 * check number is not in use
2101 */
2102 if (rdev->desc_nr < 0) {
2103 int choice = 0;
2104 if (mddev->pers) choice = mddev->raid_disks;
2105 while (find_rdev_nr(mddev, choice))
2106 choice++;
2107 rdev->desc_nr = choice;
2108 } else {
2109 if (find_rdev_nr(mddev, rdev->desc_nr))
2110 return -EBUSY;
2111 }
NeilBrownde01dfa2009-02-06 18:02:46 +11002112 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2113 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2114 mdname(mddev), mddev->max_disks);
2115 return -EBUSY;
2116 }
NeilBrown19133a42005-11-08 21:39:35 -08002117 bdevname(rdev->bdev,b);
Greg Kroah-Hartman649316b2007-12-17 23:05:35 -07002118 while ( (s=strchr(b, '/')) != NULL)
Neil Brown1edf80d2006-01-12 01:05:23 -08002119 *s = '!';
Greg Kroah-Hartman649316b2007-12-17 23:05:35 -07002120
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 rdev->mddev = mddev;
NeilBrown19133a42005-11-08 21:39:35 -08002122 printk(KERN_INFO "md: bind<%s>\n", b);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002123
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -07002124 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
NeilBrown5e55e2f2007-03-26 21:32:14 -08002125 goto fail;
NeilBrown86e6ffd2005-11-08 21:39:24 -08002126
Tejun Heo0762b8b2008-08-25 19:56:12 +09002127 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
NeilBrown00bcb4a2010-06-01 19:37:23 +10002128 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2129 /* failure here is OK */;
2130 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
NeilBrown3c0ee632008-10-21 13:25:28 +11002131
NeilBrown4b809912008-07-21 17:05:25 +10002132 list_add_rcu(&rdev->same_set, &mddev->disks);
Tejun Heoe09b4572010-11-13 11:55:17 +01002133 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
NeilBrown4044ba52009-01-09 08:31:11 +11002134
2135 /* May as well allow recovery to be retried once */
NeilBrown53890422011-07-27 11:00:36 +10002136 mddev->recovery_disabled++;
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11002137
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 return 0;
NeilBrown5e55e2f2007-03-26 21:32:14 -08002139
2140 fail:
2141 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2142 b, mdname(mddev));
2143 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144}
2145
NeilBrown177a99b2008-02-06 01:39:56 -08002146static void md_delayed_delete(struct work_struct *ws)
NeilBrown5792a282007-04-04 19:08:18 -07002147{
NeilBrown3cb03002011-10-11 16:45:26 +11002148 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
NeilBrown5792a282007-04-04 19:08:18 -07002149 kobject_del(&rdev->kobj);
NeilBrown177a99b2008-02-06 01:39:56 -08002150 kobject_put(&rdev->kobj);
NeilBrown5792a282007-04-04 19:08:18 -07002151}
2152
NeilBrown3cb03002011-10-11 16:45:26 +11002153static void unbind_rdev_from_array(struct md_rdev * rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002154{
2155 char b[BDEVNAME_SIZE];
2156 if (!rdev->mddev) {
2157 MD_BUG();
2158 return;
2159 }
Tejun Heo49731ba2011-01-14 18:43:57 +01002160 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
NeilBrown4b809912008-07-21 17:05:25 +10002161 list_del_rcu(&rdev->same_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2163 rdev->mddev = NULL;
NeilBrown86e6ffd2005-11-08 21:39:24 -08002164 sysfs_remove_link(&rdev->kobj, "block");
NeilBrown3c0ee632008-10-21 13:25:28 +11002165 sysfs_put(rdev->sysfs_state);
2166 rdev->sysfs_state = NULL;
NeilBrown2230dfe2011-07-28 11:31:46 +10002167 rdev->badblocks.count = 0;
NeilBrown5792a282007-04-04 19:08:18 -07002168 /* We need to delay this, otherwise we can deadlock when
NeilBrown4b809912008-07-21 17:05:25 +10002169 * writing to 'remove' to "dev/state". We also need
2170 * to delay it due to rcu usage.
NeilBrown5792a282007-04-04 19:08:18 -07002171 */
NeilBrown4b809912008-07-21 17:05:25 +10002172 synchronize_rcu();
NeilBrown177a99b2008-02-06 01:39:56 -08002173 INIT_WORK(&rdev->del_work, md_delayed_delete);
2174 kobject_get(&rdev->kobj);
Tejun Heoe804ac72010-10-15 15:36:08 +02002175 queue_work(md_misc_wq, &rdev->del_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176}
2177
2178/*
2179 * prevent the device from being mounted, repartitioned or
2180 * otherwise reused by a RAID array (or any other kernel
2181 * subsystem), by bd_claiming the device.
2182 */
NeilBrown3cb03002011-10-11 16:45:26 +11002183static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184{
2185 int err = 0;
2186 struct block_device *bdev;
2187 char b[BDEVNAME_SIZE];
2188
Tejun Heod4d77622010-11-13 11:55:18 +01002189 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
NeilBrown3cb03002011-10-11 16:45:26 +11002190 shared ? (struct md_rdev *)lock_rdev : rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 if (IS_ERR(bdev)) {
2192 printk(KERN_ERR "md: could not open %s.\n",
2193 __bdevname(dev, b));
2194 return PTR_ERR(bdev);
2195 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 rdev->bdev = bdev;
2197 return err;
2198}
2199
NeilBrown3cb03002011-10-11 16:45:26 +11002200static void unlock_rdev(struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201{
2202 struct block_device *bdev = rdev->bdev;
2203 rdev->bdev = NULL;
2204 if (!bdev)
2205 MD_BUG();
Tejun Heoe525fd82010-11-13 11:55:17 +01002206 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207}
2208
2209void md_autodetect_dev(dev_t dev);
2210
NeilBrown3cb03002011-10-11 16:45:26 +11002211static void export_rdev(struct md_rdev * rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212{
2213 char b[BDEVNAME_SIZE];
2214 printk(KERN_INFO "md: export_rdev(%s)\n",
2215 bdevname(rdev->bdev,b));
2216 if (rdev->mddev)
2217 MD_BUG();
NeilBrown545c8792012-05-22 13:54:30 +10002218 md_rdev_clear(rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219#ifndef MODULE
NeilBrownd0fae182008-03-04 14:29:31 -08002220 if (test_bit(AutoDetected, &rdev->flags))
2221 md_autodetect_dev(rdev->bdev->bd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222#endif
2223 unlock_rdev(rdev);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002224 kobject_put(&rdev->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002225}
2226
NeilBrown3cb03002011-10-11 16:45:26 +11002227static void kick_rdev_from_array(struct md_rdev * rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228{
2229 unbind_rdev_from_array(rdev);
2230 export_rdev(rdev);
2231}
2232
NeilBrownfd01b882011-10-11 16:47:53 +11002233static void export_array(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234{
NeilBrown3cb03002011-10-11 16:45:26 +11002235 struct md_rdev *rdev, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236
NeilBrowndafb20f2012-03-19 12:46:39 +11002237 rdev_for_each_safe(rdev, tmp, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 if (!rdev->mddev) {
2239 MD_BUG();
2240 continue;
2241 }
2242 kick_rdev_from_array(rdev);
2243 }
2244 if (!list_empty(&mddev->disks))
2245 MD_BUG();
2246 mddev->raid_disks = 0;
2247 mddev->major_version = 0;
2248}
2249
2250static void print_desc(mdp_disk_t *desc)
2251{
2252 printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
2253 desc->major,desc->minor,desc->raid_disk,desc->state);
2254}
2255
Cheng Renquancd2ac932009-01-09 08:31:08 +11002256static void print_sb_90(mdp_super_t *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257{
2258 int i;
2259
2260 printk(KERN_INFO
2261 "md: SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
2262 sb->major_version, sb->minor_version, sb->patch_version,
2263 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
2264 sb->ctime);
2265 printk(KERN_INFO "md: L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
2266 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
2267 sb->md_minor, sb->layout, sb->chunk_size);
2268 printk(KERN_INFO "md: UT:%08x ST:%d AD:%d WD:%d"
2269 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
2270 sb->utime, sb->state, sb->active_disks, sb->working_disks,
2271 sb->failed_disks, sb->spare_disks,
2272 sb->sb_csum, (unsigned long)sb->events_lo);
2273
2274 printk(KERN_INFO);
2275 for (i = 0; i < MD_SB_DISKS; i++) {
2276 mdp_disk_t *desc;
2277
2278 desc = sb->disks + i;
2279 if (desc->number || desc->major || desc->minor ||
2280 desc->raid_disk || (desc->state && (desc->state != 4))) {
2281 printk(" D %2d: ", i);
2282 print_desc(desc);
2283 }
2284 }
2285 printk(KERN_INFO "md: THIS: ");
2286 print_desc(&sb->this_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287}
2288
Cheng Renquancd2ac932009-01-09 08:31:08 +11002289static void print_sb_1(struct mdp_superblock_1 *sb)
2290{
2291 __u8 *uuid;
2292
2293 uuid = sb->set_uuid;
Joe Perchesad361c92009-07-06 13:05:40 -07002294 printk(KERN_INFO
Joe Perches7b75c2f2009-12-14 18:01:12 -08002295 "md: SB: (V:%u) (F:0x%08x) Array-ID:<%pU>\n"
Joe Perchesad361c92009-07-06 13:05:40 -07002296 "md: Name: \"%s\" CT:%llu\n",
Cheng Renquancd2ac932009-01-09 08:31:08 +11002297 le32_to_cpu(sb->major_version),
2298 le32_to_cpu(sb->feature_map),
Joe Perches7b75c2f2009-12-14 18:01:12 -08002299 uuid,
Cheng Renquancd2ac932009-01-09 08:31:08 +11002300 sb->set_name,
2301 (unsigned long long)le64_to_cpu(sb->ctime)
2302 & MD_SUPERBLOCK_1_TIME_SEC_MASK);
2303
2304 uuid = sb->device_uuid;
Joe Perchesad361c92009-07-06 13:05:40 -07002305 printk(KERN_INFO
2306 "md: L%u SZ%llu RD:%u LO:%u CS:%u DO:%llu DS:%llu SO:%llu"
Cheng Renquancd2ac932009-01-09 08:31:08 +11002307 " RO:%llu\n"
Joe Perches7b75c2f2009-12-14 18:01:12 -08002308 "md: Dev:%08x UUID: %pU\n"
Joe Perchesad361c92009-07-06 13:05:40 -07002309 "md: (F:0x%08x) UT:%llu Events:%llu ResyncOffset:%llu CSUM:0x%08x\n"
2310 "md: (MaxDev:%u) \n",
Cheng Renquancd2ac932009-01-09 08:31:08 +11002311 le32_to_cpu(sb->level),
2312 (unsigned long long)le64_to_cpu(sb->size),
2313 le32_to_cpu(sb->raid_disks),
2314 le32_to_cpu(sb->layout),
2315 le32_to_cpu(sb->chunksize),
2316 (unsigned long long)le64_to_cpu(sb->data_offset),
2317 (unsigned long long)le64_to_cpu(sb->data_size),
2318 (unsigned long long)le64_to_cpu(sb->super_offset),
2319 (unsigned long long)le64_to_cpu(sb->recovery_offset),
2320 le32_to_cpu(sb->dev_number),
Joe Perches7b75c2f2009-12-14 18:01:12 -08002321 uuid,
Cheng Renquancd2ac932009-01-09 08:31:08 +11002322 sb->devflags,
2323 (unsigned long long)le64_to_cpu(sb->utime) & MD_SUPERBLOCK_1_TIME_SEC_MASK,
2324 (unsigned long long)le64_to_cpu(sb->events),
2325 (unsigned long long)le64_to_cpu(sb->resync_offset),
2326 le32_to_cpu(sb->sb_csum),
2327 le32_to_cpu(sb->max_dev)
2328 );
2329}
2330
NeilBrown3cb03002011-10-11 16:45:26 +11002331static void print_rdev(struct md_rdev *rdev, int major_version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332{
2333 char b[BDEVNAME_SIZE];
Andre Nolldd8ac332009-03-31 14:33:13 +11002334 printk(KERN_INFO "md: rdev %s, Sect:%08llu F:%d S:%d DN:%u\n",
2335 bdevname(rdev->bdev, b), (unsigned long long)rdev->sectors,
NeilBrownb2d444d2005-11-08 21:39:31 -08002336 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
2337 rdev->desc_nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 if (rdev->sb_loaded) {
Cheng Renquancd2ac932009-01-09 08:31:08 +11002339 printk(KERN_INFO "md: rdev superblock (MJ:%d):\n", major_version);
2340 switch (major_version) {
2341 case 0:
Namhyung Kim65a06f02011-07-27 11:00:36 +10002342 print_sb_90(page_address(rdev->sb_page));
Cheng Renquancd2ac932009-01-09 08:31:08 +11002343 break;
2344 case 1:
Namhyung Kim65a06f02011-07-27 11:00:36 +10002345 print_sb_1(page_address(rdev->sb_page));
Cheng Renquancd2ac932009-01-09 08:31:08 +11002346 break;
2347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 } else
2349 printk(KERN_INFO "md: no rdev superblock!\n");
2350}
2351
Adrian Bunk5e563412006-06-26 00:27:42 -07002352static void md_print_devices(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353{
Cheng Renquan159ec1f2009-01-09 08:31:08 +11002354 struct list_head *tmp;
NeilBrown3cb03002011-10-11 16:45:26 +11002355 struct md_rdev *rdev;
NeilBrownfd01b882011-10-11 16:47:53 +11002356 struct mddev *mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 char b[BDEVNAME_SIZE];
2358
2359 printk("\n");
2360 printk("md: **********************************\n");
2361 printk("md: * <COMPLETE RAID STATE PRINTOUT> *\n");
2362 printk("md: **********************************\n");
NeilBrown29ac4aa2008-02-06 01:39:58 -08002363 for_each_mddev(mddev, tmp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364
NeilBrown32a76272005-06-21 17:17:14 -07002365 if (mddev->bitmap)
2366 bitmap_print_sb(mddev->bitmap);
2367 else
2368 printk("%s: ", mdname(mddev));
NeilBrowndafb20f2012-03-19 12:46:39 +11002369 rdev_for_each(rdev, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 printk("<%s>", bdevname(rdev->bdev,b));
2371 printk("\n");
2372
NeilBrowndafb20f2012-03-19 12:46:39 +11002373 rdev_for_each(rdev, mddev)
Cheng Renquancd2ac932009-01-09 08:31:08 +11002374 print_rdev(rdev, mddev->major_version);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002375 }
2376 printk("md: **********************************\n");
2377 printk("\n");
2378}
2379
2380
NeilBrownfd01b882011-10-11 16:47:53 +11002381static void sync_sbs(struct mddev * mddev, int nospares)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382{
NeilBrown42543762006-06-26 00:27:57 -07002383 /* Update each superblock (in-memory image), but
2384 * if we are allowed to, skip spares which already
2385 * have the right event counter, or have one earlier
2386 * (which would mean they aren't being marked as dirty
2387 * with the rest of the array)
2388 */
NeilBrown3cb03002011-10-11 16:45:26 +11002389 struct md_rdev *rdev;
NeilBrowndafb20f2012-03-19 12:46:39 +11002390 rdev_for_each(rdev, mddev) {
NeilBrown42543762006-06-26 00:27:57 -07002391 if (rdev->sb_events == mddev->events ||
2392 (nospares &&
2393 rdev->raid_disk < 0 &&
NeilBrown42543762006-06-26 00:27:57 -07002394 rdev->sb_events+1 == mddev->events)) {
2395 /* Don't update this superblock */
2396 rdev->sb_loaded = 2;
2397 } else {
Jonathan Brassow076f9682011-06-07 17:51:30 -05002398 sync_super(mddev, rdev);
NeilBrown42543762006-06-26 00:27:57 -07002399 rdev->sb_loaded = 1;
2400 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 }
2402}
2403
NeilBrownfd01b882011-10-11 16:47:53 +11002404static void md_update_sb(struct mddev * mddev, int force_change)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405{
NeilBrown3cb03002011-10-11 16:45:26 +11002406 struct md_rdev *rdev;
NeilBrown06d91a52005-06-21 17:17:12 -07002407 int sync_req;
NeilBrown42543762006-06-26 00:27:57 -07002408 int nospares = 0;
NeilBrown2699b672011-07-28 11:31:47 +10002409 int any_badblocks_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411repeat:
NeilBrown3a3a5dd2010-08-16 18:09:31 +10002412 /* First make sure individual recovery_offsets are correct */
NeilBrowndafb20f2012-03-19 12:46:39 +11002413 rdev_for_each(rdev, mddev) {
NeilBrown3a3a5dd2010-08-16 18:09:31 +10002414 if (rdev->raid_disk >= 0 &&
2415 mddev->delta_disks >= 0 &&
2416 !test_bit(In_sync, &rdev->flags) &&
2417 mddev->curr_resync_completed > rdev->recovery_offset)
2418 rdev->recovery_offset = mddev->curr_resync_completed;
2419
2420 }
Dan Williamsbd52b742010-08-30 17:33:33 +10002421 if (!mddev->persistent) {
NeilBrown070dc6d2010-08-30 17:33:34 +10002422 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrown3a3a5dd2010-08-16 18:09:31 +10002423 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownde393cd2011-07-28 11:31:48 +10002424 if (!mddev->external) {
NeilBrownd97a41d2010-10-28 17:30:20 +11002425 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrowndafb20f2012-03-19 12:46:39 +11002426 rdev_for_each(rdev, mddev) {
NeilBrownde393cd2011-07-28 11:31:48 +10002427 if (rdev->badblocks.changed) {
NeilBrownd0962932012-03-19 12:46:41 +11002428 rdev->badblocks.changed = 0;
NeilBrownde393cd2011-07-28 11:31:48 +10002429 md_ack_all_badblocks(&rdev->badblocks);
2430 md_error(mddev, rdev);
2431 }
2432 clear_bit(Blocked, &rdev->flags);
2433 clear_bit(BlockedBadBlocks, &rdev->flags);
2434 wake_up(&rdev->blocked_wait);
2435 }
2436 }
NeilBrown3a3a5dd2010-08-16 18:09:31 +10002437 wake_up(&mddev->sb_wait);
2438 return;
2439 }
2440
NeilBrowna9701a32005-11-08 21:39:34 -08002441 spin_lock_irq(&mddev->write_lock);
NeilBrown84692192006-08-27 01:23:49 -07002442
NeilBrown3a3a5dd2010-08-16 18:09:31 +10002443 mddev->utime = get_seconds();
2444
NeilBrown850b2b42006-10-03 01:15:46 -07002445 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2446 force_change = 1;
2447 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2448 /* just a clean<-> dirty transition, possibly leave spares alone,
2449 * though if events isn't the right even/odd, we will have to do
2450 * spares after all
2451 */
2452 nospares = 1;
2453 if (force_change)
2454 nospares = 0;
2455 if (mddev->degraded)
NeilBrown84692192006-08-27 01:23:49 -07002456 /* If the array is degraded, then skipping spares is both
2457 * dangerous and fairly pointless.
2458 * Dangerous because a device that was removed from the array
2459 * might have a event_count that still looks up-to-date,
2460 * so it can be re-added without a resync.
2461 * Pointless because if there are any spares to skip,
2462 * then a recovery will happen and soon that array won't
2463 * be degraded any more and the spare can go back to sleep then.
2464 */
NeilBrown850b2b42006-10-03 01:15:46 -07002465 nospares = 0;
NeilBrown84692192006-08-27 01:23:49 -07002466
NeilBrown06d91a52005-06-21 17:17:12 -07002467 sync_req = mddev->in_sync;
NeilBrown42543762006-06-26 00:27:57 -07002468
2469 /* If this is just a dirty<->clean transition, and the array is clean
2470 * and 'events' is odd, we can roll back to the previous clean state */
NeilBrown850b2b42006-10-03 01:15:46 -07002471 if (nospares
NeilBrown42543762006-06-26 00:27:57 -07002472 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
NeilBrowna8707c02010-05-18 09:28:43 +10002473 && mddev->can_decrease_events
2474 && mddev->events != 1) {
NeilBrown42543762006-06-26 00:27:57 -07002475 mddev->events--;
NeilBrowna8707c02010-05-18 09:28:43 +10002476 mddev->can_decrease_events = 0;
2477 } else {
NeilBrown42543762006-06-26 00:27:57 -07002478 /* otherwise we have to go forward and ... */
2479 mddev->events ++;
NeilBrowna8707c02010-05-18 09:28:43 +10002480 mddev->can_decrease_events = nospares;
NeilBrown42543762006-06-26 00:27:57 -07002481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482
2483 if (!mddev->events) {
2484 /*
2485 * oops, this 64-bit counter should never wrap.
2486 * Either we are in around ~1 trillion A.C., assuming
2487 * 1 reboot per second, or we have a bug:
2488 */
2489 MD_BUG();
2490 mddev->events --;
2491 }
NeilBrown2699b672011-07-28 11:31:47 +10002492
NeilBrowndafb20f2012-03-19 12:46:39 +11002493 rdev_for_each(rdev, mddev) {
NeilBrown2699b672011-07-28 11:31:47 +10002494 if (rdev->badblocks.changed)
2495 any_badblocks_changed++;
NeilBrownde393cd2011-07-28 11:31:48 +10002496 if (test_bit(Faulty, &rdev->flags))
2497 set_bit(FaultRecorded, &rdev->flags);
2498 }
NeilBrown2699b672011-07-28 11:31:47 +10002499
NeilBrowne6910632008-02-06 01:39:51 -08002500 sync_sbs(mddev, nospares);
NeilBrowna9701a32005-11-08 21:39:34 -08002501 spin_unlock_irq(&mddev->write_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502
NeilBrown36a4e1f2011-10-07 14:23:17 +11002503 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2504 mdname(mddev), mddev->in_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505
NeilBrown4ad13662007-07-17 04:06:13 -07002506 bitmap_update_sb(mddev->bitmap);
NeilBrowndafb20f2012-03-19 12:46:39 +11002507 rdev_for_each(rdev, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 char b[BDEVNAME_SIZE];
NeilBrown36a4e1f2011-10-07 14:23:17 +11002509
NeilBrown42543762006-06-26 00:27:57 -07002510 if (rdev->sb_loaded != 1)
2511 continue; /* no noise on spare devices */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002512
Andrei Warkentind70ed2e2011-10-18 12:16:48 +11002513 if (!test_bit(Faulty, &rdev->flags) &&
2514 rdev->saved_raid_disk == -1) {
NeilBrown7bfa19f2005-06-21 17:17:28 -07002515 md_super_write(mddev,rdev,
Andre Noll0f420352008-07-11 22:02:23 +10002516 rdev->sb_start, rdev->sb_size,
NeilBrown7bfa19f2005-06-21 17:17:28 -07002517 rdev->sb_page);
NeilBrown36a4e1f2011-10-07 14:23:17 +11002518 pr_debug("md: (write) %s's sb offset: %llu\n",
2519 bdevname(rdev->bdev, b),
2520 (unsigned long long)rdev->sb_start);
NeilBrown42543762006-06-26 00:27:57 -07002521 rdev->sb_events = mddev->events;
NeilBrown2699b672011-07-28 11:31:47 +10002522 if (rdev->badblocks.size) {
2523 md_super_write(mddev, rdev,
2524 rdev->badblocks.sector,
2525 rdev->badblocks.size << 9,
2526 rdev->bb_page);
2527 rdev->badblocks.size = 0;
2528 }
NeilBrown7bfa19f2005-06-21 17:17:28 -07002529
Andrei Warkentind70ed2e2011-10-18 12:16:48 +11002530 } else if (test_bit(Faulty, &rdev->flags))
NeilBrown36a4e1f2011-10-07 14:23:17 +11002531 pr_debug("md: %s (skipping faulty)\n",
2532 bdevname(rdev->bdev, b));
Andrei Warkentind70ed2e2011-10-18 12:16:48 +11002533 else
2534 pr_debug("(skipping incremental s/r ");
2535
NeilBrown7bfa19f2005-06-21 17:17:28 -07002536 if (mddev->level == LEVEL_MULTIPATH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 /* only need to write one superblock... */
2538 break;
2539 }
NeilBrowna9701a32005-11-08 21:39:34 -08002540 md_super_wait(mddev);
NeilBrown850b2b42006-10-03 01:15:46 -07002541 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
NeilBrown7bfa19f2005-06-21 17:17:28 -07002542
NeilBrowna9701a32005-11-08 21:39:34 -08002543 spin_lock_irq(&mddev->write_lock);
NeilBrown850b2b42006-10-03 01:15:46 -07002544 if (mddev->in_sync != sync_req ||
2545 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
NeilBrown06d91a52005-06-21 17:17:12 -07002546 /* have to write it out again */
NeilBrowna9701a32005-11-08 21:39:34 -08002547 spin_unlock_irq(&mddev->write_lock);
NeilBrown06d91a52005-06-21 17:17:12 -07002548 goto repeat;
2549 }
NeilBrown850b2b42006-10-03 01:15:46 -07002550 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrowna9701a32005-11-08 21:39:34 -08002551 spin_unlock_irq(&mddev->write_lock);
NeilBrown3d310eb2005-06-21 17:17:26 -07002552 wake_up(&mddev->sb_wait);
NeilBrownacb180b2009-04-14 16:28:34 +10002553 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2554 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown06d91a52005-06-21 17:17:12 -07002555
NeilBrowndafb20f2012-03-19 12:46:39 +11002556 rdev_for_each(rdev, mddev) {
NeilBrownde393cd2011-07-28 11:31:48 +10002557 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2558 clear_bit(Blocked, &rdev->flags);
2559
2560 if (any_badblocks_changed)
NeilBrown2699b672011-07-28 11:31:47 +10002561 md_ack_all_badblocks(&rdev->badblocks);
NeilBrownde393cd2011-07-28 11:31:48 +10002562 clear_bit(BlockedBadBlocks, &rdev->flags);
2563 wake_up(&rdev->blocked_wait);
2564 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565}
2566
Andre Noll7f6ce762008-03-23 18:34:54 +01002567/* words written to sysfs files may, or may not, be \n terminated.
NeilBrownbce74da2006-01-06 00:20:41 -08002568 * We want to accept with case. For this we use cmd_match.
2569 */
2570static int cmd_match(const char *cmd, const char *str)
2571{
2572 /* See if cmd, written into a sysfs file, matches
2573 * str. They must either be the same, or cmd can
2574 * have a trailing newline
2575 */
2576 while (*cmd && *str && *cmd == *str) {
2577 cmd++;
2578 str++;
2579 }
2580 if (*cmd == '\n')
2581 cmd++;
2582 if (*str || *cmd)
2583 return 0;
2584 return 1;
2585}
2586
NeilBrown86e6ffd2005-11-08 21:39:24 -08002587struct rdev_sysfs_entry {
2588 struct attribute attr;
NeilBrown3cb03002011-10-11 16:45:26 +11002589 ssize_t (*show)(struct md_rdev *, char *);
2590 ssize_t (*store)(struct md_rdev *, const char *, size_t);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002591};
2592
2593static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002594state_show(struct md_rdev *rdev, char *page)
NeilBrown86e6ffd2005-11-08 21:39:24 -08002595{
2596 char *sep = "";
NeilBrown20a49ff2008-02-06 01:39:57 -08002597 size_t len = 0;
NeilBrown86e6ffd2005-11-08 21:39:24 -08002598
NeilBrownde393cd2011-07-28 11:31:48 +10002599 if (test_bit(Faulty, &rdev->flags) ||
2600 rdev->badblocks.unacked_exist) {
NeilBrown86e6ffd2005-11-08 21:39:24 -08002601 len+= sprintf(page+len, "%sfaulty",sep);
2602 sep = ",";
2603 }
NeilBrownb2d444d2005-11-08 21:39:31 -08002604 if (test_bit(In_sync, &rdev->flags)) {
NeilBrown86e6ffd2005-11-08 21:39:24 -08002605 len += sprintf(page+len, "%sin_sync",sep);
2606 sep = ",";
2607 }
NeilBrownf6556752006-06-26 00:28:01 -07002608 if (test_bit(WriteMostly, &rdev->flags)) {
2609 len += sprintf(page+len, "%swrite_mostly",sep);
2610 sep = ",";
2611 }
NeilBrownde393cd2011-07-28 11:31:48 +10002612 if (test_bit(Blocked, &rdev->flags) ||
NeilBrown52c64152011-12-08 16:22:48 +11002613 (rdev->badblocks.unacked_exist
2614 && !test_bit(Faulty, &rdev->flags))) {
Dan Williams6bfe0b42008-04-30 00:52:32 -07002615 len += sprintf(page+len, "%sblocked", sep);
2616 sep = ",";
2617 }
NeilBrownb2d444d2005-11-08 21:39:31 -08002618 if (!test_bit(Faulty, &rdev->flags) &&
2619 !test_bit(In_sync, &rdev->flags)) {
NeilBrown86e6ffd2005-11-08 21:39:24 -08002620 len += sprintf(page+len, "%sspare", sep);
2621 sep = ",";
2622 }
NeilBrownd7a9d442011-07-28 11:31:48 +10002623 if (test_bit(WriteErrorSeen, &rdev->flags)) {
2624 len += sprintf(page+len, "%swrite_error", sep);
2625 sep = ",";
2626 }
NeilBrown2d78f8c2011-12-23 10:17:51 +11002627 if (test_bit(WantReplacement, &rdev->flags)) {
2628 len += sprintf(page+len, "%swant_replacement", sep);
2629 sep = ",";
2630 }
2631 if (test_bit(Replacement, &rdev->flags)) {
2632 len += sprintf(page+len, "%sreplacement", sep);
2633 sep = ",";
2634 }
2635
NeilBrown86e6ffd2005-11-08 21:39:24 -08002636 return len+sprintf(page+len, "\n");
2637}
2638
NeilBrown45dc2de2006-06-26 00:27:58 -07002639static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002640state_store(struct md_rdev *rdev, const char *buf, size_t len)
NeilBrown45dc2de2006-06-26 00:27:58 -07002641{
2642 /* can write
NeilBrownde393cd2011-07-28 11:31:48 +10002643 * faulty - simulates an error
NeilBrown45dc2de2006-06-26 00:27:58 -07002644 * remove - disconnects the device
NeilBrownf6556752006-06-26 00:28:01 -07002645 * writemostly - sets write_mostly
2646 * -writemostly - clears write_mostly
NeilBrownde393cd2011-07-28 11:31:48 +10002647 * blocked - sets the Blocked flags
2648 * -blocked - clears the Blocked and possibly simulates an error
NeilBrown6d56e272009-04-14 12:01:57 +10002649 * insync - sets Insync providing device isn't active
NeilBrownd7a9d442011-07-28 11:31:48 +10002650 * write_error - sets WriteErrorSeen
2651 * -write_error - clears WriteErrorSeen
NeilBrown45dc2de2006-06-26 00:27:58 -07002652 */
2653 int err = -EINVAL;
2654 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2655 md_error(rdev->mddev, rdev);
NeilBrown5ef56c82011-08-25 14:42:51 +10002656 if (test_bit(Faulty, &rdev->flags))
2657 err = 0;
2658 else
2659 err = -EBUSY;
NeilBrown45dc2de2006-06-26 00:27:58 -07002660 } else if (cmd_match(buf, "remove")) {
2661 if (rdev->raid_disk >= 0)
2662 err = -EBUSY;
2663 else {
NeilBrownfd01b882011-10-11 16:47:53 +11002664 struct mddev *mddev = rdev->mddev;
NeilBrown45dc2de2006-06-26 00:27:58 -07002665 kick_rdev_from_array(rdev);
NeilBrown3f9d7b02006-12-22 01:11:41 -08002666 if (mddev->pers)
2667 md_update_sb(mddev, 1);
NeilBrown45dc2de2006-06-26 00:27:58 -07002668 md_new_event(mddev);
2669 err = 0;
2670 }
NeilBrownf6556752006-06-26 00:28:01 -07002671 } else if (cmd_match(buf, "writemostly")) {
2672 set_bit(WriteMostly, &rdev->flags);
2673 err = 0;
2674 } else if (cmd_match(buf, "-writemostly")) {
2675 clear_bit(WriteMostly, &rdev->flags);
2676 err = 0;
Dan Williams6bfe0b42008-04-30 00:52:32 -07002677 } else if (cmd_match(buf, "blocked")) {
2678 set_bit(Blocked, &rdev->flags);
2679 err = 0;
2680 } else if (cmd_match(buf, "-blocked")) {
NeilBrownde393cd2011-07-28 11:31:48 +10002681 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown7da64a02011-08-30 16:20:17 +10002682 rdev->badblocks.unacked_exist) {
NeilBrownde393cd2011-07-28 11:31:48 +10002683 /* metadata handler doesn't understand badblocks,
2684 * so we need to fail the device
2685 */
2686 md_error(rdev->mddev, rdev);
2687 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07002688 clear_bit(Blocked, &rdev->flags);
NeilBrownde393cd2011-07-28 11:31:48 +10002689 clear_bit(BlockedBadBlocks, &rdev->flags);
Dan Williams6bfe0b42008-04-30 00:52:32 -07002690 wake_up(&rdev->blocked_wait);
2691 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2692 md_wakeup_thread(rdev->mddev->thread);
2693
2694 err = 0;
NeilBrown6d56e272009-04-14 12:01:57 +10002695 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2696 set_bit(In_sync, &rdev->flags);
2697 err = 0;
NeilBrownd7a9d442011-07-28 11:31:48 +10002698 } else if (cmd_match(buf, "write_error")) {
2699 set_bit(WriteErrorSeen, &rdev->flags);
2700 err = 0;
2701 } else if (cmd_match(buf, "-write_error")) {
2702 clear_bit(WriteErrorSeen, &rdev->flags);
2703 err = 0;
NeilBrown2d78f8c2011-12-23 10:17:51 +11002704 } else if (cmd_match(buf, "want_replacement")) {
2705 /* Any non-spare device that is not a replacement can
2706 * become want_replacement at any time, but we then need to
2707 * check if recovery is needed.
2708 */
2709 if (rdev->raid_disk >= 0 &&
2710 !test_bit(Replacement, &rdev->flags))
2711 set_bit(WantReplacement, &rdev->flags);
2712 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2713 md_wakeup_thread(rdev->mddev->thread);
2714 err = 0;
2715 } else if (cmd_match(buf, "-want_replacement")) {
2716 /* Clearing 'want_replacement' is always allowed.
2717 * Once replacements starts it is too late though.
2718 */
2719 err = 0;
2720 clear_bit(WantReplacement, &rdev->flags);
2721 } else if (cmd_match(buf, "replacement")) {
2722 /* Can only set a device as a replacement when array has not
2723 * yet been started. Once running, replacement is automatic
2724 * from spares, or by assigning 'slot'.
2725 */
2726 if (rdev->mddev->pers)
2727 err = -EBUSY;
2728 else {
2729 set_bit(Replacement, &rdev->flags);
2730 err = 0;
2731 }
2732 } else if (cmd_match(buf, "-replacement")) {
2733 /* Similarly, can only clear Replacement before start */
2734 if (rdev->mddev->pers)
2735 err = -EBUSY;
2736 else {
2737 clear_bit(Replacement, &rdev->flags);
2738 err = 0;
2739 }
NeilBrown45dc2de2006-06-26 00:27:58 -07002740 }
NeilBrown00bcb4a2010-06-01 19:37:23 +10002741 if (!err)
2742 sysfs_notify_dirent_safe(rdev->sysfs_state);
NeilBrown45dc2de2006-06-26 00:27:58 -07002743 return err ? err : len;
2744}
NeilBrown80ca3a42006-07-10 04:44:18 -07002745static struct rdev_sysfs_entry rdev_state =
2746__ATTR(state, S_IRUGO|S_IWUSR, state_show, state_store);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002747
2748static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002749errors_show(struct md_rdev *rdev, char *page)
NeilBrown4dbcdc72006-01-06 00:20:52 -08002750{
2751 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2752}
2753
2754static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002755errors_store(struct md_rdev *rdev, const char *buf, size_t len)
NeilBrown4dbcdc72006-01-06 00:20:52 -08002756{
2757 char *e;
2758 unsigned long n = simple_strtoul(buf, &e, 10);
2759 if (*buf && (*e == 0 || *e == '\n')) {
2760 atomic_set(&rdev->corrected_errors, n);
2761 return len;
2762 }
2763 return -EINVAL;
2764}
2765static struct rdev_sysfs_entry rdev_errors =
NeilBrown80ca3a42006-07-10 04:44:18 -07002766__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
NeilBrown4dbcdc72006-01-06 00:20:52 -08002767
NeilBrown014236d2006-01-06 00:20:55 -08002768static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002769slot_show(struct md_rdev *rdev, char *page)
NeilBrown014236d2006-01-06 00:20:55 -08002770{
2771 if (rdev->raid_disk < 0)
2772 return sprintf(page, "none\n");
2773 else
2774 return sprintf(page, "%d\n", rdev->raid_disk);
2775}
2776
2777static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002778slot_store(struct md_rdev *rdev, const char *buf, size_t len)
NeilBrown014236d2006-01-06 00:20:55 -08002779{
2780 char *e;
NeilBrownc303da62008-02-06 01:39:51 -08002781 int err;
NeilBrown014236d2006-01-06 00:20:55 -08002782 int slot = simple_strtoul(buf, &e, 10);
2783 if (strncmp(buf, "none", 4)==0)
2784 slot = -1;
2785 else if (e==buf || (*e && *e!= '\n'))
2786 return -EINVAL;
Neil Brown6c2fce22008-06-28 08:31:31 +10002787 if (rdev->mddev->pers && slot == -1) {
NeilBrownc303da62008-02-06 01:39:51 -08002788 /* Setting 'slot' on an active array requires also
2789 * updating the 'rd%d' link, and communicating
2790 * with the personality with ->hot_*_disk.
2791 * For now we only support removing
2792 * failed/spare devices. This normally happens automatically,
2793 * but not when the metadata is externally managed.
2794 */
NeilBrownc303da62008-02-06 01:39:51 -08002795 if (rdev->raid_disk == -1)
2796 return -EEXIST;
2797 /* personality does all needed checks */
Namhyung Kim01393f32011-06-09 11:42:54 +10002798 if (rdev->mddev->pers->hot_remove_disk == NULL)
NeilBrownc303da62008-02-06 01:39:51 -08002799 return -EINVAL;
2800 err = rdev->mddev->pers->
NeilBrownb8321b62011-12-23 10:17:51 +11002801 hot_remove_disk(rdev->mddev, rdev);
NeilBrownc303da62008-02-06 01:39:51 -08002802 if (err)
2803 return err;
Namhyung Kim36fad852011-07-27 11:00:36 +10002804 sysfs_unlink_rdev(rdev->mddev, rdev);
Maciej Trelab7103102010-04-14 16:58:16 +10002805 rdev->raid_disk = -1;
NeilBrownc303da62008-02-06 01:39:51 -08002806 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2807 md_wakeup_thread(rdev->mddev->thread);
Neil Brown6c2fce22008-06-28 08:31:31 +10002808 } else if (rdev->mddev->pers) {
Neil Brown6c2fce22008-06-28 08:31:31 +10002809 /* Activating a spare .. or possibly reactivating
NeilBrown6d56e272009-04-14 12:01:57 +10002810 * if we ever get bitmaps working here.
Neil Brown6c2fce22008-06-28 08:31:31 +10002811 */
2812
2813 if (rdev->raid_disk != -1)
2814 return -EBUSY;
2815
NeilBrownc6751b22011-02-02 11:57:13 +11002816 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2817 return -EBUSY;
2818
Neil Brown6c2fce22008-06-28 08:31:31 +10002819 if (rdev->mddev->pers->hot_add_disk == NULL)
2820 return -EINVAL;
2821
NeilBrownba1b41b2011-01-14 09:14:34 +11002822 if (slot >= rdev->mddev->raid_disks &&
2823 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2824 return -ENOSPC;
2825
Neil Brown6c2fce22008-06-28 08:31:31 +10002826 rdev->raid_disk = slot;
2827 if (test_bit(In_sync, &rdev->flags))
2828 rdev->saved_raid_disk = slot;
2829 else
2830 rdev->saved_raid_disk = -1;
NeilBrownd30519f2011-10-18 12:13:47 +11002831 clear_bit(In_sync, &rdev->flags);
Neil Brown6c2fce22008-06-28 08:31:31 +10002832 err = rdev->mddev->pers->
2833 hot_add_disk(rdev->mddev, rdev);
Neil Brown199050e2008-06-28 08:31:33 +10002834 if (err) {
Neil Brown6c2fce22008-06-28 08:31:31 +10002835 rdev->raid_disk = -1;
Neil Brown6c2fce22008-06-28 08:31:31 +10002836 return err;
Neil Brown52664732008-06-28 08:31:44 +10002837 } else
NeilBrown00bcb4a2010-06-01 19:37:23 +10002838 sysfs_notify_dirent_safe(rdev->sysfs_state);
Namhyung Kim36fad852011-07-27 11:00:36 +10002839 if (sysfs_link_rdev(rdev->mddev, rdev))
NeilBrown00bcb4a2010-06-01 19:37:23 +10002840 /* failure here is OK */;
Neil Brown6c2fce22008-06-28 08:31:31 +10002841 /* don't wakeup anyone, leave that to userspace. */
NeilBrownc303da62008-02-06 01:39:51 -08002842 } else {
NeilBrownba1b41b2011-01-14 09:14:34 +11002843 if (slot >= rdev->mddev->raid_disks &&
2844 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
NeilBrownc303da62008-02-06 01:39:51 -08002845 return -ENOSPC;
2846 rdev->raid_disk = slot;
2847 /* assume it is working */
NeilBrownc5d79ad2008-02-06 01:39:54 -08002848 clear_bit(Faulty, &rdev->flags);
2849 clear_bit(WriteMostly, &rdev->flags);
NeilBrownc303da62008-02-06 01:39:51 -08002850 set_bit(In_sync, &rdev->flags);
NeilBrown00bcb4a2010-06-01 19:37:23 +10002851 sysfs_notify_dirent_safe(rdev->sysfs_state);
NeilBrownc303da62008-02-06 01:39:51 -08002852 }
NeilBrown014236d2006-01-06 00:20:55 -08002853 return len;
2854}
2855
2856
2857static struct rdev_sysfs_entry rdev_slot =
NeilBrown80ca3a42006-07-10 04:44:18 -07002858__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
NeilBrown014236d2006-01-06 00:20:55 -08002859
NeilBrown93c8cad2006-01-06 00:20:56 -08002860static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002861offset_show(struct md_rdev *rdev, char *page)
NeilBrown93c8cad2006-01-06 00:20:56 -08002862{
Andrew Morton6961ece2006-01-06 00:20:59 -08002863 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
NeilBrown93c8cad2006-01-06 00:20:56 -08002864}
2865
2866static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002867offset_store(struct md_rdev *rdev, const char *buf, size_t len)
NeilBrown93c8cad2006-01-06 00:20:56 -08002868{
NeilBrownc6563a82012-05-21 09:27:00 +10002869 unsigned long long offset;
2870 if (strict_strtoull(buf, 10, &offset) < 0)
NeilBrown93c8cad2006-01-06 00:20:56 -08002871 return -EINVAL;
Neil Brown8ed0a522008-06-28 08:31:29 +10002872 if (rdev->mddev->pers && rdev->raid_disk >= 0)
NeilBrown93c8cad2006-01-06 00:20:56 -08002873 return -EBUSY;
Andre Nolldd8ac332009-03-31 14:33:13 +11002874 if (rdev->sectors && rdev->mddev->external)
NeilBrownc5d79ad2008-02-06 01:39:54 -08002875 /* Must set offset before size, so overlap checks
2876 * can be sane */
2877 return -EBUSY;
NeilBrown93c8cad2006-01-06 00:20:56 -08002878 rdev->data_offset = offset;
NeilBrown25f7fd42012-07-19 15:59:18 +10002879 rdev->new_data_offset = offset;
NeilBrown93c8cad2006-01-06 00:20:56 -08002880 return len;
2881}
2882
2883static struct rdev_sysfs_entry rdev_offset =
NeilBrown80ca3a42006-07-10 04:44:18 -07002884__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
NeilBrown93c8cad2006-01-06 00:20:56 -08002885
NeilBrownc6563a82012-05-21 09:27:00 +10002886static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2887{
2888 return sprintf(page, "%llu\n",
2889 (unsigned long long)rdev->new_data_offset);
2890}
2891
2892static ssize_t new_offset_store(struct md_rdev *rdev,
2893 const char *buf, size_t len)
2894{
2895 unsigned long long new_offset;
2896 struct mddev *mddev = rdev->mddev;
2897
2898 if (strict_strtoull(buf, 10, &new_offset) < 0)
2899 return -EINVAL;
2900
2901 if (mddev->sync_thread)
2902 return -EBUSY;
2903 if (new_offset == rdev->data_offset)
2904 /* reset is always permitted */
2905 ;
2906 else if (new_offset > rdev->data_offset) {
2907 /* must not push array size beyond rdev_sectors */
2908 if (new_offset - rdev->data_offset
2909 + mddev->dev_sectors > rdev->sectors)
2910 return -E2BIG;
2911 }
2912 /* Metadata worries about other space details. */
2913
2914 /* decreasing the offset is inconsistent with a backwards
2915 * reshape.
2916 */
2917 if (new_offset < rdev->data_offset &&
2918 mddev->reshape_backwards)
2919 return -EINVAL;
2920 /* Increasing offset is inconsistent with forwards
2921 * reshape. reshape_direction should be set to
2922 * 'backwards' first.
2923 */
2924 if (new_offset > rdev->data_offset &&
2925 !mddev->reshape_backwards)
2926 return -EINVAL;
2927
2928 if (mddev->pers && mddev->persistent &&
2929 !super_types[mddev->major_version]
2930 .allow_new_offset(rdev, new_offset))
2931 return -E2BIG;
2932 rdev->new_data_offset = new_offset;
2933 if (new_offset > rdev->data_offset)
2934 mddev->reshape_backwards = 1;
2935 else if (new_offset < rdev->data_offset)
2936 mddev->reshape_backwards = 0;
2937
2938 return len;
2939}
2940static struct rdev_sysfs_entry rdev_new_offset =
2941__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2942
NeilBrown83303b62006-01-06 00:21:06 -08002943static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002944rdev_size_show(struct md_rdev *rdev, char *page)
NeilBrown83303b62006-01-06 00:21:06 -08002945{
Andre Nolldd8ac332009-03-31 14:33:13 +11002946 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
NeilBrown83303b62006-01-06 00:21:06 -08002947}
2948
NeilBrownc5d79ad2008-02-06 01:39:54 -08002949static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2950{
2951 /* check if two start/length pairs overlap */
2952 if (s1+l1 <= s2)
2953 return 0;
2954 if (s2+l2 <= s1)
2955 return 0;
2956 return 1;
2957}
2958
Dan Williamsb522adc2009-03-31 15:00:31 +11002959static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2960{
2961 unsigned long long blocks;
2962 sector_t new;
2963
2964 if (strict_strtoull(buf, 10, &blocks) < 0)
2965 return -EINVAL;
2966
2967 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2968 return -EINVAL; /* sector conversion overflow */
2969
2970 new = blocks * 2;
2971 if (new != blocks * 2)
2972 return -EINVAL; /* unsigned long long to sector_t overflow */
2973
2974 *sectors = new;
2975 return 0;
2976}
2977
NeilBrown83303b62006-01-06 00:21:06 -08002978static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002979rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
NeilBrown83303b62006-01-06 00:21:06 -08002980{
NeilBrownfd01b882011-10-11 16:47:53 +11002981 struct mddev *my_mddev = rdev->mddev;
Andre Nolldd8ac332009-03-31 14:33:13 +11002982 sector_t oldsectors = rdev->sectors;
Dan Williamsb522adc2009-03-31 15:00:31 +11002983 sector_t sectors;
NeilBrown27c529b2008-03-04 14:29:33 -08002984
Dan Williamsb522adc2009-03-31 15:00:31 +11002985 if (strict_blocks_to_sectors(buf, &sectors) < 0)
Neil Brownd7027452008-07-12 10:37:50 +10002986 return -EINVAL;
NeilBrownc6563a82012-05-21 09:27:00 +10002987 if (rdev->data_offset != rdev->new_data_offset)
2988 return -EINVAL; /* too confusing */
Chris Webb0cd17fe2008-06-28 08:31:46 +10002989 if (my_mddev->pers && rdev->raid_disk >= 0) {
Neil Brownd7027452008-07-12 10:37:50 +10002990 if (my_mddev->persistent) {
Andre Nolldd8ac332009-03-31 14:33:13 +11002991 sectors = super_types[my_mddev->major_version].
2992 rdev_size_change(rdev, sectors);
2993 if (!sectors)
Chris Webb0cd17fe2008-06-28 08:31:46 +10002994 return -EBUSY;
Andre Nolldd8ac332009-03-31 14:33:13 +11002995 } else if (!sectors)
Mike Snitzer77304d22010-11-08 14:39:12 +01002996 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
Andre Nolldd8ac332009-03-31 14:33:13 +11002997 rdev->data_offset;
Chris Webb0cd17fe2008-06-28 08:31:46 +10002998 }
Andre Nolldd8ac332009-03-31 14:33:13 +11002999 if (sectors < my_mddev->dev_sectors)
Chris Webb7d3c6f82008-10-13 11:55:11 +11003000 return -EINVAL; /* component must fit device */
Chris Webb0cd17fe2008-06-28 08:31:46 +10003001
Andre Nolldd8ac332009-03-31 14:33:13 +11003002 rdev->sectors = sectors;
3003 if (sectors > oldsectors && my_mddev->external) {
NeilBrownc5d79ad2008-02-06 01:39:54 -08003004 /* need to check that all other rdevs with the same ->bdev
3005 * do not overlap. We need to unlock the mddev to avoid
Andre Nolldd8ac332009-03-31 14:33:13 +11003006 * a deadlock. We have already changed rdev->sectors, and if
NeilBrownc5d79ad2008-02-06 01:39:54 -08003007 * we have to change it back, we will have the lock again.
3008 */
NeilBrownfd01b882011-10-11 16:47:53 +11003009 struct mddev *mddev;
NeilBrownc5d79ad2008-02-06 01:39:54 -08003010 int overlap = 0;
Cheng Renquan159ec1f2009-01-09 08:31:08 +11003011 struct list_head *tmp;
NeilBrownc5d79ad2008-02-06 01:39:54 -08003012
NeilBrown27c529b2008-03-04 14:29:33 -08003013 mddev_unlock(my_mddev);
NeilBrown29ac4aa2008-02-06 01:39:58 -08003014 for_each_mddev(mddev, tmp) {
NeilBrown3cb03002011-10-11 16:45:26 +11003015 struct md_rdev *rdev2;
NeilBrownc5d79ad2008-02-06 01:39:54 -08003016
3017 mddev_lock(mddev);
NeilBrowndafb20f2012-03-19 12:46:39 +11003018 rdev_for_each(rdev2, mddev)
NeilBrownf21e9ff2011-01-31 12:10:09 +11003019 if (rdev->bdev == rdev2->bdev &&
3020 rdev != rdev2 &&
3021 overlaps(rdev->data_offset, rdev->sectors,
3022 rdev2->data_offset,
3023 rdev2->sectors)) {
NeilBrownc5d79ad2008-02-06 01:39:54 -08003024 overlap = 1;
3025 break;
3026 }
3027 mddev_unlock(mddev);
3028 if (overlap) {
3029 mddev_put(mddev);
3030 break;
3031 }
3032 }
NeilBrown27c529b2008-03-04 14:29:33 -08003033 mddev_lock(my_mddev);
NeilBrownc5d79ad2008-02-06 01:39:54 -08003034 if (overlap) {
3035 /* Someone else could have slipped in a size
3036 * change here, but doing so is just silly.
Andre Nolldd8ac332009-03-31 14:33:13 +11003037 * We put oldsectors back because we *know* it is
NeilBrownc5d79ad2008-02-06 01:39:54 -08003038 * safe, and trust userspace not to race with
3039 * itself
3040 */
Andre Nolldd8ac332009-03-31 14:33:13 +11003041 rdev->sectors = oldsectors;
NeilBrownc5d79ad2008-02-06 01:39:54 -08003042 return -EBUSY;
3043 }
3044 }
NeilBrown83303b62006-01-06 00:21:06 -08003045 return len;
3046}
3047
3048static struct rdev_sysfs_entry rdev_size =
NeilBrown80ca3a42006-07-10 04:44:18 -07003049__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
NeilBrown83303b62006-01-06 00:21:06 -08003050
Dan Williams06e3c812009-12-12 21:17:12 -07003051
NeilBrown3cb03002011-10-11 16:45:26 +11003052static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
Dan Williams06e3c812009-12-12 21:17:12 -07003053{
3054 unsigned long long recovery_start = rdev->recovery_offset;
3055
3056 if (test_bit(In_sync, &rdev->flags) ||
3057 recovery_start == MaxSector)
3058 return sprintf(page, "none\n");
3059
3060 return sprintf(page, "%llu\n", recovery_start);
3061}
3062
NeilBrown3cb03002011-10-11 16:45:26 +11003063static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
Dan Williams06e3c812009-12-12 21:17:12 -07003064{
3065 unsigned long long recovery_start;
3066
3067 if (cmd_match(buf, "none"))
3068 recovery_start = MaxSector;
3069 else if (strict_strtoull(buf, 10, &recovery_start))
3070 return -EINVAL;
3071
3072 if (rdev->mddev->pers &&
3073 rdev->raid_disk >= 0)
3074 return -EBUSY;
3075
3076 rdev->recovery_offset = recovery_start;
3077 if (recovery_start == MaxSector)
3078 set_bit(In_sync, &rdev->flags);
3079 else
3080 clear_bit(In_sync, &rdev->flags);
3081 return len;
3082}
3083
3084static struct rdev_sysfs_entry rdev_recovery_start =
3085__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
3086
NeilBrown16c791a2011-07-28 11:31:47 +10003087
3088static ssize_t
3089badblocks_show(struct badblocks *bb, char *page, int unack);
3090static ssize_t
3091badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
3092
NeilBrown3cb03002011-10-11 16:45:26 +11003093static ssize_t bb_show(struct md_rdev *rdev, char *page)
NeilBrown16c791a2011-07-28 11:31:47 +10003094{
3095 return badblocks_show(&rdev->badblocks, page, 0);
3096}
NeilBrown3cb03002011-10-11 16:45:26 +11003097static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
NeilBrown16c791a2011-07-28 11:31:47 +10003098{
NeilBrownde393cd2011-07-28 11:31:48 +10003099 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
3100 /* Maybe that ack was all we needed */
3101 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
3102 wake_up(&rdev->blocked_wait);
3103 return rv;
NeilBrown16c791a2011-07-28 11:31:47 +10003104}
3105static struct rdev_sysfs_entry rdev_bad_blocks =
3106__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
3107
3108
NeilBrown3cb03002011-10-11 16:45:26 +11003109static ssize_t ubb_show(struct md_rdev *rdev, char *page)
NeilBrown16c791a2011-07-28 11:31:47 +10003110{
3111 return badblocks_show(&rdev->badblocks, page, 1);
3112}
NeilBrown3cb03002011-10-11 16:45:26 +11003113static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
NeilBrown16c791a2011-07-28 11:31:47 +10003114{
3115 return badblocks_store(&rdev->badblocks, page, len, 1);
3116}
3117static struct rdev_sysfs_entry rdev_unack_bad_blocks =
3118__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
3119
NeilBrown86e6ffd2005-11-08 21:39:24 -08003120static struct attribute *rdev_default_attrs[] = {
3121 &rdev_state.attr,
NeilBrown4dbcdc72006-01-06 00:20:52 -08003122 &rdev_errors.attr,
NeilBrown014236d2006-01-06 00:20:55 -08003123 &rdev_slot.attr,
NeilBrown93c8cad2006-01-06 00:20:56 -08003124 &rdev_offset.attr,
NeilBrownc6563a82012-05-21 09:27:00 +10003125 &rdev_new_offset.attr,
NeilBrown83303b62006-01-06 00:21:06 -08003126 &rdev_size.attr,
Dan Williams06e3c812009-12-12 21:17:12 -07003127 &rdev_recovery_start.attr,
NeilBrown16c791a2011-07-28 11:31:47 +10003128 &rdev_bad_blocks.attr,
3129 &rdev_unack_bad_blocks.attr,
NeilBrown86e6ffd2005-11-08 21:39:24 -08003130 NULL,
3131};
3132static ssize_t
3133rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
3134{
3135 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
NeilBrown3cb03002011-10-11 16:45:26 +11003136 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
NeilBrownfd01b882011-10-11 16:47:53 +11003137 struct mddev *mddev = rdev->mddev;
NeilBrown27c529b2008-03-04 14:29:33 -08003138 ssize_t rv;
NeilBrown86e6ffd2005-11-08 21:39:24 -08003139
3140 if (!entry->show)
3141 return -EIO;
NeilBrown27c529b2008-03-04 14:29:33 -08003142
3143 rv = mddev ? mddev_lock(mddev) : -EBUSY;
3144 if (!rv) {
3145 if (rdev->mddev == NULL)
3146 rv = -EBUSY;
3147 else
3148 rv = entry->show(rdev, page);
3149 mddev_unlock(mddev);
3150 }
3151 return rv;
NeilBrown86e6ffd2005-11-08 21:39:24 -08003152}
3153
3154static ssize_t
3155rdev_attr_store(struct kobject *kobj, struct attribute *attr,
3156 const char *page, size_t length)
3157{
3158 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
NeilBrown3cb03002011-10-11 16:45:26 +11003159 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
NeilBrown27c529b2008-03-04 14:29:33 -08003160 ssize_t rv;
NeilBrownfd01b882011-10-11 16:47:53 +11003161 struct mddev *mddev = rdev->mddev;
NeilBrown86e6ffd2005-11-08 21:39:24 -08003162
3163 if (!entry->store)
3164 return -EIO;
NeilBrown67463ac2006-07-10 04:44:19 -07003165 if (!capable(CAP_SYS_ADMIN))
3166 return -EACCES;
NeilBrown27c529b2008-03-04 14:29:33 -08003167 rv = mddev ? mddev_lock(mddev): -EBUSY;
NeilBrownca388052008-02-06 01:39:55 -08003168 if (!rv) {
NeilBrown27c529b2008-03-04 14:29:33 -08003169 if (rdev->mddev == NULL)
3170 rv = -EBUSY;
3171 else
3172 rv = entry->store(rdev, page, length);
Dan Williams6a518302008-04-30 00:52:28 -07003173 mddev_unlock(mddev);
NeilBrownca388052008-02-06 01:39:55 -08003174 }
3175 return rv;
NeilBrown86e6ffd2005-11-08 21:39:24 -08003176}
3177
3178static void rdev_free(struct kobject *ko)
3179{
NeilBrown3cb03002011-10-11 16:45:26 +11003180 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
NeilBrown86e6ffd2005-11-08 21:39:24 -08003181 kfree(rdev);
3182}
Emese Revfy52cf25d2010-01-19 02:58:23 +01003183static const struct sysfs_ops rdev_sysfs_ops = {
NeilBrown86e6ffd2005-11-08 21:39:24 -08003184 .show = rdev_attr_show,
3185 .store = rdev_attr_store,
3186};
3187static struct kobj_type rdev_ktype = {
3188 .release = rdev_free,
3189 .sysfs_ops = &rdev_sysfs_ops,
3190 .default_attrs = rdev_default_attrs,
3191};
3192
NeilBrown3cb03002011-10-11 16:45:26 +11003193int md_rdev_init(struct md_rdev *rdev)
NeilBrowne8bb9a82010-06-01 19:37:26 +10003194{
3195 rdev->desc_nr = -1;
3196 rdev->saved_raid_disk = -1;
3197 rdev->raid_disk = -1;
3198 rdev->flags = 0;
3199 rdev->data_offset = 0;
NeilBrownc6563a82012-05-21 09:27:00 +10003200 rdev->new_data_offset = 0;
NeilBrowne8bb9a82010-06-01 19:37:26 +10003201 rdev->sb_events = 0;
3202 rdev->last_read_error.tv_sec = 0;
3203 rdev->last_read_error.tv_nsec = 0;
NeilBrown2699b672011-07-28 11:31:47 +10003204 rdev->sb_loaded = 0;
3205 rdev->bb_page = NULL;
NeilBrowne8bb9a82010-06-01 19:37:26 +10003206 atomic_set(&rdev->nr_pending, 0);
3207 atomic_set(&rdev->read_errors, 0);
3208 atomic_set(&rdev->corrected_errors, 0);
3209
3210 INIT_LIST_HEAD(&rdev->same_set);
3211 init_waitqueue_head(&rdev->blocked_wait);
NeilBrown2230dfe2011-07-28 11:31:46 +10003212
3213 /* Add space to store bad block list.
3214 * This reserves the space even on arrays where it cannot
3215 * be used - I wonder if that matters
3216 */
3217 rdev->badblocks.count = 0;
3218 rdev->badblocks.shift = 0;
3219 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3220 seqlock_init(&rdev->badblocks.lock);
3221 if (rdev->badblocks.page == NULL)
3222 return -ENOMEM;
3223
3224 return 0;
NeilBrowne8bb9a82010-06-01 19:37:26 +10003225}
3226EXPORT_SYMBOL_GPL(md_rdev_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003227/*
3228 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3229 *
3230 * mark the device faulty if:
3231 *
3232 * - the device is nonexistent (zero size)
3233 * - the device has no valid superblock
3234 *
3235 * a faulty rdev _never_ has rdev->sb set.
3236 */
NeilBrown3cb03002011-10-11 16:45:26 +11003237static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238{
3239 char b[BDEVNAME_SIZE];
3240 int err;
NeilBrown3cb03002011-10-11 16:45:26 +11003241 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242 sector_t size;
3243
NeilBrown9ffae0c2006-01-06 00:20:32 -08003244 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003245 if (!rdev) {
3246 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3247 return ERR_PTR(-ENOMEM);
3248 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249
NeilBrown2230dfe2011-07-28 11:31:46 +10003250 err = md_rdev_init(rdev);
3251 if (err)
3252 goto abort_free;
3253 err = alloc_disk_sb(rdev);
3254 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255 goto abort_free;
3256
NeilBrownc5d79ad2008-02-06 01:39:54 -08003257 err = lock_rdev(rdev, newdev, super_format == -2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258 if (err)
3259 goto abort_free;
3260
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -07003261 kobject_init(&rdev->kobj, &rdev_ktype);
NeilBrown86e6ffd2005-11-08 21:39:24 -08003262
Mike Snitzer77304d22010-11-08 14:39:12 +01003263 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003264 if (!size) {
3265 printk(KERN_WARNING
3266 "md: %s has zero or unknown size, marking faulty!\n",
3267 bdevname(rdev->bdev,b));
3268 err = -EINVAL;
3269 goto abort_free;
3270 }
3271
3272 if (super_format >= 0) {
3273 err = super_types[super_format].
3274 load_super(rdev, NULL, super_minor);
3275 if (err == -EINVAL) {
NeilBrowndf968c42007-07-17 04:06:11 -07003276 printk(KERN_WARNING
3277 "md: %s does not have a valid v%d.%d "
3278 "superblock, not importing!\n",
3279 bdevname(rdev->bdev,b),
3280 super_format, super_minor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281 goto abort_free;
3282 }
3283 if (err < 0) {
3284 printk(KERN_WARNING
3285 "md: could not read %s's sb, not importing!\n",
3286 bdevname(rdev->bdev,b));
3287 goto abort_free;
3288 }
3289 }
NeilBrown9f2f3832011-07-28 11:31:47 +10003290 if (super_format == -1)
3291 /* hot-add for 0.90, or non-persistent: so no badblocks */
3292 rdev->badblocks.shift = -1;
Dan Williams6bfe0b42008-04-30 00:52:32 -07003293
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 return rdev;
3295
3296abort_free:
NeilBrown2699b672011-07-28 11:31:47 +10003297 if (rdev->bdev)
3298 unlock_rdev(rdev);
NeilBrown545c8792012-05-22 13:54:30 +10003299 md_rdev_clear(rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300 kfree(rdev);
3301 return ERR_PTR(err);
3302}
3303
3304/*
3305 * Check a full RAID array for plausibility
3306 */
3307
3308
NeilBrownfd01b882011-10-11 16:47:53 +11003309static void analyze_sbs(struct mddev * mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310{
3311 int i;
NeilBrown3cb03002011-10-11 16:45:26 +11003312 struct md_rdev *rdev, *freshest, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003313 char b[BDEVNAME_SIZE];
3314
3315 freshest = NULL;
NeilBrowndafb20f2012-03-19 12:46:39 +11003316 rdev_for_each_safe(rdev, tmp, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003317 switch (super_types[mddev->major_version].
3318 load_super(rdev, freshest, mddev->minor_version)) {
3319 case 1:
3320 freshest = rdev;
3321 break;
3322 case 0:
3323 break;
3324 default:
3325 printk( KERN_ERR \
3326 "md: fatal superblock inconsistency in %s"
3327 " -- removing from array\n",
3328 bdevname(rdev->bdev,b));
3329 kick_rdev_from_array(rdev);
3330 }
3331
3332
3333 super_types[mddev->major_version].
3334 validate_super(mddev, freshest);
3335
3336 i = 0;
NeilBrowndafb20f2012-03-19 12:46:39 +11003337 rdev_for_each_safe(rdev, tmp, mddev) {
NeilBrown233fca32010-04-14 17:02:09 +10003338 if (mddev->max_disks &&
3339 (rdev->desc_nr >= mddev->max_disks ||
3340 i > mddev->max_disks)) {
NeilBrownde01dfa2009-02-06 18:02:46 +11003341 printk(KERN_WARNING
3342 "md: %s: %s: only %d devices permitted\n",
3343 mdname(mddev), bdevname(rdev->bdev, b),
3344 mddev->max_disks);
3345 kick_rdev_from_array(rdev);
3346 continue;
3347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003348 if (rdev != freshest)
3349 if (super_types[mddev->major_version].
3350 validate_super(mddev, rdev)) {
3351 printk(KERN_WARNING "md: kicking non-fresh %s"
3352 " from array!\n",
3353 bdevname(rdev->bdev,b));
3354 kick_rdev_from_array(rdev);
3355 continue;
3356 }
3357 if (mddev->level == LEVEL_MULTIPATH) {
3358 rdev->desc_nr = i++;
3359 rdev->raid_disk = rdev->desc_nr;
NeilBrownb2d444d2005-11-08 21:39:31 -08003360 set_bit(In_sync, &rdev->flags);
NeilBrown5e5e3e72009-10-16 16:35:30 +11003361 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
NeilBrowna778b732007-05-23 13:58:10 -07003362 rdev->raid_disk = -1;
3363 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 }
3365 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003366}
3367
NeilBrown72e02072009-12-14 12:49:55 +11003368/* Read a fixed-point number.
3369 * Numbers in sysfs attributes should be in "standard" units where
3370 * possible, so time should be in seconds.
3371 * However we internally use a a much smaller unit such as
3372 * milliseconds or jiffies.
3373 * This function takes a decimal number with a possible fractional
3374 * component, and produces an integer which is the result of
3375 * multiplying that number by 10^'scale'.
3376 * all without any floating-point arithmetic.
3377 */
3378int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3379{
3380 unsigned long result = 0;
3381 long decimals = -1;
3382 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3383 if (*cp == '.')
3384 decimals = 0;
3385 else if (decimals < scale) {
3386 unsigned int value;
3387 value = *cp - '0';
3388 result = result * 10 + value;
3389 if (decimals >= 0)
3390 decimals++;
3391 }
3392 cp++;
3393 }
3394 if (*cp == '\n')
3395 cp++;
3396 if (*cp)
3397 return -EINVAL;
3398 if (decimals < 0)
3399 decimals = 0;
3400 while (decimals < scale) {
3401 result *= 10;
3402 decimals ++;
3403 }
3404 *res = result;
3405 return 0;
3406}
3407
3408
NeilBrown19052c02008-08-05 15:54:13 +10003409static void md_safemode_timeout(unsigned long data);
3410
NeilBrowneae17012005-11-08 21:39:23 -08003411static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003412safe_delay_show(struct mddev *mddev, char *page)
NeilBrown16f17b32006-06-26 00:27:37 -07003413{
3414 int msec = (mddev->safemode_delay*1000)/HZ;
3415 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3416}
3417static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003418safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
NeilBrown16f17b32006-06-26 00:27:37 -07003419{
NeilBrown16f17b32006-06-26 00:27:37 -07003420 unsigned long msec;
Dan Williams97ce0a72008-09-24 22:48:19 -07003421
NeilBrown72e02072009-12-14 12:49:55 +11003422 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
NeilBrown16f17b32006-06-26 00:27:37 -07003423 return -EINVAL;
NeilBrown16f17b32006-06-26 00:27:37 -07003424 if (msec == 0)
3425 mddev->safemode_delay = 0;
3426 else {
NeilBrown19052c02008-08-05 15:54:13 +10003427 unsigned long old_delay = mddev->safemode_delay;
NeilBrown16f17b32006-06-26 00:27:37 -07003428 mddev->safemode_delay = (msec*HZ)/1000;
3429 if (mddev->safemode_delay == 0)
3430 mddev->safemode_delay = 1;
NeilBrown19052c02008-08-05 15:54:13 +10003431 if (mddev->safemode_delay < old_delay)
3432 md_safemode_timeout((unsigned long)mddev);
NeilBrown16f17b32006-06-26 00:27:37 -07003433 }
3434 return len;
3435}
3436static struct md_sysfs_entry md_safe_delay =
NeilBrown80ca3a42006-07-10 04:44:18 -07003437__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
NeilBrown16f17b32006-06-26 00:27:37 -07003438
3439static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003440level_show(struct mddev *mddev, char *page)
NeilBrowneae17012005-11-08 21:39:23 -08003441{
NeilBrown84fc4b52011-10-11 16:49:58 +11003442 struct md_personality *p = mddev->pers;
NeilBrownd9d166c2006-01-06 00:20:51 -08003443 if (p)
NeilBrowneae17012005-11-08 21:39:23 -08003444 return sprintf(page, "%s\n", p->name);
NeilBrownd9d166c2006-01-06 00:20:51 -08003445 else if (mddev->clevel[0])
3446 return sprintf(page, "%s\n", mddev->clevel);
3447 else if (mddev->level != LEVEL_NONE)
3448 return sprintf(page, "%d\n", mddev->level);
3449 else
3450 return 0;
NeilBrowneae17012005-11-08 21:39:23 -08003451}
3452
NeilBrownd9d166c2006-01-06 00:20:51 -08003453static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003454level_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownd9d166c2006-01-06 00:20:51 -08003455{
Dan Williamsf2859af2010-05-02 10:04:16 -07003456 char clevel[16];
NeilBrown20a49ff2008-02-06 01:39:57 -08003457 ssize_t rv = len;
NeilBrown84fc4b52011-10-11 16:49:58 +11003458 struct md_personality *pers;
Dan Williamsf2859af2010-05-02 10:04:16 -07003459 long level;
NeilBrown245f46c2009-03-31 14:39:39 +11003460 void *priv;
NeilBrown3cb03002011-10-11 16:45:26 +11003461 struct md_rdev *rdev;
NeilBrown245f46c2009-03-31 14:39:39 +11003462
3463 if (mddev->pers == NULL) {
3464 if (len == 0)
3465 return 0;
3466 if (len >= sizeof(mddev->clevel))
3467 return -ENOSPC;
3468 strncpy(mddev->clevel, buf, len);
3469 if (mddev->clevel[len-1] == '\n')
3470 len--;
3471 mddev->clevel[len] = 0;
3472 mddev->level = LEVEL_NONE;
3473 return rv;
3474 }
3475
3476 /* request to change the personality. Need to ensure:
3477 * - array is not engaged in resync/recovery/reshape
3478 * - old personality can be suspended
3479 * - new personality will access other array.
3480 */
3481
NeilBrownbb4f1e92010-08-08 21:18:03 +10003482 if (mddev->sync_thread ||
3483 mddev->reshape_position != MaxSector ||
3484 mddev->sysfs_active)
NeilBrownd9d166c2006-01-06 00:20:51 -08003485 return -EBUSY;
NeilBrown245f46c2009-03-31 14:39:39 +11003486
3487 if (!mddev->pers->quiesce) {
3488 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3489 mdname(mddev), mddev->pers->name);
3490 return -EINVAL;
3491 }
3492
3493 /* Now find the new personality */
Dan Williamsf2859af2010-05-02 10:04:16 -07003494 if (len == 0 || len >= sizeof(clevel))
NeilBrown245f46c2009-03-31 14:39:39 +11003495 return -EINVAL;
Dan Williamsf2859af2010-05-02 10:04:16 -07003496 strncpy(clevel, buf, len);
3497 if (clevel[len-1] == '\n')
NeilBrownd9d166c2006-01-06 00:20:51 -08003498 len--;
Dan Williamsf2859af2010-05-02 10:04:16 -07003499 clevel[len] = 0;
3500 if (strict_strtol(clevel, 10, &level))
3501 level = LEVEL_NONE;
NeilBrown245f46c2009-03-31 14:39:39 +11003502
Dan Williamsf2859af2010-05-02 10:04:16 -07003503 if (request_module("md-%s", clevel) != 0)
3504 request_module("md-level-%s", clevel);
NeilBrown245f46c2009-03-31 14:39:39 +11003505 spin_lock(&pers_lock);
Dan Williamsf2859af2010-05-02 10:04:16 -07003506 pers = find_pers(level, clevel);
NeilBrown245f46c2009-03-31 14:39:39 +11003507 if (!pers || !try_module_get(pers->owner)) {
3508 spin_unlock(&pers_lock);
Dan Williamsf2859af2010-05-02 10:04:16 -07003509 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
NeilBrown245f46c2009-03-31 14:39:39 +11003510 return -EINVAL;
3511 }
3512 spin_unlock(&pers_lock);
3513
3514 if (pers == mddev->pers) {
3515 /* Nothing to do! */
3516 module_put(pers->owner);
3517 return rv;
3518 }
3519 if (!pers->takeover) {
3520 module_put(pers->owner);
3521 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
Dan Williamsf2859af2010-05-02 10:04:16 -07003522 mdname(mddev), clevel);
NeilBrown245f46c2009-03-31 14:39:39 +11003523 return -EINVAL;
3524 }
3525
NeilBrowndafb20f2012-03-19 12:46:39 +11003526 rdev_for_each(rdev, mddev)
NeilBrowne93f68a2010-06-15 09:36:03 +01003527 rdev->new_raid_disk = rdev->raid_disk;
3528
NeilBrown245f46c2009-03-31 14:39:39 +11003529 /* ->takeover must set new_* and/or delta_disks
3530 * if it succeeds, and may set them when it fails.
3531 */
3532 priv = pers->takeover(mddev);
3533 if (IS_ERR(priv)) {
3534 mddev->new_level = mddev->level;
3535 mddev->new_layout = mddev->layout;
Andre Noll664e7c42009-06-18 08:45:27 +10003536 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown245f46c2009-03-31 14:39:39 +11003537 mddev->raid_disks -= mddev->delta_disks;
3538 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10003539 mddev->reshape_backwards = 0;
NeilBrown245f46c2009-03-31 14:39:39 +11003540 module_put(pers->owner);
3541 printk(KERN_WARNING "md: %s: %s would not accept array\n",
Dan Williamsf2859af2010-05-02 10:04:16 -07003542 mdname(mddev), clevel);
NeilBrown245f46c2009-03-31 14:39:39 +11003543 return PTR_ERR(priv);
3544 }
3545
3546 /* Looks like we have a winner */
3547 mddev_suspend(mddev);
3548 mddev->pers->stop(mddev);
NeilBrowna64c8762010-04-14 17:15:37 +10003549
3550 if (mddev->pers->sync_request == NULL &&
3551 pers->sync_request != NULL) {
3552 /* need to add the md_redundancy_group */
3553 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3554 printk(KERN_WARNING
3555 "md: cannot register extra attributes for %s\n",
3556 mdname(mddev));
NeilBrown19fdb9e2010-05-22 08:31:36 +10003557 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, NULL, "sync_action");
NeilBrowna64c8762010-04-14 17:15:37 +10003558 }
3559 if (mddev->pers->sync_request != NULL &&
3560 pers->sync_request == NULL) {
3561 /* need to remove the md_redundancy_group */
3562 if (mddev->to_remove == NULL)
3563 mddev->to_remove = &md_redundancy_group;
3564 }
3565
Trela Maciej54071b32010-03-08 16:02:42 +11003566 if (mddev->pers->sync_request == NULL &&
3567 mddev->external) {
3568 /* We are converting from a no-redundancy array
3569 * to a redundancy array and metadata is managed
3570 * externally so we need to be sure that writes
3571 * won't block due to a need to transition
3572 * clean->dirty
3573 * until external management is started.
3574 */
3575 mddev->in_sync = 0;
3576 mddev->safemode_delay = 0;
3577 mddev->safemode = 0;
3578 }
3579
NeilBrowndafb20f2012-03-19 12:46:39 +11003580 rdev_for_each(rdev, mddev) {
NeilBrowne93f68a2010-06-15 09:36:03 +01003581 if (rdev->raid_disk < 0)
3582 continue;
NeilBrownbf2cb0d2011-01-14 09:14:34 +11003583 if (rdev->new_raid_disk >= mddev->raid_disks)
NeilBrowne93f68a2010-06-15 09:36:03 +01003584 rdev->new_raid_disk = -1;
3585 if (rdev->new_raid_disk == rdev->raid_disk)
3586 continue;
Namhyung Kim36fad852011-07-27 11:00:36 +10003587 sysfs_unlink_rdev(mddev, rdev);
NeilBrowne93f68a2010-06-15 09:36:03 +01003588 }
NeilBrowndafb20f2012-03-19 12:46:39 +11003589 rdev_for_each(rdev, mddev) {
NeilBrowne93f68a2010-06-15 09:36:03 +01003590 if (rdev->raid_disk < 0)
3591 continue;
3592 if (rdev->new_raid_disk == rdev->raid_disk)
3593 continue;
3594 rdev->raid_disk = rdev->new_raid_disk;
3595 if (rdev->raid_disk < 0)
NeilBrown3a981b02009-08-03 10:59:55 +10003596 clear_bit(In_sync, &rdev->flags);
NeilBrowne93f68a2010-06-15 09:36:03 +01003597 else {
Namhyung Kim36fad852011-07-27 11:00:36 +10003598 if (sysfs_link_rdev(mddev, rdev))
3599 printk(KERN_WARNING "md: cannot register rd%d"
3600 " for %s after level change\n",
3601 rdev->raid_disk, mdname(mddev));
NeilBrown3a981b02009-08-03 10:59:55 +10003602 }
NeilBrowne93f68a2010-06-15 09:36:03 +01003603 }
3604
3605 module_put(mddev->pers->owner);
NeilBrown245f46c2009-03-31 14:39:39 +11003606 mddev->pers = pers;
3607 mddev->private = priv;
3608 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3609 mddev->level = mddev->new_level;
3610 mddev->layout = mddev->new_layout;
Andre Noll664e7c42009-06-18 08:45:27 +10003611 mddev->chunk_sectors = mddev->new_chunk_sectors;
NeilBrown245f46c2009-03-31 14:39:39 +11003612 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10003613 mddev->reshape_backwards = 0;
Krzysztof Wojcikfee68722011-04-20 15:39:53 +10003614 mddev->degraded = 0;
Trela, Maciej9af204c2010-03-08 16:02:44 +11003615 if (mddev->pers->sync_request == NULL) {
3616 /* this is now an array without redundancy, so
3617 * it must always be in_sync
3618 */
3619 mddev->in_sync = 1;
3620 del_timer_sync(&mddev->safemode_timer);
3621 }
NeilBrown245f46c2009-03-31 14:39:39 +11003622 pers->run(mddev);
NeilBrown245f46c2009-03-31 14:39:39 +11003623 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Jonathan Brassow47525e52012-05-22 13:55:29 +10003624 mddev_resume(mddev);
Maciej Trela5cac78612010-04-14 17:17:39 +10003625 sysfs_notify(&mddev->kobj, NULL, "level");
Dan Williamsbb7f8d22010-05-01 18:14:57 -07003626 md_new_event(mddev);
NeilBrownd9d166c2006-01-06 00:20:51 -08003627 return rv;
3628}
3629
3630static struct md_sysfs_entry md_level =
NeilBrown80ca3a42006-07-10 04:44:18 -07003631__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
NeilBrowneae17012005-11-08 21:39:23 -08003632
NeilBrownd4dbd022006-06-26 00:27:59 -07003633
3634static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003635layout_show(struct mddev *mddev, char *page)
NeilBrownd4dbd022006-06-26 00:27:59 -07003636{
3637 /* just a number, not meaningful for all levels */
NeilBrown08a02ec2007-05-09 02:35:38 -07003638 if (mddev->reshape_position != MaxSector &&
3639 mddev->layout != mddev->new_layout)
3640 return sprintf(page, "%d (%d)\n",
3641 mddev->new_layout, mddev->layout);
NeilBrownd4dbd022006-06-26 00:27:59 -07003642 return sprintf(page, "%d\n", mddev->layout);
3643}
3644
3645static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003646layout_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownd4dbd022006-06-26 00:27:59 -07003647{
3648 char *e;
3649 unsigned long n = simple_strtoul(buf, &e, 10);
NeilBrownd4dbd022006-06-26 00:27:59 -07003650
3651 if (!*buf || (*e && *e != '\n'))
3652 return -EINVAL;
3653
NeilBrownb3546032009-03-31 14:56:41 +11003654 if (mddev->pers) {
3655 int err;
NeilBrown50ac1682009-06-18 08:47:55 +10003656 if (mddev->pers->check_reshape == NULL)
NeilBrownb3546032009-03-31 14:56:41 +11003657 return -EBUSY;
NeilBrown597a7112009-06-18 08:47:42 +10003658 mddev->new_layout = n;
NeilBrown50ac1682009-06-18 08:47:55 +10003659 err = mddev->pers->check_reshape(mddev);
NeilBrown597a7112009-06-18 08:47:42 +10003660 if (err) {
3661 mddev->new_layout = mddev->layout;
NeilBrownb3546032009-03-31 14:56:41 +11003662 return err;
NeilBrown597a7112009-06-18 08:47:42 +10003663 }
NeilBrownb3546032009-03-31 14:56:41 +11003664 } else {
NeilBrown08a02ec2007-05-09 02:35:38 -07003665 mddev->new_layout = n;
NeilBrownb3546032009-03-31 14:56:41 +11003666 if (mddev->reshape_position == MaxSector)
3667 mddev->layout = n;
3668 }
NeilBrownd4dbd022006-06-26 00:27:59 -07003669 return len;
3670}
3671static struct md_sysfs_entry md_layout =
NeilBrown80ca3a42006-07-10 04:44:18 -07003672__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
NeilBrownd4dbd022006-06-26 00:27:59 -07003673
3674
NeilBrowneae17012005-11-08 21:39:23 -08003675static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003676raid_disks_show(struct mddev *mddev, char *page)
NeilBrowneae17012005-11-08 21:39:23 -08003677{
NeilBrownbb636542005-11-08 21:39:45 -08003678 if (mddev->raid_disks == 0)
3679 return 0;
NeilBrown08a02ec2007-05-09 02:35:38 -07003680 if (mddev->reshape_position != MaxSector &&
3681 mddev->delta_disks != 0)
3682 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3683 mddev->raid_disks - mddev->delta_disks);
NeilBrowneae17012005-11-08 21:39:23 -08003684 return sprintf(page, "%d\n", mddev->raid_disks);
3685}
3686
NeilBrownfd01b882011-10-11 16:47:53 +11003687static int update_raid_disks(struct mddev *mddev, int raid_disks);
NeilBrownda943b992006-01-06 00:20:54 -08003688
3689static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003690raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownda943b992006-01-06 00:20:54 -08003691{
NeilBrownda943b992006-01-06 00:20:54 -08003692 char *e;
3693 int rv = 0;
3694 unsigned long n = simple_strtoul(buf, &e, 10);
3695
3696 if (!*buf || (*e && *e != '\n'))
3697 return -EINVAL;
3698
3699 if (mddev->pers)
3700 rv = update_raid_disks(mddev, n);
NeilBrown08a02ec2007-05-09 02:35:38 -07003701 else if (mddev->reshape_position != MaxSector) {
NeilBrownc6563a82012-05-21 09:27:00 +10003702 struct md_rdev *rdev;
NeilBrown08a02ec2007-05-09 02:35:38 -07003703 int olddisks = mddev->raid_disks - mddev->delta_disks;
NeilBrownc6563a82012-05-21 09:27:00 +10003704
3705 rdev_for_each(rdev, mddev) {
3706 if (olddisks < n &&
3707 rdev->data_offset < rdev->new_data_offset)
3708 return -EINVAL;
3709 if (olddisks > n &&
3710 rdev->data_offset > rdev->new_data_offset)
3711 return -EINVAL;
3712 }
NeilBrown08a02ec2007-05-09 02:35:38 -07003713 mddev->delta_disks = n - olddisks;
3714 mddev->raid_disks = n;
NeilBrown2c810cd2012-05-21 09:27:00 +10003715 mddev->reshape_backwards = (mddev->delta_disks < 0);
NeilBrown08a02ec2007-05-09 02:35:38 -07003716 } else
NeilBrownda943b992006-01-06 00:20:54 -08003717 mddev->raid_disks = n;
3718 return rv ? rv : len;
3719}
3720static struct md_sysfs_entry md_raid_disks =
NeilBrown80ca3a42006-07-10 04:44:18 -07003721__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
NeilBrowneae17012005-11-08 21:39:23 -08003722
NeilBrown24dd4692005-11-08 21:39:26 -08003723static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003724chunk_size_show(struct mddev *mddev, char *page)
NeilBrown3b343802006-01-06 00:20:47 -08003725{
NeilBrown08a02ec2007-05-09 02:35:38 -07003726 if (mddev->reshape_position != MaxSector &&
Andre Noll664e7c42009-06-18 08:45:27 +10003727 mddev->chunk_sectors != mddev->new_chunk_sectors)
3728 return sprintf(page, "%d (%d)\n",
3729 mddev->new_chunk_sectors << 9,
Andre Noll9d8f0362009-06-18 08:45:01 +10003730 mddev->chunk_sectors << 9);
3731 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
NeilBrown3b343802006-01-06 00:20:47 -08003732}
3733
3734static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003735chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown3b343802006-01-06 00:20:47 -08003736{
NeilBrown3b343802006-01-06 00:20:47 -08003737 char *e;
3738 unsigned long n = simple_strtoul(buf, &e, 10);
3739
NeilBrown3b343802006-01-06 00:20:47 -08003740 if (!*buf || (*e && *e != '\n'))
3741 return -EINVAL;
3742
NeilBrownb3546032009-03-31 14:56:41 +11003743 if (mddev->pers) {
3744 int err;
NeilBrown50ac1682009-06-18 08:47:55 +10003745 if (mddev->pers->check_reshape == NULL)
NeilBrownb3546032009-03-31 14:56:41 +11003746 return -EBUSY;
NeilBrown597a7112009-06-18 08:47:42 +10003747 mddev->new_chunk_sectors = n >> 9;
NeilBrown50ac1682009-06-18 08:47:55 +10003748 err = mddev->pers->check_reshape(mddev);
NeilBrown597a7112009-06-18 08:47:42 +10003749 if (err) {
3750 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrownb3546032009-03-31 14:56:41 +11003751 return err;
NeilBrown597a7112009-06-18 08:47:42 +10003752 }
NeilBrownb3546032009-03-31 14:56:41 +11003753 } else {
Andre Noll664e7c42009-06-18 08:45:27 +10003754 mddev->new_chunk_sectors = n >> 9;
NeilBrownb3546032009-03-31 14:56:41 +11003755 if (mddev->reshape_position == MaxSector)
Andre Noll9d8f0362009-06-18 08:45:01 +10003756 mddev->chunk_sectors = n >> 9;
NeilBrownb3546032009-03-31 14:56:41 +11003757 }
NeilBrown3b343802006-01-06 00:20:47 -08003758 return len;
3759}
3760static struct md_sysfs_entry md_chunk_size =
NeilBrown80ca3a42006-07-10 04:44:18 -07003761__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
NeilBrown3b343802006-01-06 00:20:47 -08003762
NeilBrowna94213b2006-06-26 00:28:00 -07003763static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003764resync_start_show(struct mddev *mddev, char *page)
NeilBrowna94213b2006-06-26 00:28:00 -07003765{
NeilBrownd1a7c502009-03-31 15:24:32 +11003766 if (mddev->recovery_cp == MaxSector)
3767 return sprintf(page, "none\n");
NeilBrowna94213b2006-06-26 00:28:00 -07003768 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3769}
3770
3771static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003772resync_start_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrowna94213b2006-06-26 00:28:00 -07003773{
NeilBrowna94213b2006-06-26 00:28:00 -07003774 char *e;
3775 unsigned long long n = simple_strtoull(buf, &e, 10);
3776
NeilBrownb0986362011-05-11 15:52:21 +10003777 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
NeilBrowna94213b2006-06-26 00:28:00 -07003778 return -EBUSY;
Dan Williams06e3c812009-12-12 21:17:12 -07003779 if (cmd_match(buf, "none"))
3780 n = MaxSector;
3781 else if (!*buf || (*e && *e != '\n'))
NeilBrowna94213b2006-06-26 00:28:00 -07003782 return -EINVAL;
3783
3784 mddev->recovery_cp = n;
3785 return len;
3786}
3787static struct md_sysfs_entry md_resync_start =
NeilBrown80ca3a42006-07-10 04:44:18 -07003788__ATTR(resync_start, S_IRUGO|S_IWUSR, resync_start_show, resync_start_store);
NeilBrowna94213b2006-06-26 00:28:00 -07003789
NeilBrown9e653b62006-06-26 00:27:58 -07003790/*
3791 * The array state can be:
3792 *
3793 * clear
3794 * No devices, no size, no level
3795 * Equivalent to STOP_ARRAY ioctl
3796 * inactive
3797 * May have some settings, but array is not active
3798 * all IO results in error
3799 * When written, doesn't tear down array, but just stops it
3800 * suspended (not supported yet)
3801 * All IO requests will block. The array can be reconfigured.
Andre Noll910d8cb2008-03-25 21:00:53 +01003802 * Writing this, if accepted, will block until array is quiescent
NeilBrown9e653b62006-06-26 00:27:58 -07003803 * readonly
3804 * no resync can happen. no superblocks get written.
3805 * write requests fail
3806 * read-auto
3807 * like readonly, but behaves like 'clean' on a write request.
3808 *
3809 * clean - no pending writes, but otherwise active.
3810 * When written to inactive array, starts without resync
3811 * If a write request arrives then
3812 * if metadata is known, mark 'dirty' and switch to 'active'.
3813 * if not known, block and switch to write-pending
3814 * If written to an active array that has pending writes, then fails.
3815 * active
3816 * fully active: IO and resync can be happening.
3817 * When written to inactive array, starts with resync
3818 *
3819 * write-pending
3820 * clean, but writes are blocked waiting for 'active' to be written.
3821 *
3822 * active-idle
3823 * like active, but no writes have been seen for a while (100msec).
3824 *
3825 */
3826enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3827 write_pending, active_idle, bad_word};
Adrian Bunk05381952006-06-26 00:28:01 -07003828static char *array_states[] = {
NeilBrown9e653b62006-06-26 00:27:58 -07003829 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3830 "write-pending", "active-idle", NULL };
3831
3832static int match_word(const char *word, char **list)
3833{
3834 int n;
3835 for (n=0; list[n]; n++)
3836 if (cmd_match(word, list[n]))
3837 break;
3838 return n;
3839}
3840
3841static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003842array_state_show(struct mddev *mddev, char *page)
NeilBrown9e653b62006-06-26 00:27:58 -07003843{
3844 enum array_state st = inactive;
3845
3846 if (mddev->pers)
3847 switch(mddev->ro) {
3848 case 1:
3849 st = readonly;
3850 break;
3851 case 2:
3852 st = read_auto;
3853 break;
3854 case 0:
3855 if (mddev->in_sync)
3856 st = clean;
NeilBrown070dc6d2010-08-30 17:33:34 +10003857 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
NeilBrowne6910632008-02-06 01:39:51 -08003858 st = write_pending;
NeilBrown9e653b62006-06-26 00:27:58 -07003859 else if (mddev->safemode)
3860 st = active_idle;
3861 else
3862 st = active;
3863 }
3864 else {
3865 if (list_empty(&mddev->disks) &&
3866 mddev->raid_disks == 0 &&
Andre Noll58c0fed2009-03-31 14:33:13 +11003867 mddev->dev_sectors == 0)
NeilBrown9e653b62006-06-26 00:27:58 -07003868 st = clear;
3869 else
3870 st = inactive;
3871 }
3872 return sprintf(page, "%s\n", array_states[st]);
3873}
3874
NeilBrowna05b7ea2012-07-19 15:59:18 +10003875static int do_md_stop(struct mddev * mddev, int ro, struct block_device *bdev);
3876static int md_set_readonly(struct mddev * mddev, struct block_device *bdev);
NeilBrownfd01b882011-10-11 16:47:53 +11003877static int do_md_run(struct mddev * mddev);
3878static int restart_array(struct mddev *mddev);
NeilBrown9e653b62006-06-26 00:27:58 -07003879
3880static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003881array_state_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown9e653b62006-06-26 00:27:58 -07003882{
3883 int err = -EINVAL;
3884 enum array_state st = match_word(buf, array_states);
3885 switch(st) {
3886 case bad_word:
3887 break;
3888 case clear:
3889 /* stopping an active array */
NeilBrowna05b7ea2012-07-19 15:59:18 +10003890 err = do_md_stop(mddev, 0, NULL);
NeilBrown9e653b62006-06-26 00:27:58 -07003891 break;
3892 case inactive:
3893 /* stopping an active array */
NeilBrown90cf1952012-07-31 10:04:55 +10003894 if (mddev->pers)
NeilBrowna05b7ea2012-07-19 15:59:18 +10003895 err = do_md_stop(mddev, 2, NULL);
NeilBrown90cf1952012-07-31 10:04:55 +10003896 else
NeilBrowne6910632008-02-06 01:39:51 -08003897 err = 0; /* already inactive */
NeilBrown9e653b62006-06-26 00:27:58 -07003898 break;
3899 case suspended:
3900 break; /* not supported yet */
3901 case readonly:
3902 if (mddev->pers)
NeilBrowna05b7ea2012-07-19 15:59:18 +10003903 err = md_set_readonly(mddev, NULL);
NeilBrown9e653b62006-06-26 00:27:58 -07003904 else {
3905 mddev->ro = 1;
NeilBrown648b6292008-04-30 00:52:30 -07003906 set_disk_ro(mddev->gendisk, 1);
NeilBrown9e653b62006-06-26 00:27:58 -07003907 err = do_md_run(mddev);
3908 }
3909 break;
3910 case read_auto:
NeilBrown9e653b62006-06-26 00:27:58 -07003911 if (mddev->pers) {
NeilBrown80268ee2008-10-13 11:55:12 +11003912 if (mddev->ro == 0)
NeilBrowna05b7ea2012-07-19 15:59:18 +10003913 err = md_set_readonly(mddev, NULL);
NeilBrown80268ee2008-10-13 11:55:12 +11003914 else if (mddev->ro == 1)
NeilBrown648b6292008-04-30 00:52:30 -07003915 err = restart_array(mddev);
3916 if (err == 0) {
3917 mddev->ro = 2;
3918 set_disk_ro(mddev->gendisk, 0);
3919 }
NeilBrown9e653b62006-06-26 00:27:58 -07003920 } else {
3921 mddev->ro = 2;
3922 err = do_md_run(mddev);
3923 }
3924 break;
3925 case clean:
3926 if (mddev->pers) {
3927 restart_array(mddev);
3928 spin_lock_irq(&mddev->write_lock);
3929 if (atomic_read(&mddev->writes_pending) == 0) {
NeilBrowne6910632008-02-06 01:39:51 -08003930 if (mddev->in_sync == 0) {
3931 mddev->in_sync = 1;
NeilBrown31a59e32008-04-30 00:52:30 -07003932 if (mddev->safemode == 1)
3933 mddev->safemode = 0;
NeilBrown070dc6d2010-08-30 17:33:34 +10003934 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrowne6910632008-02-06 01:39:51 -08003935 }
3936 err = 0;
3937 } else
3938 err = -EBUSY;
NeilBrown9e653b62006-06-26 00:27:58 -07003939 spin_unlock_irq(&mddev->write_lock);
NeilBrown5bf29592009-05-07 12:50:57 +10003940 } else
3941 err = -EINVAL;
NeilBrown9e653b62006-06-26 00:27:58 -07003942 break;
3943 case active:
3944 if (mddev->pers) {
3945 restart_array(mddev);
NeilBrown070dc6d2010-08-30 17:33:34 +10003946 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrown9e653b62006-06-26 00:27:58 -07003947 wake_up(&mddev->sb_wait);
3948 err = 0;
3949 } else {
3950 mddev->ro = 0;
NeilBrown648b6292008-04-30 00:52:30 -07003951 set_disk_ro(mddev->gendisk, 0);
NeilBrown9e653b62006-06-26 00:27:58 -07003952 err = do_md_run(mddev);
3953 }
3954 break;
3955 case write_pending:
3956 case active_idle:
3957 /* these cannot be set */
3958 break;
3959 }
3960 if (err)
3961 return err;
Neil Brown0fd62b82008-06-28 08:31:36 +10003962 else {
NeilBrown1d23f172011-12-08 15:49:12 +11003963 if (mddev->hold_active == UNTIL_IOCTL)
3964 mddev->hold_active = 0;
NeilBrown00bcb4a2010-06-01 19:37:23 +10003965 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrown9e653b62006-06-26 00:27:58 -07003966 return len;
Neil Brown0fd62b82008-06-28 08:31:36 +10003967 }
NeilBrown9e653b62006-06-26 00:27:58 -07003968}
NeilBrown80ca3a42006-07-10 04:44:18 -07003969static struct md_sysfs_entry md_array_state =
3970__ATTR(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
NeilBrown9e653b62006-06-26 00:27:58 -07003971
NeilBrown6d7ff732006-01-06 00:21:16 -08003972static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003973max_corrected_read_errors_show(struct mddev *mddev, char *page) {
Robert Becker1e509152009-12-14 12:49:58 +11003974 return sprintf(page, "%d\n",
3975 atomic_read(&mddev->max_corr_read_errors));
3976}
3977
3978static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003979max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
Robert Becker1e509152009-12-14 12:49:58 +11003980{
3981 char *e;
3982 unsigned long n = simple_strtoul(buf, &e, 10);
3983
3984 if (*buf && (*e == 0 || *e == '\n')) {
3985 atomic_set(&mddev->max_corr_read_errors, n);
3986 return len;
3987 }
3988 return -EINVAL;
3989}
3990
3991static struct md_sysfs_entry max_corr_read_errors =
3992__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3993 max_corrected_read_errors_store);
3994
3995static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003996null_show(struct mddev *mddev, char *page)
NeilBrown6d7ff732006-01-06 00:21:16 -08003997{
3998 return -EINVAL;
3999}
4000
4001static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004002new_dev_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown6d7ff732006-01-06 00:21:16 -08004003{
4004 /* buf must be %d:%d\n? giving major and minor numbers */
4005 /* The new device is added to the array.
4006 * If the array has a persistent superblock, we read the
4007 * superblock to initialise info and check validity.
4008 * Otherwise, only checking done is that in bind_rdev_to_array,
4009 * which mainly checks size.
4010 */
4011 char *e;
4012 int major = simple_strtoul(buf, &e, 10);
4013 int minor;
4014 dev_t dev;
NeilBrown3cb03002011-10-11 16:45:26 +11004015 struct md_rdev *rdev;
NeilBrown6d7ff732006-01-06 00:21:16 -08004016 int err;
4017
4018 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
4019 return -EINVAL;
4020 minor = simple_strtoul(e+1, &e, 10);
4021 if (*e && *e != '\n')
4022 return -EINVAL;
4023 dev = MKDEV(major, minor);
4024 if (major != MAJOR(dev) ||
4025 minor != MINOR(dev))
4026 return -EOVERFLOW;
4027
4028
4029 if (mddev->persistent) {
4030 rdev = md_import_device(dev, mddev->major_version,
4031 mddev->minor_version);
4032 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
NeilBrown3cb03002011-10-11 16:45:26 +11004033 struct md_rdev *rdev0
4034 = list_entry(mddev->disks.next,
4035 struct md_rdev, same_set);
NeilBrown6d7ff732006-01-06 00:21:16 -08004036 err = super_types[mddev->major_version]
4037 .load_super(rdev, rdev0, mddev->minor_version);
4038 if (err < 0)
4039 goto out;
4040 }
NeilBrownc5d79ad2008-02-06 01:39:54 -08004041 } else if (mddev->external)
4042 rdev = md_import_device(dev, -2, -1);
4043 else
NeilBrown6d7ff732006-01-06 00:21:16 -08004044 rdev = md_import_device(dev, -1, -1);
4045
4046 if (IS_ERR(rdev))
4047 return PTR_ERR(rdev);
4048 err = bind_rdev_to_array(rdev, mddev);
4049 out:
4050 if (err)
4051 export_rdev(rdev);
4052 return err ? err : len;
4053}
4054
4055static struct md_sysfs_entry md_new_device =
NeilBrown80ca3a42006-07-10 04:44:18 -07004056__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
NeilBrown3b343802006-01-06 00:20:47 -08004057
4058static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004059bitmap_store(struct mddev *mddev, const char *buf, size_t len)
Paul Clements9b1d1da2006-10-03 01:15:49 -07004060{
4061 char *end;
4062 unsigned long chunk, end_chunk;
4063
4064 if (!mddev->bitmap)
4065 goto out;
4066 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
4067 while (*buf) {
4068 chunk = end_chunk = simple_strtoul(buf, &end, 0);
4069 if (buf == end) break;
4070 if (*end == '-') { /* range */
4071 buf = end + 1;
4072 end_chunk = simple_strtoul(buf, &end, 0);
4073 if (buf == end) break;
4074 }
4075 if (*end && !isspace(*end)) break;
4076 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
André Goddard Rosae7d28602009-12-14 18:01:06 -08004077 buf = skip_spaces(end);
Paul Clements9b1d1da2006-10-03 01:15:49 -07004078 }
4079 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
4080out:
4081 return len;
4082}
4083
4084static struct md_sysfs_entry md_bitmap =
4085__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
4086
4087static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004088size_show(struct mddev *mddev, char *page)
NeilBrowna35b0d62006-01-06 00:20:49 -08004089{
Andre Noll58c0fed2009-03-31 14:33:13 +11004090 return sprintf(page, "%llu\n",
4091 (unsigned long long)mddev->dev_sectors / 2);
NeilBrowna35b0d62006-01-06 00:20:49 -08004092}
4093
NeilBrownfd01b882011-10-11 16:47:53 +11004094static int update_size(struct mddev *mddev, sector_t num_sectors);
NeilBrowna35b0d62006-01-06 00:20:49 -08004095
4096static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004097size_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrowna35b0d62006-01-06 00:20:49 -08004098{
4099 /* If array is inactive, we can reduce the component size, but
4100 * not increase it (except from 0).
4101 * If array is active, we can try an on-line resize
4102 */
Dan Williamsb522adc2009-03-31 15:00:31 +11004103 sector_t sectors;
4104 int err = strict_blocks_to_sectors(buf, &sectors);
NeilBrowna35b0d62006-01-06 00:20:49 -08004105
Andre Noll58c0fed2009-03-31 14:33:13 +11004106 if (err < 0)
4107 return err;
NeilBrowna35b0d62006-01-06 00:20:49 -08004108 if (mddev->pers) {
Andre Noll58c0fed2009-03-31 14:33:13 +11004109 err = update_size(mddev, sectors);
NeilBrown850b2b42006-10-03 01:15:46 -07004110 md_update_sb(mddev, 1);
NeilBrowna35b0d62006-01-06 00:20:49 -08004111 } else {
Andre Noll58c0fed2009-03-31 14:33:13 +11004112 if (mddev->dev_sectors == 0 ||
4113 mddev->dev_sectors > sectors)
4114 mddev->dev_sectors = sectors;
NeilBrowna35b0d62006-01-06 00:20:49 -08004115 else
4116 err = -ENOSPC;
4117 }
4118 return err ? err : len;
4119}
4120
4121static struct md_sysfs_entry md_size =
NeilBrown80ca3a42006-07-10 04:44:18 -07004122__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
NeilBrowna35b0d62006-01-06 00:20:49 -08004123
NeilBrown8bb93aa2006-01-06 00:20:50 -08004124
4125/* Metdata version.
NeilBrowne6910632008-02-06 01:39:51 -08004126 * This is one of
4127 * 'none' for arrays with no metadata (good luck...)
4128 * 'external' for arrays with externally managed metadata,
NeilBrown8bb93aa2006-01-06 00:20:50 -08004129 * or N.M for internally known formats
4130 */
4131static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004132metadata_show(struct mddev *mddev, char *page)
NeilBrown8bb93aa2006-01-06 00:20:50 -08004133{
4134 if (mddev->persistent)
4135 return sprintf(page, "%d.%d\n",
4136 mddev->major_version, mddev->minor_version);
NeilBrowne6910632008-02-06 01:39:51 -08004137 else if (mddev->external)
4138 return sprintf(page, "external:%s\n", mddev->metadata_type);
NeilBrown8bb93aa2006-01-06 00:20:50 -08004139 else
4140 return sprintf(page, "none\n");
4141}
4142
4143static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004144metadata_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown8bb93aa2006-01-06 00:20:50 -08004145{
4146 int major, minor;
4147 char *e;
NeilBrownea43ddd2008-10-13 11:55:11 +11004148 /* Changing the details of 'external' metadata is
4149 * always permitted. Otherwise there must be
4150 * no devices attached to the array.
4151 */
4152 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4153 ;
4154 else if (!list_empty(&mddev->disks))
NeilBrown8bb93aa2006-01-06 00:20:50 -08004155 return -EBUSY;
4156
4157 if (cmd_match(buf, "none")) {
4158 mddev->persistent = 0;
NeilBrowne6910632008-02-06 01:39:51 -08004159 mddev->external = 0;
4160 mddev->major_version = 0;
4161 mddev->minor_version = 90;
4162 return len;
4163 }
4164 if (strncmp(buf, "external:", 9) == 0) {
NeilBrown20a49ff2008-02-06 01:39:57 -08004165 size_t namelen = len-9;
NeilBrowne6910632008-02-06 01:39:51 -08004166 if (namelen >= sizeof(mddev->metadata_type))
4167 namelen = sizeof(mddev->metadata_type)-1;
4168 strncpy(mddev->metadata_type, buf+9, namelen);
4169 mddev->metadata_type[namelen] = 0;
4170 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4171 mddev->metadata_type[--namelen] = 0;
4172 mddev->persistent = 0;
4173 mddev->external = 1;
NeilBrown8bb93aa2006-01-06 00:20:50 -08004174 mddev->major_version = 0;
4175 mddev->minor_version = 90;
4176 return len;
4177 }
4178 major = simple_strtoul(buf, &e, 10);
4179 if (e==buf || *e != '.')
4180 return -EINVAL;
4181 buf = e+1;
4182 minor = simple_strtoul(buf, &e, 10);
NeilBrown3f9d7b02006-12-22 01:11:41 -08004183 if (e==buf || (*e && *e != '\n') )
NeilBrown8bb93aa2006-01-06 00:20:50 -08004184 return -EINVAL;
Ahmed S. Darwish50511da2007-05-09 02:35:34 -07004185 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
NeilBrown8bb93aa2006-01-06 00:20:50 -08004186 return -ENOENT;
4187 mddev->major_version = major;
4188 mddev->minor_version = minor;
4189 mddev->persistent = 1;
NeilBrowne6910632008-02-06 01:39:51 -08004190 mddev->external = 0;
NeilBrown8bb93aa2006-01-06 00:20:50 -08004191 return len;
4192}
4193
4194static struct md_sysfs_entry md_metadata =
NeilBrown80ca3a42006-07-10 04:44:18 -07004195__ATTR(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
NeilBrown8bb93aa2006-01-06 00:20:50 -08004196
NeilBrowna35b0d62006-01-06 00:20:49 -08004197static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004198action_show(struct mddev *mddev, char *page)
NeilBrown24dd4692005-11-08 21:39:26 -08004199{
NeilBrown7eec3142005-11-08 21:39:44 -08004200 char *type = "idle";
NeilBrownb6a9ce62009-05-26 09:41:17 +10004201 if (test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
4202 type = "frozen";
4203 else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
NeilBrown2b12ab62007-10-16 23:30:53 -07004204 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))) {
NeilBrownccfcc3c2006-03-27 01:18:09 -08004205 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4206 type = "reshape";
4207 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
NeilBrown24dd4692005-11-08 21:39:26 -08004208 if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4209 type = "resync";
4210 else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
4211 type = "check";
4212 else
4213 type = "repair";
Neil Brown72a23c22008-06-28 08:31:41 +10004214 } else if (test_bit(MD_RECOVERY_RECOVER, &mddev->recovery))
NeilBrown24dd4692005-11-08 21:39:26 -08004215 type = "recover";
4216 }
4217 return sprintf(page, "%s\n", type);
4218}
4219
NeilBrownfd01b882011-10-11 16:47:53 +11004220static void reap_sync_thread(struct mddev *mddev);
NeilBrown7ebc0be2011-01-14 09:14:33 +11004221
NeilBrown24dd4692005-11-08 21:39:26 -08004222static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004223action_store(struct mddev *mddev, const char *page, size_t len)
NeilBrown24dd4692005-11-08 21:39:26 -08004224{
NeilBrown7eec3142005-11-08 21:39:44 -08004225 if (!mddev->pers || !mddev->pers->sync_request)
4226 return -EINVAL;
4227
NeilBrownb6a9ce62009-05-26 09:41:17 +10004228 if (cmd_match(page, "frozen"))
4229 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4230 else
4231 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4232
4233 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
NeilBrown7eec3142005-11-08 21:39:44 -08004234 if (mddev->sync_thread) {
4235 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrown7ebc0be2011-01-14 09:14:33 +11004236 reap_sync_thread(mddev);
NeilBrown7eec3142005-11-08 21:39:44 -08004237 }
NeilBrown03c902e2006-01-06 00:20:46 -08004238 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4239 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
NeilBrown24dd4692005-11-08 21:39:26 -08004240 return -EBUSY;
Neil Brown72a23c22008-06-28 08:31:41 +10004241 else if (cmd_match(page, "resync"))
NeilBrown7eec3142005-11-08 21:39:44 -08004242 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
Neil Brown72a23c22008-06-28 08:31:41 +10004243 else if (cmd_match(page, "recover")) {
4244 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4245 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4246 } else if (cmd_match(page, "reshape")) {
NeilBrown16484bf2006-03-27 01:18:13 -08004247 int err;
4248 if (mddev->pers->start_reshape == NULL)
4249 return -EINVAL;
4250 err = mddev->pers->start_reshape(mddev);
4251 if (err)
4252 return err;
Neil Browna99ac972008-06-28 08:31:43 +10004253 sysfs_notify(&mddev->kobj, NULL, "degraded");
NeilBrown16484bf2006-03-27 01:18:13 -08004254 } else {
NeilBrownbce74da2006-01-06 00:20:41 -08004255 if (cmd_match(page, "check"))
NeilBrown7eec3142005-11-08 21:39:44 -08004256 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
NeilBrown2adc7d42006-05-20 14:59:57 -07004257 else if (!cmd_match(page, "repair"))
NeilBrown7eec3142005-11-08 21:39:44 -08004258 return -EINVAL;
4259 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4260 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
NeilBrown7eec3142005-11-08 21:39:44 -08004261 }
NeilBrown03c902e2006-01-06 00:20:46 -08004262 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrown24dd4692005-11-08 21:39:26 -08004263 md_wakeup_thread(mddev->thread);
NeilBrown00bcb4a2010-06-01 19:37:23 +10004264 sysfs_notify_dirent_safe(mddev->sysfs_action);
NeilBrown24dd4692005-11-08 21:39:26 -08004265 return len;
4266}
4267
NeilBrown9d888832005-11-08 21:39:26 -08004268static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004269mismatch_cnt_show(struct mddev *mddev, char *page)
NeilBrown9d888832005-11-08 21:39:26 -08004270{
4271 return sprintf(page, "%llu\n",
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11004272 (unsigned long long)
4273 atomic64_read(&mddev->resync_mismatches));
NeilBrown9d888832005-11-08 21:39:26 -08004274}
4275
NeilBrown80ca3a42006-07-10 04:44:18 -07004276static struct md_sysfs_entry md_scan_mode =
4277__ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
NeilBrown24dd4692005-11-08 21:39:26 -08004278
NeilBrown96de1e62005-11-08 21:39:39 -08004279
NeilBrown80ca3a42006-07-10 04:44:18 -07004280static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
NeilBrown9d888832005-11-08 21:39:26 -08004281
NeilBrown88202a02006-01-06 00:21:36 -08004282static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004283sync_min_show(struct mddev *mddev, char *page)
NeilBrown88202a02006-01-06 00:21:36 -08004284{
4285 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4286 mddev->sync_speed_min ? "local": "system");
4287}
4288
4289static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004290sync_min_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown88202a02006-01-06 00:21:36 -08004291{
4292 int min;
4293 char *e;
4294 if (strncmp(buf, "system", 6)==0) {
4295 mddev->sync_speed_min = 0;
4296 return len;
4297 }
4298 min = simple_strtoul(buf, &e, 10);
4299 if (buf == e || (*e && *e != '\n') || min <= 0)
4300 return -EINVAL;
4301 mddev->sync_speed_min = min;
4302 return len;
4303}
4304
4305static struct md_sysfs_entry md_sync_min =
4306__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4307
4308static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004309sync_max_show(struct mddev *mddev, char *page)
NeilBrown88202a02006-01-06 00:21:36 -08004310{
4311 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4312 mddev->sync_speed_max ? "local": "system");
4313}
4314
4315static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004316sync_max_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown88202a02006-01-06 00:21:36 -08004317{
4318 int max;
4319 char *e;
4320 if (strncmp(buf, "system", 6)==0) {
4321 mddev->sync_speed_max = 0;
4322 return len;
4323 }
4324 max = simple_strtoul(buf, &e, 10);
4325 if (buf == e || (*e && *e != '\n') || max <= 0)
4326 return -EINVAL;
4327 mddev->sync_speed_max = max;
4328 return len;
4329}
4330
4331static struct md_sysfs_entry md_sync_max =
4332__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4333
Iustin Popd7f3d292007-10-16 23:30:54 -07004334static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004335degraded_show(struct mddev *mddev, char *page)
Iustin Popd7f3d292007-10-16 23:30:54 -07004336{
4337 return sprintf(page, "%d\n", mddev->degraded);
4338}
4339static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
NeilBrown88202a02006-01-06 00:21:36 -08004340
4341static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004342sync_force_parallel_show(struct mddev *mddev, char *page)
Bernd Schubert90b08712008-05-23 13:04:38 -07004343{
4344 return sprintf(page, "%d\n", mddev->parallel_resync);
4345}
4346
4347static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004348sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
Bernd Schubert90b08712008-05-23 13:04:38 -07004349{
4350 long n;
4351
4352 if (strict_strtol(buf, 10, &n))
4353 return -EINVAL;
4354
4355 if (n != 0 && n != 1)
4356 return -EINVAL;
4357
4358 mddev->parallel_resync = n;
4359
4360 if (mddev->sync_thread)
4361 wake_up(&resync_wait);
4362
4363 return len;
4364}
4365
4366/* force parallel resync, even with shared block devices */
4367static struct md_sysfs_entry md_sync_force_parallel =
4368__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4369 sync_force_parallel_show, sync_force_parallel_store);
4370
4371static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004372sync_speed_show(struct mddev *mddev, char *page)
NeilBrown88202a02006-01-06 00:21:36 -08004373{
4374 unsigned long resync, dt, db;
NeilBrownd1a7c502009-03-31 15:24:32 +11004375 if (mddev->curr_resync == 0)
4376 return sprintf(page, "none\n");
Andre Noll9687a602008-03-25 22:24:09 +01004377 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4378 dt = (jiffies - mddev->resync_mark) / HZ;
NeilBrown88202a02006-01-06 00:21:36 -08004379 if (!dt) dt++;
Andre Noll9687a602008-03-25 22:24:09 +01004380 db = resync - mddev->resync_mark_cnt;
4381 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
NeilBrown88202a02006-01-06 00:21:36 -08004382}
4383
NeilBrown80ca3a42006-07-10 04:44:18 -07004384static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
NeilBrown88202a02006-01-06 00:21:36 -08004385
4386static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004387sync_completed_show(struct mddev *mddev, char *page)
NeilBrown88202a02006-01-06 00:21:36 -08004388{
Rémi Rérolle13ae8642011-01-14 09:14:34 +11004389 unsigned long long max_sectors, resync;
NeilBrown88202a02006-01-06 00:21:36 -08004390
NeilBrownacb180b2009-04-14 16:28:34 +10004391 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4392 return sprintf(page, "none\n");
4393
NeilBrownc804cde2012-05-21 09:28:33 +10004394 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4395 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
Andre Noll58c0fed2009-03-31 14:33:13 +11004396 max_sectors = mddev->resync_max_sectors;
NeilBrown88202a02006-01-06 00:21:36 -08004397 else
Andre Noll58c0fed2009-03-31 14:33:13 +11004398 max_sectors = mddev->dev_sectors;
NeilBrown88202a02006-01-06 00:21:36 -08004399
NeilBrownacb180b2009-04-14 16:28:34 +10004400 resync = mddev->curr_resync_completed;
Rémi Rérolle13ae8642011-01-14 09:14:34 +11004401 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
NeilBrown88202a02006-01-06 00:21:36 -08004402}
4403
NeilBrown80ca3a42006-07-10 04:44:18 -07004404static struct md_sysfs_entry md_sync_completed = __ATTR_RO(sync_completed);
NeilBrown88202a02006-01-06 00:21:36 -08004405
NeilBrowne464eaf2006-03-27 01:18:14 -08004406static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004407min_sync_show(struct mddev *mddev, char *page)
Neil Brown5e96ee62008-06-28 08:31:24 +10004408{
4409 return sprintf(page, "%llu\n",
4410 (unsigned long long)mddev->resync_min);
4411}
4412static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004413min_sync_store(struct mddev *mddev, const char *buf, size_t len)
Neil Brown5e96ee62008-06-28 08:31:24 +10004414{
4415 unsigned long long min;
4416 if (strict_strtoull(buf, 10, &min))
4417 return -EINVAL;
4418 if (min > mddev->resync_max)
4419 return -EINVAL;
4420 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4421 return -EBUSY;
4422
4423 /* Must be a multiple of chunk_size */
Andre Noll9d8f0362009-06-18 08:45:01 +10004424 if (mddev->chunk_sectors) {
raz ben yehuda2ac06c32009-06-16 17:01:42 +10004425 sector_t temp = min;
Andre Noll9d8f0362009-06-18 08:45:01 +10004426 if (sector_div(temp, mddev->chunk_sectors))
Neil Brown5e96ee62008-06-28 08:31:24 +10004427 return -EINVAL;
4428 }
4429 mddev->resync_min = min;
4430
4431 return len;
4432}
4433
4434static struct md_sysfs_entry md_min_sync =
4435__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4436
4437static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004438max_sync_show(struct mddev *mddev, char *page)
NeilBrownc6207272008-02-06 01:39:52 -08004439{
4440 if (mddev->resync_max == MaxSector)
4441 return sprintf(page, "max\n");
4442 else
4443 return sprintf(page, "%llu\n",
4444 (unsigned long long)mddev->resync_max);
4445}
4446static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004447max_sync_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownc6207272008-02-06 01:39:52 -08004448{
4449 if (strncmp(buf, "max", 3) == 0)
4450 mddev->resync_max = MaxSector;
4451 else {
Neil Brown5e96ee62008-06-28 08:31:24 +10004452 unsigned long long max;
4453 if (strict_strtoull(buf, 10, &max))
4454 return -EINVAL;
4455 if (max < mddev->resync_min)
NeilBrownc6207272008-02-06 01:39:52 -08004456 return -EINVAL;
4457 if (max < mddev->resync_max &&
NeilBrown4d484a42009-08-13 10:41:50 +10004458 mddev->ro == 0 &&
NeilBrownc6207272008-02-06 01:39:52 -08004459 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4460 return -EBUSY;
4461
4462 /* Must be a multiple of chunk_size */
Andre Noll9d8f0362009-06-18 08:45:01 +10004463 if (mddev->chunk_sectors) {
raz ben yehuda2ac06c32009-06-16 17:01:42 +10004464 sector_t temp = max;
Andre Noll9d8f0362009-06-18 08:45:01 +10004465 if (sector_div(temp, mddev->chunk_sectors))
NeilBrownc6207272008-02-06 01:39:52 -08004466 return -EINVAL;
4467 }
4468 mddev->resync_max = max;
4469 }
4470 wake_up(&mddev->recovery_wait);
4471 return len;
4472}
4473
4474static struct md_sysfs_entry md_max_sync =
4475__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4476
4477static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004478suspend_lo_show(struct mddev *mddev, char *page)
NeilBrowne464eaf2006-03-27 01:18:14 -08004479{
4480 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4481}
4482
4483static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004484suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrowne464eaf2006-03-27 01:18:14 -08004485{
4486 char *e;
4487 unsigned long long new = simple_strtoull(buf, &e, 10);
NeilBrown23ddff32011-01-14 09:14:34 +11004488 unsigned long long old = mddev->suspend_lo;
NeilBrowne464eaf2006-03-27 01:18:14 -08004489
NeilBrownb8d966e2009-07-01 11:14:04 +10004490 if (mddev->pers == NULL ||
4491 mddev->pers->quiesce == NULL)
NeilBrowne464eaf2006-03-27 01:18:14 -08004492 return -EINVAL;
4493 if (buf == e || (*e && *e != '\n'))
4494 return -EINVAL;
NeilBrown23ddff32011-01-14 09:14:34 +11004495
4496 mddev->suspend_lo = new;
4497 if (new >= old)
4498 /* Shrinking suspended region */
NeilBrowne464eaf2006-03-27 01:18:14 -08004499 mddev->pers->quiesce(mddev, 2);
NeilBrown23ddff32011-01-14 09:14:34 +11004500 else {
4501 /* Expanding suspended region - need to wait */
4502 mddev->pers->quiesce(mddev, 1);
4503 mddev->pers->quiesce(mddev, 0);
4504 }
4505 return len;
NeilBrowne464eaf2006-03-27 01:18:14 -08004506}
4507static struct md_sysfs_entry md_suspend_lo =
4508__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4509
4510
4511static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004512suspend_hi_show(struct mddev *mddev, char *page)
NeilBrowne464eaf2006-03-27 01:18:14 -08004513{
4514 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4515}
4516
4517static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004518suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrowne464eaf2006-03-27 01:18:14 -08004519{
4520 char *e;
4521 unsigned long long new = simple_strtoull(buf, &e, 10);
NeilBrown23ddff32011-01-14 09:14:34 +11004522 unsigned long long old = mddev->suspend_hi;
NeilBrowne464eaf2006-03-27 01:18:14 -08004523
NeilBrownb8d966e2009-07-01 11:14:04 +10004524 if (mddev->pers == NULL ||
4525 mddev->pers->quiesce == NULL)
NeilBrowne464eaf2006-03-27 01:18:14 -08004526 return -EINVAL;
4527 if (buf == e || (*e && *e != '\n'))
4528 return -EINVAL;
NeilBrown23ddff32011-01-14 09:14:34 +11004529
4530 mddev->suspend_hi = new;
4531 if (new <= old)
4532 /* Shrinking suspended region */
4533 mddev->pers->quiesce(mddev, 2);
4534 else {
4535 /* Expanding suspended region - need to wait */
NeilBrowne464eaf2006-03-27 01:18:14 -08004536 mddev->pers->quiesce(mddev, 1);
4537 mddev->pers->quiesce(mddev, 0);
NeilBrown23ddff32011-01-14 09:14:34 +11004538 }
4539 return len;
NeilBrowne464eaf2006-03-27 01:18:14 -08004540}
4541static struct md_sysfs_entry md_suspend_hi =
4542__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4543
NeilBrown08a02ec2007-05-09 02:35:38 -07004544static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004545reshape_position_show(struct mddev *mddev, char *page)
NeilBrown08a02ec2007-05-09 02:35:38 -07004546{
4547 if (mddev->reshape_position != MaxSector)
4548 return sprintf(page, "%llu\n",
4549 (unsigned long long)mddev->reshape_position);
4550 strcpy(page, "none\n");
4551 return 5;
4552}
4553
4554static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004555reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown08a02ec2007-05-09 02:35:38 -07004556{
NeilBrownc6563a82012-05-21 09:27:00 +10004557 struct md_rdev *rdev;
NeilBrown08a02ec2007-05-09 02:35:38 -07004558 char *e;
4559 unsigned long long new = simple_strtoull(buf, &e, 10);
4560 if (mddev->pers)
4561 return -EBUSY;
4562 if (buf == e || (*e && *e != '\n'))
4563 return -EINVAL;
4564 mddev->reshape_position = new;
4565 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10004566 mddev->reshape_backwards = 0;
NeilBrown08a02ec2007-05-09 02:35:38 -07004567 mddev->new_level = mddev->level;
4568 mddev->new_layout = mddev->layout;
Andre Noll664e7c42009-06-18 08:45:27 +10004569 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrownc6563a82012-05-21 09:27:00 +10004570 rdev_for_each(rdev, mddev)
4571 rdev->new_data_offset = rdev->data_offset;
NeilBrown08a02ec2007-05-09 02:35:38 -07004572 return len;
4573}
4574
4575static struct md_sysfs_entry md_reshape_position =
4576__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4577 reshape_position_store);
4578
Dan Williamsb522adc2009-03-31 15:00:31 +11004579static ssize_t
NeilBrown2c810cd2012-05-21 09:27:00 +10004580reshape_direction_show(struct mddev *mddev, char *page)
4581{
4582 return sprintf(page, "%s\n",
4583 mddev->reshape_backwards ? "backwards" : "forwards");
4584}
4585
4586static ssize_t
4587reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4588{
4589 int backwards = 0;
4590 if (cmd_match(buf, "forwards"))
4591 backwards = 0;
4592 else if (cmd_match(buf, "backwards"))
4593 backwards = 1;
4594 else
4595 return -EINVAL;
4596 if (mddev->reshape_backwards == backwards)
4597 return len;
4598
4599 /* check if we are allowed to change */
4600 if (mddev->delta_disks)
4601 return -EBUSY;
4602
4603 if (mddev->persistent &&
4604 mddev->major_version == 0)
4605 return -EINVAL;
4606
4607 mddev->reshape_backwards = backwards;
4608 return len;
4609}
4610
4611static struct md_sysfs_entry md_reshape_direction =
4612__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4613 reshape_direction_store);
4614
4615static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004616array_size_show(struct mddev *mddev, char *page)
Dan Williamsb522adc2009-03-31 15:00:31 +11004617{
4618 if (mddev->external_size)
4619 return sprintf(page, "%llu\n",
4620 (unsigned long long)mddev->array_sectors/2);
4621 else
4622 return sprintf(page, "default\n");
4623}
4624
4625static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004626array_size_store(struct mddev *mddev, const char *buf, size_t len)
Dan Williamsb522adc2009-03-31 15:00:31 +11004627{
4628 sector_t sectors;
4629
4630 if (strncmp(buf, "default", 7) == 0) {
4631 if (mddev->pers)
4632 sectors = mddev->pers->size(mddev, 0, 0);
4633 else
4634 sectors = mddev->array_sectors;
4635
4636 mddev->external_size = 0;
4637 } else {
4638 if (strict_blocks_to_sectors(buf, &sectors) < 0)
4639 return -EINVAL;
4640 if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
NeilBrown2b69c832009-05-26 09:41:17 +10004641 return -E2BIG;
Dan Williamsb522adc2009-03-31 15:00:31 +11004642
4643 mddev->external_size = 1;
4644 }
4645
4646 mddev->array_sectors = sectors;
NeilBrowncbe6ef12011-02-16 13:58:38 +11004647 if (mddev->pers) {
4648 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10004649 revalidate_disk(mddev->gendisk);
NeilBrowncbe6ef12011-02-16 13:58:38 +11004650 }
Dan Williamsb522adc2009-03-31 15:00:31 +11004651 return len;
4652}
4653
4654static struct md_sysfs_entry md_array_size =
4655__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4656 array_size_store);
NeilBrowne464eaf2006-03-27 01:18:14 -08004657
NeilBrowneae17012005-11-08 21:39:23 -08004658static struct attribute *md_default_attrs[] = {
4659 &md_level.attr,
NeilBrownd4dbd022006-06-26 00:27:59 -07004660 &md_layout.attr,
NeilBrowneae17012005-11-08 21:39:23 -08004661 &md_raid_disks.attr,
NeilBrown3b343802006-01-06 00:20:47 -08004662 &md_chunk_size.attr,
NeilBrowna35b0d62006-01-06 00:20:49 -08004663 &md_size.attr,
NeilBrowna94213b2006-06-26 00:28:00 -07004664 &md_resync_start.attr,
NeilBrown8bb93aa2006-01-06 00:20:50 -08004665 &md_metadata.attr,
NeilBrown6d7ff732006-01-06 00:21:16 -08004666 &md_new_device.attr,
NeilBrown16f17b32006-06-26 00:27:37 -07004667 &md_safe_delay.attr,
NeilBrown9e653b62006-06-26 00:27:58 -07004668 &md_array_state.attr,
NeilBrown08a02ec2007-05-09 02:35:38 -07004669 &md_reshape_position.attr,
NeilBrown2c810cd2012-05-21 09:27:00 +10004670 &md_reshape_direction.attr,
Dan Williamsb522adc2009-03-31 15:00:31 +11004671 &md_array_size.attr,
Robert Becker1e509152009-12-14 12:49:58 +11004672 &max_corr_read_errors.attr,
NeilBrown411036f2005-11-08 21:39:40 -08004673 NULL,
4674};
4675
4676static struct attribute *md_redundancy_attrs[] = {
NeilBrown24dd4692005-11-08 21:39:26 -08004677 &md_scan_mode.attr,
NeilBrown9d888832005-11-08 21:39:26 -08004678 &md_mismatches.attr,
NeilBrown88202a02006-01-06 00:21:36 -08004679 &md_sync_min.attr,
4680 &md_sync_max.attr,
4681 &md_sync_speed.attr,
Bernd Schubert90b08712008-05-23 13:04:38 -07004682 &md_sync_force_parallel.attr,
NeilBrown88202a02006-01-06 00:21:36 -08004683 &md_sync_completed.attr,
Neil Brown5e96ee62008-06-28 08:31:24 +10004684 &md_min_sync.attr,
NeilBrownc6207272008-02-06 01:39:52 -08004685 &md_max_sync.attr,
NeilBrowne464eaf2006-03-27 01:18:14 -08004686 &md_suspend_lo.attr,
4687 &md_suspend_hi.attr,
Paul Clements9b1d1da2006-10-03 01:15:49 -07004688 &md_bitmap.attr,
Iustin Popd7f3d292007-10-16 23:30:54 -07004689 &md_degraded.attr,
NeilBrowneae17012005-11-08 21:39:23 -08004690 NULL,
4691};
NeilBrown411036f2005-11-08 21:39:40 -08004692static struct attribute_group md_redundancy_group = {
4693 .name = NULL,
4694 .attrs = md_redundancy_attrs,
4695};
4696
NeilBrowneae17012005-11-08 21:39:23 -08004697
4698static ssize_t
4699md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4700{
4701 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
NeilBrownfd01b882011-10-11 16:47:53 +11004702 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
NeilBrown96de1e62005-11-08 21:39:39 -08004703 ssize_t rv;
NeilBrowneae17012005-11-08 21:39:23 -08004704
4705 if (!entry->show)
4706 return -EIO;
NeilBrownaf8a2432011-12-08 15:49:46 +11004707 spin_lock(&all_mddevs_lock);
4708 if (list_empty(&mddev->all_mddevs)) {
4709 spin_unlock(&all_mddevs_lock);
4710 return -EBUSY;
4711 }
4712 mddev_get(mddev);
4713 spin_unlock(&all_mddevs_lock);
4714
Ingo Molnar5dc5cf72006-04-20 02:43:23 -07004715 rv = mddev_lock(mddev);
4716 if (!rv) {
4717 rv = entry->show(mddev, page);
4718 mddev_unlock(mddev);
4719 }
NeilBrownaf8a2432011-12-08 15:49:46 +11004720 mddev_put(mddev);
NeilBrown96de1e62005-11-08 21:39:39 -08004721 return rv;
NeilBrowneae17012005-11-08 21:39:23 -08004722}
4723
4724static ssize_t
4725md_attr_store(struct kobject *kobj, struct attribute *attr,
4726 const char *page, size_t length)
4727{
4728 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
NeilBrownfd01b882011-10-11 16:47:53 +11004729 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
NeilBrown96de1e62005-11-08 21:39:39 -08004730 ssize_t rv;
NeilBrowneae17012005-11-08 21:39:23 -08004731
4732 if (!entry->store)
4733 return -EIO;
NeilBrown67463ac2006-07-10 04:44:19 -07004734 if (!capable(CAP_SYS_ADMIN))
4735 return -EACCES;
NeilBrownaf8a2432011-12-08 15:49:46 +11004736 spin_lock(&all_mddevs_lock);
4737 if (list_empty(&mddev->all_mddevs)) {
4738 spin_unlock(&all_mddevs_lock);
4739 return -EBUSY;
4740 }
4741 mddev_get(mddev);
4742 spin_unlock(&all_mddevs_lock);
Ingo Molnar5dc5cf72006-04-20 02:43:23 -07004743 rv = mddev_lock(mddev);
4744 if (!rv) {
4745 rv = entry->store(mddev, page, length);
4746 mddev_unlock(mddev);
4747 }
NeilBrownaf8a2432011-12-08 15:49:46 +11004748 mddev_put(mddev);
NeilBrown96de1e62005-11-08 21:39:39 -08004749 return rv;
NeilBrowneae17012005-11-08 21:39:23 -08004750}
4751
4752static void md_free(struct kobject *ko)
4753{
NeilBrownfd01b882011-10-11 16:47:53 +11004754 struct mddev *mddev = container_of(ko, struct mddev, kobj);
NeilBrowna21d1502009-01-09 08:31:09 +11004755
4756 if (mddev->sysfs_state)
4757 sysfs_put(mddev->sysfs_state);
4758
4759 if (mddev->gendisk) {
4760 del_gendisk(mddev->gendisk);
4761 put_disk(mddev->gendisk);
4762 }
4763 if (mddev->queue)
4764 blk_cleanup_queue(mddev->queue);
4765
NeilBrowneae17012005-11-08 21:39:23 -08004766 kfree(mddev);
4767}
4768
Emese Revfy52cf25d2010-01-19 02:58:23 +01004769static const struct sysfs_ops md_sysfs_ops = {
NeilBrowneae17012005-11-08 21:39:23 -08004770 .show = md_attr_show,
4771 .store = md_attr_store,
4772};
4773static struct kobj_type md_ktype = {
4774 .release = md_free,
4775 .sysfs_ops = &md_sysfs_ops,
4776 .default_attrs = md_default_attrs,
4777};
4778
Linus Torvalds1da177e2005-04-16 15:20:36 -07004779int mdp_major = 0;
4780
Dan Williams5fd3a172009-03-04 00:57:25 -07004781static void mddev_delayed_delete(struct work_struct *ws)
4782{
NeilBrownfd01b882011-10-11 16:47:53 +11004783 struct mddev *mddev = container_of(ws, struct mddev, del_work);
Dan Williams5fd3a172009-03-04 00:57:25 -07004784
NeilBrown43a70502009-12-14 12:49:55 +11004785 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
Dan Williams5fd3a172009-03-04 00:57:25 -07004786 kobject_del(&mddev->kobj);
4787 kobject_put(&mddev->kobj);
4788}
4789
NeilBrownefeb53c2009-01-09 08:31:10 +11004790static int md_alloc(dev_t dev, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004791{
Arjan van de Ven48c9c272006-03-27 01:18:20 -08004792 static DEFINE_MUTEX(disks_mutex);
NeilBrownfd01b882011-10-11 16:47:53 +11004793 struct mddev *mddev = mddev_find(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004794 struct gendisk *disk;
NeilBrownefeb53c2009-01-09 08:31:10 +11004795 int partitioned;
4796 int shift;
4797 int unit;
Greg Kroah-Hartman3830c622007-12-17 15:54:39 -04004798 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004799
4800 if (!mddev)
NeilBrownefeb53c2009-01-09 08:31:10 +11004801 return -ENODEV;
4802
4803 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4804 shift = partitioned ? MdpMinorShift : 0;
4805 unit = MINOR(mddev->unit) >> shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004806
Tejun Heoe804ac72010-10-15 15:36:08 +02004807 /* wait for any previous instance of this device to be
4808 * completely removed (mddev_delayed_delete).
NeilBrownd3374822009-01-09 08:31:10 +11004809 */
Tejun Heoe804ac72010-10-15 15:36:08 +02004810 flush_workqueue(md_misc_wq);
NeilBrownd3374822009-01-09 08:31:10 +11004811
Arjan van de Ven48c9c272006-03-27 01:18:20 -08004812 mutex_lock(&disks_mutex);
NeilBrown0909dc42009-07-01 12:27:21 +10004813 error = -EEXIST;
4814 if (mddev->gendisk)
4815 goto abort;
NeilBrownefeb53c2009-01-09 08:31:10 +11004816
4817 if (name) {
4818 /* Need to ensure that 'name' is not a duplicate.
4819 */
NeilBrownfd01b882011-10-11 16:47:53 +11004820 struct mddev *mddev2;
NeilBrownefeb53c2009-01-09 08:31:10 +11004821 spin_lock(&all_mddevs_lock);
4822
4823 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4824 if (mddev2->gendisk &&
4825 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4826 spin_unlock(&all_mddevs_lock);
NeilBrown0909dc42009-07-01 12:27:21 +10004827 goto abort;
NeilBrownefeb53c2009-01-09 08:31:10 +11004828 }
4829 spin_unlock(&all_mddevs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004830 }
NeilBrown8b765392009-01-09 08:31:08 +11004831
NeilBrown0909dc42009-07-01 12:27:21 +10004832 error = -ENOMEM;
NeilBrown8b765392009-01-09 08:31:08 +11004833 mddev->queue = blk_alloc_queue(GFP_KERNEL);
NeilBrown0909dc42009-07-01 12:27:21 +10004834 if (!mddev->queue)
4835 goto abort;
NeilBrown409c57f2009-03-31 14:39:39 +11004836 mddev->queue->queuedata = mddev;
4837
NeilBrown409c57f2009-03-31 14:39:39 +11004838 blk_queue_make_request(mddev->queue, md_make_request);
Martin K. Petersenb1bd0552012-01-11 16:27:11 +01004839 blk_set_stacking_limits(&mddev->queue->limits);
NeilBrown8b765392009-01-09 08:31:08 +11004840
Linus Torvalds1da177e2005-04-16 15:20:36 -07004841 disk = alloc_disk(1 << shift);
4842 if (!disk) {
NeilBrown8b765392009-01-09 08:31:08 +11004843 blk_cleanup_queue(mddev->queue);
4844 mddev->queue = NULL;
NeilBrown0909dc42009-07-01 12:27:21 +10004845 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004846 }
NeilBrownefeb53c2009-01-09 08:31:10 +11004847 disk->major = MAJOR(mddev->unit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004848 disk->first_minor = unit << shift;
NeilBrownefeb53c2009-01-09 08:31:10 +11004849 if (name)
4850 strcpy(disk->disk_name, name);
4851 else if (partitioned)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004852 sprintf(disk->disk_name, "md_d%d", unit);
Greg Kroah-Hartmance7b0f42005-06-20 21:15:16 -07004853 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07004854 sprintf(disk->disk_name, "md%d", unit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004855 disk->fops = &md_fops;
4856 disk->private_data = mddev;
4857 disk->queue = mddev->queue;
NeilBrownb0140892011-05-10 17:49:01 +10004858 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
NeilBrown92850bb2008-10-21 13:25:32 +11004859 /* Allow extended partitions. This makes the
NeilBrownd3374822009-01-09 08:31:10 +11004860 * 'mdp' device redundant, but we can't really
NeilBrown92850bb2008-10-21 13:25:32 +11004861 * remove it now.
4862 */
4863 disk->flags |= GENHD_FL_EXT_DEVT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004864 mddev->gendisk = disk;
NeilBrownb0140892011-05-10 17:49:01 +10004865 /* As soon as we call add_disk(), another thread could get
4866 * through to md_open, so make sure it doesn't get too far
4867 */
4868 mutex_lock(&mddev->open_mutex);
4869 add_disk(disk);
4870
Tejun Heoed9e1982008-08-25 19:56:05 +09004871 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4872 &disk_to_dev(disk)->kobj, "%s", "md");
NeilBrown0909dc42009-07-01 12:27:21 +10004873 if (error) {
4874 /* This isn't possible, but as kobject_init_and_add is marked
4875 * __must_check, we must do something with the result
4876 */
NeilBrown5e55e2f2007-03-26 21:32:14 -08004877 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4878 disk->disk_name);
NeilBrown0909dc42009-07-01 12:27:21 +10004879 error = 0;
4880 }
NeilBrown00bcb4a2010-06-01 19:37:23 +10004881 if (mddev->kobj.sd &&
4882 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
NeilBrown43a70502009-12-14 12:49:55 +11004883 printk(KERN_DEBUG "pointless warning\n");
NeilBrownb0140892011-05-10 17:49:01 +10004884 mutex_unlock(&mddev->open_mutex);
NeilBrown0909dc42009-07-01 12:27:21 +10004885 abort:
4886 mutex_unlock(&disks_mutex);
NeilBrown00bcb4a2010-06-01 19:37:23 +10004887 if (!error && mddev->kobj.sd) {
Greg Kroah-Hartman3830c622007-12-17 15:54:39 -04004888 kobject_uevent(&mddev->kobj, KOBJ_ADD);
NeilBrown00bcb4a2010-06-01 19:37:23 +10004889 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
NeilBrownb62b7592008-10-21 13:25:21 +11004890 }
NeilBrownd3374822009-01-09 08:31:10 +11004891 mddev_put(mddev);
NeilBrown0909dc42009-07-01 12:27:21 +10004892 return error;
NeilBrownefeb53c2009-01-09 08:31:10 +11004893}
4894
4895static struct kobject *md_probe(dev_t dev, int *part, void *data)
4896{
4897 md_alloc(dev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004898 return NULL;
4899}
4900
NeilBrownefeb53c2009-01-09 08:31:10 +11004901static int add_named_array(const char *val, struct kernel_param *kp)
4902{
4903 /* val must be "md_*" where * is not all digits.
4904 * We allocate an array with a large free minor number, and
4905 * set the name to val. val must not already be an active name.
4906 */
4907 int len = strlen(val);
4908 char buf[DISK_NAME_LEN];
4909
4910 while (len && val[len-1] == '\n')
4911 len--;
4912 if (len >= DISK_NAME_LEN)
4913 return -E2BIG;
4914 strlcpy(buf, val, len+1);
4915 if (strncmp(buf, "md_", 3) != 0)
4916 return -EINVAL;
4917 return md_alloc(0, buf);
4918}
4919
Linus Torvalds1da177e2005-04-16 15:20:36 -07004920static void md_safemode_timeout(unsigned long data)
4921{
NeilBrownfd01b882011-10-11 16:47:53 +11004922 struct mddev *mddev = (struct mddev *) data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004923
Neil Brown0fd62b82008-06-28 08:31:36 +10004924 if (!atomic_read(&mddev->writes_pending)) {
4925 mddev->safemode = 1;
4926 if (mddev->external)
NeilBrown00bcb4a2010-06-01 19:37:23 +10004927 sysfs_notify_dirent_safe(mddev->sysfs_state);
Neil Brown0fd62b82008-06-28 08:31:36 +10004928 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004929 md_wakeup_thread(mddev->thread);
4930}
4931
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08004932static int start_dirty_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004933
NeilBrownfd01b882011-10-11 16:47:53 +11004934int md_run(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004935{
NeilBrown2604b702006-01-06 00:20:36 -08004936 int err;
NeilBrown3cb03002011-10-11 16:45:26 +11004937 struct md_rdev *rdev;
NeilBrown84fc4b52011-10-11 16:49:58 +11004938 struct md_personality *pers;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004939
NeilBrowna757e642005-04-16 15:26:42 -07004940 if (list_empty(&mddev->disks))
4941 /* cannot run an array with no devices.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004942 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004943
4944 if (mddev->pers)
4945 return -EBUSY;
NeilBrownbb4f1e92010-08-08 21:18:03 +10004946 /* Cannot run until previous stop completes properly */
4947 if (mddev->sysfs_active)
4948 return -EBUSY;
NeilBrownb6eb1272010-04-15 10:13:47 +10004949
Linus Torvalds1da177e2005-04-16 15:20:36 -07004950 /*
4951 * Analyze all RAID superblock(s)
4952 */
NeilBrown1ec4a932008-02-06 01:39:53 -08004953 if (!mddev->raid_disks) {
4954 if (!mddev->persistent)
4955 return -EINVAL;
NeilBrowna757e642005-04-16 15:26:42 -07004956 analyze_sbs(mddev);
NeilBrown1ec4a932008-02-06 01:39:53 -08004957 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004958
NeilBrownd9d166c2006-01-06 00:20:51 -08004959 if (mddev->level != LEVEL_NONE)
4960 request_module("md-level-%d", mddev->level);
4961 else if (mddev->clevel[0])
4962 request_module("md-%s", mddev->clevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004963
4964 /*
4965 * Drop all container device buffers, from now on
4966 * the only valid external interface is through the md
4967 * device.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004968 */
NeilBrowndafb20f2012-03-19 12:46:39 +11004969 rdev_for_each(rdev, mddev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08004970 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004971 continue;
4972 sync_blockdev(rdev->bdev);
Peter Zijlstraf98393a2007-05-06 14:49:54 -07004973 invalidate_bdev(rdev->bdev);
NeilBrownf0d76d72007-07-17 04:06:12 -07004974
4975 /* perform some consistency tests on the device.
4976 * We don't want the data to overlap the metadata,
Andre Noll58c0fed2009-03-31 14:33:13 +11004977 * Internal Bitmap issues have been handled elsewhere.
NeilBrownf0d76d72007-07-17 04:06:12 -07004978 */
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +11004979 if (rdev->meta_bdev) {
4980 /* Nothing to check */;
4981 } else if (rdev->data_offset < rdev->sb_start) {
Andre Noll58c0fed2009-03-31 14:33:13 +11004982 if (mddev->dev_sectors &&
4983 rdev->data_offset + mddev->dev_sectors
Andre Noll0f420352008-07-11 22:02:23 +10004984 > rdev->sb_start) {
NeilBrownf0d76d72007-07-17 04:06:12 -07004985 printk("md: %s: data overlaps metadata\n",
4986 mdname(mddev));
4987 return -EINVAL;
4988 }
4989 } else {
Andre Noll0f420352008-07-11 22:02:23 +10004990 if (rdev->sb_start + rdev->sb_size/512
NeilBrownf0d76d72007-07-17 04:06:12 -07004991 > rdev->data_offset) {
4992 printk("md: %s: metadata overlaps data\n",
4993 mdname(mddev));
4994 return -EINVAL;
4995 }
4996 }
NeilBrown00bcb4a2010-06-01 19:37:23 +10004997 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004998 }
4999
NeilBrowna167f662010-10-26 18:31:13 +11005000 if (mddev->bio_set == NULL)
Kent Overstreet395c72a2012-09-06 15:34:55 -07005001 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
NeilBrowna167f662010-10-26 18:31:13 +11005002
Linus Torvalds1da177e2005-04-16 15:20:36 -07005003 spin_lock(&pers_lock);
NeilBrownd9d166c2006-01-06 00:20:51 -08005004 pers = find_pers(mddev->level, mddev->clevel);
NeilBrown2604b702006-01-06 00:20:36 -08005005 if (!pers || !try_module_get(pers->owner)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005006 spin_unlock(&pers_lock);
NeilBrownd9d166c2006-01-06 00:20:51 -08005007 if (mddev->level != LEVEL_NONE)
5008 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5009 mddev->level);
5010 else
5011 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5012 mddev->clevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005013 return -EINVAL;
5014 }
NeilBrown2604b702006-01-06 00:20:36 -08005015 mddev->pers = pers;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005016 spin_unlock(&pers_lock);
NeilBrown34817e82009-03-31 14:39:38 +11005017 if (mddev->level != pers->level) {
5018 mddev->level = pers->level;
5019 mddev->new_level = pers->level;
5020 }
NeilBrownd9d166c2006-01-06 00:20:51 -08005021 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005022
NeilBrownf6705572006-03-27 01:18:11 -08005023 if (mddev->reshape_position != MaxSector &&
NeilBrown63c70c42006-03-27 01:18:13 -08005024 pers->start_reshape == NULL) {
NeilBrownf6705572006-03-27 01:18:11 -08005025 /* This personality cannot handle reshaping... */
5026 mddev->pers = NULL;
5027 module_put(pers->owner);
5028 return -EINVAL;
5029 }
5030
NeilBrown7dd5e7c32007-02-28 20:11:35 -08005031 if (pers->sync_request) {
5032 /* Warn if this is a potentially silly
5033 * configuration.
5034 */
5035 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11005036 struct md_rdev *rdev2;
NeilBrown7dd5e7c32007-02-28 20:11:35 -08005037 int warned = 0;
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005038
NeilBrowndafb20f2012-03-19 12:46:39 +11005039 rdev_for_each(rdev, mddev)
5040 rdev_for_each(rdev2, mddev) {
NeilBrown7dd5e7c32007-02-28 20:11:35 -08005041 if (rdev < rdev2 &&
5042 rdev->bdev->bd_contains ==
5043 rdev2->bdev->bd_contains) {
5044 printk(KERN_WARNING
5045 "%s: WARNING: %s appears to be"
5046 " on the same physical disk as"
5047 " %s.\n",
5048 mdname(mddev),
5049 bdevname(rdev->bdev,b),
5050 bdevname(rdev2->bdev,b2));
5051 warned = 1;
5052 }
5053 }
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005054
NeilBrown7dd5e7c32007-02-28 20:11:35 -08005055 if (warned)
5056 printk(KERN_WARNING
5057 "True protection against single-disk"
5058 " failure might be compromised.\n");
5059 }
5060
NeilBrown657390d2005-08-26 18:34:16 -07005061 mddev->recovery = 0;
Andre Noll58c0fed2009-03-31 14:33:13 +11005062 /* may be over-ridden by personality */
5063 mddev->resync_max_sectors = mddev->dev_sectors;
5064
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005065 mddev->ok_start_degraded = start_dirty_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005066
NeilBrown0f9552b2009-12-30 12:08:50 +11005067 if (start_readonly && mddev->ro == 0)
NeilBrownf91de922005-11-08 21:39:36 -08005068 mddev->ro = 2; /* read-only, but switch on first write */
5069
NeilBrownb15c2e52006-01-06 00:20:16 -08005070 err = mddev->pers->run(mddev);
Andre Noll13e53df2008-03-26 00:07:03 +01005071 if (err)
5072 printk(KERN_ERR "md: pers->run() failed ...\n");
Dan Williamsb522adc2009-03-31 15:00:31 +11005073 else if (mddev->pers->size(mddev, 0, 0) < mddev->array_sectors) {
5074 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5075 " but 'external_size' not in effect?\n", __func__);
5076 printk(KERN_ERR
5077 "md: invalid array_size %llu > default size %llu\n",
5078 (unsigned long long)mddev->array_sectors / 2,
5079 (unsigned long long)mddev->pers->size(mddev, 0, 0) / 2);
5080 err = -EINVAL;
5081 mddev->pers->stop(mddev);
5082 }
NeilBrownef99bf42012-05-22 13:55:08 +10005083 if (err == 0 && mddev->pers->sync_request &&
5084 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
NeilBrownb15c2e52006-01-06 00:20:16 -08005085 err = bitmap_create(mddev);
5086 if (err) {
5087 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5088 mdname(mddev), err);
5089 mddev->pers->stop(mddev);
5090 }
5091 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005092 if (err) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005093 module_put(mddev->pers->owner);
5094 mddev->pers = NULL;
NeilBrown32a76272005-06-21 17:17:14 -07005095 bitmap_destroy(mddev);
5096 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005097 }
NeilBrown5e55e2f2007-03-26 21:32:14 -08005098 if (mddev->pers->sync_request) {
NeilBrown00bcb4a2010-06-01 19:37:23 +10005099 if (mddev->kobj.sd &&
5100 sysfs_create_group(&mddev->kobj, &md_redundancy_group))
NeilBrown5e55e2f2007-03-26 21:32:14 -08005101 printk(KERN_WARNING
5102 "md: cannot register extra attributes for %s\n",
5103 mdname(mddev));
NeilBrown00bcb4a2010-06-01 19:37:23 +10005104 mddev->sysfs_action = sysfs_get_dirent_safe(mddev->kobj.sd, "sync_action");
NeilBrown5e55e2f2007-03-26 21:32:14 -08005105 } else if (mddev->ro == 2) /* auto-readonly not meaningful */
NeilBrownfd9d49c2005-11-08 21:39:42 -08005106 mddev->ro = 0;
5107
Linus Torvalds1da177e2005-04-16 15:20:36 -07005108 atomic_set(&mddev->writes_pending,0);
Robert Becker1e509152009-12-14 12:49:58 +11005109 atomic_set(&mddev->max_corr_read_errors,
5110 MD_DEFAULT_MAX_CORRECTED_READ_ERRORS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005111 mddev->safemode = 0;
5112 mddev->safemode_timer.function = md_safemode_timeout;
5113 mddev->safemode_timer.data = (unsigned long) mddev;
NeilBrown16f17b32006-06-26 00:27:37 -07005114 mddev->safemode_delay = (200 * HZ)/1000 +1; /* 200 msec delay */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005115 mddev->in_sync = 1;
NeilBrown0ca69882011-01-14 09:14:33 +11005116 smp_wmb();
5117 mddev->ready = 1;
NeilBrowndafb20f2012-03-19 12:46:39 +11005118 rdev_for_each(rdev, mddev)
Namhyung Kim36fad852011-07-27 11:00:36 +10005119 if (rdev->raid_disk >= 0)
5120 if (sysfs_link_rdev(mddev, rdev))
NeilBrown00bcb4a2010-06-01 19:37:23 +10005121 /* failure here is OK */;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005122
5123 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5124
NeilBrown850b2b42006-10-03 01:15:46 -07005125 if (mddev->flags)
5126 md_update_sb(mddev, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005127
NeilBrownd7603b72006-01-06 00:20:30 -08005128 md_new_event(mddev);
NeilBrown00bcb4a2010-06-01 19:37:23 +10005129 sysfs_notify_dirent_safe(mddev->sysfs_state);
5130 sysfs_notify_dirent_safe(mddev->sysfs_action);
Neil Browna99ac972008-06-28 08:31:43 +10005131 sysfs_notify(&mddev->kobj, NULL, "degraded");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005132 return 0;
5133}
NeilBrown390ee602010-06-01 19:37:27 +10005134EXPORT_SYMBOL_GPL(md_run);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005135
NeilBrownfd01b882011-10-11 16:47:53 +11005136static int do_md_run(struct mddev *mddev)
NeilBrownfe60b012010-03-29 11:10:42 +11005137{
5138 int err;
5139
5140 err = md_run(mddev);
5141 if (err)
5142 goto out;
NeilBrown69e51b42010-06-01 19:37:35 +10005143 err = bitmap_load(mddev);
5144 if (err) {
5145 bitmap_destroy(mddev);
5146 goto out;
5147 }
Jonathan Brassow0fd018a2011-06-07 17:49:36 -05005148
5149 md_wakeup_thread(mddev->thread);
5150 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5151
NeilBrownfe60b012010-03-29 11:10:42 +11005152 set_capacity(mddev->gendisk, mddev->array_sectors);
5153 revalidate_disk(mddev->gendisk);
NeilBrownf0b4f7e2011-02-24 17:26:41 +11005154 mddev->changed = 1;
NeilBrownfe60b012010-03-29 11:10:42 +11005155 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5156out:
5157 return err;
5158}
5159
NeilBrownfd01b882011-10-11 16:47:53 +11005160static int restart_array(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005161{
5162 struct gendisk *disk = mddev->gendisk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005163
Andre Noll80fab1d2008-07-11 22:02:21 +10005164 /* Complain if it has no devices */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005165 if (list_empty(&mddev->disks))
Andre Noll80fab1d2008-07-11 22:02:21 +10005166 return -ENXIO;
5167 if (!mddev->pers)
5168 return -EINVAL;
5169 if (!mddev->ro)
5170 return -EBUSY;
5171 mddev->safemode = 0;
5172 mddev->ro = 0;
5173 set_disk_ro(disk, 0);
5174 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5175 mdname(mddev));
5176 /* Kick recovery or resync if necessary */
5177 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5178 md_wakeup_thread(mddev->thread);
5179 md_wakeup_thread(mddev->sync_thread);
NeilBrown00bcb4a2010-06-01 19:37:23 +10005180 sysfs_notify_dirent_safe(mddev->sysfs_state);
Andre Noll80fab1d2008-07-11 22:02:21 +10005181 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005182}
5183
NeilBrownacc55e22006-06-26 00:27:47 -07005184/* similar to deny_write_access, but accounts for our holding a reference
5185 * to the file ourselves */
5186static int deny_bitmap_write_access(struct file * file)
5187{
5188 struct inode *inode = file->f_mapping->host;
5189
5190 spin_lock(&inode->i_lock);
5191 if (atomic_read(&inode->i_writecount) > 1) {
5192 spin_unlock(&inode->i_lock);
5193 return -ETXTBSY;
5194 }
5195 atomic_set(&inode->i_writecount, -1);
5196 spin_unlock(&inode->i_lock);
5197
5198 return 0;
5199}
5200
NeilBrown43a70502009-12-14 12:49:55 +11005201void restore_bitmap_write_access(struct file *file)
NeilBrownacc55e22006-06-26 00:27:47 -07005202{
5203 struct inode *inode = file->f_mapping->host;
5204
5205 spin_lock(&inode->i_lock);
5206 atomic_set(&inode->i_writecount, 1);
5207 spin_unlock(&inode->i_lock);
5208}
5209
NeilBrownfd01b882011-10-11 16:47:53 +11005210static void md_clean(struct mddev *mddev)
NeilBrown6177b472010-03-29 11:37:13 +11005211{
5212 mddev->array_sectors = 0;
5213 mddev->external_size = 0;
5214 mddev->dev_sectors = 0;
5215 mddev->raid_disks = 0;
5216 mddev->recovery_cp = 0;
5217 mddev->resync_min = 0;
5218 mddev->resync_max = MaxSector;
5219 mddev->reshape_position = MaxSector;
5220 mddev->external = 0;
5221 mddev->persistent = 0;
5222 mddev->level = LEVEL_NONE;
5223 mddev->clevel[0] = 0;
5224 mddev->flags = 0;
5225 mddev->ro = 0;
5226 mddev->metadata_type[0] = 0;
5227 mddev->chunk_sectors = 0;
5228 mddev->ctime = mddev->utime = 0;
5229 mddev->layout = 0;
5230 mddev->max_disks = 0;
5231 mddev->events = 0;
NeilBrowna8707c02010-05-18 09:28:43 +10005232 mddev->can_decrease_events = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005233 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10005234 mddev->reshape_backwards = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005235 mddev->new_level = LEVEL_NONE;
5236 mddev->new_layout = 0;
5237 mddev->new_chunk_sectors = 0;
5238 mddev->curr_resync = 0;
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11005239 atomic64_set(&mddev->resync_mismatches, 0);
NeilBrown6177b472010-03-29 11:37:13 +11005240 mddev->suspend_lo = mddev->suspend_hi = 0;
5241 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5242 mddev->recovery = 0;
5243 mddev->in_sync = 0;
NeilBrownf0b4f7e2011-02-24 17:26:41 +11005244 mddev->changed = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005245 mddev->degraded = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005246 mddev->safemode = 0;
NeilBrown050b6612012-03-19 12:46:39 +11005247 mddev->merge_check_needed = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005248 mddev->bitmap_info.offset = 0;
5249 mddev->bitmap_info.default_offset = 0;
NeilBrown6409bb02012-05-22 13:55:07 +10005250 mddev->bitmap_info.default_space = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005251 mddev->bitmap_info.chunksize = 0;
5252 mddev->bitmap_info.daemon_sleep = 0;
5253 mddev->bitmap_info.max_write_behind = 0;
5254}
5255
NeilBrownfd01b882011-10-11 16:47:53 +11005256static void __md_stop_writes(struct mddev *mddev)
NeilBrowna047e122010-03-29 12:07:53 +11005257{
5258 if (mddev->sync_thread) {
5259 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5260 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrown7ebc0be2011-01-14 09:14:33 +11005261 reap_sync_thread(mddev);
NeilBrowna047e122010-03-29 12:07:53 +11005262 }
5263
5264 del_timer_sync(&mddev->safemode_timer);
5265
5266 bitmap_flush(mddev);
5267 md_super_wait(mddev);
5268
5269 if (!mddev->in_sync || mddev->flags) {
5270 /* mark array as shutdown cleanly */
5271 mddev->in_sync = 1;
5272 md_update_sb(mddev, 1);
5273 }
5274}
NeilBrowndefad612011-01-14 09:14:33 +11005275
NeilBrownfd01b882011-10-11 16:47:53 +11005276void md_stop_writes(struct mddev *mddev)
NeilBrowndefad612011-01-14 09:14:33 +11005277{
5278 mddev_lock(mddev);
5279 __md_stop_writes(mddev);
5280 mddev_unlock(mddev);
5281}
NeilBrown390ee602010-06-01 19:37:27 +10005282EXPORT_SYMBOL_GPL(md_stop_writes);
NeilBrowna047e122010-03-29 12:07:53 +11005283
NeilBrownfd01b882011-10-11 16:47:53 +11005284void md_stop(struct mddev *mddev)
NeilBrown6177b472010-03-29 11:37:13 +11005285{
NeilBrown0ca69882011-01-14 09:14:33 +11005286 mddev->ready = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005287 mddev->pers->stop(mddev);
5288 if (mddev->pers->sync_request && mddev->to_remove == NULL)
5289 mddev->to_remove = &md_redundancy_group;
5290 module_put(mddev->pers->owner);
5291 mddev->pers = NULL;
NeilBrowncca9cf92010-04-01 12:08:16 +11005292 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
NeilBrown6177b472010-03-29 11:37:13 +11005293}
NeilBrown390ee602010-06-01 19:37:27 +10005294EXPORT_SYMBOL_GPL(md_stop);
NeilBrown6177b472010-03-29 11:37:13 +11005295
NeilBrowna05b7ea2012-07-19 15:59:18 +10005296static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
NeilBrowna4bd82d2010-03-29 13:23:10 +11005297{
5298 int err = 0;
5299 mutex_lock(&mddev->open_mutex);
NeilBrowna05b7ea2012-07-19 15:59:18 +10005300 if (atomic_read(&mddev->openers) > !!bdev) {
NeilBrowna4bd82d2010-03-29 13:23:10 +11005301 printk("md: %s still in use.\n",mdname(mddev));
5302 err = -EBUSY;
5303 goto out;
5304 }
NeilBrowna05b7ea2012-07-19 15:59:18 +10005305 if (bdev)
5306 sync_blockdev(bdev);
NeilBrowna4bd82d2010-03-29 13:23:10 +11005307 if (mddev->pers) {
NeilBrowndefad612011-01-14 09:14:33 +11005308 __md_stop_writes(mddev);
NeilBrowna4bd82d2010-03-29 13:23:10 +11005309
5310 err = -ENXIO;
5311 if (mddev->ro==1)
5312 goto out;
5313 mddev->ro = 1;
5314 set_disk_ro(mddev->gendisk, 1);
5315 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
NeilBrown00bcb4a2010-06-01 19:37:23 +10005316 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrowna4bd82d2010-03-29 13:23:10 +11005317 err = 0;
5318 }
5319out:
5320 mutex_unlock(&mddev->open_mutex);
5321 return err;
5322}
5323
NeilBrown9e653b62006-06-26 00:27:58 -07005324/* mode:
5325 * 0 - completely stop and dis-assemble array
NeilBrown9e653b62006-06-26 00:27:58 -07005326 * 2 - stop but do not disassemble array
5327 */
NeilBrowna05b7ea2012-07-19 15:59:18 +10005328static int do_md_stop(struct mddev * mddev, int mode,
5329 struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005330{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005331 struct gendisk *disk = mddev->gendisk;
NeilBrown3cb03002011-10-11 16:45:26 +11005332 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005333
NeilBrownc8c00a62009-08-10 12:50:52 +10005334 mutex_lock(&mddev->open_mutex);
NeilBrowna05b7ea2012-07-19 15:59:18 +10005335 if (atomic_read(&mddev->openers) > !!bdev ||
NeilBrownbb4f1e92010-08-08 21:18:03 +10005336 mddev->sysfs_active) {
Neil Browndf5b20c2008-07-11 22:02:22 +10005337 printk("md: %s still in use.\n",mdname(mddev));
NeilBrown6e17b022010-08-07 21:41:19 +10005338 mutex_unlock(&mddev->open_mutex);
5339 return -EBUSY;
5340 }
NeilBrowna05b7ea2012-07-19 15:59:18 +10005341 if (bdev)
5342 /* It is possible IO was issued on some other
5343 * open file which was closed before we took ->open_mutex.
5344 * As that was not the last close __blkdev_put will not
5345 * have called sync_blockdev, so we must.
5346 */
5347 sync_blockdev(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005348
NeilBrown6e17b022010-08-07 21:41:19 +10005349 if (mddev->pers) {
NeilBrowna4bd82d2010-03-29 13:23:10 +11005350 if (mddev->ro)
5351 set_disk_ro(disk, 0);
NeilBrown409c57f2009-03-31 14:39:39 +11005352
NeilBrowndefad612011-01-14 09:14:33 +11005353 __md_stop_writes(mddev);
NeilBrowna4bd82d2010-03-29 13:23:10 +11005354 md_stop(mddev);
5355 mddev->queue->merge_bvec_fn = NULL;
NeilBrowna4bd82d2010-03-29 13:23:10 +11005356 mddev->queue->backing_dev_info.congested_fn = NULL;
NeilBrown6177b472010-03-29 11:37:13 +11005357
NeilBrowna4bd82d2010-03-29 13:23:10 +11005358 /* tell userspace to handle 'inactive' */
NeilBrown00bcb4a2010-06-01 19:37:23 +10005359 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrown0d4ca602006-12-10 02:20:44 -08005360
NeilBrowndafb20f2012-03-19 12:46:39 +11005361 rdev_for_each(rdev, mddev)
Namhyung Kim36fad852011-07-27 11:00:36 +10005362 if (rdev->raid_disk >= 0)
5363 sysfs_unlink_rdev(mddev, rdev);
NeilBrownc4647292009-05-07 12:51:06 +10005364
NeilBrowna4bd82d2010-03-29 13:23:10 +11005365 set_capacity(disk, 0);
NeilBrown6e17b022010-08-07 21:41:19 +10005366 mutex_unlock(&mddev->open_mutex);
NeilBrownf0b4f7e2011-02-24 17:26:41 +11005367 mddev->changed = 1;
NeilBrowna4bd82d2010-03-29 13:23:10 +11005368 revalidate_disk(disk);
NeilBrown0d4ca602006-12-10 02:20:44 -08005369
NeilBrowna4bd82d2010-03-29 13:23:10 +11005370 if (mddev->ro)
5371 mddev->ro = 0;
NeilBrown6e17b022010-08-07 21:41:19 +10005372 } else
5373 mutex_unlock(&mddev->open_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005374 /*
5375 * Free resources if final stop
5376 */
NeilBrown9e653b62006-06-26 00:27:58 -07005377 if (mode == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005378 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5379
NeilBrown978f9462006-02-02 14:28:05 -08005380 bitmap_destroy(mddev);
NeilBrownc3d97142009-12-14 12:49:52 +11005381 if (mddev->bitmap_info.file) {
5382 restore_bitmap_write_access(mddev->bitmap_info.file);
5383 fput(mddev->bitmap_info.file);
5384 mddev->bitmap_info.file = NULL;
NeilBrown978f9462006-02-02 14:28:05 -08005385 }
NeilBrownc3d97142009-12-14 12:49:52 +11005386 mddev->bitmap_info.offset = 0;
NeilBrown978f9462006-02-02 14:28:05 -08005387
Linus Torvalds1da177e2005-04-16 15:20:36 -07005388 export_array(mddev);
5389
NeilBrown6177b472010-03-29 11:37:13 +11005390 md_clean(mddev);
NeilBrown934d9c22008-10-28 17:01:23 +11005391 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
NeilBrownefeb53c2009-01-09 08:31:10 +11005392 if (mddev->hold_active == UNTIL_STOP)
5393 mddev->hold_active = 0;
NeilBrowna4bd82d2010-03-29 13:23:10 +11005394 }
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11005395 blk_integrity_unregister(disk);
NeilBrownd7603b72006-01-06 00:20:30 -08005396 md_new_event(mddev);
NeilBrown00bcb4a2010-06-01 19:37:23 +10005397 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrown6e17b022010-08-07 21:41:19 +10005398 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005399}
5400
Jeff Garzikfdee8ae2006-12-10 02:20:50 -08005401#ifndef MODULE
NeilBrownfd01b882011-10-11 16:47:53 +11005402static void autorun_array(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005403{
NeilBrown3cb03002011-10-11 16:45:26 +11005404 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005405 int err;
5406
NeilBrowna757e642005-04-16 15:26:42 -07005407 if (list_empty(&mddev->disks))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005408 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005409
5410 printk(KERN_INFO "md: running: ");
5411
NeilBrowndafb20f2012-03-19 12:46:39 +11005412 rdev_for_each(rdev, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005413 char b[BDEVNAME_SIZE];
5414 printk("<%s>", bdevname(rdev->bdev,b));
5415 }
5416 printk("\n");
5417
NeilBrownd710e132008-10-13 11:55:12 +11005418 err = do_md_run(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005419 if (err) {
5420 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
NeilBrowna05b7ea2012-07-19 15:59:18 +10005421 do_md_stop(mddev, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005422 }
5423}
5424
5425/*
5426 * lets try to run arrays based on all disks that have arrived
5427 * until now. (those are in pending_raid_disks)
5428 *
5429 * the method: pick the first pending disk, collect all disks with
5430 * the same UUID, remove all from the pending list and put them into
5431 * the 'same_array' list. Then order this list based on superblock
5432 * update time (freshest comes first), kick out 'old' disks and
5433 * compare superblocks. If everything's fine then run it.
5434 *
5435 * If "unit" is allocated, then bump its reference count
5436 */
5437static void autorun_devices(int part)
5438{
NeilBrown3cb03002011-10-11 16:45:26 +11005439 struct md_rdev *rdev0, *rdev, *tmp;
NeilBrownfd01b882011-10-11 16:47:53 +11005440 struct mddev *mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005441 char b[BDEVNAME_SIZE];
5442
5443 printk(KERN_INFO "md: autorun ...\n");
5444 while (!list_empty(&pending_raid_disks)) {
NeilBrowne8703fe2006-10-03 01:15:59 -07005445 int unit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005446 dev_t dev;
NeilBrownad01c9e2006-03-27 01:18:07 -08005447 LIST_HEAD(candidates);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005448 rdev0 = list_entry(pending_raid_disks.next,
NeilBrown3cb03002011-10-11 16:45:26 +11005449 struct md_rdev, same_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005450
5451 printk(KERN_INFO "md: considering %s ...\n",
5452 bdevname(rdev0->bdev,b));
5453 INIT_LIST_HEAD(&candidates);
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005454 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005455 if (super_90_load(rdev, rdev0, 0) >= 0) {
5456 printk(KERN_INFO "md: adding %s ...\n",
5457 bdevname(rdev->bdev,b));
5458 list_move(&rdev->same_set, &candidates);
5459 }
5460 /*
5461 * now we have a set of devices, with all of them having
5462 * mostly sane superblocks. It's time to allocate the
5463 * mddev.
5464 */
NeilBrowne8703fe2006-10-03 01:15:59 -07005465 if (part) {
5466 dev = MKDEV(mdp_major,
5467 rdev0->preferred_minor << MdpMinorShift);
5468 unit = MINOR(dev) >> MdpMinorShift;
5469 } else {
5470 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5471 unit = MINOR(dev);
5472 }
5473 if (rdev0->preferred_minor != unit) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005474 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5475 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5476 break;
5477 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005478
5479 md_probe(dev, NULL, NULL);
5480 mddev = mddev_find(dev);
Neil Brown9bbbca32008-06-28 08:31:17 +10005481 if (!mddev || !mddev->gendisk) {
5482 if (mddev)
5483 mddev_put(mddev);
5484 printk(KERN_ERR
Linus Torvalds1da177e2005-04-16 15:20:36 -07005485 "md: cannot allocate memory for md drive.\n");
5486 break;
5487 }
5488 if (mddev_lock(mddev))
5489 printk(KERN_WARNING "md: %s locked, cannot run\n",
5490 mdname(mddev));
5491 else if (mddev->raid_disks || mddev->major_version
5492 || !list_empty(&mddev->disks)) {
5493 printk(KERN_WARNING
5494 "md: %s already running, cannot run %s\n",
5495 mdname(mddev), bdevname(rdev0->bdev,b));
5496 mddev_unlock(mddev);
5497 } else {
5498 printk(KERN_INFO "md: created %s\n", mdname(mddev));
NeilBrown1ec4a932008-02-06 01:39:53 -08005499 mddev->persistent = 1;
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005500 rdev_for_each_list(rdev, tmp, &candidates) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005501 list_del_init(&rdev->same_set);
5502 if (bind_rdev_to_array(rdev, mddev))
5503 export_rdev(rdev);
5504 }
5505 autorun_array(mddev);
5506 mddev_unlock(mddev);
5507 }
5508 /* on success, candidates will be empty, on error
5509 * it won't...
5510 */
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005511 rdev_for_each_list(rdev, tmp, &candidates) {
NeilBrown4b809912008-07-21 17:05:25 +10005512 list_del_init(&rdev->same_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005513 export_rdev(rdev);
NeilBrown4b809912008-07-21 17:05:25 +10005514 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005515 mddev_put(mddev);
5516 }
5517 printk(KERN_INFO "md: ... autorun DONE.\n");
5518}
Jeff Garzikfdee8ae2006-12-10 02:20:50 -08005519#endif /* !MODULE */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005520
Linus Torvalds1da177e2005-04-16 15:20:36 -07005521static int get_version(void __user * arg)
5522{
5523 mdu_version_t ver;
5524
5525 ver.major = MD_MAJOR_VERSION;
5526 ver.minor = MD_MINOR_VERSION;
5527 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5528
5529 if (copy_to_user(arg, &ver, sizeof(ver)))
5530 return -EFAULT;
5531
5532 return 0;
5533}
5534
NeilBrownfd01b882011-10-11 16:47:53 +11005535static int get_array_info(struct mddev * mddev, void __user * arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005536{
5537 mdu_array_info_t info;
NeilBrowna9f326e2009-09-23 18:06:41 +10005538 int nr,working,insync,failed,spare;
NeilBrown3cb03002011-10-11 16:45:26 +11005539 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005540
NeilBrown1ca69c42012-10-11 13:37:33 +11005541 nr = working = insync = failed = spare = 0;
5542 rcu_read_lock();
5543 rdev_for_each_rcu(rdev, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005544 nr++;
NeilBrownb2d444d2005-11-08 21:39:31 -08005545 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005546 failed++;
5547 else {
5548 working++;
NeilBrownb2d444d2005-11-08 21:39:31 -08005549 if (test_bit(In_sync, &rdev->flags))
NeilBrowna9f326e2009-09-23 18:06:41 +10005550 insync++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005551 else
5552 spare++;
5553 }
5554 }
NeilBrown1ca69c42012-10-11 13:37:33 +11005555 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005556
5557 info.major_version = mddev->major_version;
5558 info.minor_version = mddev->minor_version;
5559 info.patch_version = MD_PATCHLEVEL_VERSION;
5560 info.ctime = mddev->ctime;
5561 info.level = mddev->level;
Andre Noll58c0fed2009-03-31 14:33:13 +11005562 info.size = mddev->dev_sectors / 2;
5563 if (info.size != mddev->dev_sectors / 2) /* overflow */
NeilBrown284ae7c2006-02-03 03:03:40 -08005564 info.size = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005565 info.nr_disks = nr;
5566 info.raid_disks = mddev->raid_disks;
5567 info.md_minor = mddev->md_minor;
5568 info.not_persistent= !mddev->persistent;
5569
5570 info.utime = mddev->utime;
5571 info.state = 0;
5572 if (mddev->in_sync)
5573 info.state = (1<<MD_SB_CLEAN);
NeilBrownc3d97142009-12-14 12:49:52 +11005574 if (mddev->bitmap && mddev->bitmap_info.offset)
NeilBrown36fa3062005-09-09 16:23:45 -07005575 info.state = (1<<MD_SB_BITMAP_PRESENT);
NeilBrowna9f326e2009-09-23 18:06:41 +10005576 info.active_disks = insync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005577 info.working_disks = working;
5578 info.failed_disks = failed;
5579 info.spare_disks = spare;
5580
5581 info.layout = mddev->layout;
Andre Noll9d8f0362009-06-18 08:45:01 +10005582 info.chunk_size = mddev->chunk_sectors << 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005583
5584 if (copy_to_user(arg, &info, sizeof(info)))
5585 return -EFAULT;
5586
5587 return 0;
5588}
5589
NeilBrownfd01b882011-10-11 16:47:53 +11005590static int get_bitmap_file(struct mddev * mddev, void __user * arg)
NeilBrown32a76272005-06-21 17:17:14 -07005591{
5592 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
5593 char *ptr, *buf = NULL;
5594 int err = -ENOMEM;
5595
Dan Williamsb5470dc2008-06-27 21:44:04 -07005596 if (md_allow_write(mddev))
5597 file = kmalloc(sizeof(*file), GFP_NOIO);
5598 else
5599 file = kmalloc(sizeof(*file), GFP_KERNEL);
NeilBrown2a2275d2007-01-26 00:57:11 -08005600
NeilBrown32a76272005-06-21 17:17:14 -07005601 if (!file)
5602 goto out;
5603
5604 /* bitmap disabled, zero the first byte and copy out */
NeilBrown1ec885c2012-05-22 13:55:10 +10005605 if (!mddev->bitmap || !mddev->bitmap->storage.file) {
NeilBrown32a76272005-06-21 17:17:14 -07005606 file->pathname[0] = '\0';
5607 goto copy_out;
5608 }
5609
5610 buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
5611 if (!buf)
5612 goto out;
5613
NeilBrown1ec885c2012-05-22 13:55:10 +10005614 ptr = d_path(&mddev->bitmap->storage.file->f_path,
5615 buf, sizeof(file->pathname));
Christoph Hellwig6bcfd602008-05-23 13:04:34 -07005616 if (IS_ERR(ptr))
NeilBrown32a76272005-06-21 17:17:14 -07005617 goto out;
5618
5619 strcpy(file->pathname, ptr);
5620
5621copy_out:
5622 err = 0;
5623 if (copy_to_user(arg, file, sizeof(*file)))
5624 err = -EFAULT;
5625out:
5626 kfree(buf);
5627 kfree(file);
5628 return err;
5629}
5630
NeilBrownfd01b882011-10-11 16:47:53 +11005631static int get_disk_info(struct mddev * mddev, void __user * arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005632{
5633 mdu_disk_info_t info;
NeilBrown3cb03002011-10-11 16:45:26 +11005634 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005635
5636 if (copy_from_user(&info, arg, sizeof(info)))
5637 return -EFAULT;
5638
NeilBrown1ca69c42012-10-11 13:37:33 +11005639 rcu_read_lock();
5640 rdev = find_rdev_nr_rcu(mddev, info.number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005641 if (rdev) {
5642 info.major = MAJOR(rdev->bdev->bd_dev);
5643 info.minor = MINOR(rdev->bdev->bd_dev);
5644 info.raid_disk = rdev->raid_disk;
5645 info.state = 0;
NeilBrownb2d444d2005-11-08 21:39:31 -08005646 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005647 info.state |= (1<<MD_DISK_FAULTY);
NeilBrownb2d444d2005-11-08 21:39:31 -08005648 else if (test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005649 info.state |= (1<<MD_DISK_ACTIVE);
5650 info.state |= (1<<MD_DISK_SYNC);
5651 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -07005652 if (test_bit(WriteMostly, &rdev->flags))
5653 info.state |= (1<<MD_DISK_WRITEMOSTLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005654 } else {
5655 info.major = info.minor = 0;
5656 info.raid_disk = -1;
5657 info.state = (1<<MD_DISK_REMOVED);
5658 }
NeilBrown1ca69c42012-10-11 13:37:33 +11005659 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005660
5661 if (copy_to_user(arg, &info, sizeof(info)))
5662 return -EFAULT;
5663
5664 return 0;
5665}
5666
NeilBrownfd01b882011-10-11 16:47:53 +11005667static int add_new_disk(struct mddev * mddev, mdu_disk_info_t *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005668{
5669 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11005670 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005671 dev_t dev = MKDEV(info->major,info->minor);
5672
5673 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5674 return -EOVERFLOW;
5675
5676 if (!mddev->raid_disks) {
5677 int err;
5678 /* expecting a device which has a superblock */
5679 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5680 if (IS_ERR(rdev)) {
5681 printk(KERN_WARNING
5682 "md: md_import_device returned %ld\n",
5683 PTR_ERR(rdev));
5684 return PTR_ERR(rdev);
5685 }
5686 if (!list_empty(&mddev->disks)) {
NeilBrown3cb03002011-10-11 16:45:26 +11005687 struct md_rdev *rdev0
5688 = list_entry(mddev->disks.next,
5689 struct md_rdev, same_set);
NeilBrowna9f326e2009-09-23 18:06:41 +10005690 err = super_types[mddev->major_version]
Linus Torvalds1da177e2005-04-16 15:20:36 -07005691 .load_super(rdev, rdev0, mddev->minor_version);
5692 if (err < 0) {
5693 printk(KERN_WARNING
5694 "md: %s has different UUID to %s\n",
5695 bdevname(rdev->bdev,b),
5696 bdevname(rdev0->bdev,b2));
5697 export_rdev(rdev);
5698 return -EINVAL;
5699 }
5700 }
5701 err = bind_rdev_to_array(rdev, mddev);
5702 if (err)
5703 export_rdev(rdev);
5704 return err;
5705 }
5706
5707 /*
5708 * add_new_disk can be used once the array is assembled
5709 * to add "hot spares". They must already have a superblock
5710 * written
5711 */
5712 if (mddev->pers) {
5713 int err;
5714 if (!mddev->pers->hot_add_disk) {
5715 printk(KERN_WARNING
5716 "%s: personality does not support diskops!\n",
5717 mdname(mddev));
5718 return -EINVAL;
5719 }
NeilBrown7b1e35f2005-09-09 16:23:50 -07005720 if (mddev->persistent)
5721 rdev = md_import_device(dev, mddev->major_version,
5722 mddev->minor_version);
5723 else
5724 rdev = md_import_device(dev, -1, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005725 if (IS_ERR(rdev)) {
5726 printk(KERN_WARNING
5727 "md: md_import_device returned %ld\n",
5728 PTR_ERR(rdev));
5729 return PTR_ERR(rdev);
5730 }
NeilBrown1a855a02010-12-09 16:36:28 +11005731 /* set saved_raid_disk if appropriate */
NeilBrown41158c72005-06-21 17:17:25 -07005732 if (!mddev->persistent) {
5733 if (info->state & (1<<MD_DISK_SYNC) &&
NeilBrownbf572542011-01-12 09:03:35 +11005734 info->raid_disk < mddev->raid_disks) {
NeilBrown41158c72005-06-21 17:17:25 -07005735 rdev->raid_disk = info->raid_disk;
NeilBrownbf572542011-01-12 09:03:35 +11005736 set_bit(In_sync, &rdev->flags);
5737 } else
NeilBrown41158c72005-06-21 17:17:25 -07005738 rdev->raid_disk = -1;
5739 } else
5740 super_types[mddev->major_version].
5741 validate_super(mddev, rdev);
NeilBrownbedd86b2011-05-11 14:26:20 +10005742 if ((info->state & (1<<MD_DISK_SYNC)) &&
NeilBrownf4563092012-07-03 15:59:06 +10005743 rdev->raid_disk != info->raid_disk) {
NeilBrownbedd86b2011-05-11 14:26:20 +10005744 /* This was a hot-add request, but events doesn't
5745 * match, so reject it.
5746 */
5747 export_rdev(rdev);
5748 return -EINVAL;
5749 }
5750
NeilBrown1a855a02010-12-09 16:36:28 +11005751 if (test_bit(In_sync, &rdev->flags))
5752 rdev->saved_raid_disk = rdev->raid_disk;
5753 else
5754 rdev->saved_raid_disk = -1;
NeilBrown41158c72005-06-21 17:17:25 -07005755
NeilBrownb2d444d2005-11-08 21:39:31 -08005756 clear_bit(In_sync, &rdev->flags); /* just to be sure */
NeilBrown8ddf9ef2005-09-09 16:23:45 -07005757 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5758 set_bit(WriteMostly, &rdev->flags);
NeilBrown575a80f2009-03-31 14:33:13 +11005759 else
5760 clear_bit(WriteMostly, &rdev->flags);
NeilBrown8ddf9ef2005-09-09 16:23:45 -07005761
Linus Torvalds1da177e2005-04-16 15:20:36 -07005762 rdev->raid_disk = -1;
5763 err = bind_rdev_to_array(rdev, mddev);
NeilBrown7c7546c2006-06-26 00:27:41 -07005764 if (!err && !mddev->pers->hot_remove_disk) {
5765 /* If there is hot_add_disk but no hot_remove_disk
5766 * then added disks for geometry changes,
5767 * and should be added immediately.
5768 */
5769 super_types[mddev->major_version].
5770 validate_super(mddev, rdev);
5771 err = mddev->pers->hot_add_disk(mddev, rdev);
5772 if (err)
5773 unbind_rdev_from_array(rdev);
5774 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005775 if (err)
5776 export_rdev(rdev);
Neil Brown52664732008-06-28 08:31:44 +10005777 else
NeilBrown00bcb4a2010-06-01 19:37:23 +10005778 sysfs_notify_dirent_safe(rdev->sysfs_state);
NeilBrownc3617772005-06-21 17:17:10 -07005779
NeilBrown17571282006-12-10 02:20:52 -08005780 md_update_sb(mddev, 1);
Neil Brown72a23c22008-06-28 08:31:41 +10005781 if (mddev->degraded)
5782 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
NeilBrownc3617772005-06-21 17:17:10 -07005783 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
马建朋9864c002011-06-09 11:42:48 +10005784 if (!err)
5785 md_new_event(mddev);
NeilBrown005eca52005-08-22 13:11:08 -07005786 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005787 return err;
5788 }
5789
5790 /* otherwise, add_new_disk is only allowed
5791 * for major_version==0 superblocks
5792 */
5793 if (mddev->major_version != 0) {
5794 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5795 mdname(mddev));
5796 return -EINVAL;
5797 }
5798
5799 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5800 int err;
NeilBrownd710e132008-10-13 11:55:12 +11005801 rdev = md_import_device(dev, -1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005802 if (IS_ERR(rdev)) {
5803 printk(KERN_WARNING
5804 "md: error, md_import_device() returned %ld\n",
5805 PTR_ERR(rdev));
5806 return PTR_ERR(rdev);
5807 }
5808 rdev->desc_nr = info->number;
5809 if (info->raid_disk < mddev->raid_disks)
5810 rdev->raid_disk = info->raid_disk;
5811 else
5812 rdev->raid_disk = -1;
5813
Linus Torvalds1da177e2005-04-16 15:20:36 -07005814 if (rdev->raid_disk < mddev->raid_disks)
NeilBrownb2d444d2005-11-08 21:39:31 -08005815 if (info->state & (1<<MD_DISK_SYNC))
5816 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005817
NeilBrown8ddf9ef2005-09-09 16:23:45 -07005818 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5819 set_bit(WriteMostly, &rdev->flags);
5820
Linus Torvalds1da177e2005-04-16 15:20:36 -07005821 if (!mddev->persistent) {
5822 printk(KERN_INFO "md: nonpersistent superblock ...\n");
Mike Snitzer77304d22010-11-08 14:39:12 +01005823 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5824 } else
Jonathan Brassow57b2caa2011-01-14 09:14:33 +11005825 rdev->sb_start = calc_dev_sboffset(rdev);
NeilBrown8190e752009-06-18 08:48:58 +10005826 rdev->sectors = rdev->sb_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005827
NeilBrown2bf071b2006-01-06 00:20:55 -08005828 err = bind_rdev_to_array(rdev, mddev);
5829 if (err) {
5830 export_rdev(rdev);
5831 return err;
5832 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005833 }
5834
5835 return 0;
5836}
5837
NeilBrownfd01b882011-10-11 16:47:53 +11005838static int hot_remove_disk(struct mddev * mddev, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005839{
5840 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11005841 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005842
Linus Torvalds1da177e2005-04-16 15:20:36 -07005843 rdev = find_rdev(mddev, dev);
5844 if (!rdev)
5845 return -ENXIO;
5846
5847 if (rdev->raid_disk >= 0)
5848 goto busy;
5849
5850 kick_rdev_from_array(rdev);
NeilBrown850b2b42006-10-03 01:15:46 -07005851 md_update_sb(mddev, 1);
NeilBrownd7603b72006-01-06 00:20:30 -08005852 md_new_event(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005853
5854 return 0;
5855busy:
Nick Andrewfdefa4d2008-04-21 22:42:58 +00005856 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005857 bdevname(rdev->bdev,b), mdname(mddev));
5858 return -EBUSY;
5859}
5860
NeilBrownfd01b882011-10-11 16:47:53 +11005861static int hot_add_disk(struct mddev * mddev, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005862{
5863 char b[BDEVNAME_SIZE];
5864 int err;
NeilBrown3cb03002011-10-11 16:45:26 +11005865 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005866
5867 if (!mddev->pers)
5868 return -ENODEV;
5869
5870 if (mddev->major_version != 0) {
5871 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5872 " version-0 superblocks.\n",
5873 mdname(mddev));
5874 return -EINVAL;
5875 }
5876 if (!mddev->pers->hot_add_disk) {
5877 printk(KERN_WARNING
5878 "%s: personality does not support diskops!\n",
5879 mdname(mddev));
5880 return -EINVAL;
5881 }
5882
NeilBrownd710e132008-10-13 11:55:12 +11005883 rdev = md_import_device(dev, -1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005884 if (IS_ERR(rdev)) {
5885 printk(KERN_WARNING
5886 "md: error, md_import_device() returned %ld\n",
5887 PTR_ERR(rdev));
5888 return -EINVAL;
5889 }
5890
5891 if (mddev->persistent)
Jonathan Brassow57b2caa2011-01-14 09:14:33 +11005892 rdev->sb_start = calc_dev_sboffset(rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005893 else
Mike Snitzer77304d22010-11-08 14:39:12 +01005894 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005895
NeilBrown8190e752009-06-18 08:48:58 +10005896 rdev->sectors = rdev->sb_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005897
NeilBrownb2d444d2005-11-08 21:39:31 -08005898 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005899 printk(KERN_WARNING
5900 "md: can not hot-add faulty %s disk to %s!\n",
5901 bdevname(rdev->bdev,b), mdname(mddev));
5902 err = -EINVAL;
5903 goto abort_export;
5904 }
NeilBrownb2d444d2005-11-08 21:39:31 -08005905 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005906 rdev->desc_nr = -1;
NeilBrown58427302006-10-06 00:44:04 -07005907 rdev->saved_raid_disk = -1;
NeilBrown2bf071b2006-01-06 00:20:55 -08005908 err = bind_rdev_to_array(rdev, mddev);
5909 if (err)
5910 goto abort_export;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005911
5912 /*
5913 * The rest should better be atomic, we can have disk failures
5914 * noticed in interrupt contexts ...
5915 */
5916
Linus Torvalds1da177e2005-04-16 15:20:36 -07005917 rdev->raid_disk = -1;
5918
NeilBrown850b2b42006-10-03 01:15:46 -07005919 md_update_sb(mddev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005920
5921 /*
5922 * Kick recovery, maybe this spare has to be added to the
5923 * array immediately.
5924 */
5925 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5926 md_wakeup_thread(mddev->thread);
NeilBrownd7603b72006-01-06 00:20:30 -08005927 md_new_event(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005928 return 0;
5929
Linus Torvalds1da177e2005-04-16 15:20:36 -07005930abort_export:
5931 export_rdev(rdev);
5932 return err;
5933}
5934
NeilBrownfd01b882011-10-11 16:47:53 +11005935static int set_bitmap_file(struct mddev *mddev, int fd)
NeilBrown32a76272005-06-21 17:17:14 -07005936{
5937 int err;
5938
NeilBrown36fa3062005-09-09 16:23:45 -07005939 if (mddev->pers) {
5940 if (!mddev->pers->quiesce)
5941 return -EBUSY;
5942 if (mddev->recovery || mddev->sync_thread)
5943 return -EBUSY;
5944 /* we should be able to change the bitmap.. */
NeilBrown32a76272005-06-21 17:17:14 -07005945 }
5946
NeilBrown36fa3062005-09-09 16:23:45 -07005947
5948 if (fd >= 0) {
5949 if (mddev->bitmap)
5950 return -EEXIST; /* cannot add when bitmap is present */
NeilBrownc3d97142009-12-14 12:49:52 +11005951 mddev->bitmap_info.file = fget(fd);
NeilBrown36fa3062005-09-09 16:23:45 -07005952
NeilBrownc3d97142009-12-14 12:49:52 +11005953 if (mddev->bitmap_info.file == NULL) {
NeilBrown36fa3062005-09-09 16:23:45 -07005954 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
5955 mdname(mddev));
5956 return -EBADF;
5957 }
5958
NeilBrownc3d97142009-12-14 12:49:52 +11005959 err = deny_bitmap_write_access(mddev->bitmap_info.file);
NeilBrown36fa3062005-09-09 16:23:45 -07005960 if (err) {
5961 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
5962 mdname(mddev));
NeilBrownc3d97142009-12-14 12:49:52 +11005963 fput(mddev->bitmap_info.file);
5964 mddev->bitmap_info.file = NULL;
NeilBrown36fa3062005-09-09 16:23:45 -07005965 return err;
5966 }
NeilBrownc3d97142009-12-14 12:49:52 +11005967 mddev->bitmap_info.offset = 0; /* file overrides offset */
NeilBrown36fa3062005-09-09 16:23:45 -07005968 } else if (mddev->bitmap == NULL)
5969 return -ENOENT; /* cannot remove what isn't there */
5970 err = 0;
5971 if (mddev->pers) {
5972 mddev->pers->quiesce(mddev, 1);
NeilBrown69e51b42010-06-01 19:37:35 +10005973 if (fd >= 0) {
NeilBrown36fa3062005-09-09 16:23:45 -07005974 err = bitmap_create(mddev);
NeilBrown69e51b42010-06-01 19:37:35 +10005975 if (!err)
5976 err = bitmap_load(mddev);
5977 }
NeilBrownd7375ab2006-06-26 00:27:43 -07005978 if (fd < 0 || err) {
NeilBrown36fa3062005-09-09 16:23:45 -07005979 bitmap_destroy(mddev);
NeilBrownd7375ab2006-06-26 00:27:43 -07005980 fd = -1; /* make sure to put the file */
5981 }
NeilBrown36fa3062005-09-09 16:23:45 -07005982 mddev->pers->quiesce(mddev, 0);
NeilBrownd7375ab2006-06-26 00:27:43 -07005983 }
5984 if (fd < 0) {
NeilBrownc3d97142009-12-14 12:49:52 +11005985 if (mddev->bitmap_info.file) {
5986 restore_bitmap_write_access(mddev->bitmap_info.file);
5987 fput(mddev->bitmap_info.file);
NeilBrownacc55e22006-06-26 00:27:47 -07005988 }
NeilBrownc3d97142009-12-14 12:49:52 +11005989 mddev->bitmap_info.file = NULL;
NeilBrown36fa3062005-09-09 16:23:45 -07005990 }
5991
NeilBrown32a76272005-06-21 17:17:14 -07005992 return err;
5993}
5994
Linus Torvalds1da177e2005-04-16 15:20:36 -07005995/*
5996 * set_array_info is used two different ways
5997 * The original usage is when creating a new array.
5998 * In this usage, raid_disks is > 0 and it together with
5999 * level, size, not_persistent,layout,chunksize determine the
6000 * shape of the array.
6001 * This will always create an array with a type-0.90.0 superblock.
6002 * The newer usage is when assembling an array.
6003 * In this case raid_disks will be 0, and the major_version field is
6004 * use to determine which style super-blocks are to be found on the devices.
6005 * The minor and patch _version numbers are also kept incase the
6006 * super_block handler wishes to interpret them.
6007 */
NeilBrownfd01b882011-10-11 16:47:53 +11006008static int set_array_info(struct mddev * mddev, mdu_array_info_t *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006009{
6010
6011 if (info->raid_disks == 0) {
6012 /* just setting version number for superblock loading */
6013 if (info->major_version < 0 ||
Ahmed S. Darwish50511da2007-05-09 02:35:34 -07006014 info->major_version >= ARRAY_SIZE(super_types) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07006015 super_types[info->major_version].name == NULL) {
6016 /* maybe try to auto-load a module? */
6017 printk(KERN_INFO
6018 "md: superblock version %d not known\n",
6019 info->major_version);
6020 return -EINVAL;
6021 }
6022 mddev->major_version = info->major_version;
6023 mddev->minor_version = info->minor_version;
6024 mddev->patch_version = info->patch_version;
NeilBrown3f9d7b02006-12-22 01:11:41 -08006025 mddev->persistent = !info->not_persistent;
NeilBrowncbd19982009-12-30 12:08:49 +11006026 /* ensure mddev_put doesn't delete this now that there
6027 * is some minimal configuration.
6028 */
6029 mddev->ctime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006030 return 0;
6031 }
6032 mddev->major_version = MD_MAJOR_VERSION;
6033 mddev->minor_version = MD_MINOR_VERSION;
6034 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6035 mddev->ctime = get_seconds();
6036
6037 mddev->level = info->level;
NeilBrown17115e02006-01-16 22:14:57 -08006038 mddev->clevel[0] = 0;
Andre Noll58c0fed2009-03-31 14:33:13 +11006039 mddev->dev_sectors = 2 * (sector_t)info->size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006040 mddev->raid_disks = info->raid_disks;
6041 /* don't set md_minor, it is determined by which /dev/md* was
6042 * openned
6043 */
6044 if (info->state & (1<<MD_SB_CLEAN))
6045 mddev->recovery_cp = MaxSector;
6046 else
6047 mddev->recovery_cp = 0;
6048 mddev->persistent = ! info->not_persistent;
NeilBrowne6910632008-02-06 01:39:51 -08006049 mddev->external = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006050
6051 mddev->layout = info->layout;
Andre Noll9d8f0362009-06-18 08:45:01 +10006052 mddev->chunk_sectors = info->chunk_size >> 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006053
6054 mddev->max_disks = MD_SB_DISKS;
6055
NeilBrowne6910632008-02-06 01:39:51 -08006056 if (mddev->persistent)
6057 mddev->flags = 0;
NeilBrown850b2b42006-10-03 01:15:46 -07006058 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006059
NeilBrownc3d97142009-12-14 12:49:52 +11006060 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
NeilBrown6409bb02012-05-22 13:55:07 +10006061 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
NeilBrownc3d97142009-12-14 12:49:52 +11006062 mddev->bitmap_info.offset = 0;
NeilBrownb2a27032005-11-28 13:44:12 -08006063
NeilBrownf6705572006-03-27 01:18:11 -08006064 mddev->reshape_position = MaxSector;
6065
Linus Torvalds1da177e2005-04-16 15:20:36 -07006066 /*
6067 * Generate a 128 bit UUID
6068 */
6069 get_random_bytes(mddev->uuid, 16);
6070
NeilBrownf6705572006-03-27 01:18:11 -08006071 mddev->new_level = mddev->level;
Andre Noll664e7c42009-06-18 08:45:27 +10006072 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08006073 mddev->new_layout = mddev->layout;
6074 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10006075 mddev->reshape_backwards = 0;
NeilBrownf6705572006-03-27 01:18:11 -08006076
Linus Torvalds1da177e2005-04-16 15:20:36 -07006077 return 0;
6078}
6079
NeilBrownfd01b882011-10-11 16:47:53 +11006080void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
Dan Williams1f403622009-03-31 14:59:03 +11006081{
Dan Williamsb522adc2009-03-31 15:00:31 +11006082 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6083
6084 if (mddev->external_size)
6085 return;
6086
Dan Williams1f403622009-03-31 14:59:03 +11006087 mddev->array_sectors = array_sectors;
6088}
6089EXPORT_SYMBOL(md_set_array_sectors);
6090
NeilBrownfd01b882011-10-11 16:47:53 +11006091static int update_size(struct mddev *mddev, sector_t num_sectors)
NeilBrowna35b0d62006-01-06 00:20:49 -08006092{
NeilBrown3cb03002011-10-11 16:45:26 +11006093 struct md_rdev *rdev;
NeilBrowna35b0d62006-01-06 00:20:49 -08006094 int rv;
Andre Nolld71f9f82008-07-11 22:02:22 +10006095 int fit = (num_sectors == 0);
NeilBrowna35b0d62006-01-06 00:20:49 -08006096
6097 if (mddev->pers->resize == NULL)
6098 return -EINVAL;
Andre Nolld71f9f82008-07-11 22:02:22 +10006099 /* The "num_sectors" is the number of sectors of each device that
6100 * is used. This can only make sense for arrays with redundancy.
6101 * linear and raid0 always use whatever space is available. We can only
6102 * consider changing this number if no resync or reconstruction is
6103 * happening, and if the new size is acceptable. It must fit before the
Andre Noll0f420352008-07-11 22:02:23 +10006104 * sb_start or, if that is <data_offset, it must fit before the size
Andre Nolld71f9f82008-07-11 22:02:22 +10006105 * of each device. If num_sectors is zero, we find the largest size
6106 * that fits.
NeilBrowna35b0d62006-01-06 00:20:49 -08006107 */
6108 if (mddev->sync_thread)
6109 return -EBUSY;
NeilBrowna4a61252012-05-22 13:55:27 +10006110
NeilBrowndafb20f2012-03-19 12:46:39 +11006111 rdev_for_each(rdev, mddev) {
Andre Nolldd8ac332009-03-31 14:33:13 +11006112 sector_t avail = rdev->sectors;
NeilBrown01ab5662006-10-28 10:38:30 -07006113
Andre Nolld71f9f82008-07-11 22:02:22 +10006114 if (fit && (num_sectors == 0 || num_sectors > avail))
6115 num_sectors = avail;
6116 if (avail < num_sectors)
NeilBrowna35b0d62006-01-06 00:20:49 -08006117 return -ENOSPC;
6118 }
Andre Nolld71f9f82008-07-11 22:02:22 +10006119 rv = mddev->pers->resize(mddev, num_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10006120 if (!rv)
6121 revalidate_disk(mddev->gendisk);
NeilBrowna35b0d62006-01-06 00:20:49 -08006122 return rv;
6123}
6124
NeilBrownfd01b882011-10-11 16:47:53 +11006125static int update_raid_disks(struct mddev *mddev, int raid_disks)
NeilBrownda943b992006-01-06 00:20:54 -08006126{
6127 int rv;
NeilBrownc6563a82012-05-21 09:27:00 +10006128 struct md_rdev *rdev;
NeilBrownda943b992006-01-06 00:20:54 -08006129 /* change the number of raid disks */
NeilBrown63c70c42006-03-27 01:18:13 -08006130 if (mddev->pers->check_reshape == NULL)
NeilBrownda943b992006-01-06 00:20:54 -08006131 return -EINVAL;
6132 if (raid_disks <= 0 ||
NeilBrown233fca32010-04-14 17:02:09 +10006133 (mddev->max_disks && raid_disks >= mddev->max_disks))
NeilBrownda943b992006-01-06 00:20:54 -08006134 return -EINVAL;
NeilBrown63c70c42006-03-27 01:18:13 -08006135 if (mddev->sync_thread || mddev->reshape_position != MaxSector)
NeilBrownda943b992006-01-06 00:20:54 -08006136 return -EBUSY;
NeilBrownc6563a82012-05-21 09:27:00 +10006137
6138 rdev_for_each(rdev, mddev) {
6139 if (mddev->raid_disks < raid_disks &&
6140 rdev->data_offset < rdev->new_data_offset)
6141 return -EINVAL;
6142 if (mddev->raid_disks > raid_disks &&
6143 rdev->data_offset > rdev->new_data_offset)
6144 return -EINVAL;
6145 }
6146
NeilBrown63c70c42006-03-27 01:18:13 -08006147 mddev->delta_disks = raid_disks - mddev->raid_disks;
NeilBrown2c810cd2012-05-21 09:27:00 +10006148 if (mddev->delta_disks < 0)
6149 mddev->reshape_backwards = 1;
6150 else if (mddev->delta_disks > 0)
6151 mddev->reshape_backwards = 0;
NeilBrown63c70c42006-03-27 01:18:13 -08006152
6153 rv = mddev->pers->check_reshape(mddev);
NeilBrown2c810cd2012-05-21 09:27:00 +10006154 if (rv < 0) {
NeilBrownde171cb2011-01-31 11:57:42 +11006155 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10006156 mddev->reshape_backwards = 0;
6157 }
NeilBrownda943b992006-01-06 00:20:54 -08006158 return rv;
6159}
6160
6161
Linus Torvalds1da177e2005-04-16 15:20:36 -07006162/*
6163 * update_array_info is used to change the configuration of an
6164 * on-line array.
6165 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6166 * fields in the info are checked against the array.
6167 * Any differences that cannot be handled will cause an error.
6168 * Normally, only one change can be managed at a time.
6169 */
NeilBrownfd01b882011-10-11 16:47:53 +11006170static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006171{
6172 int rv = 0;
6173 int cnt = 0;
NeilBrown36fa3062005-09-09 16:23:45 -07006174 int state = 0;
6175
6176 /* calculate expected state,ignoring low bits */
NeilBrownc3d97142009-12-14 12:49:52 +11006177 if (mddev->bitmap && mddev->bitmap_info.offset)
NeilBrown36fa3062005-09-09 16:23:45 -07006178 state |= (1 << MD_SB_BITMAP_PRESENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006179
6180 if (mddev->major_version != info->major_version ||
6181 mddev->minor_version != info->minor_version ||
6182/* mddev->patch_version != info->patch_version || */
6183 mddev->ctime != info->ctime ||
6184 mddev->level != info->level ||
6185/* mddev->layout != info->layout || */
6186 !mddev->persistent != info->not_persistent||
Andre Noll9d8f0362009-06-18 08:45:01 +10006187 mddev->chunk_sectors != info->chunk_size >> 9 ||
NeilBrown36fa3062005-09-09 16:23:45 -07006188 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6189 ((state^info->state) & 0xfffffe00)
6190 )
Linus Torvalds1da177e2005-04-16 15:20:36 -07006191 return -EINVAL;
6192 /* Check there is only one change */
Andre Noll58c0fed2009-03-31 14:33:13 +11006193 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6194 cnt++;
6195 if (mddev->raid_disks != info->raid_disks)
6196 cnt++;
6197 if (mddev->layout != info->layout)
6198 cnt++;
6199 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6200 cnt++;
6201 if (cnt == 0)
6202 return 0;
6203 if (cnt > 1)
6204 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006205
6206 if (mddev->layout != info->layout) {
6207 /* Change layout
6208 * we don't need to do anything at the md level, the
6209 * personality will take care of it all.
6210 */
NeilBrown50ac1682009-06-18 08:47:55 +10006211 if (mddev->pers->check_reshape == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006212 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10006213 else {
6214 mddev->new_layout = info->layout;
NeilBrown50ac1682009-06-18 08:47:55 +10006215 rv = mddev->pers->check_reshape(mddev);
NeilBrown597a7112009-06-18 08:47:42 +10006216 if (rv)
6217 mddev->new_layout = mddev->layout;
6218 return rv;
6219 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006220 }
Andre Noll58c0fed2009-03-31 14:33:13 +11006221 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
Andre Nolld71f9f82008-07-11 22:02:22 +10006222 rv = update_size(mddev, (sector_t)info->size * 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006223
NeilBrownda943b992006-01-06 00:20:54 -08006224 if (mddev->raid_disks != info->raid_disks)
6225 rv = update_raid_disks(mddev, info->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006226
NeilBrown36fa3062005-09-09 16:23:45 -07006227 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
6228 if (mddev->pers->quiesce == NULL)
6229 return -EINVAL;
6230 if (mddev->recovery || mddev->sync_thread)
6231 return -EBUSY;
6232 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6233 /* add the bitmap */
6234 if (mddev->bitmap)
6235 return -EEXIST;
NeilBrownc3d97142009-12-14 12:49:52 +11006236 if (mddev->bitmap_info.default_offset == 0)
NeilBrown36fa3062005-09-09 16:23:45 -07006237 return -EINVAL;
NeilBrownc3d97142009-12-14 12:49:52 +11006238 mddev->bitmap_info.offset =
6239 mddev->bitmap_info.default_offset;
NeilBrown6409bb02012-05-22 13:55:07 +10006240 mddev->bitmap_info.space =
6241 mddev->bitmap_info.default_space;
NeilBrown36fa3062005-09-09 16:23:45 -07006242 mddev->pers->quiesce(mddev, 1);
6243 rv = bitmap_create(mddev);
NeilBrown69e51b42010-06-01 19:37:35 +10006244 if (!rv)
6245 rv = bitmap_load(mddev);
NeilBrown36fa3062005-09-09 16:23:45 -07006246 if (rv)
6247 bitmap_destroy(mddev);
6248 mddev->pers->quiesce(mddev, 0);
6249 } else {
6250 /* remove the bitmap */
6251 if (!mddev->bitmap)
6252 return -ENOENT;
NeilBrown1ec885c2012-05-22 13:55:10 +10006253 if (mddev->bitmap->storage.file)
NeilBrown36fa3062005-09-09 16:23:45 -07006254 return -EINVAL;
6255 mddev->pers->quiesce(mddev, 1);
6256 bitmap_destroy(mddev);
6257 mddev->pers->quiesce(mddev, 0);
NeilBrownc3d97142009-12-14 12:49:52 +11006258 mddev->bitmap_info.offset = 0;
NeilBrown36fa3062005-09-09 16:23:45 -07006259 }
6260 }
NeilBrown850b2b42006-10-03 01:15:46 -07006261 md_update_sb(mddev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006262 return rv;
6263}
6264
NeilBrownfd01b882011-10-11 16:47:53 +11006265static int set_disk_faulty(struct mddev *mddev, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006266{
NeilBrown3cb03002011-10-11 16:45:26 +11006267 struct md_rdev *rdev;
NeilBrown1ca69c42012-10-11 13:37:33 +11006268 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006269
6270 if (mddev->pers == NULL)
6271 return -ENODEV;
6272
NeilBrown1ca69c42012-10-11 13:37:33 +11006273 rcu_read_lock();
6274 rdev = find_rdev_rcu(mddev, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006275 if (!rdev)
NeilBrown1ca69c42012-10-11 13:37:33 +11006276 err = -ENODEV;
6277 else {
6278 md_error(mddev, rdev);
6279 if (!test_bit(Faulty, &rdev->flags))
6280 err = -EBUSY;
6281 }
6282 rcu_read_unlock();
6283 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006284}
6285
Andre Noll2f9618c2008-04-25 18:57:58 +02006286/*
6287 * We have a problem here : there is no easy way to give a CHS
6288 * virtual geometry. We currently pretend that we have a 2 heads
6289 * 4 sectors (with a BIG number of cylinders...). This drives
6290 * dosfs just mad... ;-)
6291 */
Christoph Hellwiga885c8c2006-01-08 01:02:50 -08006292static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6293{
NeilBrownfd01b882011-10-11 16:47:53 +11006294 struct mddev *mddev = bdev->bd_disk->private_data;
Christoph Hellwiga885c8c2006-01-08 01:02:50 -08006295
6296 geo->heads = 2;
6297 geo->sectors = 4;
NeilBrown49ce6ce2010-03-29 10:51:42 +11006298 geo->cylinders = mddev->array_sectors / 8;
Christoph Hellwiga885c8c2006-01-08 01:02:50 -08006299 return 0;
6300}
6301
Al Viroa39907f2008-03-02 10:31:15 -05006302static int md_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006303 unsigned int cmd, unsigned long arg)
6304{
6305 int err = 0;
6306 void __user *argp = (void __user *)arg;
NeilBrownfd01b882011-10-11 16:47:53 +11006307 struct mddev *mddev = NULL;
Dan Williamse2218352010-05-12 08:25:37 +10006308 int ro;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006309
NeilBrown506c9e42011-12-23 10:17:26 +11006310 switch (cmd) {
6311 case RAID_VERSION:
6312 case GET_ARRAY_INFO:
6313 case GET_DISK_INFO:
6314 break;
6315 default:
6316 if (!capable(CAP_SYS_ADMIN))
6317 return -EACCES;
6318 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006319
6320 /*
6321 * Commands dealing with the RAID driver but not any
6322 * particular array:
6323 */
6324 switch (cmd)
6325 {
6326 case RAID_VERSION:
6327 err = get_version(argp);
6328 goto done;
6329
6330 case PRINT_RAID_DEBUG:
6331 err = 0;
6332 md_print_devices();
6333 goto done;
6334
6335#ifndef MODULE
6336 case RAID_AUTORUN:
6337 err = 0;
6338 autostart_arrays(arg);
6339 goto done;
6340#endif
6341 default:;
6342 }
6343
6344 /*
6345 * Commands creating/starting a new array:
6346 */
6347
Al Viroa39907f2008-03-02 10:31:15 -05006348 mddev = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006349
6350 if (!mddev) {
6351 BUG();
6352 goto abort;
6353 }
6354
NeilBrown1ca69c42012-10-11 13:37:33 +11006355 /* Some actions do not requires the mutex */
6356 switch (cmd) {
6357 case GET_ARRAY_INFO:
6358 if (!mddev->raid_disks && !mddev->external)
6359 err = -ENODEV;
6360 else
6361 err = get_array_info(mddev, argp);
6362 goto abort;
6363
6364 case GET_DISK_INFO:
6365 if (!mddev->raid_disks && !mddev->external)
6366 err = -ENODEV;
6367 else
6368 err = get_disk_info(mddev, argp);
6369 goto abort;
6370
6371 case SET_DISK_FAULTY:
6372 err = set_disk_faulty(mddev, new_decode_dev(arg));
6373 goto abort;
6374 }
6375
Linus Torvalds1da177e2005-04-16 15:20:36 -07006376 err = mddev_lock(mddev);
6377 if (err) {
6378 printk(KERN_INFO
6379 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6380 err, cmd);
6381 goto abort;
6382 }
6383
6384 switch (cmd)
6385 {
6386 case SET_ARRAY_INFO:
6387 {
6388 mdu_array_info_t info;
6389 if (!arg)
6390 memset(&info, 0, sizeof(info));
6391 else if (copy_from_user(&info, argp, sizeof(info))) {
6392 err = -EFAULT;
6393 goto abort_unlock;
6394 }
6395 if (mddev->pers) {
6396 err = update_array_info(mddev, &info);
6397 if (err) {
6398 printk(KERN_WARNING "md: couldn't update"
6399 " array info. %d\n", err);
6400 goto abort_unlock;
6401 }
6402 goto done_unlock;
6403 }
6404 if (!list_empty(&mddev->disks)) {
6405 printk(KERN_WARNING
6406 "md: array %s already has disks!\n",
6407 mdname(mddev));
6408 err = -EBUSY;
6409 goto abort_unlock;
6410 }
6411 if (mddev->raid_disks) {
6412 printk(KERN_WARNING
6413 "md: array %s already initialised!\n",
6414 mdname(mddev));
6415 err = -EBUSY;
6416 goto abort_unlock;
6417 }
6418 err = set_array_info(mddev, &info);
6419 if (err) {
6420 printk(KERN_WARNING "md: couldn't set"
6421 " array info. %d\n", err);
6422 goto abort_unlock;
6423 }
6424 }
6425 goto done_unlock;
6426
6427 default:;
6428 }
6429
6430 /*
6431 * Commands querying/configuring an existing array:
6432 */
NeilBrown32a76272005-06-21 17:17:14 -07006433 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
NeilBrown3f9d7b02006-12-22 01:11:41 -08006434 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
NeilBrowna17184a2008-02-06 01:39:55 -08006435 if ((!mddev->raid_disks && !mddev->external)
6436 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6437 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6438 && cmd != GET_BITMAP_FILE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006439 err = -ENODEV;
6440 goto abort_unlock;
6441 }
6442
6443 /*
6444 * Commands even a read-only array can execute:
6445 */
6446 switch (cmd)
6447 {
NeilBrown32a76272005-06-21 17:17:14 -07006448 case GET_BITMAP_FILE:
viro@ZenIV.linux.org.uk87162a22005-09-09 20:36:43 +01006449 err = get_bitmap_file(mddev, argp);
NeilBrown32a76272005-06-21 17:17:14 -07006450 goto done_unlock;
6451
Linus Torvalds1da177e2005-04-16 15:20:36 -07006452 case RESTART_ARRAY_RW:
6453 err = restart_array(mddev);
6454 goto done_unlock;
6455
6456 case STOP_ARRAY:
NeilBrowna05b7ea2012-07-19 15:59:18 +10006457 err = do_md_stop(mddev, 0, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006458 goto done_unlock;
6459
6460 case STOP_ARRAY_RO:
NeilBrowna05b7ea2012-07-19 15:59:18 +10006461 err = md_set_readonly(mddev, bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006462 goto done_unlock;
6463
Dan Williamse2218352010-05-12 08:25:37 +10006464 case BLKROSET:
6465 if (get_user(ro, (int __user *)(arg))) {
6466 err = -EFAULT;
6467 goto done_unlock;
6468 }
6469 err = -EINVAL;
6470
6471 /* if the bdev is going readonly the value of mddev->ro
6472 * does not matter, no writes are coming
6473 */
6474 if (ro)
6475 goto done_unlock;
6476
6477 /* are we are already prepared for writes? */
6478 if (mddev->ro != 1)
6479 goto done_unlock;
6480
6481 /* transitioning to readauto need only happen for
6482 * arrays that call md_write_start
6483 */
6484 if (mddev->pers) {
6485 err = restart_array(mddev);
6486 if (err == 0) {
6487 mddev->ro = 2;
6488 set_disk_ro(mddev->gendisk, 0);
6489 }
6490 }
6491 goto done_unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006492 }
6493
6494 /*
6495 * The remaining ioctls are changing the state of the
NeilBrownf91de922005-11-08 21:39:36 -08006496 * superblock, so we do not allow them on read-only arrays.
6497 * However non-MD ioctls (e.g. get-size) will still come through
6498 * here and hit the 'default' below, so only disallow
6499 * 'md' ioctls, and switch to rw mode if started auto-readonly.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006500 */
Andre Nollbb57fc62008-04-25 19:06:35 +02006501 if (_IOC_TYPE(cmd) == MD_MAJOR && mddev->ro && mddev->pers) {
NeilBrownf91de922005-11-08 21:39:36 -08006502 if (mddev->ro == 2) {
6503 mddev->ro = 0;
NeilBrown00bcb4a2010-06-01 19:37:23 +10006504 sysfs_notify_dirent_safe(mddev->sysfs_state);
Neil Brown0fd62b82008-06-28 08:31:36 +10006505 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6506 md_wakeup_thread(mddev->thread);
NeilBrownf91de922005-11-08 21:39:36 -08006507 } else {
6508 err = -EROFS;
6509 goto abort_unlock;
6510 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006511 }
6512
6513 switch (cmd)
6514 {
6515 case ADD_NEW_DISK:
6516 {
6517 mdu_disk_info_t info;
6518 if (copy_from_user(&info, argp, sizeof(info)))
6519 err = -EFAULT;
6520 else
6521 err = add_new_disk(mddev, &info);
6522 goto done_unlock;
6523 }
6524
6525 case HOT_REMOVE_DISK:
6526 err = hot_remove_disk(mddev, new_decode_dev(arg));
6527 goto done_unlock;
6528
6529 case HOT_ADD_DISK:
6530 err = hot_add_disk(mddev, new_decode_dev(arg));
6531 goto done_unlock;
6532
Linus Torvalds1da177e2005-04-16 15:20:36 -07006533 case RUN_ARRAY:
NeilBrownd710e132008-10-13 11:55:12 +11006534 err = do_md_run(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006535 goto done_unlock;
6536
NeilBrown32a76272005-06-21 17:17:14 -07006537 case SET_BITMAP_FILE:
6538 err = set_bitmap_file(mddev, (int)arg);
6539 goto done_unlock;
6540
Linus Torvalds1da177e2005-04-16 15:20:36 -07006541 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006542 err = -EINVAL;
6543 goto abort_unlock;
6544 }
6545
6546done_unlock:
6547abort_unlock:
NeilBrownd3374822009-01-09 08:31:10 +11006548 if (mddev->hold_active == UNTIL_IOCTL &&
6549 err != -EINVAL)
6550 mddev->hold_active = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006551 mddev_unlock(mddev);
6552
6553 return err;
6554done:
6555 if (err)
6556 MD_BUG();
6557abort:
6558 return err;
6559}
Arnd Bergmannaa98aa32009-12-14 12:50:05 +11006560#ifdef CONFIG_COMPAT
6561static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6562 unsigned int cmd, unsigned long arg)
6563{
6564 switch (cmd) {
6565 case HOT_REMOVE_DISK:
6566 case HOT_ADD_DISK:
6567 case SET_DISK_FAULTY:
6568 case SET_BITMAP_FILE:
6569 /* These take in integer arg, do not convert */
6570 break;
6571 default:
6572 arg = (unsigned long)compat_ptr(arg);
6573 break;
6574 }
6575
6576 return md_ioctl(bdev, mode, cmd, arg);
6577}
6578#endif /* CONFIG_COMPAT */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006579
Al Viroa39907f2008-03-02 10:31:15 -05006580static int md_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006581{
6582 /*
6583 * Succeed if we can lock the mddev, which confirms that
6584 * it isn't being stopped right now.
6585 */
NeilBrownfd01b882011-10-11 16:47:53 +11006586 struct mddev *mddev = mddev_find(bdev->bd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006587 int err;
6588
Yuanhan Liu0c098222012-05-22 13:55:32 +10006589 if (!mddev)
6590 return -ENODEV;
6591
NeilBrownd3374822009-01-09 08:31:10 +11006592 if (mddev->gendisk != bdev->bd_disk) {
6593 /* we are racing with mddev_put which is discarding this
6594 * bd_disk.
6595 */
6596 mddev_put(mddev);
6597 /* Wait until bdev->bd_disk is definitely gone */
Tejun Heoe804ac72010-10-15 15:36:08 +02006598 flush_workqueue(md_misc_wq);
NeilBrownd3374822009-01-09 08:31:10 +11006599 /* Then retry the open from the top */
6600 return -ERESTARTSYS;
6601 }
6602 BUG_ON(mddev != bdev->bd_disk->private_data);
6603
NeilBrownc8c00a62009-08-10 12:50:52 +10006604 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006605 goto out;
6606
6607 err = 0;
NeilBrownf2ea68c2008-07-21 17:05:25 +10006608 atomic_inc(&mddev->openers);
NeilBrownc8c00a62009-08-10 12:50:52 +10006609 mutex_unlock(&mddev->open_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006610
NeilBrownf0b4f7e2011-02-24 17:26:41 +11006611 check_disk_change(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006612 out:
6613 return err;
6614}
6615
Al Viroa39907f2008-03-02 10:31:15 -05006616static int md_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006617{
NeilBrownfd01b882011-10-11 16:47:53 +11006618 struct mddev *mddev = disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006619
Eric Sesterhenn52e5f9d2006-10-03 23:33:23 +02006620 BUG_ON(!mddev);
NeilBrownf2ea68c2008-07-21 17:05:25 +10006621 atomic_dec(&mddev->openers);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006622 mddev_put(mddev);
6623
6624 return 0;
6625}
NeilBrownf0b4f7e2011-02-24 17:26:41 +11006626
6627static int md_media_changed(struct gendisk *disk)
6628{
NeilBrownfd01b882011-10-11 16:47:53 +11006629 struct mddev *mddev = disk->private_data;
NeilBrownf0b4f7e2011-02-24 17:26:41 +11006630
6631 return mddev->changed;
6632}
6633
6634static int md_revalidate(struct gendisk *disk)
6635{
NeilBrownfd01b882011-10-11 16:47:53 +11006636 struct mddev *mddev = disk->private_data;
NeilBrownf0b4f7e2011-02-24 17:26:41 +11006637
6638 mddev->changed = 0;
6639 return 0;
6640}
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07006641static const struct block_device_operations md_fops =
Linus Torvalds1da177e2005-04-16 15:20:36 -07006642{
6643 .owner = THIS_MODULE,
Al Viroa39907f2008-03-02 10:31:15 -05006644 .open = md_open,
6645 .release = md_release,
NeilBrownb492b852009-05-26 12:57:36 +10006646 .ioctl = md_ioctl,
Arnd Bergmannaa98aa32009-12-14 12:50:05 +11006647#ifdef CONFIG_COMPAT
6648 .compat_ioctl = md_compat_ioctl,
6649#endif
Christoph Hellwiga885c8c2006-01-08 01:02:50 -08006650 .getgeo = md_getgeo,
NeilBrownf0b4f7e2011-02-24 17:26:41 +11006651 .media_changed = md_media_changed,
6652 .revalidate_disk= md_revalidate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006653};
6654
Adrian Bunk75c96f82005-05-05 16:16:09 -07006655static int md_thread(void * arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006656{
NeilBrown2b8bf342011-10-11 16:48:23 +11006657 struct md_thread *thread = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006658
Linus Torvalds1da177e2005-04-16 15:20:36 -07006659 /*
6660 * md_thread is a 'system-thread', it's priority should be very
6661 * high. We avoid resource deadlocks individually in each
6662 * raid personality. (RAID5 does preallocation) We also use RR and
6663 * the very same RT priority as kswapd, thus we will never get
6664 * into a priority inversion deadlock.
6665 *
6666 * we definitely have to have equal or higher priority than
6667 * bdflush, otherwise bdflush will deadlock if there are too
6668 * many dirty RAID5 blocks.
6669 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006670
NeilBrown6985c432005-10-19 21:23:47 -07006671 allow_signal(SIGKILL);
NeilBrowna6fb0932005-09-09 16:23:56 -07006672 while (!kthread_should_stop()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006673
NeilBrown93588e22005-11-15 00:09:12 -08006674 /* We need to wait INTERRUPTIBLE so that
6675 * we don't add to the load-average.
6676 * That means we need to be sure no signals are
6677 * pending
6678 */
6679 if (signal_pending(current))
6680 flush_signals(current);
6681
6682 wait_event_interruptible_timeout
6683 (thread->wqueue,
6684 test_bit(THREAD_WAKEUP, &thread->flags)
6685 || kthread_should_stop(),
6686 thread->timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006687
NeilBrown6c987912011-01-14 09:13:53 +11006688 clear_bit(THREAD_WAKEUP, &thread->flags);
6689 if (!kthread_should_stop())
Shaohua Li4ed87312012-10-11 13:34:00 +11006690 thread->run(thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006691 }
NeilBrowna6fb0932005-09-09 16:23:56 -07006692
Linus Torvalds1da177e2005-04-16 15:20:36 -07006693 return 0;
6694}
6695
NeilBrown2b8bf342011-10-11 16:48:23 +11006696void md_wakeup_thread(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006697{
6698 if (thread) {
NeilBrown36a4e1f2011-10-07 14:23:17 +11006699 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006700 set_bit(THREAD_WAKEUP, &thread->flags);
6701 wake_up(&thread->wqueue);
6702 }
6703}
6704
Shaohua Li4ed87312012-10-11 13:34:00 +11006705struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6706 struct mddev *mddev, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006707{
NeilBrown2b8bf342011-10-11 16:48:23 +11006708 struct md_thread *thread;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006709
NeilBrown2b8bf342011-10-11 16:48:23 +11006710 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006711 if (!thread)
6712 return NULL;
6713
Linus Torvalds1da177e2005-04-16 15:20:36 -07006714 init_waitqueue_head(&thread->wqueue);
6715
Linus Torvalds1da177e2005-04-16 15:20:36 -07006716 thread->run = run;
6717 thread->mddev = mddev;
NeilBrown32a76272005-06-21 17:17:14 -07006718 thread->timeout = MAX_SCHEDULE_TIMEOUT;
NeilBrown0da3c612009-09-23 18:09:45 +10006719 thread->tsk = kthread_run(md_thread, thread,
6720 "%s_%s",
6721 mdname(thread->mddev),
NeilBrown02326052012-07-03 15:56:52 +10006722 name);
NeilBrowna6fb0932005-09-09 16:23:56 -07006723 if (IS_ERR(thread->tsk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006724 kfree(thread);
6725 return NULL;
6726 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006727 return thread;
6728}
6729
NeilBrown2b8bf342011-10-11 16:48:23 +11006730void md_unregister_thread(struct md_thread **threadp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006731{
NeilBrown2b8bf342011-10-11 16:48:23 +11006732 struct md_thread *thread = *threadp;
NeilBrowne0cf8f02009-03-31 14:39:39 +11006733 if (!thread)
6734 return;
NeilBrown36a4e1f2011-10-07 14:23:17 +11006735 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
NeilBrown01f96c02011-09-21 15:30:20 +10006736 /* Locking ensures that mddev_unlock does not wake_up a
6737 * non-existent thread
6738 */
6739 spin_lock(&pers_lock);
6740 *threadp = NULL;
6741 spin_unlock(&pers_lock);
NeilBrowna6fb0932005-09-09 16:23:56 -07006742
6743 kthread_stop(thread->tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006744 kfree(thread);
6745}
6746
NeilBrownfd01b882011-10-11 16:47:53 +11006747void md_error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006748{
6749 if (!mddev) {
6750 MD_BUG();
6751 return;
6752 }
6753
NeilBrownb2d444d2005-11-08 21:39:31 -08006754 if (!rdev || test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006755 return;
Dan Williams6bfe0b42008-04-30 00:52:32 -07006756
NeilBrownde393cd2011-07-28 11:31:48 +10006757 if (!mddev->pers || !mddev->pers->error_handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006758 return;
6759 mddev->pers->error_handler(mddev,rdev);
Neil Brown72a23c22008-06-28 08:31:41 +10006760 if (mddev->degraded)
6761 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
NeilBrown00bcb4a2010-06-01 19:37:23 +10006762 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006763 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6764 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6765 md_wakeup_thread(mddev->thread);
NeilBrown768a4182010-07-26 11:49:55 +10006766 if (mddev->event_work.func)
Tejun Heoe804ac72010-10-15 15:36:08 +02006767 queue_work(md_misc_wq, &mddev->event_work);
NeilBrownc331eb02006-05-30 21:27:13 -07006768 md_new_event_inintr(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006769}
6770
6771/* seq_file implementation /proc/mdstat */
6772
6773static void status_unused(struct seq_file *seq)
6774{
6775 int i = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11006776 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006777
6778 seq_printf(seq, "unused devices: ");
6779
Cheng Renquan159ec1f2009-01-09 08:31:08 +11006780 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006781 char b[BDEVNAME_SIZE];
6782 i++;
6783 seq_printf(seq, "%s ",
6784 bdevname(rdev->bdev,b));
6785 }
6786 if (!i)
6787 seq_printf(seq, "<none>");
6788
6789 seq_printf(seq, "\n");
6790}
6791
6792
NeilBrownfd01b882011-10-11 16:47:53 +11006793static void status_resync(struct seq_file *seq, struct mddev * mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006794{
NeilBrowndd71cf62009-05-07 12:49:35 +10006795 sector_t max_sectors, resync, res;
6796 unsigned long dt, db;
6797 sector_t rt;
NeilBrown4588b422006-03-27 01:18:04 -08006798 int scale;
6799 unsigned int per_milli;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006800
NeilBrowndd71cf62009-05-07 12:49:35 +10006801 resync = mddev->curr_resync - atomic_read(&mddev->recovery_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006802
NeilBrownc804cde2012-05-21 09:28:33 +10006803 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6804 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
NeilBrowndd71cf62009-05-07 12:49:35 +10006805 max_sectors = mddev->resync_max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006806 else
NeilBrowndd71cf62009-05-07 12:49:35 +10006807 max_sectors = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006808
6809 /*
6810 * Should not happen.
6811 */
NeilBrowndd71cf62009-05-07 12:49:35 +10006812 if (!max_sectors) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006813 MD_BUG();
6814 return;
6815 }
NeilBrown4588b422006-03-27 01:18:04 -08006816 /* Pick 'scale' such that (resync>>scale)*1000 will fit
NeilBrowndd71cf62009-05-07 12:49:35 +10006817 * in a sector_t, and (max_sectors>>scale) will fit in a
NeilBrown4588b422006-03-27 01:18:04 -08006818 * u32, as those are the requirements for sector_div.
6819 * Thus 'scale' must be at least 10
6820 */
6821 scale = 10;
6822 if (sizeof(sector_t) > sizeof(unsigned long)) {
NeilBrowndd71cf62009-05-07 12:49:35 +10006823 while ( max_sectors/2 > (1ULL<<(scale+32)))
NeilBrown4588b422006-03-27 01:18:04 -08006824 scale++;
6825 }
6826 res = (resync>>scale)*1000;
NeilBrowndd71cf62009-05-07 12:49:35 +10006827 sector_div(res, (u32)((max_sectors>>scale)+1));
NeilBrown4588b422006-03-27 01:18:04 -08006828
6829 per_milli = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006830 {
NeilBrown4588b422006-03-27 01:18:04 -08006831 int i, x = per_milli/50, y = 20-x;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006832 seq_printf(seq, "[");
6833 for (i = 0; i < x; i++)
6834 seq_printf(seq, "=");
6835 seq_printf(seq, ">");
6836 for (i = 0; i < y; i++)
6837 seq_printf(seq, ".");
6838 seq_printf(seq, "] ");
6839 }
NeilBrown4588b422006-03-27 01:18:04 -08006840 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
NeilBrownccfcc3c2006-03-27 01:18:09 -08006841 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6842 "reshape" :
NeilBrown61df9d92006-10-03 01:15:57 -07006843 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6844 "check" :
6845 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6846 "resync" : "recovery"))),
6847 per_milli/10, per_milli % 10,
NeilBrowndd71cf62009-05-07 12:49:35 +10006848 (unsigned long long) resync/2,
6849 (unsigned long long) max_sectors/2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006850
6851 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006852 * dt: time from mark until now
6853 * db: blocks written from mark until now
6854 * rt: remaining time
NeilBrowndd71cf62009-05-07 12:49:35 +10006855 *
6856 * rt is a sector_t, so could be 32bit or 64bit.
6857 * So we divide before multiply in case it is 32bit and close
6858 * to the limit.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03006859 * We scale the divisor (db) by 32 to avoid losing precision
NeilBrowndd71cf62009-05-07 12:49:35 +10006860 * near the end of resync when the number of remaining sectors
6861 * is close to 'db'.
6862 * We then divide rt by 32 after multiplying by db to compensate.
6863 * The '+1' avoids division by zero if db is very small.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006864 */
6865 dt = ((jiffies - mddev->resync_mark) / HZ);
6866 if (!dt) dt++;
NeilBrownff4e8d92006-07-10 04:44:16 -07006867 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6868 - mddev->resync_mark_cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006869
NeilBrowndd71cf62009-05-07 12:49:35 +10006870 rt = max_sectors - resync; /* number of remaining sectors */
6871 sector_div(rt, db/32+1);
6872 rt *= dt;
6873 rt >>= 5;
6874
6875 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
6876 ((unsigned long)rt % 60)/6);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006877
NeilBrownff4e8d92006-07-10 04:44:16 -07006878 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006879}
6880
6881static void *md_seq_start(struct seq_file *seq, loff_t *pos)
6882{
6883 struct list_head *tmp;
6884 loff_t l = *pos;
NeilBrownfd01b882011-10-11 16:47:53 +11006885 struct mddev *mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006886
6887 if (l >= 0x10000)
6888 return NULL;
6889 if (!l--)
6890 /* header */
6891 return (void*)1;
6892
6893 spin_lock(&all_mddevs_lock);
6894 list_for_each(tmp,&all_mddevs)
6895 if (!l--) {
NeilBrownfd01b882011-10-11 16:47:53 +11006896 mddev = list_entry(tmp, struct mddev, all_mddevs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006897 mddev_get(mddev);
6898 spin_unlock(&all_mddevs_lock);
6899 return mddev;
6900 }
6901 spin_unlock(&all_mddevs_lock);
6902 if (!l--)
6903 return (void*)2;/* tail */
6904 return NULL;
6905}
6906
6907static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
6908{
6909 struct list_head *tmp;
NeilBrownfd01b882011-10-11 16:47:53 +11006910 struct mddev *next_mddev, *mddev = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006911
6912 ++*pos;
6913 if (v == (void*)2)
6914 return NULL;
6915
6916 spin_lock(&all_mddevs_lock);
6917 if (v == (void*)1)
6918 tmp = all_mddevs.next;
6919 else
6920 tmp = mddev->all_mddevs.next;
6921 if (tmp != &all_mddevs)
NeilBrownfd01b882011-10-11 16:47:53 +11006922 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
Linus Torvalds1da177e2005-04-16 15:20:36 -07006923 else {
6924 next_mddev = (void*)2;
6925 *pos = 0x10000;
6926 }
6927 spin_unlock(&all_mddevs_lock);
6928
6929 if (v != (void*)1)
6930 mddev_put(mddev);
6931 return next_mddev;
6932
6933}
6934
6935static void md_seq_stop(struct seq_file *seq, void *v)
6936{
NeilBrownfd01b882011-10-11 16:47:53 +11006937 struct mddev *mddev = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006938
6939 if (mddev && v != (void*)1 && v != (void*)2)
6940 mddev_put(mddev);
6941}
6942
6943static int md_seq_show(struct seq_file *seq, void *v)
6944{
NeilBrownfd01b882011-10-11 16:47:53 +11006945 struct mddev *mddev = v;
Andre Nolldd8ac332009-03-31 14:33:13 +11006946 sector_t sectors;
NeilBrown3cb03002011-10-11 16:45:26 +11006947 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006948
6949 if (v == (void*)1) {
NeilBrown84fc4b52011-10-11 16:49:58 +11006950 struct md_personality *pers;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006951 seq_printf(seq, "Personalities : ");
6952 spin_lock(&pers_lock);
NeilBrown2604b702006-01-06 00:20:36 -08006953 list_for_each_entry(pers, &pers_list, list)
6954 seq_printf(seq, "[%s] ", pers->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006955
6956 spin_unlock(&pers_lock);
6957 seq_printf(seq, "\n");
Kay Sieversf1514632011-07-12 20:48:39 +02006958 seq->poll_event = atomic_read(&md_event_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006959 return 0;
6960 }
6961 if (v == (void*)2) {
6962 status_unused(seq);
6963 return 0;
6964 }
6965
Ingo Molnar5dc5cf72006-04-20 02:43:23 -07006966 if (mddev_lock(mddev) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006967 return -EINTR;
Ingo Molnar5dc5cf72006-04-20 02:43:23 -07006968
Linus Torvalds1da177e2005-04-16 15:20:36 -07006969 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
6970 seq_printf(seq, "%s : %sactive", mdname(mddev),
6971 mddev->pers ? "" : "in");
6972 if (mddev->pers) {
NeilBrownf91de922005-11-08 21:39:36 -08006973 if (mddev->ro==1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006974 seq_printf(seq, " (read-only)");
NeilBrownf91de922005-11-08 21:39:36 -08006975 if (mddev->ro==2)
NeilBrown52720ae2008-03-10 11:43:47 -07006976 seq_printf(seq, " (auto-read-only)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07006977 seq_printf(seq, " %s", mddev->pers->name);
6978 }
6979
Andre Nolldd8ac332009-03-31 14:33:13 +11006980 sectors = 0;
NeilBrowndafb20f2012-03-19 12:46:39 +11006981 rdev_for_each(rdev, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006982 char b[BDEVNAME_SIZE];
6983 seq_printf(seq, " %s[%d]",
6984 bdevname(rdev->bdev,b), rdev->desc_nr);
NeilBrown8ddf9ef2005-09-09 16:23:45 -07006985 if (test_bit(WriteMostly, &rdev->flags))
6986 seq_printf(seq, "(W)");
NeilBrownb2d444d2005-11-08 21:39:31 -08006987 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006988 seq_printf(seq, "(F)");
6989 continue;
NeilBrown2d78f8c2011-12-23 10:17:51 +11006990 }
6991 if (rdev->raid_disk < 0)
NeilBrownb325a322005-09-09 16:24:00 -07006992 seq_printf(seq, "(S)"); /* spare */
NeilBrown2d78f8c2011-12-23 10:17:51 +11006993 if (test_bit(Replacement, &rdev->flags))
6994 seq_printf(seq, "(R)");
Andre Nolldd8ac332009-03-31 14:33:13 +11006995 sectors += rdev->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006996 }
6997
6998 if (!list_empty(&mddev->disks)) {
6999 if (mddev->pers)
7000 seq_printf(seq, "\n %llu blocks",
Andre Nollf233ea52008-07-21 17:05:22 +10007001 (unsigned long long)
7002 mddev->array_sectors / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007003 else
7004 seq_printf(seq, "\n %llu blocks",
Andre Nolldd8ac332009-03-31 14:33:13 +11007005 (unsigned long long)sectors / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007006 }
NeilBrown1cd6bf12005-09-09 16:24:00 -07007007 if (mddev->persistent) {
7008 if (mddev->major_version != 0 ||
7009 mddev->minor_version != 90) {
7010 seq_printf(seq," super %d.%d",
7011 mddev->major_version,
7012 mddev->minor_version);
7013 }
NeilBrowne6910632008-02-06 01:39:51 -08007014 } else if (mddev->external)
7015 seq_printf(seq, " super external:%s",
7016 mddev->metadata_type);
7017 else
NeilBrown1cd6bf12005-09-09 16:24:00 -07007018 seq_printf(seq, " super non-persistent");
Linus Torvalds1da177e2005-04-16 15:20:36 -07007019
7020 if (mddev->pers) {
NeilBrownd710e132008-10-13 11:55:12 +11007021 mddev->pers->status(seq, mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007022 seq_printf(seq, "\n ");
NeilBrown8e1b39d2005-11-08 21:39:41 -08007023 if (mddev->pers->sync_request) {
7024 if (mddev->curr_resync > 2) {
NeilBrownd710e132008-10-13 11:55:12 +11007025 status_resync(seq, mddev);
NeilBrown8e1b39d2005-11-08 21:39:41 -08007026 seq_printf(seq, "\n ");
7027 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
7028 seq_printf(seq, "\tresync=DELAYED\n ");
7029 else if (mddev->recovery_cp < MaxSector)
7030 seq_printf(seq, "\tresync=PENDING\n ");
7031 }
NeilBrown32a76272005-06-21 17:17:14 -07007032 } else
7033 seq_printf(seq, "\n ");
7034
NeilBrown57148962012-03-19 12:46:40 +11007035 bitmap_status(seq, mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007036
7037 seq_printf(seq, "\n");
7038 }
7039 mddev_unlock(mddev);
7040
7041 return 0;
7042}
7043
Jan Engelhardt110518b2009-05-07 12:49:37 +10007044static const struct seq_operations md_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007045 .start = md_seq_start,
7046 .next = md_seq_next,
7047 .stop = md_seq_stop,
7048 .show = md_seq_show,
7049};
7050
7051static int md_seq_open(struct inode *inode, struct file *file)
7052{
Kay Sieversf1514632011-07-12 20:48:39 +02007053 struct seq_file *seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007054 int error;
7055
7056 error = seq_open(file, &md_seq_ops);
NeilBrownd7603b72006-01-06 00:20:30 -08007057 if (error)
Kay Sieversf1514632011-07-12 20:48:39 +02007058 return error;
7059
7060 seq = file->private_data;
7061 seq->poll_event = atomic_read(&md_event_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007062 return error;
7063}
7064
NeilBrownd7603b72006-01-06 00:20:30 -08007065static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7066{
Kay Sieversf1514632011-07-12 20:48:39 +02007067 struct seq_file *seq = filp->private_data;
NeilBrownd7603b72006-01-06 00:20:30 -08007068 int mask;
7069
7070 poll_wait(filp, &md_event_waiters, wait);
7071
7072 /* always allow read */
7073 mask = POLLIN | POLLRDNORM;
7074
Kay Sieversf1514632011-07-12 20:48:39 +02007075 if (seq->poll_event != atomic_read(&md_event_count))
NeilBrownd7603b72006-01-06 00:20:30 -08007076 mask |= POLLERR | POLLPRI;
7077 return mask;
7078}
7079
Arjan van de Venfa027c22007-02-12 00:55:33 -08007080static const struct file_operations md_seq_fops = {
Akinobu Mitae24650c2006-10-17 00:09:38 -07007081 .owner = THIS_MODULE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007082 .open = md_seq_open,
7083 .read = seq_read,
7084 .llseek = seq_lseek,
Martin Peschkec3f94b402007-05-09 02:35:35 -07007085 .release = seq_release_private,
NeilBrownd7603b72006-01-06 00:20:30 -08007086 .poll = mdstat_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007087};
7088
NeilBrown84fc4b52011-10-11 16:49:58 +11007089int register_md_personality(struct md_personality *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007090{
Linus Torvalds1da177e2005-04-16 15:20:36 -07007091 spin_lock(&pers_lock);
NeilBrown2604b702006-01-06 00:20:36 -08007092 list_add_tail(&p->list, &pers_list);
7093 printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007094 spin_unlock(&pers_lock);
7095 return 0;
7096}
7097
NeilBrown84fc4b52011-10-11 16:49:58 +11007098int unregister_md_personality(struct md_personality *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007099{
NeilBrown2604b702006-01-06 00:20:36 -08007100 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007101 spin_lock(&pers_lock);
NeilBrown2604b702006-01-06 00:20:36 -08007102 list_del_init(&p->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007103 spin_unlock(&pers_lock);
7104 return 0;
7105}
7106
NeilBrownfd01b882011-10-11 16:47:53 +11007107static int is_mddev_idle(struct mddev *mddev, int init)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007108{
NeilBrown3cb03002011-10-11 16:45:26 +11007109 struct md_rdev * rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007110 int idle;
NeilBrowneea1bf32009-03-31 14:27:02 +11007111 int curr_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007112
7113 idle = 1;
NeilBrown4b809912008-07-21 17:05:25 +10007114 rcu_read_lock();
7115 rdev_for_each_rcu(rdev, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007116 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
NeilBrowneea1bf32009-03-31 14:27:02 +11007117 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7118 (int)part_stat_read(&disk->part0, sectors[1]) -
7119 atomic_read(&disk->sync_io);
NeilBrown713f6ab2007-07-17 04:06:12 -07007120 /* sync IO will cause sync_io to increase before the disk_stats
7121 * as sync_io is counted when a request starts, and
7122 * disk_stats is counted when it completes.
7123 * So resync activity will cause curr_events to be smaller than
7124 * when there was no such activity.
7125 * non-sync IO will cause disk_stat to increase without
7126 * increasing sync_io so curr_events will (eventually)
7127 * be larger than it was before. Once it becomes
7128 * substantially larger, the test below will cause
7129 * the array to appear non-idle, and resync will slow
7130 * down.
7131 * If there is a lot of outstanding resync activity when
7132 * we set last_event to curr_events, then all that activity
7133 * completing might cause the array to appear non-idle
7134 * and resync will be slowed down even though there might
7135 * not have been non-resync activity. This will only
7136 * happen once though. 'last_events' will soon reflect
7137 * the state where there is little or no outstanding
7138 * resync requests, and further resync activity will
7139 * always make curr_events less than last_events.
NeilBrownc0e48522005-11-18 01:11:01 -08007140 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07007141 */
NeilBrowneea1bf32009-03-31 14:27:02 +11007142 if (init || curr_events - rdev->last_events > 64) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007143 rdev->last_events = curr_events;
7144 idle = 0;
7145 }
7146 }
NeilBrown4b809912008-07-21 17:05:25 +10007147 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007148 return idle;
7149}
7150
NeilBrownfd01b882011-10-11 16:47:53 +11007151void md_done_sync(struct mddev *mddev, int blocks, int ok)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007152{
7153 /* another "blocks" (512byte) blocks have been synced */
7154 atomic_sub(blocks, &mddev->recovery_active);
7155 wake_up(&mddev->recovery_wait);
7156 if (!ok) {
NeilBrowndfc70642008-05-23 13:04:39 -07007157 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007158 md_wakeup_thread(mddev->thread);
7159 // stop recovery, signal do_sync ....
7160 }
7161}
7162
7163
NeilBrown06d91a52005-06-21 17:17:12 -07007164/* md_write_start(mddev, bi)
7165 * If we need to update some array metadata (e.g. 'active' flag
NeilBrown3d310eb2005-06-21 17:17:26 -07007166 * in superblock) before writing, schedule a superblock update
7167 * and wait for it to complete.
NeilBrown06d91a52005-06-21 17:17:12 -07007168 */
NeilBrownfd01b882011-10-11 16:47:53 +11007169void md_write_start(struct mddev *mddev, struct bio *bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007170{
Neil Brown0fd62b82008-06-28 08:31:36 +10007171 int did_change = 0;
NeilBrown06d91a52005-06-21 17:17:12 -07007172 if (bio_data_dir(bi) != WRITE)
NeilBrown3d310eb2005-06-21 17:17:26 -07007173 return;
NeilBrown06d91a52005-06-21 17:17:12 -07007174
NeilBrownf91de922005-11-08 21:39:36 -08007175 BUG_ON(mddev->ro == 1);
7176 if (mddev->ro == 2) {
7177 /* need to switch to read/write */
7178 mddev->ro = 0;
7179 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7180 md_wakeup_thread(mddev->thread);
NeilBrown25156192008-03-04 14:29:32 -08007181 md_wakeup_thread(mddev->sync_thread);
Neil Brown0fd62b82008-06-28 08:31:36 +10007182 did_change = 1;
NeilBrownf91de922005-11-08 21:39:36 -08007183 }
NeilBrown06d91a52005-06-21 17:17:12 -07007184 atomic_inc(&mddev->writes_pending);
NeilBrown31a59e32008-04-30 00:52:30 -07007185 if (mddev->safemode == 1)
7186 mddev->safemode = 0;
NeilBrown06d91a52005-06-21 17:17:12 -07007187 if (mddev->in_sync) {
NeilBrowna9701a32005-11-08 21:39:34 -08007188 spin_lock_irq(&mddev->write_lock);
NeilBrown3d310eb2005-06-21 17:17:26 -07007189 if (mddev->in_sync) {
7190 mddev->in_sync = 0;
NeilBrown850b2b42006-10-03 01:15:46 -07007191 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrown070dc6d2010-08-30 17:33:34 +10007192 set_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrown3d310eb2005-06-21 17:17:26 -07007193 md_wakeup_thread(mddev->thread);
Neil Brown0fd62b82008-06-28 08:31:36 +10007194 did_change = 1;
NeilBrown3d310eb2005-06-21 17:17:26 -07007195 }
NeilBrowna9701a32005-11-08 21:39:34 -08007196 spin_unlock_irq(&mddev->write_lock);
NeilBrown06d91a52005-06-21 17:17:12 -07007197 }
Neil Brown0fd62b82008-06-28 08:31:36 +10007198 if (did_change)
NeilBrown00bcb4a2010-06-01 19:37:23 +10007199 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrown09a44cc2008-05-23 13:04:36 -07007200 wait_event(mddev->sb_wait,
NeilBrown09a44cc2008-05-23 13:04:36 -07007201 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007202}
7203
NeilBrownfd01b882011-10-11 16:47:53 +11007204void md_write_end(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007205{
7206 if (atomic_dec_and_test(&mddev->writes_pending)) {
7207 if (mddev->safemode == 2)
7208 md_wakeup_thread(mddev->thread);
NeilBrown16f17b32006-06-26 00:27:37 -07007209 else if (mddev->safemode_delay)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007210 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7211 }
7212}
7213
NeilBrown2a2275d2007-01-26 00:57:11 -08007214/* md_allow_write(mddev)
7215 * Calling this ensures that the array is marked 'active' so that writes
7216 * may proceed without blocking. It is important to call this before
7217 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7218 * Must be called with mddev_lock held.
Dan Williamsb5470dc2008-06-27 21:44:04 -07007219 *
7220 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7221 * is dropped, so return -EAGAIN after notifying userspace.
NeilBrown2a2275d2007-01-26 00:57:11 -08007222 */
NeilBrownfd01b882011-10-11 16:47:53 +11007223int md_allow_write(struct mddev *mddev)
NeilBrown2a2275d2007-01-26 00:57:11 -08007224{
7225 if (!mddev->pers)
Dan Williamsb5470dc2008-06-27 21:44:04 -07007226 return 0;
NeilBrown2a2275d2007-01-26 00:57:11 -08007227 if (mddev->ro)
Dan Williamsb5470dc2008-06-27 21:44:04 -07007228 return 0;
Neil Brown1a0fd492008-06-28 08:31:27 +10007229 if (!mddev->pers->sync_request)
Dan Williamsb5470dc2008-06-27 21:44:04 -07007230 return 0;
NeilBrown2a2275d2007-01-26 00:57:11 -08007231
7232 spin_lock_irq(&mddev->write_lock);
7233 if (mddev->in_sync) {
7234 mddev->in_sync = 0;
7235 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrown070dc6d2010-08-30 17:33:34 +10007236 set_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrown2a2275d2007-01-26 00:57:11 -08007237 if (mddev->safemode_delay &&
7238 mddev->safemode == 0)
7239 mddev->safemode = 1;
7240 spin_unlock_irq(&mddev->write_lock);
7241 md_update_sb(mddev, 0);
NeilBrown00bcb4a2010-06-01 19:37:23 +10007242 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrown2a2275d2007-01-26 00:57:11 -08007243 } else
7244 spin_unlock_irq(&mddev->write_lock);
Dan Williamsb5470dc2008-06-27 21:44:04 -07007245
NeilBrown070dc6d2010-08-30 17:33:34 +10007246 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
Dan Williamsb5470dc2008-06-27 21:44:04 -07007247 return -EAGAIN;
7248 else
7249 return 0;
NeilBrown2a2275d2007-01-26 00:57:11 -08007250}
7251EXPORT_SYMBOL_GPL(md_allow_write);
7252
Linus Torvalds1da177e2005-04-16 15:20:36 -07007253#define SYNC_MARKS 10
7254#define SYNC_MARK_STEP (3*HZ)
Shaohua Li4ed87312012-10-11 13:34:00 +11007255void md_do_sync(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007256{
Shaohua Li4ed87312012-10-11 13:34:00 +11007257 struct mddev *mddev = thread->mddev;
NeilBrownfd01b882011-10-11 16:47:53 +11007258 struct mddev *mddev2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007259 unsigned int currspeed = 0,
7260 window;
NeilBrown57afd892005-06-21 17:17:13 -07007261 sector_t max_sectors,j, io_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007262 unsigned long mark[SYNC_MARKS];
7263 sector_t mark_cnt[SYNC_MARKS];
7264 int last_mark,m;
7265 struct list_head *tmp;
7266 sector_t last_check;
NeilBrown57afd892005-06-21 17:17:13 -07007267 int skipped = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11007268 struct md_rdev *rdev;
NeilBrown61df9d92006-10-03 01:15:57 -07007269 char *desc;
majianpeng7c2c57c2012-07-03 12:12:26 +10007270 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007271
7272 /* just incase thread restarts... */
7273 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7274 return;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007275 if (mddev->ro) /* never try to sync a read-only array */
7276 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007277
NeilBrown61df9d92006-10-03 01:15:57 -07007278 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7279 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
7280 desc = "data-check";
7281 else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7282 desc = "requested-resync";
7283 else
7284 desc = "resync";
7285 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7286 desc = "reshape";
7287 else
7288 desc = "recovery";
7289
Linus Torvalds1da177e2005-04-16 15:20:36 -07007290 /* we overload curr_resync somewhat here.
7291 * 0 == not engaged in resync at all
7292 * 2 == checking that there is no conflict with another sync
7293 * 1 == like 2, but have yielded to allow conflicting resync to
7294 * commense
7295 * other == active in resync - this many blocks
7296 *
7297 * Before starting a resync we must have set curr_resync to
7298 * 2, and then checked that every "conflicting" array has curr_resync
7299 * less than ours. When we find one that is the same or higher
7300 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7301 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7302 * This will mean we have to start checking from the beginning again.
7303 *
7304 */
7305
7306 do {
7307 mddev->curr_resync = 2;
7308
7309 try_again:
NeilBrown404e4b42009-12-30 15:25:23 +11007310 if (kthread_should_stop())
NeilBrown6985c432005-10-19 21:23:47 -07007311 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrown404e4b42009-12-30 15:25:23 +11007312
7313 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -07007314 goto skip;
NeilBrown29ac4aa2008-02-06 01:39:58 -08007315 for_each_mddev(mddev2, tmp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007316 if (mddev2 == mddev)
7317 continue;
Bernd Schubert90b08712008-05-23 13:04:38 -07007318 if (!mddev->parallel_resync
7319 && mddev2->curr_resync
7320 && match_mddev_units(mddev, mddev2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007321 DEFINE_WAIT(wq);
7322 if (mddev < mddev2 && mddev->curr_resync == 2) {
7323 /* arbitrarily yield */
7324 mddev->curr_resync = 1;
7325 wake_up(&resync_wait);
7326 }
7327 if (mddev > mddev2 && mddev->curr_resync == 1)
7328 /* no need to wait here, we can wait the next
7329 * time 'round when curr_resync == 2
7330 */
7331 continue;
NeilBrown97441972008-09-19 11:49:54 +10007332 /* We need to wait 'interruptible' so as not to
7333 * contribute to the load average, and not to
7334 * be caught by 'softlockup'
7335 */
7336 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
NeilBrown787453c2005-11-08 21:39:43 -08007337 if (!kthread_should_stop() &&
NeilBrown8712e552005-10-26 01:58:58 -07007338 mddev2->curr_resync >= mddev->curr_resync) {
NeilBrown61df9d92006-10-03 01:15:57 -07007339 printk(KERN_INFO "md: delaying %s of %s"
7340 " until %s has finished (they"
Linus Torvalds1da177e2005-04-16 15:20:36 -07007341 " share one or more physical units)\n",
NeilBrown61df9d92006-10-03 01:15:57 -07007342 desc, mdname(mddev), mdname(mddev2));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007343 mddev_put(mddev2);
NeilBrown97441972008-09-19 11:49:54 +10007344 if (signal_pending(current))
7345 flush_signals(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007346 schedule();
7347 finish_wait(&resync_wait, &wq);
7348 goto try_again;
7349 }
7350 finish_wait(&resync_wait, &wq);
7351 }
7352 }
7353 } while (mddev->curr_resync < 2);
7354
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007355 j = 0;
NeilBrown9d888832005-11-08 21:39:26 -08007356 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007357 /* resync follows the size requested by the personality,
NeilBrown57afd892005-06-21 17:17:13 -07007358 * which defaults to physical size, but can be virtual size
Linus Torvalds1da177e2005-04-16 15:20:36 -07007359 */
7360 max_sectors = mddev->resync_max_sectors;
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11007361 atomic64_set(&mddev->resync_mismatches, 0);
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007362 /* we don't use the checkpoint if there's a bitmap */
Neil Brown5e96ee62008-06-28 08:31:24 +10007363 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7364 j = mddev->resync_min;
7365 else if (!mddev->bitmap)
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007366 j = mddev->recovery_cp;
Neil Brown5e96ee62008-06-28 08:31:24 +10007367
NeilBrownccfcc3c2006-03-27 01:18:09 -08007368 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
NeilBrownc804cde2012-05-21 09:28:33 +10007369 max_sectors = mddev->resync_max_sectors;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007370 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007371 /* recovery follows the physical size of devices */
Andre Noll58c0fed2009-03-31 14:33:13 +11007372 max_sectors = mddev->dev_sectors;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007373 j = MaxSector;
Dan Williams4e59ca72009-12-12 21:17:06 -07007374 rcu_read_lock();
NeilBrowndafb20f2012-03-19 12:46:39 +11007375 rdev_for_each_rcu(rdev, mddev)
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007376 if (rdev->raid_disk >= 0 &&
7377 !test_bit(Faulty, &rdev->flags) &&
7378 !test_bit(In_sync, &rdev->flags) &&
7379 rdev->recovery_offset < j)
7380 j = rdev->recovery_offset;
Dan Williams4e59ca72009-12-12 21:17:06 -07007381 rcu_read_unlock();
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007382 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007383
NeilBrown61df9d92006-10-03 01:15:57 -07007384 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7385 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7386 " %d KB/sec/disk.\n", speed_min(mddev));
Adrian Bunk338cec32005-09-10 00:26:54 -07007387 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
NeilBrown61df9d92006-10-03 01:15:57 -07007388 "(but not more than %d KB/sec) for %s.\n",
7389 speed_max(mddev), desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007390
NeilBrowneea1bf32009-03-31 14:27:02 +11007391 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007392
NeilBrown57afd892005-06-21 17:17:13 -07007393 io_sectors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007394 for (m = 0; m < SYNC_MARKS; m++) {
7395 mark[m] = jiffies;
NeilBrown57afd892005-06-21 17:17:13 -07007396 mark_cnt[m] = io_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007397 }
7398 last_mark = 0;
7399 mddev->resync_mark = mark[last_mark];
7400 mddev->resync_mark_cnt = mark_cnt[last_mark];
7401
7402 /*
7403 * Tune reconstruction:
7404 */
7405 window = 32*(PAGE_SIZE/512);
Jonathan Brassowac424502011-06-07 17:48:35 -05007406 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7407 window/2, (unsigned long long)max_sectors/2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007408
7409 atomic_set(&mddev->recovery_active, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007410 last_check = 0;
7411
7412 if (j>2) {
7413 printk(KERN_INFO
NeilBrown61df9d92006-10-03 01:15:57 -07007414 "md: resuming %s of %s from checkpoint.\n",
7415 desc, mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007416 mddev->curr_resync = j;
7417 }
NeilBrown75d3da42011-01-14 09:14:34 +11007418 mddev->curr_resync_completed = j;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007419
majianpeng7c2c57c2012-07-03 12:12:26 +10007420 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007421 while (j < max_sectors) {
NeilBrown57afd892005-06-21 17:17:13 -07007422 sector_t sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007423
NeilBrown57afd892005-06-21 17:17:13 -07007424 skipped = 0;
NeilBrown97e4f422009-03-31 14:33:13 +11007425
NeilBrown7a91ee12009-05-26 12:57:21 +10007426 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7427 ((mddev->curr_resync > mddev->curr_resync_completed &&
7428 (mddev->curr_resync - mddev->curr_resync_completed)
7429 > (max_sectors >> 4)) ||
7430 (j - mddev->curr_resync_completed)*2
7431 >= mddev->resync_max - mddev->curr_resync_completed
7432 )) {
NeilBrown97e4f422009-03-31 14:33:13 +11007433 /* time to update curr_resync_completed */
NeilBrown97e4f422009-03-31 14:33:13 +11007434 wait_event(mddev->recovery_wait,
7435 atomic_read(&mddev->recovery_active) == 0);
NeilBrown75d3da42011-01-14 09:14:34 +11007436 mddev->curr_resync_completed = j;
NeilBrown070dc6d2010-08-30 17:33:34 +10007437 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrownacb180b2009-04-14 16:28:34 +10007438 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown97e4f422009-03-31 14:33:13 +11007439 }
NeilBrownacb180b2009-04-14 16:28:34 +10007440
NeilBrowne62e58a2009-07-01 13:15:35 +10007441 while (j >= mddev->resync_max && !kthread_should_stop()) {
7442 /* As this condition is controlled by user-space,
7443 * we can block indefinitely, so use '_interruptible'
7444 * to avoid triggering warnings.
7445 */
7446 flush_signals(current); /* just in case */
7447 wait_event_interruptible(mddev->recovery_wait,
7448 mddev->resync_max > j
7449 || kthread_should_stop());
7450 }
NeilBrownacb180b2009-04-14 16:28:34 +10007451
7452 if (kthread_should_stop())
7453 goto interrupted;
7454
NeilBrown57afd892005-06-21 17:17:13 -07007455 sectors = mddev->pers->sync_request(mddev, j, &skipped,
NeilBrownc6207272008-02-06 01:39:52 -08007456 currspeed < speed_min(mddev));
NeilBrown57afd892005-06-21 17:17:13 -07007457 if (sectors == 0) {
NeilBrowndfc70642008-05-23 13:04:39 -07007458 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007459 goto out;
7460 }
NeilBrown57afd892005-06-21 17:17:13 -07007461
7462 if (!skipped) { /* actual IO requested */
7463 io_sectors += sectors;
7464 atomic_add(sectors, &mddev->recovery_active);
7465 }
7466
NeilBrowne875ece2011-07-28 11:39:24 +10007467 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7468 break;
7469
Linus Torvalds1da177e2005-04-16 15:20:36 -07007470 j += sectors;
7471 if (j>1) mddev->curr_resync = j;
NeilBrownff4e8d92006-07-10 04:44:16 -07007472 mddev->curr_mark_cnt = io_sectors;
NeilBrownd7603b72006-01-06 00:20:30 -08007473 if (last_check == 0)
NeilBrowne875ece2011-07-28 11:39:24 +10007474 /* this is the earliest that rebuild will be
NeilBrownd7603b72006-01-06 00:20:30 -08007475 * visible in /proc/mdstat
7476 */
7477 md_new_event(mddev);
NeilBrown57afd892005-06-21 17:17:13 -07007478
7479 if (last_check + window > io_sectors || j == max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007480 continue;
7481
NeilBrown57afd892005-06-21 17:17:13 -07007482 last_check = io_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007483 repeat:
7484 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7485 /* step marks */
7486 int next = (last_mark+1) % SYNC_MARKS;
7487
7488 mddev->resync_mark = mark[next];
7489 mddev->resync_mark_cnt = mark_cnt[next];
7490 mark[next] = jiffies;
NeilBrown57afd892005-06-21 17:17:13 -07007491 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007492 last_mark = next;
7493 }
7494
7495
NeilBrownc6207272008-02-06 01:39:52 -08007496 if (kthread_should_stop())
7497 goto interrupted;
7498
Linus Torvalds1da177e2005-04-16 15:20:36 -07007499
7500 /*
7501 * this loop exits only if either when we are slower than
7502 * the 'hard' speed limit, or the system was IO-idle for
7503 * a jiffy.
7504 * the system might be non-idle CPU-wise, but we only care
7505 * about not overloading the IO subsystem. (things like an
7506 * e2fsck being done on the RAID array should execute fast)
7507 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07007508 cond_resched();
7509
NeilBrown57afd892005-06-21 17:17:13 -07007510 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
7511 /((jiffies-mddev->resync_mark)/HZ +1) +1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007512
NeilBrown88202a02006-01-06 00:21:36 -08007513 if (currspeed > speed_min(mddev)) {
7514 if ((currspeed > speed_max(mddev)) ||
NeilBrowneea1bf32009-03-31 14:27:02 +11007515 !is_mddev_idle(mddev, 0)) {
NeilBrownc0e48522005-11-18 01:11:01 -08007516 msleep(500);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007517 goto repeat;
7518 }
7519 }
7520 }
NeilBrown61df9d92006-10-03 01:15:57 -07007521 printk(KERN_INFO "md: %s: %s done.\n",mdname(mddev), desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007522 /*
7523 * this also signals 'finished resyncing' to md_stop
7524 */
7525 out:
majianpeng7c2c57c2012-07-03 12:12:26 +10007526 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007527 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7528
7529 /* tell personality that we are finished */
NeilBrown57afd892005-06-21 17:17:13 -07007530 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007531
NeilBrowndfc70642008-05-23 13:04:39 -07007532 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007533 mddev->curr_resync > 2) {
7534 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7535 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7536 if (mddev->curr_resync >= mddev->recovery_cp) {
7537 printk(KERN_INFO
NeilBrown61df9d92006-10-03 01:15:57 -07007538 "md: checkpointing %s of %s.\n",
7539 desc, mdname(mddev));
NeilBrowndb91ff52012-02-07 12:01:51 +11007540 mddev->recovery_cp =
7541 mddev->curr_resync_completed;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007542 }
7543 } else
7544 mddev->recovery_cp = MaxSector;
7545 } else {
7546 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7547 mddev->curr_resync = MaxSector;
Dan Williams4e59ca72009-12-12 21:17:06 -07007548 rcu_read_lock();
NeilBrowndafb20f2012-03-19 12:46:39 +11007549 rdev_for_each_rcu(rdev, mddev)
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007550 if (rdev->raid_disk >= 0 &&
NeilBrown70fffd02010-06-16 17:01:25 +10007551 mddev->delta_disks >= 0 &&
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007552 !test_bit(Faulty, &rdev->flags) &&
7553 !test_bit(In_sync, &rdev->flags) &&
7554 rdev->recovery_offset < mddev->curr_resync)
7555 rdev->recovery_offset = mddev->curr_resync;
Dan Williams4e59ca72009-12-12 21:17:06 -07007556 rcu_read_unlock();
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007558 }
NeilBrowndb91ff52012-02-07 12:01:51 +11007559 skip:
NeilBrown17571282006-12-10 02:20:52 -08007560 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007561
NeilBrownc07b70a2009-12-14 12:49:48 +11007562 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7563 /* We completed so min/max setting can be forgotten if used. */
7564 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7565 mddev->resync_min = 0;
7566 mddev->resync_max = MaxSector;
7567 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7568 mddev->resync_min = mddev->curr_resync_completed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007569 mddev->curr_resync = 0;
7570 wake_up(&resync_wait);
7571 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7572 md_wakeup_thread(mddev->thread);
NeilBrownc6207272008-02-06 01:39:52 -08007573 return;
7574
7575 interrupted:
7576 /*
7577 * got a signal, exit.
7578 */
7579 printk(KERN_INFO
7580 "md: md_do_sync() got signal ... exiting\n");
7581 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
7582 goto out;
7583
Linus Torvalds1da177e2005-04-16 15:20:36 -07007584}
NeilBrown29269552006-03-27 01:18:10 -08007585EXPORT_SYMBOL_GPL(md_do_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007586
NeilBrownfd01b882011-10-11 16:47:53 +11007587static int remove_and_add_spares(struct mddev *mddev)
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007588{
NeilBrown3cb03002011-10-11 16:45:26 +11007589 struct md_rdev *rdev;
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007590 int spares = 0;
NeilBrownf2a371c2012-01-09 00:46:41 +11007591 int removed = 0;
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007592
NeilBrown97e4f422009-03-31 14:33:13 +11007593 mddev->curr_resync_completed = 0;
7594
NeilBrowndafb20f2012-03-19 12:46:39 +11007595 rdev_for_each(rdev, mddev)
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007596 if (rdev->raid_disk >= 0 &&
Dan Williams6bfe0b42008-04-30 00:52:32 -07007597 !test_bit(Blocked, &rdev->flags) &&
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007598 (test_bit(Faulty, &rdev->flags) ||
7599 ! test_bit(In_sync, &rdev->flags)) &&
7600 atomic_read(&rdev->nr_pending)==0) {
7601 if (mddev->pers->hot_remove_disk(
NeilBrownb8321b62011-12-23 10:17:51 +11007602 mddev, rdev) == 0) {
Namhyung Kim36fad852011-07-27 11:00:36 +10007603 sysfs_unlink_rdev(mddev, rdev);
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007604 rdev->raid_disk = -1;
NeilBrownf2a371c2012-01-09 00:46:41 +11007605 removed++;
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007606 }
7607 }
NeilBrownf2a371c2012-01-09 00:46:41 +11007608 if (removed)
7609 sysfs_notify(&mddev->kobj, NULL,
7610 "degraded");
7611
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007612
NeilBrowndafb20f2012-03-19 12:46:39 +11007613 rdev_for_each(rdev, mddev) {
NeilBrown7bfec5f2011-12-23 10:17:53 +11007614 if (rdev->raid_disk >= 0 &&
7615 !test_bit(In_sync, &rdev->flags) &&
7616 !test_bit(Faulty, &rdev->flags))
7617 spares++;
7618 if (rdev->raid_disk < 0
7619 && !test_bit(Faulty, &rdev->flags)) {
7620 rdev->recovery_offset = 0;
7621 if (mddev->pers->
7622 hot_add_disk(mddev, rdev) == 0) {
7623 if (sysfs_link_rdev(mddev, rdev))
7624 /* failure here is OK */;
NeilBrowndfc70642008-05-23 13:04:39 -07007625 spares++;
NeilBrown7bfec5f2011-12-23 10:17:53 +11007626 md_new_event(mddev);
7627 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007628 }
NeilBrowndfc70642008-05-23 13:04:39 -07007629 }
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007630 }
7631 return spares;
7632}
NeilBrown7ebc0be2011-01-14 09:14:33 +11007633
NeilBrownfd01b882011-10-11 16:47:53 +11007634static void reap_sync_thread(struct mddev *mddev)
NeilBrown7ebc0be2011-01-14 09:14:33 +11007635{
NeilBrown3cb03002011-10-11 16:45:26 +11007636 struct md_rdev *rdev;
NeilBrown7ebc0be2011-01-14 09:14:33 +11007637
7638 /* resync has finished, collect result */
NeilBrown01f96c02011-09-21 15:30:20 +10007639 md_unregister_thread(&mddev->sync_thread);
NeilBrown7ebc0be2011-01-14 09:14:33 +11007640 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
7641 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
7642 /* success...*/
7643 /* activate any spares */
7644 if (mddev->pers->spare_active(mddev))
7645 sysfs_notify(&mddev->kobj, NULL,
7646 "degraded");
7647 }
7648 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7649 mddev->pers->finish_reshape)
7650 mddev->pers->finish_reshape(mddev);
NeilBrown7ebc0be2011-01-14 09:14:33 +11007651
Andrei Warkentind70ed2e2011-10-18 12:16:48 +11007652 /* If array is no-longer degraded, then any saved_raid_disk
7653 * information must be scrapped. Also if any device is now
7654 * In_sync we must scrape the saved_raid_disk for that device
7655 * do the superblock for an incrementally recovered device
7656 * written out.
NeilBrown7ebc0be2011-01-14 09:14:33 +11007657 */
NeilBrowndafb20f2012-03-19 12:46:39 +11007658 rdev_for_each(rdev, mddev)
Andrei Warkentind70ed2e2011-10-18 12:16:48 +11007659 if (!mddev->degraded ||
7660 test_bit(In_sync, &rdev->flags))
NeilBrown7ebc0be2011-01-14 09:14:33 +11007661 rdev->saved_raid_disk = -1;
7662
Andrei Warkentind70ed2e2011-10-18 12:16:48 +11007663 md_update_sb(mddev, 1);
NeilBrown7ebc0be2011-01-14 09:14:33 +11007664 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7665 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7666 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7667 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7668 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7669 /* flag recovery needed just to double check */
7670 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7671 sysfs_notify_dirent_safe(mddev->sysfs_action);
7672 md_new_event(mddev);
Jonathan Brassow768e587e2011-07-27 11:00:37 +10007673 if (mddev->event_work.func)
7674 queue_work(md_misc_wq, &mddev->event_work);
NeilBrown7ebc0be2011-01-14 09:14:33 +11007675}
7676
Linus Torvalds1da177e2005-04-16 15:20:36 -07007677/*
7678 * This routine is regularly called by all per-raid-array threads to
7679 * deal with generic issues like resync and super-block update.
7680 * Raid personalities that don't have a thread (linear/raid0) do not
7681 * need this as they never do any recovery or update the superblock.
7682 *
7683 * It does not do any resync itself, but rather "forks" off other threads
7684 * to do that as needed.
7685 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7686 * "->recovery" and create a thread at ->sync_thread.
NeilBrowndfc70642008-05-23 13:04:39 -07007687 * When the thread finishes it sets MD_RECOVERY_DONE
Linus Torvalds1da177e2005-04-16 15:20:36 -07007688 * and wakeups up this thread which will reap the thread and finish up.
7689 * This thread also removes any faulty devices (with nr_pending == 0).
7690 *
7691 * The overall approach is:
7692 * 1/ if the superblock needs updating, update it.
7693 * 2/ If a recovery thread is running, don't do anything else.
7694 * 3/ If recovery has finished, clean up, possibly marking spares active.
7695 * 4/ If there are any faulty devices, remove them.
7696 * 5/ If array is degraded, try to add spares devices
7697 * 6/ If array has spares or is not in-sync, start a resync thread.
7698 */
NeilBrownfd01b882011-10-11 16:47:53 +11007699void md_check_recovery(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007700{
Jonathan Brassow68866e42011-06-08 15:10:08 +10007701 if (mddev->suspended)
7702 return;
7703
NeilBrown5f404022005-06-21 17:17:16 -07007704 if (mddev->bitmap)
NeilBrownaa5cbd12009-12-14 12:49:46 +11007705 bitmap_daemon_work(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007706
NeilBrownfca4d842005-06-21 17:17:11 -07007707 if (signal_pending(current)) {
NeilBrown31a59e32008-04-30 00:52:30 -07007708 if (mddev->pers->sync_request && !mddev->external) {
NeilBrownfca4d842005-06-21 17:17:11 -07007709 printk(KERN_INFO "md: %s in immediate safe mode\n",
7710 mdname(mddev));
7711 mddev->safemode = 2;
7712 }
7713 flush_signals(current);
7714 }
7715
NeilBrownc89a8ee2008-08-05 15:54:13 +10007716 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7717 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007718 if ( ! (
NeilBrown126925c2010-09-07 17:02:47 +10007719 (mddev->flags & ~ (1<<MD_CHANGE_PENDING)) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07007720 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
NeilBrownfca4d842005-06-21 17:17:11 -07007721 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
NeilBrown31a59e32008-04-30 00:52:30 -07007722 (mddev->external == 0 && mddev->safemode == 1) ||
NeilBrownfca4d842005-06-21 17:17:11 -07007723 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7724 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007725 ))
7726 return;
NeilBrownfca4d842005-06-21 17:17:11 -07007727
NeilBrowndf5b89b2006-03-27 01:18:20 -08007728 if (mddev_trylock(mddev)) {
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007729 int spares = 0;
NeilBrownfca4d842005-06-21 17:17:11 -07007730
NeilBrownc89a8ee2008-08-05 15:54:13 +10007731 if (mddev->ro) {
7732 /* Only thing we do on a ro array is remove
7733 * failed devices.
7734 */
NeilBrown3cb03002011-10-11 16:45:26 +11007735 struct md_rdev *rdev;
NeilBrowndafb20f2012-03-19 12:46:39 +11007736 rdev_for_each(rdev, mddev)
NeilBrowna8c42c72011-01-31 13:47:13 +11007737 if (rdev->raid_disk >= 0 &&
7738 !test_bit(Blocked, &rdev->flags) &&
7739 test_bit(Faulty, &rdev->flags) &&
7740 atomic_read(&rdev->nr_pending)==0) {
7741 if (mddev->pers->hot_remove_disk(
NeilBrownb8321b62011-12-23 10:17:51 +11007742 mddev, rdev) == 0) {
Namhyung Kim36fad852011-07-27 11:00:36 +10007743 sysfs_unlink_rdev(mddev, rdev);
NeilBrowna8c42c72011-01-31 13:47:13 +11007744 rdev->raid_disk = -1;
7745 }
7746 }
NeilBrownc89a8ee2008-08-05 15:54:13 +10007747 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7748 goto unlock;
7749 }
7750
NeilBrown31a59e32008-04-30 00:52:30 -07007751 if (!mddev->external) {
Neil Brown0fd62b82008-06-28 08:31:36 +10007752 int did_change = 0;
NeilBrown31a59e32008-04-30 00:52:30 -07007753 spin_lock_irq(&mddev->write_lock);
7754 if (mddev->safemode &&
7755 !atomic_read(&mddev->writes_pending) &&
7756 !mddev->in_sync &&
7757 mddev->recovery_cp == MaxSector) {
7758 mddev->in_sync = 1;
Neil Brown0fd62b82008-06-28 08:31:36 +10007759 did_change = 1;
NeilBrown070dc6d2010-08-30 17:33:34 +10007760 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrown31a59e32008-04-30 00:52:30 -07007761 }
7762 if (mddev->safemode == 1)
7763 mddev->safemode = 0;
7764 spin_unlock_irq(&mddev->write_lock);
Neil Brown0fd62b82008-06-28 08:31:36 +10007765 if (did_change)
NeilBrown00bcb4a2010-06-01 19:37:23 +10007766 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrownfca4d842005-06-21 17:17:11 -07007767 }
NeilBrownfca4d842005-06-21 17:17:11 -07007768
NeilBrown850b2b42006-10-03 01:15:46 -07007769 if (mddev->flags)
7770 md_update_sb(mddev, 0);
NeilBrown06d91a52005-06-21 17:17:12 -07007771
Linus Torvalds1da177e2005-04-16 15:20:36 -07007772 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7773 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7774 /* resync/recovery still happening */
7775 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7776 goto unlock;
7777 }
7778 if (mddev->sync_thread) {
NeilBrown7ebc0be2011-01-14 09:14:33 +11007779 reap_sync_thread(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007780 goto unlock;
7781 }
Neil Brown72a23c22008-06-28 08:31:41 +10007782 /* Set RUNNING before clearing NEEDED to avoid
7783 * any transients in the value of "sync_action".
7784 */
7785 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
NeilBrown24dd4692005-11-08 21:39:26 -08007786 /* Clear some bits that don't mean anything, but
7787 * might be left set
7788 */
NeilBrown24dd4692005-11-08 21:39:26 -08007789 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7790 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007791
NeilBrowned209582012-04-24 10:23:14 +10007792 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7793 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007794 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007795 /* no recovery is running.
7796 * remove any failed drives, then
7797 * add spares if possible.
7798 * Spare are also removed and re-added, to allow
7799 * the personality to fail the re-add.
7800 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07007801
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007802 if (mddev->reshape_position != MaxSector) {
NeilBrown50ac1682009-06-18 08:47:55 +10007803 if (mddev->pers->check_reshape == NULL ||
7804 mddev->pers->check_reshape(mddev) != 0)
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007805 /* Cannot proceed */
7806 goto unlock;
7807 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
Neil Brown72a23c22008-06-28 08:31:41 +10007808 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007809 } else if ((spares = remove_and_add_spares(mddev))) {
NeilBrown24dd4692005-11-08 21:39:26 -08007810 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7811 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
Dan Williams56ac36d2008-08-07 10:02:47 -07007812 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
Neil Brown72a23c22008-06-28 08:31:41 +10007813 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
NeilBrown24dd4692005-11-08 21:39:26 -08007814 } else if (mddev->recovery_cp < MaxSector) {
7815 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
Neil Brown72a23c22008-06-28 08:31:41 +10007816 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
NeilBrown24dd4692005-11-08 21:39:26 -08007817 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7818 /* nothing to be done ... */
Linus Torvalds1da177e2005-04-16 15:20:36 -07007819 goto unlock;
NeilBrown24dd4692005-11-08 21:39:26 -08007820
Linus Torvalds1da177e2005-04-16 15:20:36 -07007821 if (mddev->pers->sync_request) {
NeilBrownef99bf42012-05-22 13:55:08 +10007822 if (spares) {
NeilBrowna654b9d82005-06-21 17:17:27 -07007823 /* We are adding a device or devices to an array
7824 * which has the bitmap stored on all devices.
7825 * So make sure all bitmap pages get written
7826 */
7827 bitmap_write_all(mddev->bitmap);
7828 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007829 mddev->sync_thread = md_register_thread(md_do_sync,
7830 mddev,
NeilBrown0da3c612009-09-23 18:09:45 +10007831 "resync");
Linus Torvalds1da177e2005-04-16 15:20:36 -07007832 if (!mddev->sync_thread) {
7833 printk(KERN_ERR "%s: could not start resync"
7834 " thread...\n",
7835 mdname(mddev));
7836 /* leave the spares where they are, it shouldn't hurt */
NeilBrown7ebc0be2011-01-14 09:14:33 +11007837 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7838 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7839 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7840 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7841 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
NeilBrownd7603b72006-01-06 00:20:30 -08007842 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07007843 md_wakeup_thread(mddev->sync_thread);
NeilBrown00bcb4a2010-06-01 19:37:23 +10007844 sysfs_notify_dirent_safe(mddev->sysfs_action);
NeilBrownd7603b72006-01-06 00:20:30 -08007845 md_new_event(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007846 }
7847 unlock:
Neil Brown72a23c22008-06-28 08:31:41 +10007848 if (!mddev->sync_thread) {
7849 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
7850 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7851 &mddev->recovery))
NeilBrown0c3573f2009-01-09 08:31:05 +11007852 if (mddev->sysfs_action)
NeilBrown00bcb4a2010-06-01 19:37:23 +10007853 sysfs_notify_dirent_safe(mddev->sysfs_action);
Neil Brown72a23c22008-06-28 08:31:41 +10007854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007855 mddev_unlock(mddev);
7856 }
7857}
7858
NeilBrownfd01b882011-10-11 16:47:53 +11007859void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
Dan Williams6bfe0b42008-04-30 00:52:32 -07007860{
NeilBrown00bcb4a2010-06-01 19:37:23 +10007861 sysfs_notify_dirent_safe(rdev->sysfs_state);
Dan Williams6bfe0b42008-04-30 00:52:32 -07007862 wait_event_timeout(rdev->blocked_wait,
NeilBrownde393cd2011-07-28 11:31:48 +10007863 !test_bit(Blocked, &rdev->flags) &&
7864 !test_bit(BlockedBadBlocks, &rdev->flags),
Dan Williams6bfe0b42008-04-30 00:52:32 -07007865 msecs_to_jiffies(5000));
7866 rdev_dec_pending(rdev, mddev);
7867}
7868EXPORT_SYMBOL(md_wait_for_blocked_rdev);
7869
NeilBrownc6563a82012-05-21 09:27:00 +10007870void md_finish_reshape(struct mddev *mddev)
7871{
7872 /* called be personality module when reshape completes. */
7873 struct md_rdev *rdev;
7874
7875 rdev_for_each(rdev, mddev) {
7876 if (rdev->data_offset > rdev->new_data_offset)
7877 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
7878 else
7879 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
7880 rdev->data_offset = rdev->new_data_offset;
7881 }
7882}
7883EXPORT_SYMBOL(md_finish_reshape);
NeilBrown2230dfe2011-07-28 11:31:46 +10007884
7885/* Bad block management.
7886 * We can record which blocks on each device are 'bad' and so just
7887 * fail those blocks, or that stripe, rather than the whole device.
7888 * Entries in the bad-block table are 64bits wide. This comprises:
7889 * Length of bad-range, in sectors: 0-511 for lengths 1-512
7890 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
7891 * A 'shift' can be set so that larger blocks are tracked and
7892 * consequently larger devices can be covered.
7893 * 'Acknowledged' flag - 1 bit. - the most significant bit.
7894 *
7895 * Locking of the bad-block table uses a seqlock so md_is_badblock
7896 * might need to retry if it is very unlucky.
7897 * We will sometimes want to check for bad blocks in a bi_end_io function,
7898 * so we use the write_seqlock_irq variant.
7899 *
7900 * When looking for a bad block we specify a range and want to
7901 * know if any block in the range is bad. So we binary-search
7902 * to the last range that starts at-or-before the given endpoint,
7903 * (or "before the sector after the target range")
7904 * then see if it ends after the given start.
7905 * We return
7906 * 0 if there are no known bad blocks in the range
7907 * 1 if there are known bad block which are all acknowledged
7908 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
7909 * plus the start/length of the first bad section we overlap.
7910 */
7911int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
7912 sector_t *first_bad, int *bad_sectors)
7913{
7914 int hi;
7915 int lo = 0;
7916 u64 *p = bb->page;
7917 int rv = 0;
7918 sector_t target = s + sectors;
7919 unsigned seq;
7920
7921 if (bb->shift > 0) {
7922 /* round the start down, and the end up */
7923 s >>= bb->shift;
7924 target += (1<<bb->shift) - 1;
7925 target >>= bb->shift;
7926 sectors = target - s;
7927 }
7928 /* 'target' is now the first block after the bad range */
7929
7930retry:
7931 seq = read_seqbegin(&bb->lock);
7932
7933 hi = bb->count;
7934
7935 /* Binary search between lo and hi for 'target'
7936 * i.e. for the last range that starts before 'target'
7937 */
7938 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
7939 * are known not to be the last range before target.
7940 * VARIANT: hi-lo is the number of possible
7941 * ranges, and decreases until it reaches 1
7942 */
7943 while (hi - lo > 1) {
7944 int mid = (lo + hi) / 2;
7945 sector_t a = BB_OFFSET(p[mid]);
7946 if (a < target)
7947 /* This could still be the one, earlier ranges
7948 * could not. */
7949 lo = mid;
7950 else
7951 /* This and later ranges are definitely out. */
7952 hi = mid;
7953 }
7954 /* 'lo' might be the last that started before target, but 'hi' isn't */
7955 if (hi > lo) {
7956 /* need to check all range that end after 's' to see if
7957 * any are unacknowledged.
7958 */
7959 while (lo >= 0 &&
7960 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
7961 if (BB_OFFSET(p[lo]) < target) {
7962 /* starts before the end, and finishes after
7963 * the start, so they must overlap
7964 */
7965 if (rv != -1 && BB_ACK(p[lo]))
7966 rv = 1;
7967 else
7968 rv = -1;
7969 *first_bad = BB_OFFSET(p[lo]);
7970 *bad_sectors = BB_LEN(p[lo]);
7971 }
7972 lo--;
7973 }
7974 }
7975
7976 if (read_seqretry(&bb->lock, seq))
7977 goto retry;
7978
7979 return rv;
7980}
7981EXPORT_SYMBOL_GPL(md_is_badblock);
7982
7983/*
7984 * Add a range of bad blocks to the table.
7985 * This might extend the table, or might contract it
7986 * if two adjacent ranges can be merged.
7987 * We binary-search to find the 'insertion' point, then
7988 * decide how best to handle it.
7989 */
7990static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
7991 int acknowledged)
7992{
7993 u64 *p;
7994 int lo, hi;
7995 int rv = 1;
7996
7997 if (bb->shift < 0)
7998 /* badblocks are disabled */
7999 return 0;
8000
8001 if (bb->shift) {
8002 /* round the start down, and the end up */
8003 sector_t next = s + sectors;
8004 s >>= bb->shift;
8005 next += (1<<bb->shift) - 1;
8006 next >>= bb->shift;
8007 sectors = next - s;
8008 }
8009
8010 write_seqlock_irq(&bb->lock);
8011
8012 p = bb->page;
8013 lo = 0;
8014 hi = bb->count;
8015 /* Find the last range that starts at-or-before 's' */
8016 while (hi - lo > 1) {
8017 int mid = (lo + hi) / 2;
8018 sector_t a = BB_OFFSET(p[mid]);
8019 if (a <= s)
8020 lo = mid;
8021 else
8022 hi = mid;
8023 }
8024 if (hi > lo && BB_OFFSET(p[lo]) > s)
8025 hi = lo;
8026
8027 if (hi > lo) {
8028 /* we found a range that might merge with the start
8029 * of our new range
8030 */
8031 sector_t a = BB_OFFSET(p[lo]);
8032 sector_t e = a + BB_LEN(p[lo]);
8033 int ack = BB_ACK(p[lo]);
8034 if (e >= s) {
8035 /* Yes, we can merge with a previous range */
8036 if (s == a && s + sectors >= e)
8037 /* new range covers old */
8038 ack = acknowledged;
8039 else
8040 ack = ack && acknowledged;
8041
8042 if (e < s + sectors)
8043 e = s + sectors;
8044 if (e - a <= BB_MAX_LEN) {
8045 p[lo] = BB_MAKE(a, e-a, ack);
8046 s = e;
8047 } else {
8048 /* does not all fit in one range,
8049 * make p[lo] maximal
8050 */
8051 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8052 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8053 s = a + BB_MAX_LEN;
8054 }
8055 sectors = e - s;
8056 }
8057 }
8058 if (sectors && hi < bb->count) {
8059 /* 'hi' points to the first range that starts after 's'.
8060 * Maybe we can merge with the start of that range */
8061 sector_t a = BB_OFFSET(p[hi]);
8062 sector_t e = a + BB_LEN(p[hi]);
8063 int ack = BB_ACK(p[hi]);
8064 if (a <= s + sectors) {
8065 /* merging is possible */
8066 if (e <= s + sectors) {
8067 /* full overlap */
8068 e = s + sectors;
8069 ack = acknowledged;
8070 } else
8071 ack = ack && acknowledged;
8072
8073 a = s;
8074 if (e - a <= BB_MAX_LEN) {
8075 p[hi] = BB_MAKE(a, e-a, ack);
8076 s = e;
8077 } else {
8078 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8079 s = a + BB_MAX_LEN;
8080 }
8081 sectors = e - s;
8082 lo = hi;
8083 hi++;
8084 }
8085 }
8086 if (sectors == 0 && hi < bb->count) {
8087 /* we might be able to combine lo and hi */
8088 /* Note: 's' is at the end of 'lo' */
8089 sector_t a = BB_OFFSET(p[hi]);
8090 int lolen = BB_LEN(p[lo]);
8091 int hilen = BB_LEN(p[hi]);
8092 int newlen = lolen + hilen - (s - a);
8093 if (s >= a && newlen < BB_MAX_LEN) {
8094 /* yes, we can combine them */
8095 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8096 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8097 memmove(p + hi, p + hi + 1,
8098 (bb->count - hi - 1) * 8);
8099 bb->count--;
8100 }
8101 }
8102 while (sectors) {
8103 /* didn't merge (it all).
8104 * Need to add a range just before 'hi' */
8105 if (bb->count >= MD_MAX_BADBLOCKS) {
8106 /* No room for more */
8107 rv = 0;
8108 break;
8109 } else {
8110 int this_sectors = sectors;
8111 memmove(p + hi + 1, p + hi,
8112 (bb->count - hi) * 8);
8113 bb->count++;
8114
8115 if (this_sectors > BB_MAX_LEN)
8116 this_sectors = BB_MAX_LEN;
8117 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8118 sectors -= this_sectors;
8119 s += this_sectors;
8120 }
8121 }
8122
8123 bb->changed = 1;
NeilBrownde393cd2011-07-28 11:31:48 +10008124 if (!acknowledged)
8125 bb->unacked_exist = 1;
NeilBrown2230dfe2011-07-28 11:31:46 +10008126 write_sequnlock_irq(&bb->lock);
8127
8128 return rv;
8129}
8130
NeilBrown3cb03002011-10-11 16:45:26 +11008131int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
NeilBrownc6563a82012-05-21 09:27:00 +10008132 int is_new)
NeilBrown2230dfe2011-07-28 11:31:46 +10008133{
NeilBrownc6563a82012-05-21 09:27:00 +10008134 int rv;
8135 if (is_new)
8136 s += rdev->new_data_offset;
8137 else
8138 s += rdev->data_offset;
8139 rv = md_set_badblocks(&rdev->badblocks,
8140 s, sectors, 0);
NeilBrown2230dfe2011-07-28 11:31:46 +10008141 if (rv) {
8142 /* Make sure they get written out promptly */
NeilBrown8bd2f0a2011-12-08 16:26:08 +11008143 sysfs_notify_dirent_safe(rdev->sysfs_state);
NeilBrown2230dfe2011-07-28 11:31:46 +10008144 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8145 md_wakeup_thread(rdev->mddev->thread);
8146 }
8147 return rv;
8148}
8149EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8150
8151/*
8152 * Remove a range of bad blocks from the table.
8153 * This may involve extending the table if we spilt a region,
8154 * but it must not fail. So if the table becomes full, we just
8155 * drop the remove request.
8156 */
8157static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8158{
8159 u64 *p;
8160 int lo, hi;
8161 sector_t target = s + sectors;
8162 int rv = 0;
8163
8164 if (bb->shift > 0) {
8165 /* When clearing we round the start up and the end down.
8166 * This should not matter as the shift should align with
8167 * the block size and no rounding should ever be needed.
8168 * However it is better the think a block is bad when it
8169 * isn't than to think a block is not bad when it is.
8170 */
8171 s += (1<<bb->shift) - 1;
8172 s >>= bb->shift;
8173 target >>= bb->shift;
8174 sectors = target - s;
8175 }
8176
8177 write_seqlock_irq(&bb->lock);
8178
8179 p = bb->page;
8180 lo = 0;
8181 hi = bb->count;
8182 /* Find the last range that starts before 'target' */
8183 while (hi - lo > 1) {
8184 int mid = (lo + hi) / 2;
8185 sector_t a = BB_OFFSET(p[mid]);
8186 if (a < target)
8187 lo = mid;
8188 else
8189 hi = mid;
8190 }
8191 if (hi > lo) {
8192 /* p[lo] is the last range that could overlap the
8193 * current range. Earlier ranges could also overlap,
8194 * but only this one can overlap the end of the range.
8195 */
8196 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8197 /* Partial overlap, leave the tail of this range */
8198 int ack = BB_ACK(p[lo]);
8199 sector_t a = BB_OFFSET(p[lo]);
8200 sector_t end = a + BB_LEN(p[lo]);
8201
8202 if (a < s) {
8203 /* we need to split this range */
8204 if (bb->count >= MD_MAX_BADBLOCKS) {
8205 rv = 0;
8206 goto out;
8207 }
8208 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8209 bb->count++;
8210 p[lo] = BB_MAKE(a, s-a, ack);
8211 lo++;
8212 }
8213 p[lo] = BB_MAKE(target, end - target, ack);
8214 /* there is no longer an overlap */
8215 hi = lo;
8216 lo--;
8217 }
8218 while (lo >= 0 &&
8219 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8220 /* This range does overlap */
8221 if (BB_OFFSET(p[lo]) < s) {
8222 /* Keep the early parts of this range. */
8223 int ack = BB_ACK(p[lo]);
8224 sector_t start = BB_OFFSET(p[lo]);
8225 p[lo] = BB_MAKE(start, s - start, ack);
8226 /* now low doesn't overlap, so.. */
8227 break;
8228 }
8229 lo--;
8230 }
8231 /* 'lo' is strictly before, 'hi' is strictly after,
8232 * anything between needs to be discarded
8233 */
8234 if (hi - lo > 1) {
8235 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8236 bb->count -= (hi - lo - 1);
8237 }
8238 }
8239
8240 bb->changed = 1;
8241out:
8242 write_sequnlock_irq(&bb->lock);
8243 return rv;
8244}
8245
NeilBrownc6563a82012-05-21 09:27:00 +10008246int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8247 int is_new)
NeilBrown2230dfe2011-07-28 11:31:46 +10008248{
NeilBrownc6563a82012-05-21 09:27:00 +10008249 if (is_new)
8250 s += rdev->new_data_offset;
8251 else
8252 s += rdev->data_offset;
NeilBrown2230dfe2011-07-28 11:31:46 +10008253 return md_clear_badblocks(&rdev->badblocks,
NeilBrownc6563a82012-05-21 09:27:00 +10008254 s, sectors);
NeilBrown2230dfe2011-07-28 11:31:46 +10008255}
8256EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8257
8258/*
8259 * Acknowledge all bad blocks in a list.
8260 * This only succeeds if ->changed is clear. It is used by
8261 * in-kernel metadata updates
8262 */
8263void md_ack_all_badblocks(struct badblocks *bb)
8264{
8265 if (bb->page == NULL || bb->changed)
8266 /* no point even trying */
8267 return;
8268 write_seqlock_irq(&bb->lock);
8269
majianpengecb178b2012-03-19 12:46:42 +11008270 if (bb->changed == 0 && bb->unacked_exist) {
NeilBrown2230dfe2011-07-28 11:31:46 +10008271 u64 *p = bb->page;
8272 int i;
8273 for (i = 0; i < bb->count ; i++) {
8274 if (!BB_ACK(p[i])) {
8275 sector_t start = BB_OFFSET(p[i]);
8276 int len = BB_LEN(p[i]);
8277 p[i] = BB_MAKE(start, len, 1);
8278 }
8279 }
NeilBrownde393cd2011-07-28 11:31:48 +10008280 bb->unacked_exist = 0;
NeilBrown2230dfe2011-07-28 11:31:46 +10008281 }
8282 write_sequnlock_irq(&bb->lock);
8283}
8284EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8285
NeilBrown16c791a2011-07-28 11:31:47 +10008286/* sysfs access to bad-blocks list.
8287 * We present two files.
8288 * 'bad-blocks' lists sector numbers and lengths of ranges that
8289 * are recorded as bad. The list is truncated to fit within
8290 * the one-page limit of sysfs.
8291 * Writing "sector length" to this file adds an acknowledged
8292 * bad block list.
8293 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8294 * been acknowledged. Writing to this file adds bad blocks
8295 * without acknowledging them. This is largely for testing.
8296 */
8297
8298static ssize_t
8299badblocks_show(struct badblocks *bb, char *page, int unack)
8300{
8301 size_t len;
8302 int i;
8303 u64 *p = bb->page;
8304 unsigned seq;
8305
8306 if (bb->shift < 0)
8307 return 0;
8308
8309retry:
8310 seq = read_seqbegin(&bb->lock);
8311
8312 len = 0;
8313 i = 0;
8314
8315 while (len < PAGE_SIZE && i < bb->count) {
8316 sector_t s = BB_OFFSET(p[i]);
8317 unsigned int length = BB_LEN(p[i]);
8318 int ack = BB_ACK(p[i]);
8319 i++;
8320
8321 if (unack && ack)
8322 continue;
8323
8324 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8325 (unsigned long long)s << bb->shift,
8326 length << bb->shift);
8327 }
NeilBrownde393cd2011-07-28 11:31:48 +10008328 if (unack && len == 0)
8329 bb->unacked_exist = 0;
NeilBrown16c791a2011-07-28 11:31:47 +10008330
8331 if (read_seqretry(&bb->lock, seq))
8332 goto retry;
8333
8334 return len;
8335}
8336
8337#define DO_DEBUG 1
8338
8339static ssize_t
8340badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8341{
8342 unsigned long long sector;
8343 int length;
8344 char newline;
8345#ifdef DO_DEBUG
8346 /* Allow clearing via sysfs *only* for testing/debugging.
8347 * Normally only a successful write may clear a badblock
8348 */
8349 int clear = 0;
8350 if (page[0] == '-') {
8351 clear = 1;
8352 page++;
8353 }
8354#endif /* DO_DEBUG */
8355
8356 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8357 case 3:
8358 if (newline != '\n')
8359 return -EINVAL;
8360 case 2:
8361 if (length <= 0)
8362 return -EINVAL;
8363 break;
8364 default:
8365 return -EINVAL;
8366 }
8367
8368#ifdef DO_DEBUG
8369 if (clear) {
8370 md_clear_badblocks(bb, sector, length);
8371 return len;
8372 }
8373#endif /* DO_DEBUG */
8374 if (md_set_badblocks(bb, sector, length, !unack))
8375 return len;
8376 else
8377 return -ENOSPC;
8378}
8379
Adrian Bunk75c96f82005-05-05 16:16:09 -07008380static int md_notify_reboot(struct notifier_block *this,
8381 unsigned long code, void *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07008382{
8383 struct list_head *tmp;
NeilBrownfd01b882011-10-11 16:47:53 +11008384 struct mddev *mddev;
Daniel P. Berrange2dba6a92011-09-23 10:40:45 +01008385 int need_delay = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008386
NeilBrownc744a652012-03-19 12:46:37 +11008387 for_each_mddev(mddev, tmp) {
8388 if (mddev_trylock(mddev)) {
NeilBrown30b8aa92012-04-24 10:23:16 +10008389 if (mddev->pers)
8390 __md_stop_writes(mddev);
NeilBrownc744a652012-03-19 12:46:37 +11008391 mddev->safemode = 2;
8392 mddev_unlock(mddev);
Daniel P. Berrange2dba6a92011-09-23 10:40:45 +01008393 }
NeilBrownc744a652012-03-19 12:46:37 +11008394 need_delay = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008395 }
NeilBrownc744a652012-03-19 12:46:37 +11008396 /*
8397 * certain more exotic SCSI devices are known to be
8398 * volatile wrt too early system reboots. While the
8399 * right place to handle this issue is the given
8400 * driver, we do want to have a safe RAID driver ...
8401 */
8402 if (need_delay)
8403 mdelay(1000*1);
8404
Linus Torvalds1da177e2005-04-16 15:20:36 -07008405 return NOTIFY_DONE;
8406}
8407
Adrian Bunk75c96f82005-05-05 16:16:09 -07008408static struct notifier_block md_notifier = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07008409 .notifier_call = md_notify_reboot,
8410 .next = NULL,
8411 .priority = INT_MAX, /* before any real devices */
8412};
8413
8414static void md_geninit(void)
8415{
NeilBrown36a4e1f2011-10-07 14:23:17 +11008416 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07008417
Denis V. Lunevc7705f32008-04-29 01:02:35 -07008418 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008419}
8420
Adrian Bunk75c96f82005-05-05 16:16:09 -07008421static int __init md_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07008422{
Tejun Heoe804ac72010-10-15 15:36:08 +02008423 int ret = -ENOMEM;
8424
Tejun Heoada609e2011-01-25 14:35:54 +01008425 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
Tejun Heoe804ac72010-10-15 15:36:08 +02008426 if (!md_wq)
8427 goto err_wq;
8428
8429 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8430 if (!md_misc_wq)
8431 goto err_misc_wq;
8432
8433 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8434 goto err_md;
8435
8436 if ((ret = register_blkdev(0, "mdp")) < 0)
8437 goto err_mdp;
8438 mdp_major = ret;
8439
Christoph Hellwig3dbd8c22009-03-31 14:27:02 +11008440 blk_register_region(MKDEV(MD_MAJOR, 0), 1UL<<MINORBITS, THIS_MODULE,
NeilBrowne8703fe2006-10-03 01:15:59 -07008441 md_probe, NULL, NULL);
8442 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07008443 md_probe, NULL, NULL);
8444
Linus Torvalds1da177e2005-04-16 15:20:36 -07008445 register_reboot_notifier(&md_notifier);
Eric W. Biederman0b4d4142007-02-14 00:34:09 -08008446 raid_table_header = register_sysctl_table(raid_root_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008447
8448 md_geninit();
NeilBrownd710e132008-10-13 11:55:12 +11008449 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008450
Tejun Heoe804ac72010-10-15 15:36:08 +02008451err_mdp:
8452 unregister_blkdev(MD_MAJOR, "md");
8453err_md:
8454 destroy_workqueue(md_misc_wq);
8455err_misc_wq:
8456 destroy_workqueue(md_wq);
8457err_wq:
8458 return ret;
8459}
Linus Torvalds1da177e2005-04-16 15:20:36 -07008460
8461#ifndef MODULE
8462
8463/*
8464 * Searches all registered partitions for autorun RAID arrays
8465 * at boot time.
8466 */
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008467
8468static LIST_HEAD(all_detected_devices);
8469struct detected_devices_node {
8470 struct list_head list;
8471 dev_t dev;
8472};
Linus Torvalds1da177e2005-04-16 15:20:36 -07008473
8474void md_autodetect_dev(dev_t dev)
8475{
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008476 struct detected_devices_node *node_detected_dev;
8477
8478 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8479 if (node_detected_dev) {
8480 node_detected_dev->dev = dev;
8481 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8482 } else {
8483 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8484 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8485 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07008486}
8487
8488
8489static void autostart_arrays(int part)
8490{
NeilBrown3cb03002011-10-11 16:45:26 +11008491 struct md_rdev *rdev;
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008492 struct detected_devices_node *node_detected_dev;
8493 dev_t dev;
8494 int i_scanned, i_passed;
8495
8496 i_scanned = 0;
8497 i_passed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008498
8499 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8500
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008501 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8502 i_scanned++;
8503 node_detected_dev = list_entry(all_detected_devices.next,
8504 struct detected_devices_node, list);
8505 list_del(&node_detected_dev->list);
8506 dev = node_detected_dev->dev;
8507 kfree(node_detected_dev);
NeilBrowndf968c42007-07-17 04:06:11 -07008508 rdev = md_import_device(dev,0, 90);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008509 if (IS_ERR(rdev))
8510 continue;
8511
NeilBrownb2d444d2005-11-08 21:39:31 -08008512 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07008513 MD_BUG();
8514 continue;
8515 }
NeilBrownd0fae182008-03-04 14:29:31 -08008516 set_bit(AutoDetected, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008517 list_add(&rdev->same_set, &pending_raid_disks);
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008518 i_passed++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008519 }
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008520
8521 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8522 i_scanned, i_passed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008523
8524 autorun_devices(part);
8525}
8526
Jeff Garzikfdee8ae2006-12-10 02:20:50 -08008527#endif /* !MODULE */
Linus Torvalds1da177e2005-04-16 15:20:36 -07008528
8529static __exit void md_exit(void)
8530{
NeilBrownfd01b882011-10-11 16:47:53 +11008531 struct mddev *mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008532 struct list_head *tmp;
Greg Kroah-Hartman8ab5e4c2005-06-20 21:15:16 -07008533
Christoph Hellwig3dbd8c22009-03-31 14:27:02 +11008534 blk_unregister_region(MKDEV(MD_MAJOR,0), 1U << MINORBITS);
NeilBrowne8703fe2006-10-03 01:15:59 -07008535 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008536
Christoph Hellwig3dbd8c22009-03-31 14:27:02 +11008537 unregister_blkdev(MD_MAJOR,"md");
Linus Torvalds1da177e2005-04-16 15:20:36 -07008538 unregister_blkdev(mdp_major, "mdp");
8539 unregister_reboot_notifier(&md_notifier);
8540 unregister_sysctl_table(raid_table_header);
8541 remove_proc_entry("mdstat", NULL);
NeilBrown29ac4aa2008-02-06 01:39:58 -08008542 for_each_mddev(mddev, tmp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07008543 export_array(mddev);
NeilBrownd3374822009-01-09 08:31:10 +11008544 mddev->hold_active = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008545 }
Tejun Heoe804ac72010-10-15 15:36:08 +02008546 destroy_workqueue(md_misc_wq);
8547 destroy_workqueue(md_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008548}
8549
Dan Williams685784a2007-07-09 11:56:42 -07008550subsys_initcall(md_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008551module_exit(md_exit)
8552
NeilBrownf91de922005-11-08 21:39:36 -08008553static int get_ro(char *buffer, struct kernel_param *kp)
8554{
8555 return sprintf(buffer, "%d", start_readonly);
8556}
8557static int set_ro(const char *val, struct kernel_param *kp)
8558{
8559 char *e;
8560 int num = simple_strtoul(val, &e, 10);
8561 if (*val && (*e == '\0' || *e == '\n')) {
8562 start_readonly = num;
NeilBrown4dbcdc72006-01-06 00:20:52 -08008563 return 0;
NeilBrownf91de922005-11-08 21:39:36 -08008564 }
8565 return -EINVAL;
8566}
8567
NeilBrown80ca3a42006-07-10 04:44:18 -07008568module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8569module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08008570
NeilBrownefeb53c2009-01-09 08:31:10 +11008571module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
NeilBrownf91de922005-11-08 21:39:36 -08008572
Linus Torvalds1da177e2005-04-16 15:20:36 -07008573EXPORT_SYMBOL(register_md_personality);
8574EXPORT_SYMBOL(unregister_md_personality);
8575EXPORT_SYMBOL(md_error);
8576EXPORT_SYMBOL(md_done_sync);
8577EXPORT_SYMBOL(md_write_start);
8578EXPORT_SYMBOL(md_write_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008579EXPORT_SYMBOL(md_register_thread);
8580EXPORT_SYMBOL(md_unregister_thread);
8581EXPORT_SYMBOL(md_wakeup_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008582EXPORT_SYMBOL(md_check_recovery);
8583MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11008584MODULE_DESCRIPTION("MD RAID framework");
NeilBrownaa1595e2005-08-04 12:53:32 -07008585MODULE_ALIAS("md");
NeilBrown72008652005-08-26 18:34:15 -07008586MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);