blob: e6178787ce3dd4a9b0e33b80b72ac39f73e25e33 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 md.c : Multiple Devices driver for Linux
NeilBrownf72ffdd2014-09-30 14:23:59 +10003 Copyright (C) 1998, 1999, 2000 Ingo Molnar
Linus Torvalds1da177e2005-04-16 15:20:36 -07004
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
Bernd Schubert90b08712008-05-23 13:04:38 -070069static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
Tejun Heoe804ac72010-10-15 15:36:08 +020070static struct workqueue_struct *md_wq;
71static struct workqueue_struct *md_misc_wq;
Bernd Schubert90b08712008-05-23 13:04:38 -070072
NeilBrown746d3202013-04-24 11:42:41 +100073static int remove_and_add_spares(struct mddev *mddev,
74 struct md_rdev *this);
NeilBrown5aa61f42014-12-15 12:56:57 +110075static void mddev_detach(struct mddev *mddev);
NeilBrown746d3202013-04-24 11:42:41 +100076
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
Joe Perches82592c32013-11-14 15:16:18 +1100112static struct ctl_table raid_table[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 {
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
Joe Perches82592c32013-11-14 15:16:18 +1100130static struct ctl_table raid_dir_table[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 {
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
Joe Perches82592c32013-11-14 15:16:18 +1100140static struct ctl_table raid_root_table[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 {
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183/*
NeilBrownd7603b72006-01-06 00:20:30 -0800184 * We have a system wide 'event count' that is incremented
185 * on any 'interesting' event, and readers of /proc/mdstat
186 * can use 'poll' or 'select' to find out when the event
187 * count increases.
188 *
189 * Events are:
190 * start array, stop array, error, add device, remove device,
191 * start build, activate spare
192 */
NeilBrown2989ddb2006-01-06 00:20:43 -0800193static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
NeilBrownd7603b72006-01-06 00:20:30 -0800194static atomic_t md_event_count;
NeilBrownfd01b882011-10-11 16:47:53 +1100195void md_new_event(struct mddev *mddev)
NeilBrownd7603b72006-01-06 00:20:30 -0800196{
197 atomic_inc(&md_event_count);
198 wake_up(&md_event_waiters);
199}
NeilBrown29269552006-03-27 01:18:10 -0800200EXPORT_SYMBOL_GPL(md_new_event);
NeilBrownd7603b72006-01-06 00:20:30 -0800201
NeilBrownc331eb02006-05-30 21:27:13 -0700202/* Alternate version that can be called from interrupts
203 * when calling sysfs_notify isn't needed.
204 */
NeilBrownfd01b882011-10-11 16:47:53 +1100205static void md_new_event_inintr(struct mddev *mddev)
NeilBrownc331eb02006-05-30 21:27:13 -0700206{
207 atomic_inc(&md_event_count);
208 wake_up(&md_event_waiters);
209}
210
NeilBrownd7603b72006-01-06 00:20:30 -0800211/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 * Enables to iterate over all existing md arrays
213 * all_mddevs_lock protects this list.
214 */
215static LIST_HEAD(all_mddevs);
216static DEFINE_SPINLOCK(all_mddevs_lock);
217
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218/*
219 * iterates through all used mddevs in the system.
220 * We take care to grab the all_mddevs_lock whenever navigating
221 * the list, and to always hold a refcount when unlocked.
222 * Any code which breaks out of this loop while own
223 * a reference to the current mddev and must mddev_put it.
224 */
NeilBrownfd01b882011-10-11 16:47:53 +1100225#define for_each_mddev(_mddev,_tmp) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 \
NeilBrownf72ffdd2014-09-30 14:23:59 +1000227 for (({ spin_lock(&all_mddevs_lock); \
NeilBrownfd01b882011-10-11 16:47:53 +1100228 _tmp = all_mddevs.next; \
229 _mddev = NULL;}); \
230 ({ if (_tmp != &all_mddevs) \
231 mddev_get(list_entry(_tmp, struct mddev, all_mddevs));\
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 spin_unlock(&all_mddevs_lock); \
NeilBrownfd01b882011-10-11 16:47:53 +1100233 if (_mddev) mddev_put(_mddev); \
234 _mddev = list_entry(_tmp, struct mddev, all_mddevs); \
235 _tmp != &all_mddevs;}); \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 ({ spin_lock(&all_mddevs_lock); \
NeilBrownfd01b882011-10-11 16:47:53 +1100237 _tmp = _tmp->next;}) \
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 )
239
NeilBrown409c57f2009-03-31 14:39:39 +1100240/* Rather than calling directly into the personality make_request function,
241 * IO requests come here first so that we can check if the device is
242 * being suspended pending a reconfiguration.
243 * We hold a refcount over the call to ->make_request. By the time that
244 * call has finished, the bio has been linked into some internal structure
245 * and so is visible to ->quiesce(), so we don't need the refcount any more.
246 */
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200247static void md_make_request(struct request_queue *q, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248{
NeilBrown49077322010-03-25 16:20:56 +1100249 const int rw = bio_data_dir(bio);
NeilBrownfd01b882011-10-11 16:47:53 +1100250 struct mddev *mddev = q->queuedata;
Chris Masone91ece52011-02-07 19:21:48 -0500251 unsigned int sectors;
Gu Zheng74672d02015-04-03 08:44:47 +0800252 int cpu;
NeilBrown49077322010-03-25 16:20:56 +1100253
NeilBrown0ca698862011-01-14 09:14:33 +1100254 if (mddev == NULL || mddev->pers == NULL
255 || !mddev->ready) {
NeilBrown409c57f2009-03-31 14:39:39 +1100256 bio_io_error(bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200257 return;
NeilBrown409c57f2009-03-31 14:39:39 +1100258 }
Sebastian Riemerbbfa57c2013-02-21 13:28:09 +1100259 if (mddev->ro == 1 && unlikely(rw == WRITE)) {
260 bio_endio(bio, bio_sectors(bio) == 0 ? 0 : -EROFS);
261 return;
262 }
NeilBrown0ca698862011-01-14 09:14:33 +1100263 smp_rmb(); /* Ensure implications of 'active' are visible */
NeilBrown409c57f2009-03-31 14:39:39 +1100264 rcu_read_lock();
Tejun Heoe9c74692010-09-03 11:56:18 +0200265 if (mddev->suspended) {
NeilBrown409c57f2009-03-31 14:39:39 +1100266 DEFINE_WAIT(__wait);
267 for (;;) {
268 prepare_to_wait(&mddev->sb_wait, &__wait,
269 TASK_UNINTERRUPTIBLE);
Tejun Heoe9c74692010-09-03 11:56:18 +0200270 if (!mddev->suspended)
NeilBrown409c57f2009-03-31 14:39:39 +1100271 break;
272 rcu_read_unlock();
273 schedule();
274 rcu_read_lock();
275 }
276 finish_wait(&mddev->sb_wait, &__wait);
277 }
278 atomic_inc(&mddev->active_io);
279 rcu_read_unlock();
NeilBrown49077322010-03-25 16:20:56 +1100280
Chris Masone91ece52011-02-07 19:21:48 -0500281 /*
282 * save the sectors now since our bio can
283 * go away inside make_request
284 */
285 sectors = bio_sectors(bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200286 mddev->pers->make_request(mddev, bio);
NeilBrown49077322010-03-25 16:20:56 +1100287
Gu Zheng74672d02015-04-03 08:44:47 +0800288 cpu = part_stat_lock();
289 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
290 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw], sectors);
291 part_stat_unlock();
NeilBrown49077322010-03-25 16:20:56 +1100292
NeilBrown409c57f2009-03-31 14:39:39 +1100293 if (atomic_dec_and_test(&mddev->active_io) && mddev->suspended)
294 wake_up(&mddev->sb_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295}
296
NeilBrown9e35b992010-04-06 14:23:02 +1000297/* mddev_suspend makes sure no new requests are submitted
298 * to the device, and that any requests that have been submitted
299 * are completely handled.
NeilBrownafa0f552014-12-15 12:56:58 +1100300 * Once mddev_detach() is called and completes, the module will be
301 * completely unused.
NeilBrown9e35b992010-04-06 14:23:02 +1000302 */
NeilBrownfd01b882011-10-11 16:47:53 +1100303void mddev_suspend(struct mddev *mddev)
NeilBrown409c57f2009-03-31 14:39:39 +1100304{
305 BUG_ON(mddev->suspended);
306 mddev->suspended = 1;
307 synchronize_rcu();
308 wait_event(mddev->sb_wait, atomic_read(&mddev->active_io) == 0);
309 mddev->pers->quiesce(mddev, 1);
Jonathan Brassow0d9f4f12012-05-16 04:06:14 -0500310
311 del_timer_sync(&mddev->safemode_timer);
NeilBrown409c57f2009-03-31 14:39:39 +1100312}
NeilBrown390ee602010-06-01 19:37:27 +1000313EXPORT_SYMBOL_GPL(mddev_suspend);
NeilBrown409c57f2009-03-31 14:39:39 +1100314
NeilBrownfd01b882011-10-11 16:47:53 +1100315void mddev_resume(struct mddev *mddev)
NeilBrown409c57f2009-03-31 14:39:39 +1100316{
317 mddev->suspended = 0;
318 wake_up(&mddev->sb_wait);
319 mddev->pers->quiesce(mddev, 0);
Jonathan Brassow0fd018a2011-06-07 17:49:36 -0500320
Jonathan Brassow47525e52012-05-22 13:55:29 +1000321 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
Jonathan Brassow0fd018a2011-06-07 17:49:36 -0500322 md_wakeup_thread(mddev->thread);
323 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
NeilBrown409c57f2009-03-31 14:39:39 +1100324}
NeilBrown390ee602010-06-01 19:37:27 +1000325EXPORT_SYMBOL_GPL(mddev_resume);
NeilBrown409c57f2009-03-31 14:39:39 +1100326
NeilBrownfd01b882011-10-11 16:47:53 +1100327int mddev_congested(struct mddev *mddev, int bits)
NeilBrown3fa841d2009-09-23 18:10:29 +1000328{
NeilBrown5c675f82014-12-15 12:56:56 +1100329 struct md_personality *pers = mddev->pers;
330 int ret = 0;
331
332 rcu_read_lock();
333 if (mddev->suspended)
334 ret = 1;
335 else if (pers && pers->congested)
336 ret = pers->congested(mddev, bits);
337 rcu_read_unlock();
338 return ret;
NeilBrown3fa841d2009-09-23 18:10:29 +1000339}
NeilBrown5c675f82014-12-15 12:56:56 +1100340EXPORT_SYMBOL_GPL(mddev_congested);
341static int md_congested(void *data, int bits)
342{
343 struct mddev *mddev = data;
344 return mddev_congested(mddev, bits);
345}
NeilBrown3fa841d2009-09-23 18:10:29 +1000346
NeilBrown64590f42014-12-15 12:56:57 +1100347static int md_mergeable_bvec(struct request_queue *q,
348 struct bvec_merge_data *bvm,
349 struct bio_vec *biovec)
350{
351 struct mddev *mddev = q->queuedata;
352 int ret;
353 rcu_read_lock();
354 if (mddev->suspended) {
355 /* Must always allow one vec */
356 if (bvm->bi_size == 0)
357 ret = biovec->bv_len;
358 else
359 ret = 0;
360 } else {
361 struct md_personality *pers = mddev->pers;
362 if (pers && pers->mergeable_bvec)
363 ret = pers->mergeable_bvec(mddev, bvm, biovec);
364 else
365 ret = biovec->bv_len;
366 }
367 rcu_read_unlock();
368 return ret;
369}
NeilBrowna2826aa2009-12-14 12:49:49 +1100370/*
Tejun Heoe9c74692010-09-03 11:56:18 +0200371 * Generic flush handling for md
NeilBrowna2826aa2009-12-14 12:49:49 +1100372 */
373
Tejun Heoe9c74692010-09-03 11:56:18 +0200374static void md_end_flush(struct bio *bio, int err)
NeilBrowna2826aa2009-12-14 12:49:49 +1100375{
NeilBrown3cb03002011-10-11 16:45:26 +1100376 struct md_rdev *rdev = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +1100377 struct mddev *mddev = rdev->mddev;
NeilBrowna2826aa2009-12-14 12:49:49 +1100378
379 rdev_dec_pending(rdev, mddev);
380
381 if (atomic_dec_and_test(&mddev->flush_pending)) {
Tejun Heoe9c74692010-09-03 11:56:18 +0200382 /* The pre-request flush has finished */
Tejun Heoe804ac72010-10-15 15:36:08 +0200383 queue_work(md_wq, &mddev->flush_work);
NeilBrowna2826aa2009-12-14 12:49:49 +1100384 }
385 bio_put(bio);
386}
387
NeilBrowna7a07e62010-12-09 16:04:25 +1100388static void md_submit_flush_data(struct work_struct *ws);
389
NeilBrowna035fc32010-12-09 16:17:51 +1100390static void submit_flushes(struct work_struct *ws)
NeilBrowna2826aa2009-12-14 12:49:49 +1100391{
NeilBrownfd01b882011-10-11 16:47:53 +1100392 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
NeilBrown3cb03002011-10-11 16:45:26 +1100393 struct md_rdev *rdev;
NeilBrowna2826aa2009-12-14 12:49:49 +1100394
NeilBrowna7a07e62010-12-09 16:04:25 +1100395 INIT_WORK(&mddev->flush_work, md_submit_flush_data);
396 atomic_set(&mddev->flush_pending, 1);
NeilBrowna2826aa2009-12-14 12:49:49 +1100397 rcu_read_lock();
NeilBrowndafb20f2012-03-19 12:46:39 +1100398 rdev_for_each_rcu(rdev, mddev)
NeilBrowna2826aa2009-12-14 12:49:49 +1100399 if (rdev->raid_disk >= 0 &&
400 !test_bit(Faulty, &rdev->flags)) {
401 /* Take two references, one is dropped
402 * when request finishes, one after
403 * we reclaim rcu_read_lock
404 */
405 struct bio *bi;
406 atomic_inc(&rdev->nr_pending);
407 atomic_inc(&rdev->nr_pending);
408 rcu_read_unlock();
Shaohua Lib5e1b8c2012-05-21 09:26:59 +1000409 bi = bio_alloc_mddev(GFP_NOIO, 0, mddev);
Tejun Heoe9c74692010-09-03 11:56:18 +0200410 bi->bi_end_io = md_end_flush;
NeilBrowna2826aa2009-12-14 12:49:49 +1100411 bi->bi_private = rdev;
412 bi->bi_bdev = rdev->bdev;
413 atomic_inc(&mddev->flush_pending);
Tejun Heoe9c74692010-09-03 11:56:18 +0200414 submit_bio(WRITE_FLUSH, bi);
NeilBrowna2826aa2009-12-14 12:49:49 +1100415 rcu_read_lock();
416 rdev_dec_pending(rdev, mddev);
417 }
418 rcu_read_unlock();
NeilBrowna7a07e62010-12-09 16:04:25 +1100419 if (atomic_dec_and_test(&mddev->flush_pending))
420 queue_work(md_wq, &mddev->flush_work);
NeilBrowna2826aa2009-12-14 12:49:49 +1100421}
422
Tejun Heoe9c74692010-09-03 11:56:18 +0200423static void md_submit_flush_data(struct work_struct *ws)
NeilBrowna2826aa2009-12-14 12:49:49 +1100424{
NeilBrownfd01b882011-10-11 16:47:53 +1100425 struct mddev *mddev = container_of(ws, struct mddev, flush_work);
Tejun Heoe9c74692010-09-03 11:56:18 +0200426 struct bio *bio = mddev->flush_bio;
NeilBrowna2826aa2009-12-14 12:49:49 +1100427
Kent Overstreet4f024f32013-10-11 15:44:27 -0700428 if (bio->bi_iter.bi_size == 0)
NeilBrowna2826aa2009-12-14 12:49:49 +1100429 /* an empty barrier - all done */
430 bio_endio(bio, 0);
431 else {
Tejun Heoe9c74692010-09-03 11:56:18 +0200432 bio->bi_rw &= ~REQ_FLUSH;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200433 mddev->pers->make_request(mddev, bio);
NeilBrowna2826aa2009-12-14 12:49:49 +1100434 }
NeilBrown2b74e122010-12-09 15:59:01 +1100435
436 mddev->flush_bio = NULL;
437 wake_up(&mddev->sb_wait);
NeilBrowna2826aa2009-12-14 12:49:49 +1100438}
439
NeilBrownfd01b882011-10-11 16:47:53 +1100440void md_flush_request(struct mddev *mddev, struct bio *bio)
NeilBrowna2826aa2009-12-14 12:49:49 +1100441{
NeilBrown85572d72014-12-15 12:56:56 +1100442 spin_lock_irq(&mddev->lock);
NeilBrowna2826aa2009-12-14 12:49:49 +1100443 wait_event_lock_irq(mddev->sb_wait,
Tejun Heoe9c74692010-09-03 11:56:18 +0200444 !mddev->flush_bio,
NeilBrown85572d72014-12-15 12:56:56 +1100445 mddev->lock);
Tejun Heoe9c74692010-09-03 11:56:18 +0200446 mddev->flush_bio = bio;
NeilBrown85572d72014-12-15 12:56:56 +1100447 spin_unlock_irq(&mddev->lock);
NeilBrowna2826aa2009-12-14 12:49:49 +1100448
NeilBrowna035fc32010-12-09 16:17:51 +1100449 INIT_WORK(&mddev->flush_work, submit_flushes);
450 queue_work(md_wq, &mddev->flush_work);
NeilBrowna2826aa2009-12-14 12:49:49 +1100451}
Tejun Heoe9c74692010-09-03 11:56:18 +0200452EXPORT_SYMBOL(md_flush_request);
NeilBrown409c57f2009-03-31 14:39:39 +1100453
NeilBrown74018dc2012-07-31 09:08:15 +0200454void md_unplug(struct blk_plug_cb *cb, bool from_schedule)
NeilBrown97658cd2011-04-18 18:25:42 +1000455{
NeilBrown9cbb1752012-07-31 09:08:14 +0200456 struct mddev *mddev = cb->data;
457 md_wakeup_thread(mddev->thread);
458 kfree(cb);
NeilBrown97658cd2011-04-18 18:25:42 +1000459}
NeilBrown9cbb1752012-07-31 09:08:14 +0200460EXPORT_SYMBOL(md_unplug);
NeilBrown2ac87402010-06-01 19:37:29 +1000461
NeilBrownfd01b882011-10-11 16:47:53 +1100462static inline struct mddev *mddev_get(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463{
464 atomic_inc(&mddev->active);
465 return mddev;
466}
467
Dan Williams5fd3a172009-03-04 00:57:25 -0700468static void mddev_delayed_delete(struct work_struct *ws);
NeilBrownd3374822009-01-09 08:31:10 +1100469
NeilBrownfd01b882011-10-11 16:47:53 +1100470static void mddev_put(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471{
NeilBrowna167f662010-10-26 18:31:13 +1100472 struct bio_set *bs = NULL;
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
475 return;
NeilBrownd3374822009-01-09 08:31:10 +1100476 if (!mddev->raid_disks && list_empty(&mddev->disks) &&
NeilBrowncbd19982009-12-30 12:08:49 +1100477 mddev->ctime == 0 && !mddev->hold_active) {
478 /* Array is not configured at all, and not held active,
479 * so destroy it */
NeilBrownaf8a2432011-12-08 15:49:46 +1100480 list_del_init(&mddev->all_mddevs);
NeilBrowna167f662010-10-26 18:31:13 +1100481 bs = mddev->bio_set;
482 mddev->bio_set = NULL;
NeilBrownd3374822009-01-09 08:31:10 +1100483 if (mddev->gendisk) {
Tejun Heoe804ac72010-10-15 15:36:08 +0200484 /* We did a probe so need to clean up. Call
485 * queue_work inside the spinlock so that
486 * flush_workqueue() after mddev_find will
487 * succeed in waiting for the work to be done.
NeilBrownd3374822009-01-09 08:31:10 +1100488 */
489 INIT_WORK(&mddev->del_work, mddev_delayed_delete);
Tejun Heoe804ac72010-10-15 15:36:08 +0200490 queue_work(md_misc_wq, &mddev->del_work);
NeilBrownd3374822009-01-09 08:31:10 +1100491 } else
492 kfree(mddev);
493 }
494 spin_unlock(&all_mddevs_lock);
NeilBrowna167f662010-10-26 18:31:13 +1100495 if (bs)
496 bioset_free(bs);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497}
498
NeilBrownfd01b882011-10-11 16:47:53 +1100499void mddev_init(struct mddev *mddev)
NeilBrownfafd7fb2010-04-01 15:55:30 +1100500{
501 mutex_init(&mddev->open_mutex);
502 mutex_init(&mddev->reconfig_mutex);
503 mutex_init(&mddev->bitmap_info.mutex);
504 INIT_LIST_HEAD(&mddev->disks);
505 INIT_LIST_HEAD(&mddev->all_mddevs);
506 init_timer(&mddev->safemode_timer);
507 atomic_set(&mddev->active, 1);
508 atomic_set(&mddev->openers, 0);
509 atomic_set(&mddev->active_io, 0);
NeilBrown85572d72014-12-15 12:56:56 +1100510 spin_lock_init(&mddev->lock);
NeilBrownfafd7fb2010-04-01 15:55:30 +1100511 atomic_set(&mddev->flush_pending, 0);
512 init_waitqueue_head(&mddev->sb_wait);
513 init_waitqueue_head(&mddev->recovery_wait);
514 mddev->reshape_position = MaxSector;
NeilBrown2c810cd2012-05-21 09:27:00 +1000515 mddev->reshape_backwards = 0;
Jonathan Brassowc4a39552013-06-25 01:23:59 -0500516 mddev->last_sync_action = "none";
NeilBrownfafd7fb2010-04-01 15:55:30 +1100517 mddev->resync_min = 0;
518 mddev->resync_max = MaxSector;
519 mddev->level = LEVEL_NONE;
520}
NeilBrown390ee602010-06-01 19:37:27 +1000521EXPORT_SYMBOL_GPL(mddev_init);
NeilBrownfafd7fb2010-04-01 15:55:30 +1100522
NeilBrownf72ffdd2014-09-30 14:23:59 +1000523static struct mddev *mddev_find(dev_t unit)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524{
NeilBrownfd01b882011-10-11 16:47:53 +1100525 struct mddev *mddev, *new = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
NeilBrown8f5f02c2011-02-16 13:58:51 +1100527 if (unit && MAJOR(unit) != MD_MAJOR)
528 unit &= ~((1<<MdpMinorShift)-1);
529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 retry:
531 spin_lock(&all_mddevs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532
NeilBrownefeb53c2009-01-09 08:31:10 +1100533 if (unit) {
534 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
535 if (mddev->unit == unit) {
536 mddev_get(mddev);
537 spin_unlock(&all_mddevs_lock);
538 kfree(new);
539 return mddev;
540 }
541
542 if (new) {
543 list_add(&new->all_mddevs, &all_mddevs);
544 spin_unlock(&all_mddevs_lock);
545 new->hold_active = UNTIL_IOCTL;
546 return new;
547 }
548 } else if (new) {
549 /* find an unused unit number */
550 static int next_minor = 512;
551 int start = next_minor;
552 int is_free = 0;
553 int dev = 0;
554 while (!is_free) {
555 dev = MKDEV(MD_MAJOR, next_minor);
556 next_minor++;
557 if (next_minor > MINORMASK)
558 next_minor = 0;
559 if (next_minor == start) {
560 /* Oh dear, all in use. */
561 spin_unlock(&all_mddevs_lock);
562 kfree(new);
563 return NULL;
564 }
NeilBrownf72ffdd2014-09-30 14:23:59 +1000565
NeilBrownefeb53c2009-01-09 08:31:10 +1100566 is_free = 1;
567 list_for_each_entry(mddev, &all_mddevs, all_mddevs)
568 if (mddev->unit == dev) {
569 is_free = 0;
570 break;
571 }
572 }
573 new->unit = dev;
574 new->md_minor = MINOR(dev);
575 new->hold_active = UNTIL_STOP;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 list_add(&new->all_mddevs, &all_mddevs);
577 spin_unlock(&all_mddevs_lock);
578 return new;
579 }
580 spin_unlock(&all_mddevs_lock);
581
NeilBrown9ffae0c2006-01-06 00:20:32 -0800582 new = kzalloc(sizeof(*new), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 if (!new)
584 return NULL;
585
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 new->unit = unit;
587 if (MAJOR(unit) == MD_MAJOR)
588 new->md_minor = MINOR(unit);
589 else
590 new->md_minor = MINOR(unit) >> MdpMinorShift;
591
NeilBrownfafd7fb2010-04-01 15:55:30 +1100592 mddev_init(new);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 goto retry;
595}
596
NeilBrownb6eb1272010-04-15 10:13:47 +1000597static struct attribute_group md_redundancy_group;
598
NeilBrown5c47daf2014-12-15 12:57:01 +1100599void mddev_unlock(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600{
NeilBrowna64c8762010-04-14 17:15:37 +1000601 if (mddev->to_remove) {
NeilBrownb6eb1272010-04-15 10:13:47 +1000602 /* These cannot be removed under reconfig_mutex as
603 * an access to the files will try to take reconfig_mutex
604 * while holding the file unremovable, which leads to
605 * a deadlock.
NeilBrownbb4f1e92010-08-08 21:18:03 +1000606 * So hold set sysfs_active while the remove in happeing,
607 * and anything else which might set ->to_remove or my
608 * otherwise change the sysfs namespace will fail with
609 * -EBUSY if sysfs_active is still set.
610 * We set sysfs_active under reconfig_mutex and elsewhere
611 * test it under the same mutex to ensure its correct value
612 * is seen.
NeilBrownb6eb1272010-04-15 10:13:47 +1000613 */
NeilBrowna64c8762010-04-14 17:15:37 +1000614 struct attribute_group *to_remove = mddev->to_remove;
615 mddev->to_remove = NULL;
NeilBrownbb4f1e92010-08-08 21:18:03 +1000616 mddev->sysfs_active = 1;
NeilBrownb6eb1272010-04-15 10:13:47 +1000617 mutex_unlock(&mddev->reconfig_mutex);
618
NeilBrown00bcb4a2010-06-01 19:37:23 +1000619 if (mddev->kobj.sd) {
620 if (to_remove != &md_redundancy_group)
621 sysfs_remove_group(&mddev->kobj, to_remove);
622 if (mddev->pers == NULL ||
623 mddev->pers->sync_request == NULL) {
624 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
625 if (mddev->sysfs_action)
626 sysfs_put(mddev->sysfs_action);
627 mddev->sysfs_action = NULL;
628 }
NeilBrowna64c8762010-04-14 17:15:37 +1000629 }
NeilBrownbb4f1e92010-08-08 21:18:03 +1000630 mddev->sysfs_active = 0;
NeilBrownb6eb1272010-04-15 10:13:47 +1000631 } else
632 mutex_unlock(&mddev->reconfig_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633
Chris Dunlop751e67c2011-10-19 16:48:26 +1100634 /* As we've dropped the mutex we need a spinlock to
635 * make sure the thread doesn't disappear
NeilBrown01f96c02011-09-21 15:30:20 +1000636 */
637 spin_lock(&pers_lock);
NeilBrown005eca52005-08-22 13:11:08 -0700638 md_wakeup_thread(mddev->thread);
NeilBrown01f96c02011-09-21 15:30:20 +1000639 spin_unlock(&pers_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640}
NeilBrown5c47daf2014-12-15 12:57:01 +1100641EXPORT_SYMBOL_GPL(mddev_unlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
NeilBrown1ca69c42012-10-11 13:37:33 +1100643static struct md_rdev *find_rdev_nr_rcu(struct mddev *mddev, int nr)
644{
645 struct md_rdev *rdev;
646
647 rdev_for_each_rcu(rdev, mddev)
648 if (rdev->desc_nr == nr)
649 return rdev;
650
651 return NULL;
652}
653
654static struct md_rdev *find_rdev(struct mddev *mddev, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
NeilBrown3cb03002011-10-11 16:45:26 +1100656 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657
NeilBrowndafb20f2012-03-19 12:46:39 +1100658 rdev_for_each(rdev, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 if (rdev->bdev->bd_dev == dev)
660 return rdev;
Cheng Renquan159ec1f2009-01-09 08:31:08 +1100661
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 return NULL;
663}
664
NeilBrown1ca69c42012-10-11 13:37:33 +1100665static struct md_rdev *find_rdev_rcu(struct mddev *mddev, dev_t dev)
666{
667 struct md_rdev *rdev;
668
669 rdev_for_each_rcu(rdev, mddev)
670 if (rdev->bdev->bd_dev == dev)
671 return rdev;
672
673 return NULL;
674}
675
NeilBrown84fc4b52011-10-11 16:49:58 +1100676static struct md_personality *find_pers(int level, char *clevel)
NeilBrown2604b702006-01-06 00:20:36 -0800677{
NeilBrown84fc4b52011-10-11 16:49:58 +1100678 struct md_personality *pers;
NeilBrownd9d166c2006-01-06 00:20:51 -0800679 list_for_each_entry(pers, &pers_list, list) {
680 if (level != LEVEL_NONE && pers->level == level)
NeilBrown2604b702006-01-06 00:20:36 -0800681 return pers;
NeilBrownd9d166c2006-01-06 00:20:51 -0800682 if (strcmp(pers->name, clevel)==0)
683 return pers;
684 }
NeilBrown2604b702006-01-06 00:20:36 -0800685 return NULL;
686}
687
Andre Nollb73df2d2008-07-11 22:02:23 +1000688/* return the offset of the super block in 512byte sectors */
NeilBrown3cb03002011-10-11 16:45:26 +1100689static inline sector_t calc_dev_sboffset(struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
Jonathan Brassow57b2caa2011-01-14 09:14:33 +1100691 sector_t num_sectors = i_size_read(rdev->bdev->bd_inode) / 512;
Andre Nollb73df2d2008-07-11 22:02:23 +1000692 return MD_NEW_SIZE_SECTORS(num_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694
NeilBrownf72ffdd2014-09-30 14:23:59 +1000695static int alloc_disk_sb(struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 rdev->sb_page = alloc_page(GFP_KERNEL);
698 if (!rdev->sb_page) {
699 printk(KERN_ALERT "md: out of memory.\n");
Andre Nollebc24332008-07-11 22:02:20 +1000700 return -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 }
702
703 return 0;
704}
705
NeilBrown545c8792012-05-22 13:54:30 +1000706void md_rdev_clear(struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707{
708 if (rdev->sb_page) {
NeilBrown2d1f3b52006-01-06 00:20:31 -0800709 put_page(rdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 rdev->sb_loaded = 0;
711 rdev->sb_page = NULL;
Andre Noll0f420352008-07-11 22:02:23 +1000712 rdev->sb_start = 0;
Andre Nolldd8ac332009-03-31 14:33:13 +1100713 rdev->sectors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 }
NeilBrown2699b672011-07-28 11:31:47 +1000715 if (rdev->bb_page) {
716 put_page(rdev->bb_page);
717 rdev->bb_page = NULL;
718 }
NeilBrown4fa2f322012-05-22 13:55:01 +1000719 kfree(rdev->badblocks.page);
720 rdev->badblocks.page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721}
NeilBrown545c8792012-05-22 13:54:30 +1000722EXPORT_SYMBOL_GPL(md_rdev_clear);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723
NeilBrown6712ecf2007-09-27 12:47:43 +0200724static void super_written(struct bio *bio, int error)
NeilBrown7bfa19f2005-06-21 17:17:28 -0700725{
NeilBrown3cb03002011-10-11 16:45:26 +1100726 struct md_rdev *rdev = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +1100727 struct mddev *mddev = rdev->mddev;
NeilBrown7bfa19f2005-06-21 17:17:28 -0700728
NeilBrown3a0f5bb2006-10-03 01:16:03 -0700729 if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags)) {
730 printk("md: super_written gets error=%d, uptodate=%d\n",
731 error, test_bit(BIO_UPTODATE, &bio->bi_flags));
732 WARN_ON(test_bit(BIO_UPTODATE, &bio->bi_flags));
NeilBrowna9701a32005-11-08 21:39:34 -0800733 md_error(mddev, rdev);
NeilBrown3a0f5bb2006-10-03 01:16:03 -0700734 }
NeilBrown7bfa19f2005-06-21 17:17:28 -0700735
NeilBrowna9701a32005-11-08 21:39:34 -0800736 if (atomic_dec_and_test(&mddev->pending_writes))
737 wake_up(&mddev->sb_wait);
Neil Brownf8b58ed2005-06-27 22:29:34 -0700738 bio_put(bio);
NeilBrown7bfa19f2005-06-21 17:17:28 -0700739}
740
NeilBrownfd01b882011-10-11 16:47:53 +1100741void md_super_write(struct mddev *mddev, struct md_rdev *rdev,
NeilBrown7bfa19f2005-06-21 17:17:28 -0700742 sector_t sector, int size, struct page *page)
743{
744 /* write first size bytes of page to sector of rdev
745 * Increment mddev->pending_writes before returning
746 * and decrement it on completion, waking up sb_wait
747 * if zero is reached.
748 * If an error occurred, call md_error
749 */
NeilBrowna167f662010-10-26 18:31:13 +1100750 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, mddev);
NeilBrown7bfa19f2005-06-21 17:17:28 -0700751
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100752 bio->bi_bdev = rdev->meta_bdev ? rdev->meta_bdev : rdev->bdev;
Kent Overstreet4f024f32013-10-11 15:44:27 -0700753 bio->bi_iter.bi_sector = sector;
NeilBrown7bfa19f2005-06-21 17:17:28 -0700754 bio_add_page(bio, page, size, 0);
755 bio->bi_private = rdev;
756 bio->bi_end_io = super_written;
NeilBrowna9701a32005-11-08 21:39:34 -0800757
NeilBrown7bfa19f2005-06-21 17:17:28 -0700758 atomic_inc(&mddev->pending_writes);
Namhyung Kima5bf4df2011-08-25 14:43:34 +1000759 submit_bio(WRITE_FLUSH_FUA, bio);
NeilBrowna9701a32005-11-08 21:39:34 -0800760}
761
NeilBrownfd01b882011-10-11 16:47:53 +1100762void md_super_wait(struct mddev *mddev)
NeilBrowna9701a32005-11-08 21:39:34 -0800763{
Tejun Heoe9c74692010-09-03 11:56:18 +0200764 /* wait for all superblock writes that were scheduled to complete */
NeilBrown1967cd52014-09-09 14:20:28 +1000765 wait_event(mddev->sb_wait, atomic_read(&mddev->pending_writes)==0);
NeilBrown7bfa19f2005-06-21 17:17:28 -0700766}
767
NeilBrown3cb03002011-10-11 16:45:26 +1100768int sync_page_io(struct md_rdev *rdev, sector_t sector, int size,
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100769 struct page *page, int rw, bool metadata_op)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770{
NeilBrowna167f662010-10-26 18:31:13 +1100771 struct bio *bio = bio_alloc_mddev(GFP_NOIO, 1, rdev->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772 int ret;
773
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +1100774 bio->bi_bdev = (metadata_op && rdev->meta_bdev) ?
775 rdev->meta_bdev : rdev->bdev;
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100776 if (metadata_op)
Kent Overstreet4f024f32013-10-11 15:44:27 -0700777 bio->bi_iter.bi_sector = sector + rdev->sb_start;
NeilBrown1fdd6fc92012-05-21 09:28:32 +1000778 else if (rdev->mddev->reshape_position != MaxSector &&
779 (rdev->mddev->reshape_backwards ==
780 (sector >= rdev->mddev->reshape_position)))
Kent Overstreet4f024f32013-10-11 15:44:27 -0700781 bio->bi_iter.bi_sector = sector + rdev->new_data_offset;
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100782 else
Kent Overstreet4f024f32013-10-11 15:44:27 -0700783 bio->bi_iter.bi_sector = sector + rdev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 bio_add_page(bio, page, size, 0);
Kent Overstreetc170bbb2013-11-24 16:32:22 -0700785 submit_bio_wait(rw, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786
787 ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
788 bio_put(bio);
789 return ret;
790}
NeilBrowna8745db2006-01-06 00:20:34 -0800791EXPORT_SYMBOL_GPL(sync_page_io);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792
NeilBrownf72ffdd2014-09-30 14:23:59 +1000793static int read_disk_sb(struct md_rdev *rdev, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794{
795 char b[BDEVNAME_SIZE];
NeilBrown403df472014-09-30 15:52:29 +1000796
Linus Torvalds1da177e2005-04-16 15:20:36 -0700797 if (rdev->sb_loaded)
798 return 0;
799
Jonathan Brassowccebd4c2011-01-14 09:14:33 +1100800 if (!sync_page_io(rdev, 0, size, rdev->sb_page, READ, true))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 goto fail;
802 rdev->sb_loaded = 1;
803 return 0;
804
805fail:
806 printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
807 bdevname(rdev->bdev,b));
808 return -EINVAL;
809}
810
811static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
812{
NeilBrownf72ffdd2014-09-30 14:23:59 +1000813 return sb1->set_uuid0 == sb2->set_uuid0 &&
Andre Noll05710462008-07-11 22:02:20 +1000814 sb1->set_uuid1 == sb2->set_uuid1 &&
815 sb1->set_uuid2 == sb2->set_uuid2 &&
816 sb1->set_uuid3 == sb2->set_uuid3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817}
818
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
820{
821 int ret;
822 mdp_super_t *tmp1, *tmp2;
823
824 tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
825 tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
826
827 if (!tmp1 || !tmp2) {
828 ret = 0;
Andre Noll35020f12008-03-23 15:10:33 +0100829 printk(KERN_INFO "md.c sb_equal(): failed to allocate memory!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 goto abort;
831 }
832
833 *tmp1 = *sb1;
834 *tmp2 = *sb2;
835
836 /*
837 * nr_disks is not constant
838 */
839 tmp1->nr_disks = 0;
840 tmp2->nr_disks = 0;
841
Andre Nollce0c8e02008-07-11 22:02:20 +1000842 ret = (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4) == 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843abort:
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700844 kfree(tmp1);
845 kfree(tmp2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 return ret;
847}
848
NeilBrown4d167f02007-05-09 02:35:37 -0700849static u32 md_csum_fold(u32 csum)
850{
851 csum = (csum & 0xffff) + (csum >> 16);
852 return (csum & 0xffff) + (csum >> 16);
853}
854
NeilBrownf72ffdd2014-09-30 14:23:59 +1000855static unsigned int calc_sb_csum(mdp_super_t *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856{
NeilBrown4d167f02007-05-09 02:35:37 -0700857 u64 newcsum = 0;
858 u32 *sb32 = (u32*)sb;
859 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 unsigned int disk_csum, csum;
861
862 disk_csum = sb->sb_csum;
863 sb->sb_csum = 0;
NeilBrown4d167f02007-05-09 02:35:37 -0700864
865 for (i = 0; i < MD_SB_BYTES/4 ; i++)
866 newcsum += sb32[i];
867 csum = (newcsum & 0xffffffff) + (newcsum>>32);
868
NeilBrown4d167f02007-05-09 02:35:37 -0700869#ifdef CONFIG_ALPHA
870 /* This used to use csum_partial, which was wrong for several
871 * reasons including that different results are returned on
872 * different architectures. It isn't critical that we get exactly
873 * the same return value as before (we always csum_fold before
874 * testing, and that removes any differences). However as we
875 * know that csum_partial always returned a 16bit value on
876 * alphas, do a fold to maximise conformity to previous behaviour.
877 */
878 sb->sb_csum = md_csum_fold(disk_csum);
879#else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 sb->sb_csum = disk_csum;
NeilBrown4d167f02007-05-09 02:35:37 -0700881#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 return csum;
883}
884
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885/*
886 * Handle superblock details.
887 * We want to be able to handle multiple superblock formats
888 * so we have a common interface to them all, and an array of
889 * different handlers.
890 * We rely on user-space to write the initial superblock, and support
891 * reading and updating of superblocks.
892 * Interface methods are:
NeilBrown3cb03002011-10-11 16:45:26 +1100893 * int load_super(struct md_rdev *dev, struct md_rdev *refdev, int minor_version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 * loads and validates a superblock on dev.
895 * if refdev != NULL, compare superblocks on both devices
896 * Return:
897 * 0 - dev has a superblock that is compatible with refdev
898 * 1 - dev has a superblock that is compatible and newer than refdev
899 * so dev should be used as the refdev in future
900 * -EINVAL superblock incompatible or invalid
901 * -othererror e.g. -EIO
902 *
NeilBrownfd01b882011-10-11 16:47:53 +1100903 * int validate_super(struct mddev *mddev, struct md_rdev *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 * Verify that dev is acceptable into mddev.
905 * The first time, mddev->raid_disks will be 0, and data from
906 * dev should be merged in. Subsequent calls check that dev
907 * is new enough. Return 0 or -EINVAL
908 *
NeilBrownfd01b882011-10-11 16:47:53 +1100909 * void sync_super(struct mddev *mddev, struct md_rdev *dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 * Update the superblock for rdev with data in mddev
911 * This does not write to disc.
912 *
913 */
914
915struct super_type {
Chris Webb0cd17fe2008-06-28 08:31:46 +1000916 char *name;
917 struct module *owner;
NeilBrownc6563a82012-05-21 09:27:00 +1000918 int (*load_super)(struct md_rdev *rdev,
919 struct md_rdev *refdev,
Chris Webb0cd17fe2008-06-28 08:31:46 +1000920 int minor_version);
NeilBrownc6563a82012-05-21 09:27:00 +1000921 int (*validate_super)(struct mddev *mddev,
922 struct md_rdev *rdev);
923 void (*sync_super)(struct mddev *mddev,
924 struct md_rdev *rdev);
NeilBrown3cb03002011-10-11 16:45:26 +1100925 unsigned long long (*rdev_size_change)(struct md_rdev *rdev,
Andre Noll15f4a5f2008-07-21 14:42:12 +1000926 sector_t num_sectors);
NeilBrownc6563a82012-05-21 09:27:00 +1000927 int (*allow_new_offset)(struct md_rdev *rdev,
928 unsigned long long new_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929};
930
931/*
Andre Noll0894cc32009-06-18 08:49:23 +1000932 * Check that the given mddev has no bitmap.
933 *
934 * This function is called from the run method of all personalities that do not
935 * support bitmaps. It prints an error message and returns non-zero if mddev
936 * has a bitmap. Otherwise, it returns 0.
937 *
938 */
NeilBrownfd01b882011-10-11 16:47:53 +1100939int md_check_no_bitmap(struct mddev *mddev)
Andre Noll0894cc32009-06-18 08:49:23 +1000940{
NeilBrownc3d97142009-12-14 12:49:52 +1100941 if (!mddev->bitmap_info.file && !mddev->bitmap_info.offset)
Andre Noll0894cc32009-06-18 08:49:23 +1000942 return 0;
943 printk(KERN_ERR "%s: bitmaps are not supported for %s\n",
944 mdname(mddev), mddev->pers->name);
945 return 1;
946}
947EXPORT_SYMBOL(md_check_no_bitmap);
948
949/*
NeilBrownf72ffdd2014-09-30 14:23:59 +1000950 * load_super for 0.90.0
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 */
NeilBrown3cb03002011-10-11 16:45:26 +1100952static int super_90_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953{
954 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
955 mdp_super_t *sb;
956 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
958 /*
Andre Noll0f420352008-07-11 22:02:23 +1000959 * Calculate the position of the superblock (512byte sectors),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 * it's at the end of the disk.
961 *
962 * It also happens to be a multiple of 4Kb.
963 */
Jonathan Brassow57b2caa2011-01-14 09:14:33 +1100964 rdev->sb_start = calc_dev_sboffset(rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
NeilBrown0002b272005-09-09 16:23:53 -0700966 ret = read_disk_sb(rdev, MD_SB_BYTES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 if (ret) return ret;
968
969 ret = -EINVAL;
970
971 bdevname(rdev->bdev, b);
Namhyung Kim65a06f062011-07-27 11:00:36 +1000972 sb = page_address(rdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
974 if (sb->md_magic != MD_SB_MAGIC) {
975 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
976 b);
977 goto abort;
978 }
979
980 if (sb->major_version != 0 ||
NeilBrownf6705572006-03-27 01:18:11 -0800981 sb->minor_version < 90 ||
982 sb->minor_version > 91) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
984 sb->major_version, sb->minor_version,
985 b);
986 goto abort;
987 }
988
989 if (sb->raid_disks <= 0)
990 goto abort;
991
NeilBrown4d167f02007-05-09 02:35:37 -0700992 if (md_csum_fold(calc_sb_csum(sb)) != md_csum_fold(sb->sb_csum)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
994 b);
995 goto abort;
996 }
997
998 rdev->preferred_minor = sb->md_minor;
999 rdev->data_offset = 0;
NeilBrownc6563a82012-05-21 09:27:00 +10001000 rdev->new_data_offset = 0;
NeilBrown0002b272005-09-09 16:23:53 -07001001 rdev->sb_size = MD_SB_BYTES;
NeilBrown9f2f3832011-07-28 11:31:47 +10001002 rdev->badblocks.shift = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
1004 if (sb->level == LEVEL_MULTIPATH)
1005 rdev->desc_nr = -1;
1006 else
1007 rdev->desc_nr = sb->this_disk.number;
1008
Harvey Harrison9a7b2b02008-04-28 02:15:49 -07001009 if (!refdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 ret = 1;
Harvey Harrison9a7b2b02008-04-28 02:15:49 -07001011 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 __u64 ev1, ev2;
Namhyung Kim65a06f062011-07-27 11:00:36 +10001013 mdp_super_t *refsb = page_address(refdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 if (!uuid_equal(refsb, sb)) {
1015 printk(KERN_WARNING "md: %s has different UUID to %s\n",
1016 b, bdevname(refdev->bdev,b2));
1017 goto abort;
1018 }
1019 if (!sb_equal(refsb, sb)) {
1020 printk(KERN_WARNING "md: %s has same UUID"
1021 " but different superblock to %s\n",
1022 b, bdevname(refdev->bdev, b2));
1023 goto abort;
1024 }
1025 ev1 = md_event(sb);
1026 ev2 = md_event(refsb);
1027 if (ev1 > ev2)
1028 ret = 1;
NeilBrownf72ffdd2014-09-30 14:23:59 +10001029 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 ret = 0;
1031 }
NeilBrown8190e752009-06-18 08:48:58 +10001032 rdev->sectors = rdev->sb_start;
NeilBrown667a5312012-08-16 16:46:12 +10001033 /* Limit to 4TB as metadata cannot record more than that.
1034 * (not needed for Linear and RAID0 as metadata doesn't
1035 * record this size)
1036 */
1037 if (rdev->sectors >= (2ULL << 32) && sb->level >= 1)
NeilBrown27a7b262011-09-10 17:21:28 +10001038 rdev->sectors = (2ULL << 32) - 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039
NeilBrown27a7b262011-09-10 17:21:28 +10001040 if (rdev->sectors < ((sector_t)sb->size) * 2 && sb->level >= 1)
NeilBrown2bf071b2006-01-06 00:20:55 -08001041 /* "this cannot possibly happen" ... */
1042 ret = -EINVAL;
1043
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 abort:
1045 return ret;
1046}
1047
1048/*
1049 * validate_super for 0.90.0
1050 */
NeilBrownfd01b882011-10-11 16:47:53 +11001051static int super_90_validate(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052{
1053 mdp_disk_t *desc;
Namhyung Kim65a06f062011-07-27 11:00:36 +10001054 mdp_super_t *sb = page_address(rdev->sb_page);
NeilBrown07d84d102006-06-26 00:27:56 -07001055 __u64 ev1 = md_event(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
NeilBrown41158c72005-06-21 17:17:25 -07001057 rdev->raid_disk = -1;
NeilBrownc5d79ad2008-02-06 01:39:54 -08001058 clear_bit(Faulty, &rdev->flags);
1059 clear_bit(In_sync, &rdev->flags);
NeilBrown8313b8e2013-12-12 10:13:33 +11001060 clear_bit(Bitmap_sync, &rdev->flags);
NeilBrownc5d79ad2008-02-06 01:39:54 -08001061 clear_bit(WriteMostly, &rdev->flags);
NeilBrownc5d79ad2008-02-06 01:39:54 -08001062
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063 if (mddev->raid_disks == 0) {
1064 mddev->major_version = 0;
1065 mddev->minor_version = sb->minor_version;
1066 mddev->patch_version = sb->patch_version;
NeilBrowne6910632008-02-06 01:39:51 -08001067 mddev->external = 0;
Andre Noll9d8f0362009-06-18 08:45:01 +10001068 mddev->chunk_sectors = sb->chunk_size >> 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069 mddev->ctime = sb->ctime;
1070 mddev->utime = sb->utime;
1071 mddev->level = sb->level;
NeilBrownd9d166c2006-01-06 00:20:51 -08001072 mddev->clevel[0] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 mddev->layout = sb->layout;
1074 mddev->raid_disks = sb->raid_disks;
NeilBrown27a7b262011-09-10 17:21:28 +10001075 mddev->dev_sectors = ((sector_t)sb->size) * 2;
NeilBrown07d84d102006-06-26 00:27:56 -07001076 mddev->events = ev1;
NeilBrownc3d97142009-12-14 12:49:52 +11001077 mddev->bitmap_info.offset = 0;
NeilBrown6409bb02012-05-22 13:55:07 +10001078 mddev->bitmap_info.space = 0;
1079 /* bitmap can use 60 K after the 4K superblocks */
NeilBrownc3d97142009-12-14 12:49:52 +11001080 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
NeilBrown6409bb02012-05-22 13:55:07 +10001081 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
NeilBrown2c810cd2012-05-21 09:27:00 +10001082 mddev->reshape_backwards = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083
NeilBrownf6705572006-03-27 01:18:11 -08001084 if (mddev->minor_version >= 91) {
1085 mddev->reshape_position = sb->reshape_position;
1086 mddev->delta_disks = sb->delta_disks;
1087 mddev->new_level = sb->new_level;
1088 mddev->new_layout = sb->new_layout;
Andre Noll664e7c42009-06-18 08:45:27 +10001089 mddev->new_chunk_sectors = sb->new_chunk >> 9;
NeilBrown2c810cd2012-05-21 09:27:00 +10001090 if (mddev->delta_disks < 0)
1091 mddev->reshape_backwards = 1;
NeilBrownf6705572006-03-27 01:18:11 -08001092 } else {
1093 mddev->reshape_position = MaxSector;
1094 mddev->delta_disks = 0;
1095 mddev->new_level = mddev->level;
1096 mddev->new_layout = mddev->layout;
Andre Noll664e7c42009-06-18 08:45:27 +10001097 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08001098 }
1099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 if (sb->state & (1<<MD_SB_CLEAN))
1101 mddev->recovery_cp = MaxSector;
1102 else {
NeilBrownf72ffdd2014-09-30 14:23:59 +10001103 if (sb->events_hi == sb->cp_events_hi &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 sb->events_lo == sb->cp_events_lo) {
1105 mddev->recovery_cp = sb->recovery_cp;
1106 } else
1107 mddev->recovery_cp = 0;
1108 }
1109
1110 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
1111 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
1112 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
1113 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
1114
1115 mddev->max_disks = MD_SB_DISKS;
NeilBrowna654b9d82005-06-21 17:17:27 -07001116
1117 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
NeilBrown6409bb02012-05-22 13:55:07 +10001118 mddev->bitmap_info.file == NULL) {
NeilBrownc3d97142009-12-14 12:49:52 +11001119 mddev->bitmap_info.offset =
1120 mddev->bitmap_info.default_offset;
NeilBrown6409bb02012-05-22 13:55:07 +10001121 mddev->bitmap_info.space =
Dave Jonesc9ad0202013-08-19 22:26:32 -04001122 mddev->bitmap_info.default_space;
NeilBrown6409bb02012-05-22 13:55:07 +10001123 }
NeilBrowna654b9d82005-06-21 17:17:27 -07001124
NeilBrown41158c72005-06-21 17:17:25 -07001125 } else if (mddev->pers == NULL) {
NeilBrownbe6800a2010-05-18 10:17:09 +10001126 /* Insist on good event counter while assembling, except
1127 * for spares (which don't need an event count) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 ++ev1;
NeilBrownbe6800a2010-05-18 10:17:09 +10001129 if (sb->disks[rdev->desc_nr].state & (
1130 (1<<MD_DISK_SYNC) | (1 << MD_DISK_ACTIVE)))
NeilBrownf72ffdd2014-09-30 14:23:59 +10001131 if (ev1 < mddev->events)
NeilBrownbe6800a2010-05-18 10:17:09 +10001132 return -EINVAL;
NeilBrown41158c72005-06-21 17:17:25 -07001133 } else if (mddev->bitmap) {
1134 /* if adding to array with a bitmap, then we can accept an
1135 * older device ... but not too old.
1136 */
NeilBrown41158c72005-06-21 17:17:25 -07001137 if (ev1 < mddev->bitmap->events_cleared)
1138 return 0;
NeilBrown8313b8e2013-12-12 10:13:33 +11001139 if (ev1 < mddev->events)
1140 set_bit(Bitmap_sync, &rdev->flags);
NeilBrown07d84d102006-06-26 00:27:56 -07001141 } else {
1142 if (ev1 < mddev->events)
1143 /* just a hot-add of a new device, leave raid_disk at -1 */
1144 return 0;
1145 }
NeilBrown41158c72005-06-21 17:17:25 -07001146
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 if (mddev->level != LEVEL_MULTIPATH) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 desc = sb->disks + rdev->desc_nr;
1149
1150 if (desc->state & (1<<MD_DISK_FAULTY))
NeilBrownb2d444d2005-11-08 21:39:31 -08001151 set_bit(Faulty, &rdev->flags);
NeilBrown7c7546c2006-06-26 00:27:41 -07001152 else if (desc->state & (1<<MD_DISK_SYNC) /* &&
1153 desc->raid_disk < mddev->raid_disks */) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001154 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 rdev->raid_disk = desc->raid_disk;
NeilBrownf4667222013-12-09 12:04:56 +11001156 rdev->saved_raid_disk = desc->raid_disk;
NeilBrown0261cd9f2009-11-13 17:40:48 +11001157 } else if (desc->state & (1<<MD_DISK_ACTIVE)) {
1158 /* active but not in sync implies recovery up to
1159 * reshape position. We don't know exactly where
1160 * that is, so set to zero for now */
1161 if (mddev->minor_version >= 91) {
1162 rdev->recovery_offset = 0;
1163 rdev->raid_disk = desc->raid_disk;
1164 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -07001166 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
1167 set_bit(WriteMostly, &rdev->flags);
NeilBrown41158c72005-06-21 17:17:25 -07001168 } else /* MULTIPATH are always insync */
NeilBrownb2d444d2005-11-08 21:39:31 -08001169 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170 return 0;
1171}
1172
1173/*
1174 * sync_super for 0.90.0
1175 */
NeilBrownfd01b882011-10-11 16:47:53 +11001176static void super_90_sync(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177{
1178 mdp_super_t *sb;
NeilBrown3cb03002011-10-11 16:45:26 +11001179 struct md_rdev *rdev2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 int next_spare = mddev->raid_disks;
NeilBrown19133a42005-11-08 21:39:35 -08001181
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 /* make rdev->sb match mddev data..
1183 *
1184 * 1/ zero out disks
1185 * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
1186 * 3/ any empty disks < next_spare become removed
1187 *
1188 * disks[0] gets initialised to REMOVED because
1189 * we cannot be sure from other fields if it has
1190 * been initialised or not.
1191 */
1192 int i;
1193 int active=0, working=0,failed=0,spare=0,nr_disks=0;
1194
NeilBrown61181562005-09-09 16:24:02 -07001195 rdev->sb_size = MD_SB_BYTES;
1196
Namhyung Kim65a06f062011-07-27 11:00:36 +10001197 sb = page_address(rdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198
1199 memset(sb, 0, sizeof(*sb));
1200
1201 sb->md_magic = MD_SB_MAGIC;
1202 sb->major_version = mddev->major_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203 sb->patch_version = mddev->patch_version;
1204 sb->gvalid_words = 0; /* ignored */
1205 memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
1206 memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
1207 memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
1208 memcpy(&sb->set_uuid3, mddev->uuid+12,4);
1209
1210 sb->ctime = mddev->ctime;
1211 sb->level = mddev->level;
Andre Noll58c0fed2009-03-31 14:33:13 +11001212 sb->size = mddev->dev_sectors / 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 sb->raid_disks = mddev->raid_disks;
1214 sb->md_minor = mddev->md_minor;
NeilBrowne6910632008-02-06 01:39:51 -08001215 sb->not_persistent = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001216 sb->utime = mddev->utime;
1217 sb->state = 0;
1218 sb->events_hi = (mddev->events>>32);
1219 sb->events_lo = (u32)mddev->events;
1220
NeilBrownf6705572006-03-27 01:18:11 -08001221 if (mddev->reshape_position == MaxSector)
1222 sb->minor_version = 90;
1223 else {
1224 sb->minor_version = 91;
1225 sb->reshape_position = mddev->reshape_position;
1226 sb->new_level = mddev->new_level;
1227 sb->delta_disks = mddev->delta_disks;
1228 sb->new_layout = mddev->new_layout;
Andre Noll664e7c42009-06-18 08:45:27 +10001229 sb->new_chunk = mddev->new_chunk_sectors << 9;
NeilBrownf6705572006-03-27 01:18:11 -08001230 }
1231 mddev->minor_version = sb->minor_version;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 if (mddev->in_sync)
1233 {
1234 sb->recovery_cp = mddev->recovery_cp;
1235 sb->cp_events_hi = (mddev->events>>32);
1236 sb->cp_events_lo = (u32)mddev->events;
1237 if (mddev->recovery_cp == MaxSector)
1238 sb->state = (1<< MD_SB_CLEAN);
1239 } else
1240 sb->recovery_cp = 0;
1241
1242 sb->layout = mddev->layout;
Andre Noll9d8f0362009-06-18 08:45:01 +10001243 sb->chunk_size = mddev->chunk_sectors << 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
NeilBrownc3d97142009-12-14 12:49:52 +11001245 if (mddev->bitmap && mddev->bitmap_info.file == NULL)
NeilBrowna654b9d82005-06-21 17:17:27 -07001246 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
1247
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 sb->disks[0].state = (1<<MD_DISK_REMOVED);
NeilBrowndafb20f2012-03-19 12:46:39 +11001249 rdev_for_each(rdev2, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 mdp_disk_t *d;
NeilBrown86e6ffd2005-11-08 21:39:24 -08001251 int desc_nr;
NeilBrown0261cd9f2009-11-13 17:40:48 +11001252 int is_active = test_bit(In_sync, &rdev2->flags);
1253
1254 if (rdev2->raid_disk >= 0 &&
1255 sb->minor_version >= 91)
1256 /* we have nowhere to store the recovery_offset,
1257 * but if it is not below the reshape_position,
1258 * we can piggy-back on that.
1259 */
1260 is_active = 1;
1261 if (rdev2->raid_disk < 0 ||
1262 test_bit(Faulty, &rdev2->flags))
1263 is_active = 0;
1264 if (is_active)
NeilBrown86e6ffd2005-11-08 21:39:24 -08001265 desc_nr = rdev2->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 else
NeilBrown86e6ffd2005-11-08 21:39:24 -08001267 desc_nr = next_spare++;
NeilBrown19133a42005-11-08 21:39:35 -08001268 rdev2->desc_nr = desc_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 d = &sb->disks[rdev2->desc_nr];
1270 nr_disks++;
1271 d->number = rdev2->desc_nr;
1272 d->major = MAJOR(rdev2->bdev->bd_dev);
1273 d->minor = MINOR(rdev2->bdev->bd_dev);
NeilBrown0261cd9f2009-11-13 17:40:48 +11001274 if (is_active)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 d->raid_disk = rdev2->raid_disk;
1276 else
1277 d->raid_disk = rdev2->desc_nr; /* compatibility */
NeilBrown1be78922006-03-27 01:18:03 -08001278 if (test_bit(Faulty, &rdev2->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279 d->state = (1<<MD_DISK_FAULTY);
NeilBrown0261cd9f2009-11-13 17:40:48 +11001280 else if (is_active) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 d->state = (1<<MD_DISK_ACTIVE);
NeilBrown0261cd9f2009-11-13 17:40:48 +11001282 if (test_bit(In_sync, &rdev2->flags))
1283 d->state |= (1<<MD_DISK_SYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 active++;
1285 working++;
1286 } else {
1287 d->state = 0;
1288 spare++;
1289 working++;
1290 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -07001291 if (test_bit(WriteMostly, &rdev2->flags))
1292 d->state |= (1<<MD_DISK_WRITEMOSTLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 /* now set the "removed" and "faulty" bits on any missing devices */
1295 for (i=0 ; i < mddev->raid_disks ; i++) {
1296 mdp_disk_t *d = &sb->disks[i];
1297 if (d->state == 0 && d->number == 0) {
1298 d->number = i;
1299 d->raid_disk = i;
1300 d->state = (1<<MD_DISK_REMOVED);
1301 d->state |= (1<<MD_DISK_FAULTY);
1302 failed++;
1303 }
1304 }
1305 sb->nr_disks = nr_disks;
1306 sb->active_disks = active;
1307 sb->working_disks = working;
1308 sb->failed_disks = failed;
1309 sb->spare_disks = spare;
1310
1311 sb->this_disk = sb->disks[rdev->desc_nr];
1312 sb->sb_csum = calc_sb_csum(sb);
1313}
1314
1315/*
Chris Webb0cd17fe2008-06-28 08:31:46 +10001316 * rdev_size_change for 0.90.0
1317 */
1318static unsigned long long
NeilBrown3cb03002011-10-11 16:45:26 +11001319super_90_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
Chris Webb0cd17fe2008-06-28 08:31:46 +10001320{
Andre Noll58c0fed2009-03-31 14:33:13 +11001321 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
Chris Webb0cd17fe2008-06-28 08:31:46 +10001322 return 0; /* component must fit device */
NeilBrownc3d97142009-12-14 12:49:52 +11001323 if (rdev->mddev->bitmap_info.offset)
Chris Webb0cd17fe2008-06-28 08:31:46 +10001324 return 0; /* can't move bitmap */
Jonathan Brassow57b2caa2011-01-14 09:14:33 +11001325 rdev->sb_start = calc_dev_sboffset(rdev);
Andre Noll15f4a5f2008-07-21 14:42:12 +10001326 if (!num_sectors || num_sectors > rdev->sb_start)
1327 num_sectors = rdev->sb_start;
NeilBrown27a7b262011-09-10 17:21:28 +10001328 /* Limit to 4TB as metadata cannot record more than that.
1329 * 4TB == 2^32 KB, or 2*2^32 sectors.
1330 */
NeilBrown667a5312012-08-16 16:46:12 +10001331 if (num_sectors >= (2ULL << 32) && rdev->mddev->level >= 1)
NeilBrown27a7b262011-09-10 17:21:28 +10001332 num_sectors = (2ULL << 32) - 2;
Andre Noll0f420352008-07-11 22:02:23 +10001333 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
Chris Webb0cd17fe2008-06-28 08:31:46 +10001334 rdev->sb_page);
1335 md_super_wait(rdev->mddev);
Justin Maggardc26a44e2010-11-24 16:36:17 +11001336 return num_sectors;
Chris Webb0cd17fe2008-06-28 08:31:46 +10001337}
1338
NeilBrownc6563a82012-05-21 09:27:00 +10001339static int
1340super_90_allow_new_offset(struct md_rdev *rdev, unsigned long long new_offset)
1341{
1342 /* non-zero offset changes not possible with v0.90 */
1343 return new_offset == 0;
1344}
Chris Webb0cd17fe2008-06-28 08:31:46 +10001345
1346/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 * version 1 superblock
1348 */
1349
NeilBrownf72ffdd2014-09-30 14:23:59 +10001350static __le32 calc_sb_1_csum(struct mdp_superblock_1 *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351{
NeilBrown1c05b4b2006-10-21 10:24:08 -07001352 __le32 disk_csum;
1353 u32 csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 unsigned long long newcsum;
1355 int size = 256 + le32_to_cpu(sb->max_dev)*2;
NeilBrown1c05b4b2006-10-21 10:24:08 -07001356 __le32 *isuper = (__le32*)sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357
1358 disk_csum = sb->sb_csum;
1359 sb->sb_csum = 0;
1360 newcsum = 0;
NeilBrown1f3c9902012-12-11 13:09:00 +11001361 for (; size >= 4; size -= 4)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 newcsum += le32_to_cpu(*isuper++);
1363
1364 if (size == 2)
NeilBrown1c05b4b2006-10-21 10:24:08 -07001365 newcsum += le16_to_cpu(*(__le16*) isuper);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366
1367 csum = (newcsum & 0xffffffff) + (newcsum >> 32);
1368 sb->sb_csum = disk_csum;
1369 return cpu_to_le32(csum);
1370}
1371
NeilBrown2699b672011-07-28 11:31:47 +10001372static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
1373 int acknowledged);
NeilBrown3cb03002011-10-11 16:45:26 +11001374static int super_1_load(struct md_rdev *rdev, struct md_rdev *refdev, int minor_version)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375{
1376 struct mdp_superblock_1 *sb;
1377 int ret;
Andre Noll0f420352008-07-11 22:02:23 +10001378 sector_t sb_start;
NeilBrownc6563a82012-05-21 09:27:00 +10001379 sector_t sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
NeilBrown0002b272005-09-09 16:23:53 -07001381 int bmask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382
1383 /*
Andre Noll0f420352008-07-11 22:02:23 +10001384 * Calculate the position of the superblock in 512byte sectors.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385 * It is always aligned to a 4K boundary and
1386 * depeding on minor_version, it can be:
1387 * 0: At least 8K, but less than 12K, from end of device
1388 * 1: At start of device
1389 * 2: 4K from start of device.
1390 */
1391 switch(minor_version) {
1392 case 0:
Mike Snitzer77304d22010-11-08 14:39:12 +01001393 sb_start = i_size_read(rdev->bdev->bd_inode) >> 9;
Andre Noll0f420352008-07-11 22:02:23 +10001394 sb_start -= 8*2;
1395 sb_start &= ~(sector_t)(4*2-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 break;
1397 case 1:
Andre Noll0f420352008-07-11 22:02:23 +10001398 sb_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399 break;
1400 case 2:
Andre Noll0f420352008-07-11 22:02:23 +10001401 sb_start = 8;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 break;
1403 default:
1404 return -EINVAL;
1405 }
Andre Noll0f420352008-07-11 22:02:23 +10001406 rdev->sb_start = sb_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407
NeilBrown0002b272005-09-09 16:23:53 -07001408 /* superblock is rarely larger than 1K, but it can be larger,
1409 * and it is safe to read 4k, so we do that
1410 */
1411 ret = read_disk_sb(rdev, 4096);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 if (ret) return ret;
1413
Namhyung Kim65a06f062011-07-27 11:00:36 +10001414 sb = page_address(rdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415
1416 if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1417 sb->major_version != cpu_to_le32(1) ||
1418 le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
Andre Noll0f420352008-07-11 22:02:23 +10001419 le64_to_cpu(sb->super_offset) != rdev->sb_start ||
NeilBrown71c08052005-09-09 16:23:51 -07001420 (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 return -EINVAL;
1422
1423 if (calc_sb_1_csum(sb) != sb->sb_csum) {
1424 printk("md: invalid superblock checksum on %s\n",
1425 bdevname(rdev->bdev,b));
1426 return -EINVAL;
1427 }
1428 if (le64_to_cpu(sb->data_size) < 10) {
1429 printk("md: data_size too small on %s\n",
1430 bdevname(rdev->bdev,b));
1431 return -EINVAL;
1432 }
NeilBrownc6563a82012-05-21 09:27:00 +10001433 if (sb->pad0 ||
1434 sb->pad3[0] ||
1435 memcmp(sb->pad3, sb->pad3+1, sizeof(sb->pad3) - sizeof(sb->pad3[1])))
1436 /* Some padding is non-zero, might be a new feature */
1437 return -EINVAL;
NeilBrowne11e93f2007-05-09 02:35:36 -07001438
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 rdev->preferred_minor = 0xffff;
1440 rdev->data_offset = le64_to_cpu(sb->data_offset);
NeilBrownc6563a82012-05-21 09:27:00 +10001441 rdev->new_data_offset = rdev->data_offset;
1442 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE) &&
1443 (le32_to_cpu(sb->feature_map) & MD_FEATURE_NEW_OFFSET))
1444 rdev->new_data_offset += (s32)le32_to_cpu(sb->new_offset);
NeilBrown4dbcdc72006-01-06 00:20:52 -08001445 atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
NeilBrown0002b272005-09-09 16:23:53 -07001447 rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
Martin K. Petersene1defc42009-05-22 17:17:49 -04001448 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
NeilBrown0002b272005-09-09 16:23:53 -07001449 if (rdev->sb_size & bmask)
NeilBrowna1801f82008-03-04 14:29:31 -08001450 rdev->sb_size = (rdev->sb_size | bmask) + 1;
1451
1452 if (minor_version
Andre Noll0f420352008-07-11 22:02:23 +10001453 && rdev->data_offset < sb_start + (rdev->sb_size/512))
NeilBrowna1801f82008-03-04 14:29:31 -08001454 return -EINVAL;
NeilBrownc6563a82012-05-21 09:27:00 +10001455 if (minor_version
1456 && rdev->new_data_offset < sb_start + (rdev->sb_size/512))
1457 return -EINVAL;
NeilBrown0002b272005-09-09 16:23:53 -07001458
NeilBrown31b65a02006-07-10 04:44:14 -07001459 if (sb->level == cpu_to_le32(LEVEL_MULTIPATH))
1460 rdev->desc_nr = -1;
1461 else
1462 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1463
NeilBrown2699b672011-07-28 11:31:47 +10001464 if (!rdev->bb_page) {
1465 rdev->bb_page = alloc_page(GFP_KERNEL);
1466 if (!rdev->bb_page)
1467 return -ENOMEM;
1468 }
1469 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BAD_BLOCKS) &&
1470 rdev->badblocks.count == 0) {
1471 /* need to load the bad block list.
1472 * Currently we limit it to one page.
1473 */
1474 s32 offset;
1475 sector_t bb_sector;
1476 u64 *bbp;
1477 int i;
1478 int sectors = le16_to_cpu(sb->bblog_size);
1479 if (sectors > (PAGE_SIZE / 512))
1480 return -EINVAL;
1481 offset = le32_to_cpu(sb->bblog_offset);
1482 if (offset == 0)
1483 return -EINVAL;
1484 bb_sector = (long long)offset;
1485 if (!sync_page_io(rdev, bb_sector, sectors << 9,
1486 rdev->bb_page, READ, true))
1487 return -EIO;
1488 bbp = (u64 *)page_address(rdev->bb_page);
1489 rdev->badblocks.shift = sb->bblog_shift;
1490 for (i = 0 ; i < (sectors << (9-3)) ; i++, bbp++) {
1491 u64 bb = le64_to_cpu(*bbp);
1492 int count = bb & (0x3ff);
1493 u64 sector = bb >> 10;
1494 sector <<= sb->bblog_shift;
1495 count <<= sb->bblog_shift;
1496 if (bb + 1 == 0)
1497 break;
1498 if (md_set_badblocks(&rdev->badblocks,
1499 sector, count, 1) == 0)
1500 return -EINVAL;
1501 }
NeilBrown486adf72013-04-24 11:42:44 +10001502 } else if (sb->bblog_offset != 0)
1503 rdev->badblocks.shift = 0;
NeilBrown2699b672011-07-28 11:31:47 +10001504
Harvey Harrison9a7b2b02008-04-28 02:15:49 -07001505 if (!refdev) {
NeilBrown8ed75462006-02-03 03:03:41 -08001506 ret = 1;
Harvey Harrison9a7b2b02008-04-28 02:15:49 -07001507 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 __u64 ev1, ev2;
Namhyung Kim65a06f062011-07-27 11:00:36 +10001509 struct mdp_superblock_1 *refsb = page_address(refdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510
1511 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1512 sb->level != refsb->level ||
1513 sb->layout != refsb->layout ||
1514 sb->chunksize != refsb->chunksize) {
1515 printk(KERN_WARNING "md: %s has strangely different"
1516 " superblock to %s\n",
1517 bdevname(rdev->bdev,b),
1518 bdevname(refdev->bdev,b2));
1519 return -EINVAL;
1520 }
1521 ev1 = le64_to_cpu(sb->events);
1522 ev2 = le64_to_cpu(refsb->events);
1523
1524 if (ev1 > ev2)
NeilBrown8ed75462006-02-03 03:03:41 -08001525 ret = 1;
1526 else
1527 ret = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 }
NeilBrownc6563a82012-05-21 09:27:00 +10001529 if (minor_version) {
1530 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9);
1531 sectors -= rdev->data_offset;
1532 } else
1533 sectors = rdev->sb_start;
1534 if (sectors < le64_to_cpu(sb->data_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 return -EINVAL;
Andre Nolldd8ac332009-03-31 14:33:13 +11001536 rdev->sectors = le64_to_cpu(sb->data_size);
NeilBrown8ed75462006-02-03 03:03:41 -08001537 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538}
1539
NeilBrownfd01b882011-10-11 16:47:53 +11001540static int super_1_validate(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541{
Namhyung Kim65a06f062011-07-27 11:00:36 +10001542 struct mdp_superblock_1 *sb = page_address(rdev->sb_page);
NeilBrown07d84d102006-06-26 00:27:56 -07001543 __u64 ev1 = le64_to_cpu(sb->events);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
NeilBrown41158c72005-06-21 17:17:25 -07001545 rdev->raid_disk = -1;
NeilBrownc5d79ad2008-02-06 01:39:54 -08001546 clear_bit(Faulty, &rdev->flags);
1547 clear_bit(In_sync, &rdev->flags);
NeilBrown8313b8e2013-12-12 10:13:33 +11001548 clear_bit(Bitmap_sync, &rdev->flags);
NeilBrownc5d79ad2008-02-06 01:39:54 -08001549 clear_bit(WriteMostly, &rdev->flags);
NeilBrownc5d79ad2008-02-06 01:39:54 -08001550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 if (mddev->raid_disks == 0) {
1552 mddev->major_version = 1;
1553 mddev->patch_version = 0;
NeilBrowne6910632008-02-06 01:39:51 -08001554 mddev->external = 0;
Andre Noll9d8f0362009-06-18 08:45:01 +10001555 mddev->chunk_sectors = le32_to_cpu(sb->chunksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1557 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1558 mddev->level = le32_to_cpu(sb->level);
NeilBrownd9d166c2006-01-06 00:20:51 -08001559 mddev->clevel[0] = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 mddev->layout = le32_to_cpu(sb->layout);
1561 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
Andre Noll58c0fed2009-03-31 14:33:13 +11001562 mddev->dev_sectors = le64_to_cpu(sb->size);
NeilBrown07d84d102006-06-26 00:27:56 -07001563 mddev->events = ev1;
NeilBrownc3d97142009-12-14 12:49:52 +11001564 mddev->bitmap_info.offset = 0;
NeilBrown6409bb02012-05-22 13:55:07 +10001565 mddev->bitmap_info.space = 0;
1566 /* Default location for bitmap is 1K after superblock
1567 * using 3K - total of 4K
1568 */
NeilBrownc3d97142009-12-14 12:49:52 +11001569 mddev->bitmap_info.default_offset = 1024 >> 9;
NeilBrown6409bb02012-05-22 13:55:07 +10001570 mddev->bitmap_info.default_space = (4096-1024) >> 9;
NeilBrown2c810cd2012-05-21 09:27:00 +10001571 mddev->reshape_backwards = 0;
1572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1574 memcpy(mddev->uuid, sb->set_uuid, 16);
1575
1576 mddev->max_disks = (4096-256)/2;
NeilBrowna654b9d82005-06-21 17:17:27 -07001577
NeilBrown71c08052005-09-09 16:23:51 -07001578 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
NeilBrown6409bb02012-05-22 13:55:07 +10001579 mddev->bitmap_info.file == NULL) {
NeilBrownc3d97142009-12-14 12:49:52 +11001580 mddev->bitmap_info.offset =
1581 (__s32)le32_to_cpu(sb->bitmap_offset);
NeilBrown6409bb02012-05-22 13:55:07 +10001582 /* Metadata doesn't record how much space is available.
1583 * For 1.0, we assume we can use up to the superblock
1584 * if before, else to 4K beyond superblock.
1585 * For others, assume no change is possible.
1586 */
1587 if (mddev->minor_version > 0)
1588 mddev->bitmap_info.space = 0;
1589 else if (mddev->bitmap_info.offset > 0)
1590 mddev->bitmap_info.space =
1591 8 - mddev->bitmap_info.offset;
1592 else
1593 mddev->bitmap_info.space =
1594 -mddev->bitmap_info.offset;
1595 }
NeilBrowne11e93f2007-05-09 02:35:36 -07001596
NeilBrownf6705572006-03-27 01:18:11 -08001597 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1598 mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1599 mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1600 mddev->new_level = le32_to_cpu(sb->new_level);
1601 mddev->new_layout = le32_to_cpu(sb->new_layout);
Andre Noll664e7c42009-06-18 08:45:27 +10001602 mddev->new_chunk_sectors = le32_to_cpu(sb->new_chunk);
NeilBrown2c810cd2012-05-21 09:27:00 +10001603 if (mddev->delta_disks < 0 ||
1604 (mddev->delta_disks == 0 &&
1605 (le32_to_cpu(sb->feature_map)
1606 & MD_FEATURE_RESHAPE_BACKWARDS)))
1607 mddev->reshape_backwards = 1;
NeilBrownf6705572006-03-27 01:18:11 -08001608 } else {
1609 mddev->reshape_position = MaxSector;
1610 mddev->delta_disks = 0;
1611 mddev->new_level = mddev->level;
1612 mddev->new_layout = mddev->layout;
Andre Noll664e7c42009-06-18 08:45:27 +10001613 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08001614 }
1615
NeilBrown41158c72005-06-21 17:17:25 -07001616 } else if (mddev->pers == NULL) {
NeilBrownbe6800a2010-05-18 10:17:09 +10001617 /* Insist of good event counter while assembling, except for
1618 * spares (which don't need an event count) */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 ++ev1;
NeilBrownbe6800a2010-05-18 10:17:09 +10001620 if (rdev->desc_nr >= 0 &&
1621 rdev->desc_nr < le32_to_cpu(sb->max_dev) &&
1622 le16_to_cpu(sb->dev_roles[rdev->desc_nr]) < 0xfffe)
1623 if (ev1 < mddev->events)
1624 return -EINVAL;
NeilBrown41158c72005-06-21 17:17:25 -07001625 } else if (mddev->bitmap) {
1626 /* If adding to array with a bitmap, then we can accept an
1627 * older device, but not too old.
1628 */
NeilBrown41158c72005-06-21 17:17:25 -07001629 if (ev1 < mddev->bitmap->events_cleared)
1630 return 0;
NeilBrown8313b8e2013-12-12 10:13:33 +11001631 if (ev1 < mddev->events)
1632 set_bit(Bitmap_sync, &rdev->flags);
NeilBrown07d84d102006-06-26 00:27:56 -07001633 } else {
1634 if (ev1 < mddev->events)
1635 /* just a hot-add of a new device, leave raid_disk at -1 */
1636 return 0;
1637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638 if (mddev->level != LEVEL_MULTIPATH) {
1639 int role;
NeilBrown3673f302009-08-03 10:59:56 +10001640 if (rdev->desc_nr < 0 ||
1641 rdev->desc_nr >= le32_to_cpu(sb->max_dev)) {
1642 role = 0xffff;
1643 rdev->desc_nr = -1;
1644 } else
1645 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 switch(role) {
1647 case 0xffff: /* spare */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648 break;
1649 case 0xfffe: /* faulty */
NeilBrownb2d444d2005-11-08 21:39:31 -08001650 set_bit(Faulty, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 break;
1652 default:
NeilBrownf4667222013-12-09 12:04:56 +11001653 rdev->saved_raid_disk = role;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001654 if ((le32_to_cpu(sb->feature_map) &
NeilBrownf4667222013-12-09 12:04:56 +11001655 MD_FEATURE_RECOVERY_OFFSET)) {
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001656 rdev->recovery_offset = le64_to_cpu(sb->recovery_offset);
NeilBrownf4667222013-12-09 12:04:56 +11001657 if (!(le32_to_cpu(sb->feature_map) &
1658 MD_FEATURE_RECOVERY_BITMAP))
1659 rdev->saved_raid_disk = -1;
1660 } else
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001661 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 rdev->raid_disk = role;
1663 break;
1664 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -07001665 if (sb->devflags & WriteMostly1)
1666 set_bit(WriteMostly, &rdev->flags);
NeilBrown2d78f8c2011-12-23 10:17:51 +11001667 if (le32_to_cpu(sb->feature_map) & MD_FEATURE_REPLACEMENT)
1668 set_bit(Replacement, &rdev->flags);
NeilBrown41158c72005-06-21 17:17:25 -07001669 } else /* MULTIPATH are always insync */
NeilBrownb2d444d2005-11-08 21:39:31 -08001670 set_bit(In_sync, &rdev->flags);
NeilBrown41158c72005-06-21 17:17:25 -07001671
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 return 0;
1673}
1674
NeilBrownfd01b882011-10-11 16:47:53 +11001675static void super_1_sync(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676{
1677 struct mdp_superblock_1 *sb;
NeilBrown3cb03002011-10-11 16:45:26 +11001678 struct md_rdev *rdev2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 int max_dev, i;
1680 /* make rdev->sb match mddev and rdev data. */
1681
Namhyung Kim65a06f062011-07-27 11:00:36 +10001682 sb = page_address(rdev->sb_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683
1684 sb->feature_map = 0;
1685 sb->pad0 = 0;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001686 sb->recovery_offset = cpu_to_le64(0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 memset(sb->pad3, 0, sizeof(sb->pad3));
1688
1689 sb->utime = cpu_to_le64((__u64)mddev->utime);
1690 sb->events = cpu_to_le64(mddev->events);
1691 if (mddev->in_sync)
1692 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1693 else
1694 sb->resync_offset = cpu_to_le64(0);
1695
NeilBrown1c05b4b2006-10-21 10:24:08 -07001696 sb->cnt_corrected_read = cpu_to_le32(atomic_read(&rdev->corrected_errors));
NeilBrown4dbcdc72006-01-06 00:20:52 -08001697
NeilBrownf0ca3402006-02-02 14:28:04 -08001698 sb->raid_disks = cpu_to_le32(mddev->raid_disks);
Andre Noll58c0fed2009-03-31 14:33:13 +11001699 sb->size = cpu_to_le64(mddev->dev_sectors);
Andre Noll9d8f0362009-06-18 08:45:01 +10001700 sb->chunksize = cpu_to_le32(mddev->chunk_sectors);
NeilBrown62e1e382009-05-26 09:40:59 +10001701 sb->level = cpu_to_le32(mddev->level);
1702 sb->layout = cpu_to_le32(mddev->layout);
NeilBrownf0ca3402006-02-02 14:28:04 -08001703
NeilBrownaeb9b212011-08-25 14:43:08 +10001704 if (test_bit(WriteMostly, &rdev->flags))
1705 sb->devflags |= WriteMostly1;
1706 else
1707 sb->devflags &= ~WriteMostly1;
NeilBrownc6563a82012-05-21 09:27:00 +10001708 sb->data_offset = cpu_to_le64(rdev->data_offset);
1709 sb->data_size = cpu_to_le64(rdev->sectors);
NeilBrownaeb9b212011-08-25 14:43:08 +10001710
NeilBrownc3d97142009-12-14 12:49:52 +11001711 if (mddev->bitmap && mddev->bitmap_info.file == NULL) {
1712 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_info.offset);
NeilBrown71c08052005-09-09 16:23:51 -07001713 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
NeilBrowna654b9d82005-06-21 17:17:27 -07001714 }
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001715
1716 if (rdev->raid_disk >= 0 &&
NeilBrown97e4f422009-03-31 14:33:13 +11001717 !test_bit(In_sync, &rdev->flags)) {
NeilBrown93be75f2009-12-14 12:50:06 +11001718 sb->feature_map |=
1719 cpu_to_le32(MD_FEATURE_RECOVERY_OFFSET);
1720 sb->recovery_offset =
1721 cpu_to_le64(rdev->recovery_offset);
NeilBrownf4667222013-12-09 12:04:56 +11001722 if (rdev->saved_raid_disk >= 0 && mddev->bitmap)
1723 sb->feature_map |=
1724 cpu_to_le32(MD_FEATURE_RECOVERY_BITMAP);
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001725 }
NeilBrown2d78f8c2011-12-23 10:17:51 +11001726 if (test_bit(Replacement, &rdev->flags))
1727 sb->feature_map |=
1728 cpu_to_le32(MD_FEATURE_REPLACEMENT);
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001729
NeilBrownf6705572006-03-27 01:18:11 -08001730 if (mddev->reshape_position != MaxSector) {
1731 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1732 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1733 sb->new_layout = cpu_to_le32(mddev->new_layout);
1734 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1735 sb->new_level = cpu_to_le32(mddev->new_level);
Andre Noll664e7c42009-06-18 08:45:27 +10001736 sb->new_chunk = cpu_to_le32(mddev->new_chunk_sectors);
NeilBrown2c810cd2012-05-21 09:27:00 +10001737 if (mddev->delta_disks == 0 &&
1738 mddev->reshape_backwards)
1739 sb->feature_map
1740 |= cpu_to_le32(MD_FEATURE_RESHAPE_BACKWARDS);
NeilBrownc6563a82012-05-21 09:27:00 +10001741 if (rdev->new_data_offset != rdev->data_offset) {
1742 sb->feature_map
1743 |= cpu_to_le32(MD_FEATURE_NEW_OFFSET);
1744 sb->new_offset = cpu_to_le32((__u32)(rdev->new_data_offset
1745 - rdev->data_offset));
1746 }
NeilBrownf6705572006-03-27 01:18:11 -08001747 }
NeilBrowna654b9d82005-06-21 17:17:27 -07001748
NeilBrown2699b672011-07-28 11:31:47 +10001749 if (rdev->badblocks.count == 0)
1750 /* Nothing to do for bad blocks*/ ;
1751 else if (sb->bblog_offset == 0)
1752 /* Cannot record bad blocks on this device */
1753 md_error(mddev, rdev);
1754 else {
1755 struct badblocks *bb = &rdev->badblocks;
1756 u64 *bbp = (u64 *)page_address(rdev->bb_page);
1757 u64 *p = bb->page;
1758 sb->feature_map |= cpu_to_le32(MD_FEATURE_BAD_BLOCKS);
1759 if (bb->changed) {
1760 unsigned seq;
1761
1762retry:
1763 seq = read_seqbegin(&bb->lock);
1764
1765 memset(bbp, 0xff, PAGE_SIZE);
1766
1767 for (i = 0 ; i < bb->count ; i++) {
majianpeng35f9ac22012-11-08 08:56:27 +08001768 u64 internal_bb = p[i];
NeilBrown2699b672011-07-28 11:31:47 +10001769 u64 store_bb = ((BB_OFFSET(internal_bb) << 10)
1770 | BB_LEN(internal_bb));
majianpeng35f9ac22012-11-08 08:56:27 +08001771 bbp[i] = cpu_to_le64(store_bb);
NeilBrown2699b672011-07-28 11:31:47 +10001772 }
NeilBrownd0962932012-03-19 12:46:41 +11001773 bb->changed = 0;
NeilBrown2699b672011-07-28 11:31:47 +10001774 if (read_seqretry(&bb->lock, seq))
1775 goto retry;
1776
1777 bb->sector = (rdev->sb_start +
1778 (int)le32_to_cpu(sb->bblog_offset));
1779 bb->size = le16_to_cpu(sb->bblog_size);
NeilBrown2699b672011-07-28 11:31:47 +10001780 }
1781 }
1782
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 max_dev = 0;
NeilBrowndafb20f2012-03-19 12:46:39 +11001784 rdev_for_each(rdev2, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 if (rdev2->desc_nr+1 > max_dev)
1786 max_dev = rdev2->desc_nr+1;
NeilBrowna778b732007-05-23 13:58:10 -07001787
NeilBrown70471da2009-08-03 10:59:57 +10001788 if (max_dev > le32_to_cpu(sb->max_dev)) {
1789 int bmask;
NeilBrowna778b732007-05-23 13:58:10 -07001790 sb->max_dev = cpu_to_le32(max_dev);
NeilBrown70471da2009-08-03 10:59:57 +10001791 rdev->sb_size = max_dev * 2 + 256;
1792 bmask = queue_logical_block_size(rdev->bdev->bd_disk->queue)-1;
1793 if (rdev->sb_size & bmask)
1794 rdev->sb_size = (rdev->sb_size | bmask) + 1;
NeilBrownddcf3522010-09-08 16:48:17 +10001795 } else
1796 max_dev = le32_to_cpu(sb->max_dev);
1797
Linus Torvalds1da177e2005-04-16 15:20:36 -07001798 for (i=0; i<max_dev;i++)
1799 sb->dev_roles[i] = cpu_to_le16(0xfffe);
NeilBrownf72ffdd2014-09-30 14:23:59 +10001800
NeilBrowndafb20f2012-03-19 12:46:39 +11001801 rdev_for_each(rdev2, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 i = rdev2->desc_nr;
NeilBrownb2d444d2005-11-08 21:39:31 -08001803 if (test_bit(Faulty, &rdev2->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 sb->dev_roles[i] = cpu_to_le16(0xfffe);
NeilBrownb2d444d2005-11-08 21:39:31 -08001805 else if (test_bit(In_sync, &rdev2->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
NeilBrown93be75f2009-12-14 12:50:06 +11001807 else if (rdev2->raid_disk >= 0)
NeilBrown5fd6c1d2006-06-26 00:27:40 -07001808 sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 else
1810 sb->dev_roles[i] = cpu_to_le16(0xffff);
1811 }
1812
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813 sb->sb_csum = calc_sb_1_csum(sb);
1814}
1815
Chris Webb0cd17fe2008-06-28 08:31:46 +10001816static unsigned long long
NeilBrown3cb03002011-10-11 16:45:26 +11001817super_1_rdev_size_change(struct md_rdev *rdev, sector_t num_sectors)
Chris Webb0cd17fe2008-06-28 08:31:46 +10001818{
1819 struct mdp_superblock_1 *sb;
Andre Noll15f4a5f2008-07-21 14:42:12 +10001820 sector_t max_sectors;
Andre Noll58c0fed2009-03-31 14:33:13 +11001821 if (num_sectors && num_sectors < rdev->mddev->dev_sectors)
Chris Webb0cd17fe2008-06-28 08:31:46 +10001822 return 0; /* component must fit device */
NeilBrownc6563a82012-05-21 09:27:00 +10001823 if (rdev->data_offset != rdev->new_data_offset)
1824 return 0; /* too confusing */
Andre Noll0f420352008-07-11 22:02:23 +10001825 if (rdev->sb_start < rdev->data_offset) {
Chris Webb0cd17fe2008-06-28 08:31:46 +10001826 /* minor versions 1 and 2; superblock before data */
Mike Snitzer77304d22010-11-08 14:39:12 +01001827 max_sectors = i_size_read(rdev->bdev->bd_inode) >> 9;
Andre Noll15f4a5f2008-07-21 14:42:12 +10001828 max_sectors -= rdev->data_offset;
1829 if (!num_sectors || num_sectors > max_sectors)
1830 num_sectors = max_sectors;
NeilBrownc3d97142009-12-14 12:49:52 +11001831 } else if (rdev->mddev->bitmap_info.offset) {
Chris Webb0cd17fe2008-06-28 08:31:46 +10001832 /* minor version 0 with bitmap we can't move */
1833 return 0;
1834 } else {
1835 /* minor version 0; superblock after data */
Andre Noll0f420352008-07-11 22:02:23 +10001836 sector_t sb_start;
Mike Snitzer77304d22010-11-08 14:39:12 +01001837 sb_start = (i_size_read(rdev->bdev->bd_inode) >> 9) - 8*2;
Andre Noll0f420352008-07-11 22:02:23 +10001838 sb_start &= ~(sector_t)(4*2 - 1);
Andre Nolldd8ac332009-03-31 14:33:13 +11001839 max_sectors = rdev->sectors + sb_start - rdev->sb_start;
Andre Noll15f4a5f2008-07-21 14:42:12 +10001840 if (!num_sectors || num_sectors > max_sectors)
1841 num_sectors = max_sectors;
Andre Noll0f420352008-07-11 22:02:23 +10001842 rdev->sb_start = sb_start;
Chris Webb0cd17fe2008-06-28 08:31:46 +10001843 }
Namhyung Kim65a06f062011-07-27 11:00:36 +10001844 sb = page_address(rdev->sb_page);
Andre Noll15f4a5f2008-07-21 14:42:12 +10001845 sb->data_size = cpu_to_le64(num_sectors);
Andre Noll0f420352008-07-11 22:02:23 +10001846 sb->super_offset = rdev->sb_start;
Chris Webb0cd17fe2008-06-28 08:31:46 +10001847 sb->sb_csum = calc_sb_1_csum(sb);
Andre Noll0f420352008-07-11 22:02:23 +10001848 md_super_write(rdev->mddev, rdev, rdev->sb_start, rdev->sb_size,
Chris Webb0cd17fe2008-06-28 08:31:46 +10001849 rdev->sb_page);
1850 md_super_wait(rdev->mddev);
Justin Maggardc26a44e2010-11-24 16:36:17 +11001851 return num_sectors;
NeilBrownc6563a82012-05-21 09:27:00 +10001852
1853}
1854
1855static int
1856super_1_allow_new_offset(struct md_rdev *rdev,
1857 unsigned long long new_offset)
1858{
1859 /* All necessary checks on new >= old have been done */
1860 struct bitmap *bitmap;
1861 if (new_offset >= rdev->data_offset)
1862 return 1;
1863
1864 /* with 1.0 metadata, there is no metadata to tread on
1865 * so we can always move back */
1866 if (rdev->mddev->minor_version == 0)
1867 return 1;
1868
1869 /* otherwise we must be sure not to step on
1870 * any metadata, so stay:
1871 * 36K beyond start of superblock
1872 * beyond end of badblocks
1873 * beyond write-intent bitmap
1874 */
1875 if (rdev->sb_start + (32+4)*2 > new_offset)
1876 return 0;
1877 bitmap = rdev->mddev->bitmap;
1878 if (bitmap && !rdev->mddev->bitmap_info.file &&
1879 rdev->sb_start + rdev->mddev->bitmap_info.offset +
NeilBrown1ec885c2012-05-22 13:55:10 +10001880 bitmap->storage.file_pages * (PAGE_SIZE>>9) > new_offset)
NeilBrownc6563a82012-05-21 09:27:00 +10001881 return 0;
1882 if (rdev->badblocks.sector + rdev->badblocks.size > new_offset)
1883 return 0;
1884
1885 return 1;
Chris Webb0cd17fe2008-06-28 08:31:46 +10001886}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887
Adrian Bunk75c96f82005-05-05 16:16:09 -07001888static struct super_type super_types[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 [0] = {
1890 .name = "0.90.0",
1891 .owner = THIS_MODULE,
Chris Webb0cd17fe2008-06-28 08:31:46 +10001892 .load_super = super_90_load,
1893 .validate_super = super_90_validate,
1894 .sync_super = super_90_sync,
1895 .rdev_size_change = super_90_rdev_size_change,
NeilBrownc6563a82012-05-21 09:27:00 +10001896 .allow_new_offset = super_90_allow_new_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 },
1898 [1] = {
1899 .name = "md-1",
1900 .owner = THIS_MODULE,
Chris Webb0cd17fe2008-06-28 08:31:46 +10001901 .load_super = super_1_load,
1902 .validate_super = super_1_validate,
1903 .sync_super = super_1_sync,
1904 .rdev_size_change = super_1_rdev_size_change,
NeilBrownc6563a82012-05-21 09:27:00 +10001905 .allow_new_offset = super_1_allow_new_offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 },
1907};
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908
NeilBrownfd01b882011-10-11 16:47:53 +11001909static void sync_super(struct mddev *mddev, struct md_rdev *rdev)
Jonathan Brassow076f9682011-06-07 17:51:30 -05001910{
1911 if (mddev->sync_super) {
1912 mddev->sync_super(mddev, rdev);
1913 return;
1914 }
1915
1916 BUG_ON(mddev->major_version >= ARRAY_SIZE(super_types));
1917
1918 super_types[mddev->major_version].sync_super(mddev, rdev);
1919}
1920
NeilBrownfd01b882011-10-11 16:47:53 +11001921static int match_mddev_units(struct mddev *mddev1, struct mddev *mddev2)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922{
NeilBrown3cb03002011-10-11 16:45:26 +11001923 struct md_rdev *rdev, *rdev2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924
NeilBrown4b809912008-07-21 17:05:25 +10001925 rcu_read_lock();
1926 rdev_for_each_rcu(rdev, mddev1)
1927 rdev_for_each_rcu(rdev2, mddev2)
NeilBrown7dd5e7c32007-02-28 20:11:35 -08001928 if (rdev->bdev->bd_contains ==
NeilBrown4b809912008-07-21 17:05:25 +10001929 rdev2->bdev->bd_contains) {
1930 rcu_read_unlock();
NeilBrown7dd5e7c32007-02-28 20:11:35 -08001931 return 1;
NeilBrown4b809912008-07-21 17:05:25 +10001932 }
1933 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 return 0;
1935}
1936
1937static LIST_HEAD(pending_raid_disks);
1938
Andre Nollac5e7112009-08-03 10:59:47 +10001939/*
1940 * Try to register data integrity profile for an mddev
1941 *
1942 * This is called when an array is started and after a disk has been kicked
1943 * from the array. It only succeeds if all working and active component devices
1944 * are integrity capable with matching profiles.
1945 */
NeilBrownfd01b882011-10-11 16:47:53 +11001946int md_integrity_register(struct mddev *mddev)
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11001947{
NeilBrown3cb03002011-10-11 16:45:26 +11001948 struct md_rdev *rdev, *reference = NULL;
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11001949
Andre Nollac5e7112009-08-03 10:59:47 +10001950 if (list_empty(&mddev->disks))
1951 return 0; /* nothing to do */
Jonathan Brassow629acb62011-06-08 15:10:08 +10001952 if (!mddev->gendisk || blk_get_integrity(mddev->gendisk))
1953 return 0; /* shouldn't register, or already is */
NeilBrowndafb20f2012-03-19 12:46:39 +11001954 rdev_for_each(rdev, mddev) {
Andre Nollac5e7112009-08-03 10:59:47 +10001955 /* skip spares and non-functional disks */
1956 if (test_bit(Faulty, &rdev->flags))
1957 continue;
1958 if (rdev->raid_disk < 0)
1959 continue;
Andre Nollac5e7112009-08-03 10:59:47 +10001960 if (!reference) {
1961 /* Use the first rdev as the reference */
1962 reference = rdev;
1963 continue;
1964 }
1965 /* does this rdev's profile match the reference profile? */
1966 if (blk_integrity_compare(reference->bdev->bd_disk,
1967 rdev->bdev->bd_disk) < 0)
1968 return -EINVAL;
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11001969 }
Martin K. Petersen89078d52011-03-28 20:09:12 -04001970 if (!reference || !bdev_get_integrity(reference->bdev))
1971 return 0;
Andre Nollac5e7112009-08-03 10:59:47 +10001972 /*
1973 * All component devices are integrity capable and have matching
1974 * profiles, register the common profile for the md device.
1975 */
1976 if (blk_integrity_register(mddev->gendisk,
1977 bdev_get_integrity(reference->bdev)) != 0) {
1978 printk(KERN_ERR "md: failed to register integrity for %s\n",
1979 mdname(mddev));
1980 return -EINVAL;
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11001981 }
Martin K. Petersena91a2782011-03-17 11:11:05 +01001982 printk(KERN_NOTICE "md: data integrity enabled on %s\n", mdname(mddev));
1983 if (bioset_integrity_create(mddev->bio_set, BIO_POOL_SIZE)) {
1984 printk(KERN_ERR "md: failed to create integrity pool for %s\n",
1985 mdname(mddev));
1986 return -EINVAL;
1987 }
Andre Nollac5e7112009-08-03 10:59:47 +10001988 return 0;
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11001989}
Andre Nollac5e7112009-08-03 10:59:47 +10001990EXPORT_SYMBOL(md_integrity_register);
1991
1992/* Disable data integrity if non-capable/non-matching disk is being added */
NeilBrownfd01b882011-10-11 16:47:53 +11001993void md_integrity_add_rdev(struct md_rdev *rdev, struct mddev *mddev)
Andre Nollac5e7112009-08-03 10:59:47 +10001994{
Jonathan Brassow2863b9e2012-10-11 13:38:58 +11001995 struct blk_integrity *bi_rdev;
1996 struct blk_integrity *bi_mddev;
1997
1998 if (!mddev->gendisk)
1999 return;
2000
2001 bi_rdev = bdev_get_integrity(rdev->bdev);
2002 bi_mddev = blk_get_integrity(mddev->gendisk);
Andre Nollac5e7112009-08-03 10:59:47 +10002003
2004 if (!bi_mddev) /* nothing to do */
2005 return;
2006 if (rdev->raid_disk < 0) /* skip spares */
2007 return;
2008 if (bi_rdev && blk_integrity_compare(mddev->gendisk,
2009 rdev->bdev->bd_disk) >= 0)
2010 return;
2011 printk(KERN_NOTICE "disabling data integrity on %s\n", mdname(mddev));
2012 blk_integrity_unregister(mddev->gendisk);
2013}
2014EXPORT_SYMBOL(md_integrity_add_rdev);
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11002015
NeilBrownf72ffdd2014-09-30 14:23:59 +10002016static int bind_rdev_to_array(struct md_rdev *rdev, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017{
NeilBrown7dd5e7c32007-02-28 20:11:35 -08002018 char b[BDEVNAME_SIZE];
NeilBrownf637b9f2005-11-08 21:39:37 -08002019 struct kobject *ko;
Neil Brown1edf80d2006-01-12 01:05:23 -08002020 char *s;
NeilBrown5e55e2f2007-03-26 21:32:14 -08002021 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022
Dan Williams11e2ede2008-04-30 00:52:32 -07002023 /* prevent duplicates */
2024 if (find_rdev(mddev, rdev->bdev->bd_dev))
2025 return -EEXIST;
2026
Andre Nolldd8ac332009-03-31 14:33:13 +11002027 /* make sure rdev->sectors exceeds mddev->dev_sectors */
2028 if (rdev->sectors && (mddev->dev_sectors == 0 ||
2029 rdev->sectors < mddev->dev_sectors)) {
NeilBrowna778b732007-05-23 13:58:10 -07002030 if (mddev->pers) {
2031 /* Cannot change size, so fail
2032 * If mddev->level <= 0, then we don't care
2033 * about aligning sizes (e.g. linear)
2034 */
2035 if (mddev->level > 0)
2036 return -ENOSPC;
2037 } else
Andre Nolldd8ac332009-03-31 14:33:13 +11002038 mddev->dev_sectors = rdev->sectors;
NeilBrown2bf071b2006-01-06 00:20:55 -08002039 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040
2041 /* Verify rdev->desc_nr is unique.
2042 * If it is -1, assign a free number, else
2043 * check number is not in use
2044 */
NeilBrown4878e9e2014-09-25 17:00:11 +10002045 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 if (rdev->desc_nr < 0) {
2047 int choice = 0;
NeilBrown4878e9e2014-09-25 17:00:11 +10002048 if (mddev->pers)
2049 choice = mddev->raid_disks;
2050 while (find_rdev_nr_rcu(mddev, choice))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 choice++;
2052 rdev->desc_nr = choice;
2053 } else {
NeilBrown4878e9e2014-09-25 17:00:11 +10002054 if (find_rdev_nr_rcu(mddev, rdev->desc_nr)) {
2055 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 return -EBUSY;
NeilBrown4878e9e2014-09-25 17:00:11 +10002057 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 }
NeilBrown4878e9e2014-09-25 17:00:11 +10002059 rcu_read_unlock();
NeilBrownde01dfa2009-02-06 18:02:46 +11002060 if (mddev->max_disks && rdev->desc_nr >= mddev->max_disks) {
2061 printk(KERN_WARNING "md: %s: array is limited to %d devices\n",
2062 mdname(mddev), mddev->max_disks);
2063 return -EBUSY;
2064 }
NeilBrown19133a42005-11-08 21:39:35 -08002065 bdevname(rdev->bdev,b);
Greg Kroah-Hartman649316b2007-12-17 23:05:35 -07002066 while ( (s=strchr(b, '/')) != NULL)
Neil Brown1edf80d2006-01-12 01:05:23 -08002067 *s = '!';
Greg Kroah-Hartman649316b2007-12-17 23:05:35 -07002068
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 rdev->mddev = mddev;
NeilBrown19133a42005-11-08 21:39:35 -08002070 printk(KERN_INFO "md: bind<%s>\n", b);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002071
Greg Kroah-Hartmanb2d6db52007-12-17 23:05:35 -07002072 if ((err = kobject_add(&rdev->kobj, &mddev->kobj, "dev-%s", b)))
NeilBrown5e55e2f2007-03-26 21:32:14 -08002073 goto fail;
NeilBrown86e6ffd2005-11-08 21:39:24 -08002074
Tejun Heo0762b8b2008-08-25 19:56:12 +09002075 ko = &part_to_dev(rdev->bdev->bd_part)->kobj;
NeilBrown00bcb4a2010-06-01 19:37:23 +10002076 if (sysfs_create_link(&rdev->kobj, ko, "block"))
2077 /* failure here is OK */;
2078 rdev->sysfs_state = sysfs_get_dirent_safe(rdev->kobj.sd, "state");
NeilBrown3c0ee632008-10-21 13:25:28 +11002079
NeilBrown4b809912008-07-21 17:05:25 +10002080 list_add_rcu(&rdev->same_set, &mddev->disks);
Tejun Heoe09b4572010-11-13 11:55:17 +01002081 bd_link_disk_holder(rdev->bdev, mddev->gendisk);
NeilBrown4044ba52009-01-09 08:31:11 +11002082
2083 /* May as well allow recovery to be retried once */
NeilBrown53890422011-07-27 11:00:36 +10002084 mddev->recovery_disabled++;
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11002085
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 return 0;
NeilBrown5e55e2f2007-03-26 21:32:14 -08002087
2088 fail:
2089 printk(KERN_WARNING "md: failed to register dev-%s for %s\n",
2090 b, mdname(mddev));
2091 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092}
2093
NeilBrown177a99b2008-02-06 01:39:56 -08002094static void md_delayed_delete(struct work_struct *ws)
NeilBrown5792a282007-04-04 19:08:18 -07002095{
NeilBrown3cb03002011-10-11 16:45:26 +11002096 struct md_rdev *rdev = container_of(ws, struct md_rdev, del_work);
NeilBrown5792a282007-04-04 19:08:18 -07002097 kobject_del(&rdev->kobj);
NeilBrown177a99b2008-02-06 01:39:56 -08002098 kobject_put(&rdev->kobj);
NeilBrown5792a282007-04-04 19:08:18 -07002099}
2100
NeilBrownf72ffdd2014-09-30 14:23:59 +10002101static void unbind_rdev_from_array(struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102{
2103 char b[BDEVNAME_SIZE];
NeilBrown403df472014-09-30 15:52:29 +10002104
Tejun Heo49731ba2011-01-14 18:43:57 +01002105 bd_unlink_disk_holder(rdev->bdev, rdev->mddev->gendisk);
NeilBrown4b809912008-07-21 17:05:25 +10002106 list_del_rcu(&rdev->same_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
2108 rdev->mddev = NULL;
NeilBrown86e6ffd2005-11-08 21:39:24 -08002109 sysfs_remove_link(&rdev->kobj, "block");
NeilBrown3c0ee632008-10-21 13:25:28 +11002110 sysfs_put(rdev->sysfs_state);
2111 rdev->sysfs_state = NULL;
NeilBrown2230dfe2011-07-28 11:31:46 +10002112 rdev->badblocks.count = 0;
NeilBrown5792a282007-04-04 19:08:18 -07002113 /* We need to delay this, otherwise we can deadlock when
NeilBrown4b809912008-07-21 17:05:25 +10002114 * writing to 'remove' to "dev/state". We also need
2115 * to delay it due to rcu usage.
NeilBrown5792a282007-04-04 19:08:18 -07002116 */
NeilBrown4b809912008-07-21 17:05:25 +10002117 synchronize_rcu();
NeilBrown177a99b2008-02-06 01:39:56 -08002118 INIT_WORK(&rdev->del_work, md_delayed_delete);
2119 kobject_get(&rdev->kobj);
Tejun Heoe804ac72010-10-15 15:36:08 +02002120 queue_work(md_misc_wq, &rdev->del_work);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121}
2122
2123/*
2124 * prevent the device from being mounted, repartitioned or
2125 * otherwise reused by a RAID array (or any other kernel
2126 * subsystem), by bd_claiming the device.
2127 */
NeilBrown3cb03002011-10-11 16:45:26 +11002128static int lock_rdev(struct md_rdev *rdev, dev_t dev, int shared)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129{
2130 int err = 0;
2131 struct block_device *bdev;
2132 char b[BDEVNAME_SIZE];
2133
Tejun Heod4d77622010-11-13 11:55:18 +01002134 bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL,
NeilBrown3cb03002011-10-11 16:45:26 +11002135 shared ? (struct md_rdev *)lock_rdev : rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136 if (IS_ERR(bdev)) {
2137 printk(KERN_ERR "md: could not open %s.\n",
2138 __bdevname(dev, b));
2139 return PTR_ERR(bdev);
2140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 rdev->bdev = bdev;
2142 return err;
2143}
2144
NeilBrown3cb03002011-10-11 16:45:26 +11002145static void unlock_rdev(struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146{
2147 struct block_device *bdev = rdev->bdev;
2148 rdev->bdev = NULL;
Tejun Heoe525fd82010-11-13 11:55:17 +01002149 blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150}
2151
2152void md_autodetect_dev(dev_t dev);
2153
NeilBrownf72ffdd2014-09-30 14:23:59 +10002154static void export_rdev(struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155{
2156 char b[BDEVNAME_SIZE];
NeilBrown403df472014-09-30 15:52:29 +10002157
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158 printk(KERN_INFO "md: export_rdev(%s)\n",
2159 bdevname(rdev->bdev,b));
NeilBrown545c8792012-05-22 13:54:30 +10002160 md_rdev_clear(rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161#ifndef MODULE
NeilBrownd0fae182008-03-04 14:29:31 -08002162 if (test_bit(AutoDetected, &rdev->flags))
2163 md_autodetect_dev(rdev->bdev->bd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164#endif
2165 unlock_rdev(rdev);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002166 kobject_put(&rdev->kobj);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167}
2168
NeilBrownf72ffdd2014-09-30 14:23:59 +10002169static void kick_rdev_from_array(struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170{
2171 unbind_rdev_from_array(rdev);
2172 export_rdev(rdev);
2173}
2174
NeilBrownfd01b882011-10-11 16:47:53 +11002175static void export_array(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176{
NeilBrown0638bb02014-09-25 17:43:47 +10002177 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
NeilBrown0638bb02014-09-25 17:43:47 +10002179 while (!list_empty(&mddev->disks)) {
2180 rdev = list_first_entry(&mddev->disks, struct md_rdev,
2181 same_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002182 kick_rdev_from_array(rdev);
2183 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 mddev->raid_disks = 0;
2185 mddev->major_version = 0;
2186}
2187
NeilBrownf72ffdd2014-09-30 14:23:59 +10002188static void sync_sbs(struct mddev *mddev, int nospares)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189{
NeilBrown42543762006-06-26 00:27:57 -07002190 /* Update each superblock (in-memory image), but
2191 * if we are allowed to, skip spares which already
2192 * have the right event counter, or have one earlier
2193 * (which would mean they aren't being marked as dirty
2194 * with the rest of the array)
2195 */
NeilBrown3cb03002011-10-11 16:45:26 +11002196 struct md_rdev *rdev;
NeilBrowndafb20f2012-03-19 12:46:39 +11002197 rdev_for_each(rdev, mddev) {
NeilBrown42543762006-06-26 00:27:57 -07002198 if (rdev->sb_events == mddev->events ||
2199 (nospares &&
2200 rdev->raid_disk < 0 &&
NeilBrown42543762006-06-26 00:27:57 -07002201 rdev->sb_events+1 == mddev->events)) {
2202 /* Don't update this superblock */
2203 rdev->sb_loaded = 2;
2204 } else {
Jonathan Brassow076f9682011-06-07 17:51:30 -05002205 sync_super(mddev, rdev);
NeilBrown42543762006-06-26 00:27:57 -07002206 rdev->sb_loaded = 1;
2207 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 }
2209}
2210
NeilBrownf72ffdd2014-09-30 14:23:59 +10002211static void md_update_sb(struct mddev *mddev, int force_change)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212{
NeilBrown3cb03002011-10-11 16:45:26 +11002213 struct md_rdev *rdev;
NeilBrown06d91a52005-06-21 17:17:12 -07002214 int sync_req;
NeilBrown42543762006-06-26 00:27:57 -07002215 int nospares = 0;
NeilBrown2699b672011-07-28 11:31:47 +10002216 int any_badblocks_changed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217
NeilBrownd87f0642013-04-24 11:42:40 +10002218 if (mddev->ro) {
2219 if (force_change)
2220 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2221 return;
2222 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223repeat:
NeilBrown3a3a5dd2010-08-16 18:09:31 +10002224 /* First make sure individual recovery_offsets are correct */
NeilBrowndafb20f2012-03-19 12:46:39 +11002225 rdev_for_each(rdev, mddev) {
NeilBrown3a3a5dd2010-08-16 18:09:31 +10002226 if (rdev->raid_disk >= 0 &&
2227 mddev->delta_disks >= 0 &&
2228 !test_bit(In_sync, &rdev->flags) &&
2229 mddev->curr_resync_completed > rdev->recovery_offset)
2230 rdev->recovery_offset = mddev->curr_resync_completed;
2231
NeilBrownf72ffdd2014-09-30 14:23:59 +10002232 }
Dan Williamsbd52b742010-08-30 17:33:33 +10002233 if (!mddev->persistent) {
NeilBrown070dc6d2010-08-30 17:33:34 +10002234 clear_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrown3a3a5dd2010-08-16 18:09:31 +10002235 clear_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownde393cd2011-07-28 11:31:48 +10002236 if (!mddev->external) {
NeilBrownd97a41d2010-10-28 17:30:20 +11002237 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrowndafb20f2012-03-19 12:46:39 +11002238 rdev_for_each(rdev, mddev) {
NeilBrownde393cd2011-07-28 11:31:48 +10002239 if (rdev->badblocks.changed) {
NeilBrownd0962932012-03-19 12:46:41 +11002240 rdev->badblocks.changed = 0;
NeilBrownde393cd2011-07-28 11:31:48 +10002241 md_ack_all_badblocks(&rdev->badblocks);
2242 md_error(mddev, rdev);
2243 }
2244 clear_bit(Blocked, &rdev->flags);
2245 clear_bit(BlockedBadBlocks, &rdev->flags);
2246 wake_up(&rdev->blocked_wait);
2247 }
2248 }
NeilBrown3a3a5dd2010-08-16 18:09:31 +10002249 wake_up(&mddev->sb_wait);
2250 return;
2251 }
2252
NeilBrown85572d72014-12-15 12:56:56 +11002253 spin_lock(&mddev->lock);
NeilBrown84692192006-08-27 01:23:49 -07002254
NeilBrown3a3a5dd2010-08-16 18:09:31 +10002255 mddev->utime = get_seconds();
2256
NeilBrown850b2b42006-10-03 01:15:46 -07002257 if (test_and_clear_bit(MD_CHANGE_DEVS, &mddev->flags))
2258 force_change = 1;
2259 if (test_and_clear_bit(MD_CHANGE_CLEAN, &mddev->flags))
2260 /* just a clean<-> dirty transition, possibly leave spares alone,
2261 * though if events isn't the right even/odd, we will have to do
2262 * spares after all
2263 */
2264 nospares = 1;
2265 if (force_change)
2266 nospares = 0;
2267 if (mddev->degraded)
NeilBrown84692192006-08-27 01:23:49 -07002268 /* If the array is degraded, then skipping spares is both
2269 * dangerous and fairly pointless.
2270 * Dangerous because a device that was removed from the array
2271 * might have a event_count that still looks up-to-date,
2272 * so it can be re-added without a resync.
2273 * Pointless because if there are any spares to skip,
2274 * then a recovery will happen and soon that array won't
2275 * be degraded any more and the spare can go back to sleep then.
2276 */
NeilBrown850b2b42006-10-03 01:15:46 -07002277 nospares = 0;
NeilBrown84692192006-08-27 01:23:49 -07002278
NeilBrown06d91a52005-06-21 17:17:12 -07002279 sync_req = mddev->in_sync;
NeilBrown42543762006-06-26 00:27:57 -07002280
2281 /* If this is just a dirty<->clean transition, and the array is clean
2282 * and 'events' is odd, we can roll back to the previous clean state */
NeilBrown850b2b42006-10-03 01:15:46 -07002283 if (nospares
NeilBrown42543762006-06-26 00:27:57 -07002284 && (mddev->in_sync && mddev->recovery_cp == MaxSector)
NeilBrowna8707c02010-05-18 09:28:43 +10002285 && mddev->can_decrease_events
2286 && mddev->events != 1) {
NeilBrown42543762006-06-26 00:27:57 -07002287 mddev->events--;
NeilBrowna8707c02010-05-18 09:28:43 +10002288 mddev->can_decrease_events = 0;
2289 } else {
NeilBrown42543762006-06-26 00:27:57 -07002290 /* otherwise we have to go forward and ... */
2291 mddev->events ++;
NeilBrowna8707c02010-05-18 09:28:43 +10002292 mddev->can_decrease_events = nospares;
NeilBrown42543762006-06-26 00:27:57 -07002293 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294
NeilBrown403df472014-09-30 15:52:29 +10002295 /*
2296 * This 64-bit counter should never wrap.
2297 * Either we are in around ~1 trillion A.C., assuming
2298 * 1 reboot per second, or we have a bug...
2299 */
2300 WARN_ON(mddev->events == 0);
NeilBrown2699b672011-07-28 11:31:47 +10002301
NeilBrowndafb20f2012-03-19 12:46:39 +11002302 rdev_for_each(rdev, mddev) {
NeilBrown2699b672011-07-28 11:31:47 +10002303 if (rdev->badblocks.changed)
2304 any_badblocks_changed++;
NeilBrownde393cd2011-07-28 11:31:48 +10002305 if (test_bit(Faulty, &rdev->flags))
2306 set_bit(FaultRecorded, &rdev->flags);
2307 }
NeilBrown2699b672011-07-28 11:31:47 +10002308
NeilBrowne6910632008-02-06 01:39:51 -08002309 sync_sbs(mddev, nospares);
NeilBrown85572d72014-12-15 12:56:56 +11002310 spin_unlock(&mddev->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311
NeilBrown36a4e1f2011-10-07 14:23:17 +11002312 pr_debug("md: updating %s RAID superblock on device (in sync %d)\n",
2313 mdname(mddev), mddev->in_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314
NeilBrown4ad13662007-07-17 04:06:13 -07002315 bitmap_update_sb(mddev->bitmap);
NeilBrowndafb20f2012-03-19 12:46:39 +11002316 rdev_for_each(rdev, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317 char b[BDEVNAME_SIZE];
NeilBrown36a4e1f2011-10-07 14:23:17 +11002318
NeilBrown42543762006-06-26 00:27:57 -07002319 if (rdev->sb_loaded != 1)
2320 continue; /* no noise on spare devices */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321
NeilBrownf4667222013-12-09 12:04:56 +11002322 if (!test_bit(Faulty, &rdev->flags)) {
NeilBrown7bfa19f2005-06-21 17:17:28 -07002323 md_super_write(mddev,rdev,
Andre Noll0f420352008-07-11 22:02:23 +10002324 rdev->sb_start, rdev->sb_size,
NeilBrown7bfa19f2005-06-21 17:17:28 -07002325 rdev->sb_page);
NeilBrown36a4e1f2011-10-07 14:23:17 +11002326 pr_debug("md: (write) %s's sb offset: %llu\n",
2327 bdevname(rdev->bdev, b),
2328 (unsigned long long)rdev->sb_start);
NeilBrown42543762006-06-26 00:27:57 -07002329 rdev->sb_events = mddev->events;
NeilBrown2699b672011-07-28 11:31:47 +10002330 if (rdev->badblocks.size) {
2331 md_super_write(mddev, rdev,
2332 rdev->badblocks.sector,
2333 rdev->badblocks.size << 9,
2334 rdev->bb_page);
2335 rdev->badblocks.size = 0;
2336 }
NeilBrown7bfa19f2005-06-21 17:17:28 -07002337
NeilBrownf4667222013-12-09 12:04:56 +11002338 } else
NeilBrown36a4e1f2011-10-07 14:23:17 +11002339 pr_debug("md: %s (skipping faulty)\n",
2340 bdevname(rdev->bdev, b));
Andrei Warkentind70ed2e2011-10-18 12:16:48 +11002341
NeilBrown7bfa19f2005-06-21 17:17:28 -07002342 if (mddev->level == LEVEL_MULTIPATH)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 /* only need to write one superblock... */
2344 break;
2345 }
NeilBrowna9701a32005-11-08 21:39:34 -08002346 md_super_wait(mddev);
NeilBrown850b2b42006-10-03 01:15:46 -07002347 /* if there was a failure, MD_CHANGE_DEVS was set, and we re-write super */
NeilBrown7bfa19f2005-06-21 17:17:28 -07002348
NeilBrown85572d72014-12-15 12:56:56 +11002349 spin_lock(&mddev->lock);
NeilBrown850b2b42006-10-03 01:15:46 -07002350 if (mddev->in_sync != sync_req ||
2351 test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
NeilBrown06d91a52005-06-21 17:17:12 -07002352 /* have to write it out again */
NeilBrown85572d72014-12-15 12:56:56 +11002353 spin_unlock(&mddev->lock);
NeilBrown06d91a52005-06-21 17:17:12 -07002354 goto repeat;
2355 }
NeilBrown850b2b42006-10-03 01:15:46 -07002356 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrown85572d72014-12-15 12:56:56 +11002357 spin_unlock(&mddev->lock);
NeilBrown3d310eb2005-06-21 17:17:26 -07002358 wake_up(&mddev->sb_wait);
NeilBrownacb180b2009-04-14 16:28:34 +10002359 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
2360 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown06d91a52005-06-21 17:17:12 -07002361
NeilBrowndafb20f2012-03-19 12:46:39 +11002362 rdev_for_each(rdev, mddev) {
NeilBrownde393cd2011-07-28 11:31:48 +10002363 if (test_and_clear_bit(FaultRecorded, &rdev->flags))
2364 clear_bit(Blocked, &rdev->flags);
2365
2366 if (any_badblocks_changed)
NeilBrown2699b672011-07-28 11:31:47 +10002367 md_ack_all_badblocks(&rdev->badblocks);
NeilBrownde393cd2011-07-28 11:31:48 +10002368 clear_bit(BlockedBadBlocks, &rdev->flags);
2369 wake_up(&rdev->blocked_wait);
2370 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371}
2372
Andre Noll7f6ce762008-03-23 18:34:54 +01002373/* words written to sysfs files may, or may not, be \n terminated.
NeilBrownbce74da2006-01-06 00:20:41 -08002374 * We want to accept with case. For this we use cmd_match.
2375 */
2376static int cmd_match(const char *cmd, const char *str)
2377{
2378 /* See if cmd, written into a sysfs file, matches
2379 * str. They must either be the same, or cmd can
2380 * have a trailing newline
2381 */
2382 while (*cmd && *str && *cmd == *str) {
2383 cmd++;
2384 str++;
2385 }
2386 if (*cmd == '\n')
2387 cmd++;
2388 if (*str || *cmd)
2389 return 0;
2390 return 1;
2391}
2392
NeilBrown86e6ffd2005-11-08 21:39:24 -08002393struct rdev_sysfs_entry {
2394 struct attribute attr;
NeilBrown3cb03002011-10-11 16:45:26 +11002395 ssize_t (*show)(struct md_rdev *, char *);
2396 ssize_t (*store)(struct md_rdev *, const char *, size_t);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002397};
2398
2399static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002400state_show(struct md_rdev *rdev, char *page)
NeilBrown86e6ffd2005-11-08 21:39:24 -08002401{
2402 char *sep = "";
NeilBrown20a49ff2008-02-06 01:39:57 -08002403 size_t len = 0;
NeilBrown758bfc82014-12-15 12:56:59 +11002404 unsigned long flags = ACCESS_ONCE(rdev->flags);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002405
NeilBrown758bfc82014-12-15 12:56:59 +11002406 if (test_bit(Faulty, &flags) ||
NeilBrownde393cd2011-07-28 11:31:48 +10002407 rdev->badblocks.unacked_exist) {
NeilBrown86e6ffd2005-11-08 21:39:24 -08002408 len+= sprintf(page+len, "%sfaulty",sep);
2409 sep = ",";
2410 }
NeilBrown758bfc82014-12-15 12:56:59 +11002411 if (test_bit(In_sync, &flags)) {
NeilBrown86e6ffd2005-11-08 21:39:24 -08002412 len += sprintf(page+len, "%sin_sync",sep);
2413 sep = ",";
2414 }
NeilBrown758bfc82014-12-15 12:56:59 +11002415 if (test_bit(WriteMostly, &flags)) {
NeilBrownf6556752006-06-26 00:28:01 -07002416 len += sprintf(page+len, "%swrite_mostly",sep);
2417 sep = ",";
2418 }
NeilBrown758bfc82014-12-15 12:56:59 +11002419 if (test_bit(Blocked, &flags) ||
NeilBrown52c64152011-12-08 16:22:48 +11002420 (rdev->badblocks.unacked_exist
NeilBrown758bfc82014-12-15 12:56:59 +11002421 && !test_bit(Faulty, &flags))) {
Dan Williams6bfe0b42008-04-30 00:52:32 -07002422 len += sprintf(page+len, "%sblocked", sep);
2423 sep = ",";
2424 }
NeilBrown758bfc82014-12-15 12:56:59 +11002425 if (!test_bit(Faulty, &flags) &&
2426 !test_bit(In_sync, &flags)) {
NeilBrown86e6ffd2005-11-08 21:39:24 -08002427 len += sprintf(page+len, "%sspare", sep);
2428 sep = ",";
2429 }
NeilBrown758bfc82014-12-15 12:56:59 +11002430 if (test_bit(WriteErrorSeen, &flags)) {
NeilBrownd7a9d442011-07-28 11:31:48 +10002431 len += sprintf(page+len, "%swrite_error", sep);
2432 sep = ",";
2433 }
NeilBrown758bfc82014-12-15 12:56:59 +11002434 if (test_bit(WantReplacement, &flags)) {
NeilBrown2d78f8c2011-12-23 10:17:51 +11002435 len += sprintf(page+len, "%swant_replacement", sep);
2436 sep = ",";
2437 }
NeilBrown758bfc82014-12-15 12:56:59 +11002438 if (test_bit(Replacement, &flags)) {
NeilBrown2d78f8c2011-12-23 10:17:51 +11002439 len += sprintf(page+len, "%sreplacement", sep);
2440 sep = ",";
2441 }
2442
NeilBrown86e6ffd2005-11-08 21:39:24 -08002443 return len+sprintf(page+len, "\n");
2444}
2445
NeilBrown45dc2de2006-06-26 00:27:58 -07002446static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002447state_store(struct md_rdev *rdev, const char *buf, size_t len)
NeilBrown45dc2de2006-06-26 00:27:58 -07002448{
2449 /* can write
NeilBrownde393cd2011-07-28 11:31:48 +10002450 * faulty - simulates an error
NeilBrown45dc2de2006-06-26 00:27:58 -07002451 * remove - disconnects the device
NeilBrownf6556752006-06-26 00:28:01 -07002452 * writemostly - sets write_mostly
2453 * -writemostly - clears write_mostly
NeilBrownde393cd2011-07-28 11:31:48 +10002454 * blocked - sets the Blocked flags
2455 * -blocked - clears the Blocked and possibly simulates an error
NeilBrown6d56e272009-04-14 12:01:57 +10002456 * insync - sets Insync providing device isn't active
NeilBrownf4667222013-12-09 12:04:56 +11002457 * -insync - clear Insync for a device with a slot assigned,
2458 * so that it gets rebuilt based on bitmap
NeilBrownd7a9d442011-07-28 11:31:48 +10002459 * write_error - sets WriteErrorSeen
2460 * -write_error - clears WriteErrorSeen
NeilBrown45dc2de2006-06-26 00:27:58 -07002461 */
2462 int err = -EINVAL;
2463 if (cmd_match(buf, "faulty") && rdev->mddev->pers) {
2464 md_error(rdev->mddev, rdev);
NeilBrown5ef56c82011-08-25 14:42:51 +10002465 if (test_bit(Faulty, &rdev->flags))
2466 err = 0;
2467 else
2468 err = -EBUSY;
NeilBrown45dc2de2006-06-26 00:27:58 -07002469 } else if (cmd_match(buf, "remove")) {
2470 if (rdev->raid_disk >= 0)
2471 err = -EBUSY;
2472 else {
NeilBrownfd01b882011-10-11 16:47:53 +11002473 struct mddev *mddev = rdev->mddev;
NeilBrown45dc2de2006-06-26 00:27:58 -07002474 kick_rdev_from_array(rdev);
NeilBrown3f9d7b02006-12-22 01:11:41 -08002475 if (mddev->pers)
2476 md_update_sb(mddev, 1);
NeilBrown45dc2de2006-06-26 00:27:58 -07002477 md_new_event(mddev);
2478 err = 0;
2479 }
NeilBrownf6556752006-06-26 00:28:01 -07002480 } else if (cmd_match(buf, "writemostly")) {
2481 set_bit(WriteMostly, &rdev->flags);
2482 err = 0;
2483 } else if (cmd_match(buf, "-writemostly")) {
2484 clear_bit(WriteMostly, &rdev->flags);
2485 err = 0;
Dan Williams6bfe0b42008-04-30 00:52:32 -07002486 } else if (cmd_match(buf, "blocked")) {
2487 set_bit(Blocked, &rdev->flags);
2488 err = 0;
2489 } else if (cmd_match(buf, "-blocked")) {
NeilBrownde393cd2011-07-28 11:31:48 +10002490 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown7da64a02011-08-30 16:20:17 +10002491 rdev->badblocks.unacked_exist) {
NeilBrownde393cd2011-07-28 11:31:48 +10002492 /* metadata handler doesn't understand badblocks,
2493 * so we need to fail the device
2494 */
2495 md_error(rdev->mddev, rdev);
2496 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07002497 clear_bit(Blocked, &rdev->flags);
NeilBrownde393cd2011-07-28 11:31:48 +10002498 clear_bit(BlockedBadBlocks, &rdev->flags);
Dan Williams6bfe0b42008-04-30 00:52:32 -07002499 wake_up(&rdev->blocked_wait);
2500 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2501 md_wakeup_thread(rdev->mddev->thread);
2502
2503 err = 0;
NeilBrown6d56e272009-04-14 12:01:57 +10002504 } else if (cmd_match(buf, "insync") && rdev->raid_disk == -1) {
2505 set_bit(In_sync, &rdev->flags);
2506 err = 0;
NeilBrownf4667222013-12-09 12:04:56 +11002507 } else if (cmd_match(buf, "-insync") && rdev->raid_disk >= 0) {
NeilBrowne1960f82014-09-30 15:24:25 +10002508 if (rdev->mddev->pers == NULL) {
2509 clear_bit(In_sync, &rdev->flags);
2510 rdev->saved_raid_disk = rdev->raid_disk;
2511 rdev->raid_disk = -1;
2512 err = 0;
2513 }
NeilBrownd7a9d442011-07-28 11:31:48 +10002514 } else if (cmd_match(buf, "write_error")) {
2515 set_bit(WriteErrorSeen, &rdev->flags);
2516 err = 0;
2517 } else if (cmd_match(buf, "-write_error")) {
2518 clear_bit(WriteErrorSeen, &rdev->flags);
2519 err = 0;
NeilBrown2d78f8c2011-12-23 10:17:51 +11002520 } else if (cmd_match(buf, "want_replacement")) {
2521 /* Any non-spare device that is not a replacement can
2522 * become want_replacement at any time, but we then need to
2523 * check if recovery is needed.
2524 */
2525 if (rdev->raid_disk >= 0 &&
2526 !test_bit(Replacement, &rdev->flags))
2527 set_bit(WantReplacement, &rdev->flags);
2528 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2529 md_wakeup_thread(rdev->mddev->thread);
2530 err = 0;
2531 } else if (cmd_match(buf, "-want_replacement")) {
2532 /* Clearing 'want_replacement' is always allowed.
2533 * Once replacements starts it is too late though.
2534 */
2535 err = 0;
2536 clear_bit(WantReplacement, &rdev->flags);
2537 } else if (cmd_match(buf, "replacement")) {
2538 /* Can only set a device as a replacement when array has not
2539 * yet been started. Once running, replacement is automatic
2540 * from spares, or by assigning 'slot'.
2541 */
2542 if (rdev->mddev->pers)
2543 err = -EBUSY;
2544 else {
2545 set_bit(Replacement, &rdev->flags);
2546 err = 0;
2547 }
2548 } else if (cmd_match(buf, "-replacement")) {
2549 /* Similarly, can only clear Replacement before start */
2550 if (rdev->mddev->pers)
2551 err = -EBUSY;
2552 else {
2553 clear_bit(Replacement, &rdev->flags);
2554 err = 0;
2555 }
NeilBrown45dc2de2006-06-26 00:27:58 -07002556 }
NeilBrown00bcb4a2010-06-01 19:37:23 +10002557 if (!err)
2558 sysfs_notify_dirent_safe(rdev->sysfs_state);
NeilBrown45dc2de2006-06-26 00:27:58 -07002559 return err ? err : len;
2560}
NeilBrown80ca3a42006-07-10 04:44:18 -07002561static struct rdev_sysfs_entry rdev_state =
NeilBrown750f1992014-09-30 08:53:05 +10002562__ATTR_PREALLOC(state, S_IRUGO|S_IWUSR, state_show, state_store);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002563
2564static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002565errors_show(struct md_rdev *rdev, char *page)
NeilBrown4dbcdc72006-01-06 00:20:52 -08002566{
2567 return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
2568}
2569
2570static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002571errors_store(struct md_rdev *rdev, const char *buf, size_t len)
NeilBrown4dbcdc72006-01-06 00:20:52 -08002572{
2573 char *e;
2574 unsigned long n = simple_strtoul(buf, &e, 10);
2575 if (*buf && (*e == 0 || *e == '\n')) {
2576 atomic_set(&rdev->corrected_errors, n);
2577 return len;
2578 }
2579 return -EINVAL;
2580}
2581static struct rdev_sysfs_entry rdev_errors =
NeilBrown80ca3a42006-07-10 04:44:18 -07002582__ATTR(errors, S_IRUGO|S_IWUSR, errors_show, errors_store);
NeilBrown4dbcdc72006-01-06 00:20:52 -08002583
NeilBrown014236d2006-01-06 00:20:55 -08002584static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002585slot_show(struct md_rdev *rdev, char *page)
NeilBrown014236d2006-01-06 00:20:55 -08002586{
2587 if (rdev->raid_disk < 0)
2588 return sprintf(page, "none\n");
2589 else
2590 return sprintf(page, "%d\n", rdev->raid_disk);
2591}
2592
2593static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002594slot_store(struct md_rdev *rdev, const char *buf, size_t len)
NeilBrown014236d2006-01-06 00:20:55 -08002595{
2596 char *e;
NeilBrownc303da62008-02-06 01:39:51 -08002597 int err;
NeilBrown014236d2006-01-06 00:20:55 -08002598 int slot = simple_strtoul(buf, &e, 10);
2599 if (strncmp(buf, "none", 4)==0)
2600 slot = -1;
2601 else if (e==buf || (*e && *e!= '\n'))
2602 return -EINVAL;
Neil Brown6c2fce22008-06-28 08:31:31 +10002603 if (rdev->mddev->pers && slot == -1) {
NeilBrownc303da62008-02-06 01:39:51 -08002604 /* Setting 'slot' on an active array requires also
2605 * updating the 'rd%d' link, and communicating
2606 * with the personality with ->hot_*_disk.
2607 * For now we only support removing
2608 * failed/spare devices. This normally happens automatically,
2609 * but not when the metadata is externally managed.
2610 */
NeilBrownc303da62008-02-06 01:39:51 -08002611 if (rdev->raid_disk == -1)
2612 return -EEXIST;
2613 /* personality does all needed checks */
Namhyung Kim01393f32011-06-09 11:42:54 +10002614 if (rdev->mddev->pers->hot_remove_disk == NULL)
NeilBrownc303da62008-02-06 01:39:51 -08002615 return -EINVAL;
NeilBrown746d3202013-04-24 11:42:41 +10002616 clear_bit(Blocked, &rdev->flags);
2617 remove_and_add_spares(rdev->mddev, rdev);
2618 if (rdev->raid_disk >= 0)
2619 return -EBUSY;
NeilBrownc303da62008-02-06 01:39:51 -08002620 set_bit(MD_RECOVERY_NEEDED, &rdev->mddev->recovery);
2621 md_wakeup_thread(rdev->mddev->thread);
Neil Brown6c2fce22008-06-28 08:31:31 +10002622 } else if (rdev->mddev->pers) {
Neil Brown6c2fce22008-06-28 08:31:31 +10002623 /* Activating a spare .. or possibly reactivating
NeilBrown6d56e272009-04-14 12:01:57 +10002624 * if we ever get bitmaps working here.
Neil Brown6c2fce22008-06-28 08:31:31 +10002625 */
2626
2627 if (rdev->raid_disk != -1)
2628 return -EBUSY;
2629
NeilBrownc6751b22011-02-02 11:57:13 +11002630 if (test_bit(MD_RECOVERY_RUNNING, &rdev->mddev->recovery))
2631 return -EBUSY;
2632
Neil Brown6c2fce22008-06-28 08:31:31 +10002633 if (rdev->mddev->pers->hot_add_disk == NULL)
2634 return -EINVAL;
2635
NeilBrownba1b41b2011-01-14 09:14:34 +11002636 if (slot >= rdev->mddev->raid_disks &&
2637 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
2638 return -ENOSPC;
2639
Neil Brown6c2fce22008-06-28 08:31:31 +10002640 rdev->raid_disk = slot;
2641 if (test_bit(In_sync, &rdev->flags))
2642 rdev->saved_raid_disk = slot;
2643 else
2644 rdev->saved_raid_disk = -1;
NeilBrownd30519f2011-10-18 12:13:47 +11002645 clear_bit(In_sync, &rdev->flags);
NeilBrown8313b8e2013-12-12 10:13:33 +11002646 clear_bit(Bitmap_sync, &rdev->flags);
Neil Brown6c2fce22008-06-28 08:31:31 +10002647 err = rdev->mddev->pers->
2648 hot_add_disk(rdev->mddev, rdev);
Neil Brown199050e2008-06-28 08:31:33 +10002649 if (err) {
Neil Brown6c2fce22008-06-28 08:31:31 +10002650 rdev->raid_disk = -1;
Neil Brown6c2fce22008-06-28 08:31:31 +10002651 return err;
Neil Brown52664732008-06-28 08:31:44 +10002652 } else
NeilBrown00bcb4a2010-06-01 19:37:23 +10002653 sysfs_notify_dirent_safe(rdev->sysfs_state);
Namhyung Kim36fad852011-07-27 11:00:36 +10002654 if (sysfs_link_rdev(rdev->mddev, rdev))
NeilBrown00bcb4a2010-06-01 19:37:23 +10002655 /* failure here is OK */;
Neil Brown6c2fce22008-06-28 08:31:31 +10002656 /* don't wakeup anyone, leave that to userspace. */
NeilBrownc303da62008-02-06 01:39:51 -08002657 } else {
NeilBrownba1b41b2011-01-14 09:14:34 +11002658 if (slot >= rdev->mddev->raid_disks &&
2659 slot >= rdev->mddev->raid_disks + rdev->mddev->delta_disks)
NeilBrownc303da62008-02-06 01:39:51 -08002660 return -ENOSPC;
2661 rdev->raid_disk = slot;
2662 /* assume it is working */
NeilBrownc5d79ad2008-02-06 01:39:54 -08002663 clear_bit(Faulty, &rdev->flags);
2664 clear_bit(WriteMostly, &rdev->flags);
NeilBrownc303da62008-02-06 01:39:51 -08002665 set_bit(In_sync, &rdev->flags);
NeilBrown00bcb4a2010-06-01 19:37:23 +10002666 sysfs_notify_dirent_safe(rdev->sysfs_state);
NeilBrownc303da62008-02-06 01:39:51 -08002667 }
NeilBrown014236d2006-01-06 00:20:55 -08002668 return len;
2669}
2670
NeilBrown014236d2006-01-06 00:20:55 -08002671static struct rdev_sysfs_entry rdev_slot =
NeilBrown80ca3a42006-07-10 04:44:18 -07002672__ATTR(slot, S_IRUGO|S_IWUSR, slot_show, slot_store);
NeilBrown014236d2006-01-06 00:20:55 -08002673
NeilBrown93c8cad2006-01-06 00:20:56 -08002674static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002675offset_show(struct md_rdev *rdev, char *page)
NeilBrown93c8cad2006-01-06 00:20:56 -08002676{
Andrew Morton6961ece2006-01-06 00:20:59 -08002677 return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
NeilBrown93c8cad2006-01-06 00:20:56 -08002678}
2679
2680static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002681offset_store(struct md_rdev *rdev, const char *buf, size_t len)
NeilBrown93c8cad2006-01-06 00:20:56 -08002682{
NeilBrownc6563a82012-05-21 09:27:00 +10002683 unsigned long long offset;
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002684 if (kstrtoull(buf, 10, &offset) < 0)
NeilBrown93c8cad2006-01-06 00:20:56 -08002685 return -EINVAL;
Neil Brown8ed0a522008-06-28 08:31:29 +10002686 if (rdev->mddev->pers && rdev->raid_disk >= 0)
NeilBrown93c8cad2006-01-06 00:20:56 -08002687 return -EBUSY;
Andre Nolldd8ac332009-03-31 14:33:13 +11002688 if (rdev->sectors && rdev->mddev->external)
NeilBrownc5d79ad2008-02-06 01:39:54 -08002689 /* Must set offset before size, so overlap checks
2690 * can be sane */
2691 return -EBUSY;
NeilBrown93c8cad2006-01-06 00:20:56 -08002692 rdev->data_offset = offset;
NeilBrown25f7fd42012-07-19 15:59:18 +10002693 rdev->new_data_offset = offset;
NeilBrown93c8cad2006-01-06 00:20:56 -08002694 return len;
2695}
2696
2697static struct rdev_sysfs_entry rdev_offset =
NeilBrown80ca3a42006-07-10 04:44:18 -07002698__ATTR(offset, S_IRUGO|S_IWUSR, offset_show, offset_store);
NeilBrown93c8cad2006-01-06 00:20:56 -08002699
NeilBrownc6563a82012-05-21 09:27:00 +10002700static ssize_t new_offset_show(struct md_rdev *rdev, char *page)
2701{
2702 return sprintf(page, "%llu\n",
2703 (unsigned long long)rdev->new_data_offset);
2704}
2705
2706static ssize_t new_offset_store(struct md_rdev *rdev,
2707 const char *buf, size_t len)
2708{
2709 unsigned long long new_offset;
2710 struct mddev *mddev = rdev->mddev;
2711
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002712 if (kstrtoull(buf, 10, &new_offset) < 0)
NeilBrownc6563a82012-05-21 09:27:00 +10002713 return -EINVAL;
2714
NeilBrownf851b602014-12-11 10:02:10 +11002715 if (mddev->sync_thread ||
2716 test_bit(MD_RECOVERY_RUNNING,&mddev->recovery))
NeilBrownc6563a82012-05-21 09:27:00 +10002717 return -EBUSY;
2718 if (new_offset == rdev->data_offset)
2719 /* reset is always permitted */
2720 ;
2721 else if (new_offset > rdev->data_offset) {
2722 /* must not push array size beyond rdev_sectors */
2723 if (new_offset - rdev->data_offset
2724 + mddev->dev_sectors > rdev->sectors)
2725 return -E2BIG;
2726 }
2727 /* Metadata worries about other space details. */
2728
2729 /* decreasing the offset is inconsistent with a backwards
2730 * reshape.
2731 */
2732 if (new_offset < rdev->data_offset &&
2733 mddev->reshape_backwards)
2734 return -EINVAL;
2735 /* Increasing offset is inconsistent with forwards
2736 * reshape. reshape_direction should be set to
2737 * 'backwards' first.
2738 */
2739 if (new_offset > rdev->data_offset &&
2740 !mddev->reshape_backwards)
2741 return -EINVAL;
2742
2743 if (mddev->pers && mddev->persistent &&
2744 !super_types[mddev->major_version]
2745 .allow_new_offset(rdev, new_offset))
2746 return -E2BIG;
2747 rdev->new_data_offset = new_offset;
2748 if (new_offset > rdev->data_offset)
2749 mddev->reshape_backwards = 1;
2750 else if (new_offset < rdev->data_offset)
2751 mddev->reshape_backwards = 0;
2752
2753 return len;
2754}
2755static struct rdev_sysfs_entry rdev_new_offset =
2756__ATTR(new_offset, S_IRUGO|S_IWUSR, new_offset_show, new_offset_store);
2757
NeilBrown83303b62006-01-06 00:21:06 -08002758static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002759rdev_size_show(struct md_rdev *rdev, char *page)
NeilBrown83303b62006-01-06 00:21:06 -08002760{
Andre Nolldd8ac332009-03-31 14:33:13 +11002761 return sprintf(page, "%llu\n", (unsigned long long)rdev->sectors / 2);
NeilBrown83303b62006-01-06 00:21:06 -08002762}
2763
NeilBrownc5d79ad2008-02-06 01:39:54 -08002764static int overlaps(sector_t s1, sector_t l1, sector_t s2, sector_t l2)
2765{
2766 /* check if two start/length pairs overlap */
2767 if (s1+l1 <= s2)
2768 return 0;
2769 if (s2+l2 <= s1)
2770 return 0;
2771 return 1;
2772}
2773
Dan Williamsb522adc2009-03-31 15:00:31 +11002774static int strict_blocks_to_sectors(const char *buf, sector_t *sectors)
2775{
2776 unsigned long long blocks;
2777 sector_t new;
2778
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002779 if (kstrtoull(buf, 10, &blocks) < 0)
Dan Williamsb522adc2009-03-31 15:00:31 +11002780 return -EINVAL;
2781
2782 if (blocks & 1ULL << (8 * sizeof(blocks) - 1))
2783 return -EINVAL; /* sector conversion overflow */
2784
2785 new = blocks * 2;
2786 if (new != blocks * 2)
2787 return -EINVAL; /* unsigned long long to sector_t overflow */
2788
2789 *sectors = new;
2790 return 0;
2791}
2792
NeilBrown83303b62006-01-06 00:21:06 -08002793static ssize_t
NeilBrown3cb03002011-10-11 16:45:26 +11002794rdev_size_store(struct md_rdev *rdev, const char *buf, size_t len)
NeilBrown83303b62006-01-06 00:21:06 -08002795{
NeilBrownfd01b882011-10-11 16:47:53 +11002796 struct mddev *my_mddev = rdev->mddev;
Andre Nolldd8ac332009-03-31 14:33:13 +11002797 sector_t oldsectors = rdev->sectors;
Dan Williamsb522adc2009-03-31 15:00:31 +11002798 sector_t sectors;
NeilBrown27c529b2008-03-04 14:29:33 -08002799
Dan Williamsb522adc2009-03-31 15:00:31 +11002800 if (strict_blocks_to_sectors(buf, &sectors) < 0)
Neil Brownd7027452008-07-12 10:37:50 +10002801 return -EINVAL;
NeilBrownc6563a82012-05-21 09:27:00 +10002802 if (rdev->data_offset != rdev->new_data_offset)
2803 return -EINVAL; /* too confusing */
Chris Webb0cd17fe2008-06-28 08:31:46 +10002804 if (my_mddev->pers && rdev->raid_disk >= 0) {
Neil Brownd7027452008-07-12 10:37:50 +10002805 if (my_mddev->persistent) {
Andre Nolldd8ac332009-03-31 14:33:13 +11002806 sectors = super_types[my_mddev->major_version].
2807 rdev_size_change(rdev, sectors);
2808 if (!sectors)
Chris Webb0cd17fe2008-06-28 08:31:46 +10002809 return -EBUSY;
Andre Nolldd8ac332009-03-31 14:33:13 +11002810 } else if (!sectors)
Mike Snitzer77304d22010-11-08 14:39:12 +01002811 sectors = (i_size_read(rdev->bdev->bd_inode) >> 9) -
Andre Nolldd8ac332009-03-31 14:33:13 +11002812 rdev->data_offset;
NeilBrowna6468532013-02-21 14:33:17 +11002813 if (!my_mddev->pers->resize)
2814 /* Cannot change size for RAID0 or Linear etc */
2815 return -EINVAL;
Chris Webb0cd17fe2008-06-28 08:31:46 +10002816 }
Andre Nolldd8ac332009-03-31 14:33:13 +11002817 if (sectors < my_mddev->dev_sectors)
Chris Webb7d3c6f82008-10-13 11:55:11 +11002818 return -EINVAL; /* component must fit device */
Chris Webb0cd17fe2008-06-28 08:31:46 +10002819
Andre Nolldd8ac332009-03-31 14:33:13 +11002820 rdev->sectors = sectors;
2821 if (sectors > oldsectors && my_mddev->external) {
NeilBrown8b1afc32014-09-29 15:33:20 +10002822 /* Need to check that all other rdevs with the same
2823 * ->bdev do not overlap. 'rcu' is sufficient to walk
2824 * the rdev lists safely.
2825 * This check does not provide a hard guarantee, it
2826 * just helps avoid dangerous mistakes.
NeilBrownc5d79ad2008-02-06 01:39:54 -08002827 */
NeilBrownfd01b882011-10-11 16:47:53 +11002828 struct mddev *mddev;
NeilBrownc5d79ad2008-02-06 01:39:54 -08002829 int overlap = 0;
Cheng Renquan159ec1f2009-01-09 08:31:08 +11002830 struct list_head *tmp;
NeilBrownc5d79ad2008-02-06 01:39:54 -08002831
NeilBrown8b1afc32014-09-29 15:33:20 +10002832 rcu_read_lock();
NeilBrown29ac4aa2008-02-06 01:39:58 -08002833 for_each_mddev(mddev, tmp) {
NeilBrown3cb03002011-10-11 16:45:26 +11002834 struct md_rdev *rdev2;
NeilBrownc5d79ad2008-02-06 01:39:54 -08002835
NeilBrowndafb20f2012-03-19 12:46:39 +11002836 rdev_for_each(rdev2, mddev)
NeilBrownf21e9ff2011-01-31 12:10:09 +11002837 if (rdev->bdev == rdev2->bdev &&
2838 rdev != rdev2 &&
2839 overlaps(rdev->data_offset, rdev->sectors,
2840 rdev2->data_offset,
2841 rdev2->sectors)) {
NeilBrownc5d79ad2008-02-06 01:39:54 -08002842 overlap = 1;
2843 break;
2844 }
NeilBrownc5d79ad2008-02-06 01:39:54 -08002845 if (overlap) {
2846 mddev_put(mddev);
2847 break;
2848 }
2849 }
NeilBrown8b1afc32014-09-29 15:33:20 +10002850 rcu_read_unlock();
NeilBrownc5d79ad2008-02-06 01:39:54 -08002851 if (overlap) {
2852 /* Someone else could have slipped in a size
2853 * change here, but doing so is just silly.
Andre Nolldd8ac332009-03-31 14:33:13 +11002854 * We put oldsectors back because we *know* it is
NeilBrownc5d79ad2008-02-06 01:39:54 -08002855 * safe, and trust userspace not to race with
2856 * itself
2857 */
Andre Nolldd8ac332009-03-31 14:33:13 +11002858 rdev->sectors = oldsectors;
NeilBrownc5d79ad2008-02-06 01:39:54 -08002859 return -EBUSY;
2860 }
2861 }
NeilBrown83303b62006-01-06 00:21:06 -08002862 return len;
2863}
2864
2865static struct rdev_sysfs_entry rdev_size =
NeilBrown80ca3a42006-07-10 04:44:18 -07002866__ATTR(size, S_IRUGO|S_IWUSR, rdev_size_show, rdev_size_store);
NeilBrown83303b62006-01-06 00:21:06 -08002867
NeilBrown3cb03002011-10-11 16:45:26 +11002868static ssize_t recovery_start_show(struct md_rdev *rdev, char *page)
Dan Williams06e3c812009-12-12 21:17:12 -07002869{
2870 unsigned long long recovery_start = rdev->recovery_offset;
2871
2872 if (test_bit(In_sync, &rdev->flags) ||
2873 recovery_start == MaxSector)
2874 return sprintf(page, "none\n");
2875
2876 return sprintf(page, "%llu\n", recovery_start);
2877}
2878
NeilBrown3cb03002011-10-11 16:45:26 +11002879static ssize_t recovery_start_store(struct md_rdev *rdev, const char *buf, size_t len)
Dan Williams06e3c812009-12-12 21:17:12 -07002880{
2881 unsigned long long recovery_start;
2882
2883 if (cmd_match(buf, "none"))
2884 recovery_start = MaxSector;
Jingoo Hanb29bebd2013-06-01 16:15:16 +09002885 else if (kstrtoull(buf, 10, &recovery_start))
Dan Williams06e3c812009-12-12 21:17:12 -07002886 return -EINVAL;
2887
2888 if (rdev->mddev->pers &&
2889 rdev->raid_disk >= 0)
2890 return -EBUSY;
2891
2892 rdev->recovery_offset = recovery_start;
2893 if (recovery_start == MaxSector)
2894 set_bit(In_sync, &rdev->flags);
2895 else
2896 clear_bit(In_sync, &rdev->flags);
2897 return len;
2898}
2899
2900static struct rdev_sysfs_entry rdev_recovery_start =
2901__ATTR(recovery_start, S_IRUGO|S_IWUSR, recovery_start_show, recovery_start_store);
2902
NeilBrown16c791a2011-07-28 11:31:47 +10002903static ssize_t
2904badblocks_show(struct badblocks *bb, char *page, int unack);
2905static ssize_t
2906badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack);
2907
NeilBrown3cb03002011-10-11 16:45:26 +11002908static ssize_t bb_show(struct md_rdev *rdev, char *page)
NeilBrown16c791a2011-07-28 11:31:47 +10002909{
2910 return badblocks_show(&rdev->badblocks, page, 0);
2911}
NeilBrown3cb03002011-10-11 16:45:26 +11002912static ssize_t bb_store(struct md_rdev *rdev, const char *page, size_t len)
NeilBrown16c791a2011-07-28 11:31:47 +10002913{
NeilBrownde393cd2011-07-28 11:31:48 +10002914 int rv = badblocks_store(&rdev->badblocks, page, len, 0);
2915 /* Maybe that ack was all we needed */
2916 if (test_and_clear_bit(BlockedBadBlocks, &rdev->flags))
2917 wake_up(&rdev->blocked_wait);
2918 return rv;
NeilBrown16c791a2011-07-28 11:31:47 +10002919}
2920static struct rdev_sysfs_entry rdev_bad_blocks =
2921__ATTR(bad_blocks, S_IRUGO|S_IWUSR, bb_show, bb_store);
2922
NeilBrown3cb03002011-10-11 16:45:26 +11002923static ssize_t ubb_show(struct md_rdev *rdev, char *page)
NeilBrown16c791a2011-07-28 11:31:47 +10002924{
2925 return badblocks_show(&rdev->badblocks, page, 1);
2926}
NeilBrown3cb03002011-10-11 16:45:26 +11002927static ssize_t ubb_store(struct md_rdev *rdev, const char *page, size_t len)
NeilBrown16c791a2011-07-28 11:31:47 +10002928{
2929 return badblocks_store(&rdev->badblocks, page, len, 1);
2930}
2931static struct rdev_sysfs_entry rdev_unack_bad_blocks =
2932__ATTR(unacknowledged_bad_blocks, S_IRUGO|S_IWUSR, ubb_show, ubb_store);
2933
NeilBrown86e6ffd2005-11-08 21:39:24 -08002934static struct attribute *rdev_default_attrs[] = {
2935 &rdev_state.attr,
NeilBrown4dbcdc72006-01-06 00:20:52 -08002936 &rdev_errors.attr,
NeilBrown014236d2006-01-06 00:20:55 -08002937 &rdev_slot.attr,
NeilBrown93c8cad2006-01-06 00:20:56 -08002938 &rdev_offset.attr,
NeilBrownc6563a82012-05-21 09:27:00 +10002939 &rdev_new_offset.attr,
NeilBrown83303b62006-01-06 00:21:06 -08002940 &rdev_size.attr,
Dan Williams06e3c812009-12-12 21:17:12 -07002941 &rdev_recovery_start.attr,
NeilBrown16c791a2011-07-28 11:31:47 +10002942 &rdev_bad_blocks.attr,
2943 &rdev_unack_bad_blocks.attr,
NeilBrown86e6ffd2005-11-08 21:39:24 -08002944 NULL,
2945};
2946static ssize_t
2947rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2948{
2949 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
NeilBrown3cb03002011-10-11 16:45:26 +11002950 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002951
2952 if (!entry->show)
2953 return -EIO;
NeilBrown758bfc82014-12-15 12:56:59 +11002954 if (!rdev->mddev)
2955 return -EBUSY;
2956 return entry->show(rdev, page);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002957}
2958
2959static ssize_t
2960rdev_attr_store(struct kobject *kobj, struct attribute *attr,
2961 const char *page, size_t length)
2962{
2963 struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
NeilBrown3cb03002011-10-11 16:45:26 +11002964 struct md_rdev *rdev = container_of(kobj, struct md_rdev, kobj);
NeilBrown27c529b2008-03-04 14:29:33 -08002965 ssize_t rv;
NeilBrownfd01b882011-10-11 16:47:53 +11002966 struct mddev *mddev = rdev->mddev;
NeilBrown86e6ffd2005-11-08 21:39:24 -08002967
2968 if (!entry->store)
2969 return -EIO;
NeilBrown67463ac2006-07-10 04:44:19 -07002970 if (!capable(CAP_SYS_ADMIN))
2971 return -EACCES;
NeilBrown27c529b2008-03-04 14:29:33 -08002972 rv = mddev ? mddev_lock(mddev): -EBUSY;
NeilBrownca388052008-02-06 01:39:55 -08002973 if (!rv) {
NeilBrown27c529b2008-03-04 14:29:33 -08002974 if (rdev->mddev == NULL)
2975 rv = -EBUSY;
2976 else
2977 rv = entry->store(rdev, page, length);
Dan Williams6a518302008-04-30 00:52:28 -07002978 mddev_unlock(mddev);
NeilBrownca388052008-02-06 01:39:55 -08002979 }
2980 return rv;
NeilBrown86e6ffd2005-11-08 21:39:24 -08002981}
2982
2983static void rdev_free(struct kobject *ko)
2984{
NeilBrown3cb03002011-10-11 16:45:26 +11002985 struct md_rdev *rdev = container_of(ko, struct md_rdev, kobj);
NeilBrown86e6ffd2005-11-08 21:39:24 -08002986 kfree(rdev);
2987}
Emese Revfy52cf25d2010-01-19 02:58:23 +01002988static const struct sysfs_ops rdev_sysfs_ops = {
NeilBrown86e6ffd2005-11-08 21:39:24 -08002989 .show = rdev_attr_show,
2990 .store = rdev_attr_store,
2991};
2992static struct kobj_type rdev_ktype = {
2993 .release = rdev_free,
2994 .sysfs_ops = &rdev_sysfs_ops,
2995 .default_attrs = rdev_default_attrs,
2996};
2997
NeilBrown3cb03002011-10-11 16:45:26 +11002998int md_rdev_init(struct md_rdev *rdev)
NeilBrowne8bb9a82010-06-01 19:37:26 +10002999{
3000 rdev->desc_nr = -1;
3001 rdev->saved_raid_disk = -1;
3002 rdev->raid_disk = -1;
3003 rdev->flags = 0;
3004 rdev->data_offset = 0;
NeilBrownc6563a82012-05-21 09:27:00 +10003005 rdev->new_data_offset = 0;
NeilBrowne8bb9a82010-06-01 19:37:26 +10003006 rdev->sb_events = 0;
3007 rdev->last_read_error.tv_sec = 0;
3008 rdev->last_read_error.tv_nsec = 0;
NeilBrown2699b672011-07-28 11:31:47 +10003009 rdev->sb_loaded = 0;
3010 rdev->bb_page = NULL;
NeilBrowne8bb9a82010-06-01 19:37:26 +10003011 atomic_set(&rdev->nr_pending, 0);
3012 atomic_set(&rdev->read_errors, 0);
3013 atomic_set(&rdev->corrected_errors, 0);
3014
3015 INIT_LIST_HEAD(&rdev->same_set);
3016 init_waitqueue_head(&rdev->blocked_wait);
NeilBrown2230dfe2011-07-28 11:31:46 +10003017
3018 /* Add space to store bad block list.
3019 * This reserves the space even on arrays where it cannot
3020 * be used - I wonder if that matters
3021 */
3022 rdev->badblocks.count = 0;
NeilBrown486adf72013-04-24 11:42:44 +10003023 rdev->badblocks.shift = -1; /* disabled until explicitly enabled */
NeilBrown2230dfe2011-07-28 11:31:46 +10003024 rdev->badblocks.page = kmalloc(PAGE_SIZE, GFP_KERNEL);
3025 seqlock_init(&rdev->badblocks.lock);
3026 if (rdev->badblocks.page == NULL)
3027 return -ENOMEM;
3028
3029 return 0;
NeilBrowne8bb9a82010-06-01 19:37:26 +10003030}
3031EXPORT_SYMBOL_GPL(md_rdev_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032/*
3033 * Import a device. If 'super_format' >= 0, then sanity check the superblock
3034 *
3035 * mark the device faulty if:
3036 *
3037 * - the device is nonexistent (zero size)
3038 * - the device has no valid superblock
3039 *
3040 * a faulty rdev _never_ has rdev->sb set.
3041 */
NeilBrown3cb03002011-10-11 16:45:26 +11003042static struct md_rdev *md_import_device(dev_t newdev, int super_format, int super_minor)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003043{
3044 char b[BDEVNAME_SIZE];
3045 int err;
NeilBrown3cb03002011-10-11 16:45:26 +11003046 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047 sector_t size;
3048
NeilBrown9ffae0c2006-01-06 00:20:32 -08003049 rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003050 if (!rdev) {
3051 printk(KERN_ERR "md: could not alloc mem for new device!\n");
3052 return ERR_PTR(-ENOMEM);
3053 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054
NeilBrown2230dfe2011-07-28 11:31:46 +10003055 err = md_rdev_init(rdev);
3056 if (err)
3057 goto abort_free;
3058 err = alloc_disk_sb(rdev);
3059 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060 goto abort_free;
3061
NeilBrownc5d79ad2008-02-06 01:39:54 -08003062 err = lock_rdev(rdev, newdev, super_format == -2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 if (err)
3064 goto abort_free;
3065
Greg Kroah-Hartmanf9cb0742007-12-17 23:05:35 -07003066 kobject_init(&rdev->kobj, &rdev_ktype);
NeilBrown86e6ffd2005-11-08 21:39:24 -08003067
Mike Snitzer77304d22010-11-08 14:39:12 +01003068 size = i_size_read(rdev->bdev->bd_inode) >> BLOCK_SIZE_BITS;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 if (!size) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10003070 printk(KERN_WARNING
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 "md: %s has zero or unknown size, marking faulty!\n",
3072 bdevname(rdev->bdev,b));
3073 err = -EINVAL;
3074 goto abort_free;
3075 }
3076
3077 if (super_format >= 0) {
3078 err = super_types[super_format].
3079 load_super(rdev, NULL, super_minor);
3080 if (err == -EINVAL) {
NeilBrowndf968c42007-07-17 04:06:11 -07003081 printk(KERN_WARNING
3082 "md: %s does not have a valid v%d.%d "
3083 "superblock, not importing!\n",
3084 bdevname(rdev->bdev,b),
3085 super_format, super_minor);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 goto abort_free;
3087 }
3088 if (err < 0) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10003089 printk(KERN_WARNING
Linus Torvalds1da177e2005-04-16 15:20:36 -07003090 "md: could not read %s's sb, not importing!\n",
3091 bdevname(rdev->bdev,b));
3092 goto abort_free;
3093 }
3094 }
Dan Williams6bfe0b42008-04-30 00:52:32 -07003095
Linus Torvalds1da177e2005-04-16 15:20:36 -07003096 return rdev;
3097
3098abort_free:
NeilBrown2699b672011-07-28 11:31:47 +10003099 if (rdev->bdev)
3100 unlock_rdev(rdev);
NeilBrown545c8792012-05-22 13:54:30 +10003101 md_rdev_clear(rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 kfree(rdev);
3103 return ERR_PTR(err);
3104}
3105
3106/*
3107 * Check a full RAID array for plausibility
3108 */
3109
NeilBrownf72ffdd2014-09-30 14:23:59 +10003110static void analyze_sbs(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111{
3112 int i;
NeilBrown3cb03002011-10-11 16:45:26 +11003113 struct md_rdev *rdev, *freshest, *tmp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114 char b[BDEVNAME_SIZE];
3115
3116 freshest = NULL;
NeilBrowndafb20f2012-03-19 12:46:39 +11003117 rdev_for_each_safe(rdev, tmp, mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118 switch (super_types[mddev->major_version].
3119 load_super(rdev, freshest, mddev->minor_version)) {
3120 case 1:
3121 freshest = rdev;
3122 break;
3123 case 0:
3124 break;
3125 default:
3126 printk( KERN_ERR \
3127 "md: fatal superblock inconsistency in %s"
NeilBrownf72ffdd2014-09-30 14:23:59 +10003128 " -- removing from array\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129 bdevname(rdev->bdev,b));
3130 kick_rdev_from_array(rdev);
3131 }
3132
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133 super_types[mddev->major_version].
3134 validate_super(mddev, freshest);
3135
3136 i = 0;
NeilBrowndafb20f2012-03-19 12:46:39 +11003137 rdev_for_each_safe(rdev, tmp, mddev) {
NeilBrown233fca32010-04-14 17:02:09 +10003138 if (mddev->max_disks &&
3139 (rdev->desc_nr >= mddev->max_disks ||
3140 i > mddev->max_disks)) {
NeilBrownde01dfa2009-02-06 18:02:46 +11003141 printk(KERN_WARNING
3142 "md: %s: %s: only %d devices permitted\n",
3143 mdname(mddev), bdevname(rdev->bdev, b),
3144 mddev->max_disks);
3145 kick_rdev_from_array(rdev);
3146 continue;
3147 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148 if (rdev != freshest)
3149 if (super_types[mddev->major_version].
3150 validate_super(mddev, rdev)) {
3151 printk(KERN_WARNING "md: kicking non-fresh %s"
3152 " from array!\n",
3153 bdevname(rdev->bdev,b));
3154 kick_rdev_from_array(rdev);
3155 continue;
3156 }
3157 if (mddev->level == LEVEL_MULTIPATH) {
3158 rdev->desc_nr = i++;
3159 rdev->raid_disk = rdev->desc_nr;
NeilBrownb2d444d2005-11-08 21:39:31 -08003160 set_bit(In_sync, &rdev->flags);
NeilBrown5e5e3e72009-10-16 16:35:30 +11003161 } else if (rdev->raid_disk >= (mddev->raid_disks - min(0, mddev->delta_disks))) {
NeilBrowna778b732007-05-23 13:58:10 -07003162 rdev->raid_disk = -1;
3163 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 }
3165 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166}
3167
NeilBrown72e02072009-12-14 12:49:55 +11003168/* Read a fixed-point number.
3169 * Numbers in sysfs attributes should be in "standard" units where
3170 * possible, so time should be in seconds.
NeilBrownf72ffdd2014-09-30 14:23:59 +10003171 * However we internally use a a much smaller unit such as
NeilBrown72e02072009-12-14 12:49:55 +11003172 * milliseconds or jiffies.
3173 * This function takes a decimal number with a possible fractional
3174 * component, and produces an integer which is the result of
3175 * multiplying that number by 10^'scale'.
3176 * all without any floating-point arithmetic.
3177 */
3178int strict_strtoul_scaled(const char *cp, unsigned long *res, int scale)
3179{
3180 unsigned long result = 0;
3181 long decimals = -1;
3182 while (isdigit(*cp) || (*cp == '.' && decimals < 0)) {
3183 if (*cp == '.')
3184 decimals = 0;
3185 else if (decimals < scale) {
3186 unsigned int value;
3187 value = *cp - '0';
3188 result = result * 10 + value;
3189 if (decimals >= 0)
3190 decimals++;
3191 }
3192 cp++;
3193 }
3194 if (*cp == '\n')
3195 cp++;
3196 if (*cp)
3197 return -EINVAL;
3198 if (decimals < 0)
3199 decimals = 0;
3200 while (decimals < scale) {
3201 result *= 10;
3202 decimals ++;
3203 }
3204 *res = result;
3205 return 0;
3206}
3207
NeilBrown19052c02008-08-05 15:54:13 +10003208static void md_safemode_timeout(unsigned long data);
3209
NeilBrowneae17012005-11-08 21:39:23 -08003210static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003211safe_delay_show(struct mddev *mddev, char *page)
NeilBrown16f17b32006-06-26 00:27:37 -07003212{
3213 int msec = (mddev->safemode_delay*1000)/HZ;
3214 return sprintf(page, "%d.%03d\n", msec/1000, msec%1000);
3215}
3216static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003217safe_delay_store(struct mddev *mddev, const char *cbuf, size_t len)
NeilBrown16f17b32006-06-26 00:27:37 -07003218{
NeilBrown16f17b32006-06-26 00:27:37 -07003219 unsigned long msec;
Dan Williams97ce0a72008-09-24 22:48:19 -07003220
NeilBrown72e02072009-12-14 12:49:55 +11003221 if (strict_strtoul_scaled(cbuf, &msec, 3) < 0)
NeilBrown16f17b32006-06-26 00:27:37 -07003222 return -EINVAL;
NeilBrown16f17b32006-06-26 00:27:37 -07003223 if (msec == 0)
3224 mddev->safemode_delay = 0;
3225 else {
NeilBrown19052c02008-08-05 15:54:13 +10003226 unsigned long old_delay = mddev->safemode_delay;
NeilBrown1b30e662014-12-15 12:57:00 +11003227 unsigned long new_delay = (msec*HZ)/1000;
3228
3229 if (new_delay == 0)
3230 new_delay = 1;
3231 mddev->safemode_delay = new_delay;
3232 if (new_delay < old_delay || old_delay == 0)
3233 mod_timer(&mddev->safemode_timer, jiffies+1);
NeilBrown16f17b32006-06-26 00:27:37 -07003234 }
3235 return len;
3236}
3237static struct md_sysfs_entry md_safe_delay =
NeilBrown80ca3a42006-07-10 04:44:18 -07003238__ATTR(safe_mode_delay, S_IRUGO|S_IWUSR,safe_delay_show, safe_delay_store);
NeilBrown16f17b32006-06-26 00:27:37 -07003239
3240static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003241level_show(struct mddev *mddev, char *page)
NeilBrowneae17012005-11-08 21:39:23 -08003242{
NeilBrown36d091f2014-12-15 12:56:58 +11003243 struct md_personality *p;
3244 int ret;
3245 spin_lock(&mddev->lock);
3246 p = mddev->pers;
NeilBrownd9d166c2006-01-06 00:20:51 -08003247 if (p)
NeilBrown36d091f2014-12-15 12:56:58 +11003248 ret = sprintf(page, "%s\n", p->name);
NeilBrownd9d166c2006-01-06 00:20:51 -08003249 else if (mddev->clevel[0])
NeilBrown36d091f2014-12-15 12:56:58 +11003250 ret = sprintf(page, "%s\n", mddev->clevel);
NeilBrownd9d166c2006-01-06 00:20:51 -08003251 else if (mddev->level != LEVEL_NONE)
NeilBrown36d091f2014-12-15 12:56:58 +11003252 ret = sprintf(page, "%d\n", mddev->level);
NeilBrownd9d166c2006-01-06 00:20:51 -08003253 else
NeilBrown36d091f2014-12-15 12:56:58 +11003254 ret = 0;
3255 spin_unlock(&mddev->lock);
3256 return ret;
NeilBrowneae17012005-11-08 21:39:23 -08003257}
3258
NeilBrownd9d166c2006-01-06 00:20:51 -08003259static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003260level_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownd9d166c2006-01-06 00:20:51 -08003261{
Dan Williamsf2859af2010-05-02 10:04:16 -07003262 char clevel[16];
NeilBrown67918752014-12-15 12:57:01 +11003263 ssize_t rv;
3264 size_t slen = len;
NeilBrowndb721d32014-12-15 12:56:58 +11003265 struct md_personality *pers, *oldpers;
Dan Williamsf2859af2010-05-02 10:04:16 -07003266 long level;
NeilBrowndb721d32014-12-15 12:56:58 +11003267 void *priv, *oldpriv;
NeilBrown3cb03002011-10-11 16:45:26 +11003268 struct md_rdev *rdev;
NeilBrown245f46c2009-03-31 14:39:39 +11003269
NeilBrown67918752014-12-15 12:57:01 +11003270 if (slen == 0 || slen >= sizeof(clevel))
3271 return -EINVAL;
3272
3273 rv = mddev_lock(mddev);
3274 if (rv)
NeilBrown245f46c2009-03-31 14:39:39 +11003275 return rv;
NeilBrown67918752014-12-15 12:57:01 +11003276
3277 if (mddev->pers == NULL) {
3278 strncpy(mddev->clevel, buf, slen);
3279 if (mddev->clevel[slen-1] == '\n')
3280 slen--;
3281 mddev->clevel[slen] = 0;
3282 mddev->level = LEVEL_NONE;
3283 rv = len;
3284 goto out_unlock;
NeilBrown245f46c2009-03-31 14:39:39 +11003285 }
NeilBrown67918752014-12-15 12:57:01 +11003286 rv = -EROFS;
NeilBrownbd8839e2014-05-28 13:39:21 +10003287 if (mddev->ro)
NeilBrown67918752014-12-15 12:57:01 +11003288 goto out_unlock;
NeilBrown245f46c2009-03-31 14:39:39 +11003289
3290 /* request to change the personality. Need to ensure:
3291 * - array is not engaged in resync/recovery/reshape
3292 * - old personality can be suspended
3293 * - new personality will access other array.
3294 */
3295
NeilBrown67918752014-12-15 12:57:01 +11003296 rv = -EBUSY;
NeilBrownbb4f1e92010-08-08 21:18:03 +10003297 if (mddev->sync_thread ||
NeilBrownf851b602014-12-11 10:02:10 +11003298 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
NeilBrownbb4f1e92010-08-08 21:18:03 +10003299 mddev->reshape_position != MaxSector ||
3300 mddev->sysfs_active)
NeilBrown67918752014-12-15 12:57:01 +11003301 goto out_unlock;
NeilBrown245f46c2009-03-31 14:39:39 +11003302
NeilBrown67918752014-12-15 12:57:01 +11003303 rv = -EINVAL;
NeilBrown245f46c2009-03-31 14:39:39 +11003304 if (!mddev->pers->quiesce) {
3305 printk(KERN_WARNING "md: %s: %s does not support online personality change\n",
3306 mdname(mddev), mddev->pers->name);
NeilBrown67918752014-12-15 12:57:01 +11003307 goto out_unlock;
NeilBrown245f46c2009-03-31 14:39:39 +11003308 }
3309
3310 /* Now find the new personality */
NeilBrown67918752014-12-15 12:57:01 +11003311 strncpy(clevel, buf, slen);
3312 if (clevel[slen-1] == '\n')
3313 slen--;
3314 clevel[slen] = 0;
Jingoo Hanb29bebd2013-06-01 16:15:16 +09003315 if (kstrtol(clevel, 10, &level))
Dan Williamsf2859af2010-05-02 10:04:16 -07003316 level = LEVEL_NONE;
NeilBrown245f46c2009-03-31 14:39:39 +11003317
Dan Williamsf2859af2010-05-02 10:04:16 -07003318 if (request_module("md-%s", clevel) != 0)
3319 request_module("md-level-%s", clevel);
NeilBrown245f46c2009-03-31 14:39:39 +11003320 spin_lock(&pers_lock);
Dan Williamsf2859af2010-05-02 10:04:16 -07003321 pers = find_pers(level, clevel);
NeilBrown245f46c2009-03-31 14:39:39 +11003322 if (!pers || !try_module_get(pers->owner)) {
3323 spin_unlock(&pers_lock);
Dan Williamsf2859af2010-05-02 10:04:16 -07003324 printk(KERN_WARNING "md: personality %s not loaded\n", clevel);
NeilBrown67918752014-12-15 12:57:01 +11003325 rv = -EINVAL;
3326 goto out_unlock;
NeilBrown245f46c2009-03-31 14:39:39 +11003327 }
3328 spin_unlock(&pers_lock);
3329
3330 if (pers == mddev->pers) {
3331 /* Nothing to do! */
3332 module_put(pers->owner);
NeilBrown67918752014-12-15 12:57:01 +11003333 rv = len;
3334 goto out_unlock;
NeilBrown245f46c2009-03-31 14:39:39 +11003335 }
3336 if (!pers->takeover) {
3337 module_put(pers->owner);
3338 printk(KERN_WARNING "md: %s: %s does not support personality takeover\n",
Dan Williamsf2859af2010-05-02 10:04:16 -07003339 mdname(mddev), clevel);
NeilBrown67918752014-12-15 12:57:01 +11003340 rv = -EINVAL;
3341 goto out_unlock;
NeilBrown245f46c2009-03-31 14:39:39 +11003342 }
3343
NeilBrowndafb20f2012-03-19 12:46:39 +11003344 rdev_for_each(rdev, mddev)
NeilBrowne93f68a2010-06-15 09:36:03 +01003345 rdev->new_raid_disk = rdev->raid_disk;
3346
NeilBrown245f46c2009-03-31 14:39:39 +11003347 /* ->takeover must set new_* and/or delta_disks
3348 * if it succeeds, and may set them when it fails.
3349 */
3350 priv = pers->takeover(mddev);
3351 if (IS_ERR(priv)) {
3352 mddev->new_level = mddev->level;
3353 mddev->new_layout = mddev->layout;
Andre Noll664e7c42009-06-18 08:45:27 +10003354 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown245f46c2009-03-31 14:39:39 +11003355 mddev->raid_disks -= mddev->delta_disks;
3356 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10003357 mddev->reshape_backwards = 0;
NeilBrown245f46c2009-03-31 14:39:39 +11003358 module_put(pers->owner);
3359 printk(KERN_WARNING "md: %s: %s would not accept array\n",
Dan Williamsf2859af2010-05-02 10:04:16 -07003360 mdname(mddev), clevel);
NeilBrown67918752014-12-15 12:57:01 +11003361 rv = PTR_ERR(priv);
3362 goto out_unlock;
NeilBrown245f46c2009-03-31 14:39:39 +11003363 }
3364
3365 /* Looks like we have a winner */
3366 mddev_suspend(mddev);
NeilBrown5aa61f42014-12-15 12:56:57 +11003367 mddev_detach(mddev);
NeilBrown36d091f2014-12-15 12:56:58 +11003368
3369 spin_lock(&mddev->lock);
NeilBrowndb721d32014-12-15 12:56:58 +11003370 oldpers = mddev->pers;
3371 oldpriv = mddev->private;
3372 mddev->pers = pers;
3373 mddev->private = priv;
3374 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
3375 mddev->level = mddev->new_level;
3376 mddev->layout = mddev->new_layout;
3377 mddev->chunk_sectors = mddev->new_chunk_sectors;
3378 mddev->delta_disks = 0;
3379 mddev->reshape_backwards = 0;
3380 mddev->degraded = 0;
NeilBrown36d091f2014-12-15 12:56:58 +11003381 spin_unlock(&mddev->lock);
NeilBrownf72ffdd2014-09-30 14:23:59 +10003382
NeilBrowndb721d32014-12-15 12:56:58 +11003383 if (oldpers->sync_request == NULL &&
Trela Maciej54071b32010-03-08 16:02:42 +11003384 mddev->external) {
3385 /* We are converting from a no-redundancy array
3386 * to a redundancy array and metadata is managed
3387 * externally so we need to be sure that writes
3388 * won't block due to a need to transition
3389 * clean->dirty
3390 * until external management is started.
3391 */
3392 mddev->in_sync = 0;
3393 mddev->safemode_delay = 0;
3394 mddev->safemode = 0;
3395 }
3396
NeilBrowndb721d32014-12-15 12:56:58 +11003397 oldpers->free(mddev, oldpriv);
3398
3399 if (oldpers->sync_request == NULL &&
3400 pers->sync_request != NULL) {
3401 /* need to add the md_redundancy_group */
3402 if (sysfs_create_group(&mddev->kobj, &md_redundancy_group))
3403 printk(KERN_WARNING
3404 "md: cannot register extra attributes for %s\n",
3405 mdname(mddev));
3406 mddev->sysfs_action = sysfs_get_dirent(mddev->kobj.sd, "sync_action");
3407 }
3408 if (oldpers->sync_request != NULL &&
3409 pers->sync_request == NULL) {
3410 /* need to remove the md_redundancy_group */
3411 if (mddev->to_remove == NULL)
3412 mddev->to_remove = &md_redundancy_group;
3413 }
3414
NeilBrowndafb20f2012-03-19 12:46:39 +11003415 rdev_for_each(rdev, mddev) {
NeilBrowne93f68a2010-06-15 09:36:03 +01003416 if (rdev->raid_disk < 0)
3417 continue;
NeilBrownbf2cb0d2011-01-14 09:14:34 +11003418 if (rdev->new_raid_disk >= mddev->raid_disks)
NeilBrowne93f68a2010-06-15 09:36:03 +01003419 rdev->new_raid_disk = -1;
3420 if (rdev->new_raid_disk == rdev->raid_disk)
3421 continue;
Namhyung Kim36fad852011-07-27 11:00:36 +10003422 sysfs_unlink_rdev(mddev, rdev);
NeilBrowne93f68a2010-06-15 09:36:03 +01003423 }
NeilBrowndafb20f2012-03-19 12:46:39 +11003424 rdev_for_each(rdev, mddev) {
NeilBrowne93f68a2010-06-15 09:36:03 +01003425 if (rdev->raid_disk < 0)
3426 continue;
3427 if (rdev->new_raid_disk == rdev->raid_disk)
3428 continue;
3429 rdev->raid_disk = rdev->new_raid_disk;
3430 if (rdev->raid_disk < 0)
NeilBrown3a981b02009-08-03 10:59:55 +10003431 clear_bit(In_sync, &rdev->flags);
NeilBrowne93f68a2010-06-15 09:36:03 +01003432 else {
Namhyung Kim36fad852011-07-27 11:00:36 +10003433 if (sysfs_link_rdev(mddev, rdev))
3434 printk(KERN_WARNING "md: cannot register rd%d"
3435 " for %s after level change\n",
3436 rdev->raid_disk, mdname(mddev));
NeilBrown3a981b02009-08-03 10:59:55 +10003437 }
NeilBrowne93f68a2010-06-15 09:36:03 +01003438 }
3439
NeilBrowndb721d32014-12-15 12:56:58 +11003440 if (pers->sync_request == NULL) {
Trela, Maciej9af204c2010-03-08 16:02:44 +11003441 /* this is now an array without redundancy, so
3442 * it must always be in_sync
3443 */
3444 mddev->in_sync = 1;
3445 del_timer_sync(&mddev->safemode_timer);
3446 }
NeilBrown02e5f5c2013-11-14 15:16:15 +11003447 blk_set_stacking_limits(&mddev->queue->limits);
NeilBrown245f46c2009-03-31 14:39:39 +11003448 pers->run(mddev);
NeilBrown245f46c2009-03-31 14:39:39 +11003449 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Jonathan Brassow47525e52012-05-22 13:55:29 +10003450 mddev_resume(mddev);
NeilBrown830778a2014-01-14 15:17:03 +11003451 if (!mddev->thread)
3452 md_update_sb(mddev, 1);
Maciej Trela5cac78612010-04-14 17:17:39 +10003453 sysfs_notify(&mddev->kobj, NULL, "level");
Dan Williamsbb7f8d22010-05-01 18:14:57 -07003454 md_new_event(mddev);
NeilBrown67918752014-12-15 12:57:01 +11003455 rv = len;
3456out_unlock:
3457 mddev_unlock(mddev);
NeilBrownd9d166c2006-01-06 00:20:51 -08003458 return rv;
3459}
3460
3461static struct md_sysfs_entry md_level =
NeilBrown80ca3a42006-07-10 04:44:18 -07003462__ATTR(level, S_IRUGO|S_IWUSR, level_show, level_store);
NeilBrowneae17012005-11-08 21:39:23 -08003463
NeilBrownd4dbd022006-06-26 00:27:59 -07003464static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003465layout_show(struct mddev *mddev, char *page)
NeilBrownd4dbd022006-06-26 00:27:59 -07003466{
3467 /* just a number, not meaningful for all levels */
NeilBrown08a02ec2007-05-09 02:35:38 -07003468 if (mddev->reshape_position != MaxSector &&
3469 mddev->layout != mddev->new_layout)
3470 return sprintf(page, "%d (%d)\n",
3471 mddev->new_layout, mddev->layout);
NeilBrownd4dbd022006-06-26 00:27:59 -07003472 return sprintf(page, "%d\n", mddev->layout);
3473}
3474
3475static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003476layout_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownd4dbd022006-06-26 00:27:59 -07003477{
3478 char *e;
3479 unsigned long n = simple_strtoul(buf, &e, 10);
NeilBrown67918752014-12-15 12:57:01 +11003480 int err;
NeilBrownd4dbd022006-06-26 00:27:59 -07003481
3482 if (!*buf || (*e && *e != '\n'))
3483 return -EINVAL;
NeilBrown67918752014-12-15 12:57:01 +11003484 err = mddev_lock(mddev);
3485 if (err)
3486 return err;
NeilBrownd4dbd022006-06-26 00:27:59 -07003487
NeilBrownb3546032009-03-31 14:56:41 +11003488 if (mddev->pers) {
NeilBrown50ac1682009-06-18 08:47:55 +10003489 if (mddev->pers->check_reshape == NULL)
NeilBrown67918752014-12-15 12:57:01 +11003490 err = -EBUSY;
3491 else if (mddev->ro)
3492 err = -EROFS;
3493 else {
3494 mddev->new_layout = n;
3495 err = mddev->pers->check_reshape(mddev);
3496 if (err)
3497 mddev->new_layout = mddev->layout;
NeilBrown597a7112009-06-18 08:47:42 +10003498 }
NeilBrownb3546032009-03-31 14:56:41 +11003499 } else {
NeilBrown08a02ec2007-05-09 02:35:38 -07003500 mddev->new_layout = n;
NeilBrownb3546032009-03-31 14:56:41 +11003501 if (mddev->reshape_position == MaxSector)
3502 mddev->layout = n;
3503 }
NeilBrown67918752014-12-15 12:57:01 +11003504 mddev_unlock(mddev);
3505 return err ?: len;
NeilBrownd4dbd022006-06-26 00:27:59 -07003506}
3507static struct md_sysfs_entry md_layout =
NeilBrown80ca3a42006-07-10 04:44:18 -07003508__ATTR(layout, S_IRUGO|S_IWUSR, layout_show, layout_store);
NeilBrownd4dbd022006-06-26 00:27:59 -07003509
NeilBrowneae17012005-11-08 21:39:23 -08003510static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003511raid_disks_show(struct mddev *mddev, char *page)
NeilBrowneae17012005-11-08 21:39:23 -08003512{
NeilBrownbb636542005-11-08 21:39:45 -08003513 if (mddev->raid_disks == 0)
3514 return 0;
NeilBrown08a02ec2007-05-09 02:35:38 -07003515 if (mddev->reshape_position != MaxSector &&
3516 mddev->delta_disks != 0)
3517 return sprintf(page, "%d (%d)\n", mddev->raid_disks,
3518 mddev->raid_disks - mddev->delta_disks);
NeilBrowneae17012005-11-08 21:39:23 -08003519 return sprintf(page, "%d\n", mddev->raid_disks);
3520}
3521
NeilBrownfd01b882011-10-11 16:47:53 +11003522static int update_raid_disks(struct mddev *mddev, int raid_disks);
NeilBrownda943b992006-01-06 00:20:54 -08003523
3524static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003525raid_disks_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownda943b992006-01-06 00:20:54 -08003526{
NeilBrownda943b992006-01-06 00:20:54 -08003527 char *e;
NeilBrown67918752014-12-15 12:57:01 +11003528 int err;
NeilBrownda943b992006-01-06 00:20:54 -08003529 unsigned long n = simple_strtoul(buf, &e, 10);
3530
3531 if (!*buf || (*e && *e != '\n'))
3532 return -EINVAL;
3533
NeilBrown67918752014-12-15 12:57:01 +11003534 err = mddev_lock(mddev);
3535 if (err)
3536 return err;
NeilBrownda943b992006-01-06 00:20:54 -08003537 if (mddev->pers)
NeilBrown67918752014-12-15 12:57:01 +11003538 err = update_raid_disks(mddev, n);
NeilBrown08a02ec2007-05-09 02:35:38 -07003539 else if (mddev->reshape_position != MaxSector) {
NeilBrownc6563a82012-05-21 09:27:00 +10003540 struct md_rdev *rdev;
NeilBrown08a02ec2007-05-09 02:35:38 -07003541 int olddisks = mddev->raid_disks - mddev->delta_disks;
NeilBrownc6563a82012-05-21 09:27:00 +10003542
NeilBrown67918752014-12-15 12:57:01 +11003543 err = -EINVAL;
NeilBrownc6563a82012-05-21 09:27:00 +10003544 rdev_for_each(rdev, mddev) {
3545 if (olddisks < n &&
3546 rdev->data_offset < rdev->new_data_offset)
NeilBrown67918752014-12-15 12:57:01 +11003547 goto out_unlock;
NeilBrownc6563a82012-05-21 09:27:00 +10003548 if (olddisks > n &&
3549 rdev->data_offset > rdev->new_data_offset)
NeilBrown67918752014-12-15 12:57:01 +11003550 goto out_unlock;
NeilBrownc6563a82012-05-21 09:27:00 +10003551 }
NeilBrown67918752014-12-15 12:57:01 +11003552 err = 0;
NeilBrown08a02ec2007-05-09 02:35:38 -07003553 mddev->delta_disks = n - olddisks;
3554 mddev->raid_disks = n;
NeilBrown2c810cd2012-05-21 09:27:00 +10003555 mddev->reshape_backwards = (mddev->delta_disks < 0);
NeilBrown08a02ec2007-05-09 02:35:38 -07003556 } else
NeilBrownda943b992006-01-06 00:20:54 -08003557 mddev->raid_disks = n;
NeilBrown67918752014-12-15 12:57:01 +11003558out_unlock:
3559 mddev_unlock(mddev);
3560 return err ? err : len;
NeilBrownda943b992006-01-06 00:20:54 -08003561}
3562static struct md_sysfs_entry md_raid_disks =
NeilBrown80ca3a42006-07-10 04:44:18 -07003563__ATTR(raid_disks, S_IRUGO|S_IWUSR, raid_disks_show, raid_disks_store);
NeilBrowneae17012005-11-08 21:39:23 -08003564
NeilBrown24dd4692005-11-08 21:39:26 -08003565static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003566chunk_size_show(struct mddev *mddev, char *page)
NeilBrown3b343802006-01-06 00:20:47 -08003567{
NeilBrown08a02ec2007-05-09 02:35:38 -07003568 if (mddev->reshape_position != MaxSector &&
Andre Noll664e7c42009-06-18 08:45:27 +10003569 mddev->chunk_sectors != mddev->new_chunk_sectors)
3570 return sprintf(page, "%d (%d)\n",
3571 mddev->new_chunk_sectors << 9,
Andre Noll9d8f0362009-06-18 08:45:01 +10003572 mddev->chunk_sectors << 9);
3573 return sprintf(page, "%d\n", mddev->chunk_sectors << 9);
NeilBrown3b343802006-01-06 00:20:47 -08003574}
3575
3576static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003577chunk_size_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown3b343802006-01-06 00:20:47 -08003578{
NeilBrown67918752014-12-15 12:57:01 +11003579 int err;
NeilBrown3b343802006-01-06 00:20:47 -08003580 char *e;
3581 unsigned long n = simple_strtoul(buf, &e, 10);
3582
NeilBrown3b343802006-01-06 00:20:47 -08003583 if (!*buf || (*e && *e != '\n'))
3584 return -EINVAL;
3585
NeilBrown67918752014-12-15 12:57:01 +11003586 err = mddev_lock(mddev);
3587 if (err)
3588 return err;
NeilBrownb3546032009-03-31 14:56:41 +11003589 if (mddev->pers) {
NeilBrown50ac1682009-06-18 08:47:55 +10003590 if (mddev->pers->check_reshape == NULL)
NeilBrown67918752014-12-15 12:57:01 +11003591 err = -EBUSY;
3592 else if (mddev->ro)
3593 err = -EROFS;
3594 else {
3595 mddev->new_chunk_sectors = n >> 9;
3596 err = mddev->pers->check_reshape(mddev);
3597 if (err)
3598 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown597a7112009-06-18 08:47:42 +10003599 }
NeilBrownb3546032009-03-31 14:56:41 +11003600 } else {
Andre Noll664e7c42009-06-18 08:45:27 +10003601 mddev->new_chunk_sectors = n >> 9;
NeilBrownb3546032009-03-31 14:56:41 +11003602 if (mddev->reshape_position == MaxSector)
Andre Noll9d8f0362009-06-18 08:45:01 +10003603 mddev->chunk_sectors = n >> 9;
NeilBrownb3546032009-03-31 14:56:41 +11003604 }
NeilBrown67918752014-12-15 12:57:01 +11003605 mddev_unlock(mddev);
3606 return err ?: len;
NeilBrown3b343802006-01-06 00:20:47 -08003607}
3608static struct md_sysfs_entry md_chunk_size =
NeilBrown80ca3a42006-07-10 04:44:18 -07003609__ATTR(chunk_size, S_IRUGO|S_IWUSR, chunk_size_show, chunk_size_store);
NeilBrown3b343802006-01-06 00:20:47 -08003610
NeilBrowna94213b2006-06-26 00:28:00 -07003611static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003612resync_start_show(struct mddev *mddev, char *page)
NeilBrowna94213b2006-06-26 00:28:00 -07003613{
NeilBrownd1a7c502009-03-31 15:24:32 +11003614 if (mddev->recovery_cp == MaxSector)
3615 return sprintf(page, "none\n");
NeilBrowna94213b2006-06-26 00:28:00 -07003616 return sprintf(page, "%llu\n", (unsigned long long)mddev->recovery_cp);
3617}
3618
3619static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003620resync_start_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrowna94213b2006-06-26 00:28:00 -07003621{
NeilBrown67918752014-12-15 12:57:01 +11003622 int err;
NeilBrowna94213b2006-06-26 00:28:00 -07003623 char *e;
3624 unsigned long long n = simple_strtoull(buf, &e, 10);
3625
NeilBrown67918752014-12-15 12:57:01 +11003626 err = mddev_lock(mddev);
3627 if (err)
3628 return err;
NeilBrownb0986362011-05-11 15:52:21 +10003629 if (mddev->pers && !test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
NeilBrown67918752014-12-15 12:57:01 +11003630 err = -EBUSY;
3631 else if (cmd_match(buf, "none"))
Dan Williams06e3c812009-12-12 21:17:12 -07003632 n = MaxSector;
3633 else if (!*buf || (*e && *e != '\n'))
NeilBrown67918752014-12-15 12:57:01 +11003634 err = -EINVAL;
NeilBrowna94213b2006-06-26 00:28:00 -07003635
NeilBrown67918752014-12-15 12:57:01 +11003636 if (!err) {
3637 mddev->recovery_cp = n;
3638 if (mddev->pers)
3639 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3640 }
3641 mddev_unlock(mddev);
3642 return err ?: len;
NeilBrowna94213b2006-06-26 00:28:00 -07003643}
3644static struct md_sysfs_entry md_resync_start =
NeilBrown750f1992014-09-30 08:53:05 +10003645__ATTR_PREALLOC(resync_start, S_IRUGO|S_IWUSR,
3646 resync_start_show, resync_start_store);
NeilBrowna94213b2006-06-26 00:28:00 -07003647
NeilBrown9e653b62006-06-26 00:27:58 -07003648/*
3649 * The array state can be:
3650 *
3651 * clear
3652 * No devices, no size, no level
3653 * Equivalent to STOP_ARRAY ioctl
3654 * inactive
3655 * May have some settings, but array is not active
3656 * all IO results in error
3657 * When written, doesn't tear down array, but just stops it
3658 * suspended (not supported yet)
3659 * All IO requests will block. The array can be reconfigured.
Andre Noll910d8cb2008-03-25 21:00:53 +01003660 * Writing this, if accepted, will block until array is quiescent
NeilBrown9e653b62006-06-26 00:27:58 -07003661 * readonly
3662 * no resync can happen. no superblocks get written.
3663 * write requests fail
3664 * read-auto
3665 * like readonly, but behaves like 'clean' on a write request.
3666 *
3667 * clean - no pending writes, but otherwise active.
3668 * When written to inactive array, starts without resync
3669 * If a write request arrives then
3670 * if metadata is known, mark 'dirty' and switch to 'active'.
3671 * if not known, block and switch to write-pending
3672 * If written to an active array that has pending writes, then fails.
3673 * active
3674 * fully active: IO and resync can be happening.
3675 * When written to inactive array, starts with resync
3676 *
3677 * write-pending
3678 * clean, but writes are blocked waiting for 'active' to be written.
3679 *
3680 * active-idle
3681 * like active, but no writes have been seen for a while (100msec).
3682 *
3683 */
3684enum array_state { clear, inactive, suspended, readonly, read_auto, clean, active,
3685 write_pending, active_idle, bad_word};
Adrian Bunk05381952006-06-26 00:28:01 -07003686static char *array_states[] = {
NeilBrown9e653b62006-06-26 00:27:58 -07003687 "clear", "inactive", "suspended", "readonly", "read-auto", "clean", "active",
3688 "write-pending", "active-idle", NULL };
3689
3690static int match_word(const char *word, char **list)
3691{
3692 int n;
3693 for (n=0; list[n]; n++)
3694 if (cmd_match(word, list[n]))
3695 break;
3696 return n;
3697}
3698
3699static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003700array_state_show(struct mddev *mddev, char *page)
NeilBrown9e653b62006-06-26 00:27:58 -07003701{
3702 enum array_state st = inactive;
3703
3704 if (mddev->pers)
3705 switch(mddev->ro) {
3706 case 1:
3707 st = readonly;
3708 break;
3709 case 2:
3710 st = read_auto;
3711 break;
3712 case 0:
3713 if (mddev->in_sync)
3714 st = clean;
NeilBrown070dc6d2010-08-30 17:33:34 +10003715 else if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
NeilBrowne6910632008-02-06 01:39:51 -08003716 st = write_pending;
NeilBrown9e653b62006-06-26 00:27:58 -07003717 else if (mddev->safemode)
3718 st = active_idle;
3719 else
3720 st = active;
3721 }
3722 else {
3723 if (list_empty(&mddev->disks) &&
3724 mddev->raid_disks == 0 &&
Andre Noll58c0fed2009-03-31 14:33:13 +11003725 mddev->dev_sectors == 0)
NeilBrown9e653b62006-06-26 00:27:58 -07003726 st = clear;
3727 else
3728 st = inactive;
3729 }
3730 return sprintf(page, "%s\n", array_states[st]);
3731}
3732
NeilBrownf72ffdd2014-09-30 14:23:59 +10003733static int do_md_stop(struct mddev *mddev, int ro, struct block_device *bdev);
3734static int md_set_readonly(struct mddev *mddev, struct block_device *bdev);
3735static int do_md_run(struct mddev *mddev);
NeilBrownfd01b882011-10-11 16:47:53 +11003736static int restart_array(struct mddev *mddev);
NeilBrown9e653b62006-06-26 00:27:58 -07003737
3738static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003739array_state_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown9e653b62006-06-26 00:27:58 -07003740{
NeilBrown67918752014-12-15 12:57:01 +11003741 int err;
NeilBrown9e653b62006-06-26 00:27:58 -07003742 enum array_state st = match_word(buf, array_states);
NeilBrown67918752014-12-15 12:57:01 +11003743
3744 if (mddev->pers && (st == active || st == clean) && mddev->ro != 1) {
3745 /* don't take reconfig_mutex when toggling between
3746 * clean and active
3747 */
3748 spin_lock(&mddev->lock);
3749 if (st == active) {
3750 restart_array(mddev);
3751 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
3752 wake_up(&mddev->sb_wait);
3753 err = 0;
3754 } else /* st == clean */ {
3755 restart_array(mddev);
3756 if (atomic_read(&mddev->writes_pending) == 0) {
3757 if (mddev->in_sync == 0) {
3758 mddev->in_sync = 1;
3759 if (mddev->safemode == 1)
3760 mddev->safemode = 0;
3761 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
3762 }
3763 err = 0;
3764 } else
3765 err = -EBUSY;
3766 }
3767 spin_unlock(&mddev->lock);
3768 return err;
3769 }
3770 err = mddev_lock(mddev);
3771 if (err)
3772 return err;
3773 err = -EINVAL;
NeilBrown9e653b62006-06-26 00:27:58 -07003774 switch(st) {
3775 case bad_word:
3776 break;
3777 case clear:
3778 /* stopping an active array */
NeilBrowna05b7ea2012-07-19 15:59:18 +10003779 err = do_md_stop(mddev, 0, NULL);
NeilBrown9e653b62006-06-26 00:27:58 -07003780 break;
3781 case inactive:
3782 /* stopping an active array */
NeilBrown90cf1952012-07-31 10:04:55 +10003783 if (mddev->pers)
NeilBrowna05b7ea2012-07-19 15:59:18 +10003784 err = do_md_stop(mddev, 2, NULL);
NeilBrown90cf1952012-07-31 10:04:55 +10003785 else
NeilBrowne6910632008-02-06 01:39:51 -08003786 err = 0; /* already inactive */
NeilBrown9e653b62006-06-26 00:27:58 -07003787 break;
3788 case suspended:
3789 break; /* not supported yet */
3790 case readonly:
3791 if (mddev->pers)
NeilBrowna05b7ea2012-07-19 15:59:18 +10003792 err = md_set_readonly(mddev, NULL);
NeilBrown9e653b62006-06-26 00:27:58 -07003793 else {
3794 mddev->ro = 1;
NeilBrown648b6292008-04-30 00:52:30 -07003795 set_disk_ro(mddev->gendisk, 1);
NeilBrown9e653b62006-06-26 00:27:58 -07003796 err = do_md_run(mddev);
3797 }
3798 break;
3799 case read_auto:
NeilBrown9e653b62006-06-26 00:27:58 -07003800 if (mddev->pers) {
NeilBrown80268ee2008-10-13 11:55:12 +11003801 if (mddev->ro == 0)
NeilBrowna05b7ea2012-07-19 15:59:18 +10003802 err = md_set_readonly(mddev, NULL);
NeilBrown80268ee2008-10-13 11:55:12 +11003803 else if (mddev->ro == 1)
NeilBrown648b6292008-04-30 00:52:30 -07003804 err = restart_array(mddev);
3805 if (err == 0) {
3806 mddev->ro = 2;
3807 set_disk_ro(mddev->gendisk, 0);
3808 }
NeilBrown9e653b62006-06-26 00:27:58 -07003809 } else {
3810 mddev->ro = 2;
3811 err = do_md_run(mddev);
3812 }
3813 break;
3814 case clean:
3815 if (mddev->pers) {
3816 restart_array(mddev);
NeilBrown85572d72014-12-15 12:56:56 +11003817 spin_lock(&mddev->lock);
NeilBrown9e653b62006-06-26 00:27:58 -07003818 if (atomic_read(&mddev->writes_pending) == 0) {
NeilBrowne6910632008-02-06 01:39:51 -08003819 if (mddev->in_sync == 0) {
3820 mddev->in_sync = 1;
NeilBrown31a59e32008-04-30 00:52:30 -07003821 if (mddev->safemode == 1)
3822 mddev->safemode = 0;
NeilBrown070dc6d2010-08-30 17:33:34 +10003823 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrowne6910632008-02-06 01:39:51 -08003824 }
3825 err = 0;
3826 } else
3827 err = -EBUSY;
NeilBrown85572d72014-12-15 12:56:56 +11003828 spin_unlock(&mddev->lock);
NeilBrown5bf29592009-05-07 12:50:57 +10003829 } else
3830 err = -EINVAL;
NeilBrown9e653b62006-06-26 00:27:58 -07003831 break;
3832 case active:
3833 if (mddev->pers) {
3834 restart_array(mddev);
NeilBrown070dc6d2010-08-30 17:33:34 +10003835 clear_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrown9e653b62006-06-26 00:27:58 -07003836 wake_up(&mddev->sb_wait);
3837 err = 0;
3838 } else {
3839 mddev->ro = 0;
NeilBrown648b6292008-04-30 00:52:30 -07003840 set_disk_ro(mddev->gendisk, 0);
NeilBrown9e653b62006-06-26 00:27:58 -07003841 err = do_md_run(mddev);
3842 }
3843 break;
3844 case write_pending:
3845 case active_idle:
3846 /* these cannot be set */
3847 break;
3848 }
NeilBrown67918752014-12-15 12:57:01 +11003849
3850 if (!err) {
NeilBrown1d23f172011-12-08 15:49:12 +11003851 if (mddev->hold_active == UNTIL_IOCTL)
3852 mddev->hold_active = 0;
NeilBrown00bcb4a2010-06-01 19:37:23 +10003853 sysfs_notify_dirent_safe(mddev->sysfs_state);
Neil Brown0fd62b82008-06-28 08:31:36 +10003854 }
NeilBrown67918752014-12-15 12:57:01 +11003855 mddev_unlock(mddev);
3856 return err ?: len;
NeilBrown9e653b62006-06-26 00:27:58 -07003857}
NeilBrown80ca3a42006-07-10 04:44:18 -07003858static struct md_sysfs_entry md_array_state =
NeilBrown750f1992014-09-30 08:53:05 +10003859__ATTR_PREALLOC(array_state, S_IRUGO|S_IWUSR, array_state_show, array_state_store);
NeilBrown9e653b62006-06-26 00:27:58 -07003860
NeilBrown6d7ff732006-01-06 00:21:16 -08003861static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003862max_corrected_read_errors_show(struct mddev *mddev, char *page) {
Robert Becker1e509152009-12-14 12:49:58 +11003863 return sprintf(page, "%d\n",
3864 atomic_read(&mddev->max_corr_read_errors));
3865}
3866
3867static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003868max_corrected_read_errors_store(struct mddev *mddev, const char *buf, size_t len)
Robert Becker1e509152009-12-14 12:49:58 +11003869{
3870 char *e;
3871 unsigned long n = simple_strtoul(buf, &e, 10);
3872
3873 if (*buf && (*e == 0 || *e == '\n')) {
3874 atomic_set(&mddev->max_corr_read_errors, n);
3875 return len;
3876 }
3877 return -EINVAL;
3878}
3879
3880static struct md_sysfs_entry max_corr_read_errors =
3881__ATTR(max_read_errors, S_IRUGO|S_IWUSR, max_corrected_read_errors_show,
3882 max_corrected_read_errors_store);
3883
3884static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003885null_show(struct mddev *mddev, char *page)
NeilBrown6d7ff732006-01-06 00:21:16 -08003886{
3887 return -EINVAL;
3888}
3889
3890static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003891new_dev_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown6d7ff732006-01-06 00:21:16 -08003892{
3893 /* buf must be %d:%d\n? giving major and minor numbers */
3894 /* The new device is added to the array.
3895 * If the array has a persistent superblock, we read the
3896 * superblock to initialise info and check validity.
3897 * Otherwise, only checking done is that in bind_rdev_to_array,
3898 * which mainly checks size.
3899 */
3900 char *e;
3901 int major = simple_strtoul(buf, &e, 10);
3902 int minor;
3903 dev_t dev;
NeilBrown3cb03002011-10-11 16:45:26 +11003904 struct md_rdev *rdev;
NeilBrown6d7ff732006-01-06 00:21:16 -08003905 int err;
3906
3907 if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
3908 return -EINVAL;
3909 minor = simple_strtoul(e+1, &e, 10);
3910 if (*e && *e != '\n')
3911 return -EINVAL;
3912 dev = MKDEV(major, minor);
3913 if (major != MAJOR(dev) ||
3914 minor != MINOR(dev))
3915 return -EOVERFLOW;
3916
NeilBrown67918752014-12-15 12:57:01 +11003917 flush_workqueue(md_misc_wq);
3918
3919 err = mddev_lock(mddev);
3920 if (err)
3921 return err;
NeilBrown6d7ff732006-01-06 00:21:16 -08003922 if (mddev->persistent) {
3923 rdev = md_import_device(dev, mddev->major_version,
3924 mddev->minor_version);
3925 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
NeilBrown3cb03002011-10-11 16:45:26 +11003926 struct md_rdev *rdev0
3927 = list_entry(mddev->disks.next,
3928 struct md_rdev, same_set);
NeilBrown6d7ff732006-01-06 00:21:16 -08003929 err = super_types[mddev->major_version]
3930 .load_super(rdev, rdev0, mddev->minor_version);
3931 if (err < 0)
3932 goto out;
3933 }
NeilBrownc5d79ad2008-02-06 01:39:54 -08003934 } else if (mddev->external)
3935 rdev = md_import_device(dev, -2, -1);
3936 else
NeilBrown6d7ff732006-01-06 00:21:16 -08003937 rdev = md_import_device(dev, -1, -1);
3938
3939 if (IS_ERR(rdev))
3940 return PTR_ERR(rdev);
3941 err = bind_rdev_to_array(rdev, mddev);
3942 out:
3943 if (err)
3944 export_rdev(rdev);
NeilBrown67918752014-12-15 12:57:01 +11003945 mddev_unlock(mddev);
NeilBrown6d7ff732006-01-06 00:21:16 -08003946 return err ? err : len;
3947}
3948
3949static struct md_sysfs_entry md_new_device =
NeilBrown80ca3a42006-07-10 04:44:18 -07003950__ATTR(new_dev, S_IWUSR, null_show, new_dev_store);
NeilBrown3b343802006-01-06 00:20:47 -08003951
3952static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003953bitmap_store(struct mddev *mddev, const char *buf, size_t len)
Paul Clements9b1d1da2006-10-03 01:15:49 -07003954{
3955 char *end;
3956 unsigned long chunk, end_chunk;
NeilBrown67918752014-12-15 12:57:01 +11003957 int err;
Paul Clements9b1d1da2006-10-03 01:15:49 -07003958
NeilBrown67918752014-12-15 12:57:01 +11003959 err = mddev_lock(mddev);
3960 if (err)
3961 return err;
Paul Clements9b1d1da2006-10-03 01:15:49 -07003962 if (!mddev->bitmap)
3963 goto out;
3964 /* buf should be <chunk> <chunk> ... or <chunk>-<chunk> ... (range) */
3965 while (*buf) {
3966 chunk = end_chunk = simple_strtoul(buf, &end, 0);
3967 if (buf == end) break;
3968 if (*end == '-') { /* range */
3969 buf = end + 1;
3970 end_chunk = simple_strtoul(buf, &end, 0);
3971 if (buf == end) break;
3972 }
3973 if (*end && !isspace(*end)) break;
3974 bitmap_dirty_bits(mddev->bitmap, chunk, end_chunk);
André Goddard Rosae7d28602009-12-14 18:01:06 -08003975 buf = skip_spaces(end);
Paul Clements9b1d1da2006-10-03 01:15:49 -07003976 }
3977 bitmap_unplug(mddev->bitmap); /* flush the bits to disk */
3978out:
NeilBrown67918752014-12-15 12:57:01 +11003979 mddev_unlock(mddev);
Paul Clements9b1d1da2006-10-03 01:15:49 -07003980 return len;
3981}
3982
3983static struct md_sysfs_entry md_bitmap =
3984__ATTR(bitmap_set_bits, S_IWUSR, null_show, bitmap_store);
3985
3986static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003987size_show(struct mddev *mddev, char *page)
NeilBrowna35b0d692006-01-06 00:20:49 -08003988{
Andre Noll58c0fed2009-03-31 14:33:13 +11003989 return sprintf(page, "%llu\n",
3990 (unsigned long long)mddev->dev_sectors / 2);
NeilBrowna35b0d692006-01-06 00:20:49 -08003991}
3992
NeilBrownfd01b882011-10-11 16:47:53 +11003993static int update_size(struct mddev *mddev, sector_t num_sectors);
NeilBrowna35b0d692006-01-06 00:20:49 -08003994
3995static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11003996size_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrowna35b0d692006-01-06 00:20:49 -08003997{
3998 /* If array is inactive, we can reduce the component size, but
3999 * not increase it (except from 0).
4000 * If array is active, we can try an on-line resize
4001 */
Dan Williamsb522adc2009-03-31 15:00:31 +11004002 sector_t sectors;
4003 int err = strict_blocks_to_sectors(buf, &sectors);
NeilBrowna35b0d692006-01-06 00:20:49 -08004004
Andre Noll58c0fed2009-03-31 14:33:13 +11004005 if (err < 0)
4006 return err;
NeilBrown67918752014-12-15 12:57:01 +11004007 err = mddev_lock(mddev);
4008 if (err)
4009 return err;
NeilBrowna35b0d692006-01-06 00:20:49 -08004010 if (mddev->pers) {
Andre Noll58c0fed2009-03-31 14:33:13 +11004011 err = update_size(mddev, sectors);
NeilBrown850b2b42006-10-03 01:15:46 -07004012 md_update_sb(mddev, 1);
NeilBrowna35b0d692006-01-06 00:20:49 -08004013 } else {
Andre Noll58c0fed2009-03-31 14:33:13 +11004014 if (mddev->dev_sectors == 0 ||
4015 mddev->dev_sectors > sectors)
4016 mddev->dev_sectors = sectors;
NeilBrowna35b0d692006-01-06 00:20:49 -08004017 else
4018 err = -ENOSPC;
4019 }
NeilBrown67918752014-12-15 12:57:01 +11004020 mddev_unlock(mddev);
NeilBrowna35b0d692006-01-06 00:20:49 -08004021 return err ? err : len;
4022}
4023
4024static struct md_sysfs_entry md_size =
NeilBrown80ca3a42006-07-10 04:44:18 -07004025__ATTR(component_size, S_IRUGO|S_IWUSR, size_show, size_store);
NeilBrowna35b0d692006-01-06 00:20:49 -08004026
Masanari Iida83f0d772012-10-30 00:18:08 +09004027/* Metadata version.
NeilBrowne6910632008-02-06 01:39:51 -08004028 * This is one of
4029 * 'none' for arrays with no metadata (good luck...)
4030 * 'external' for arrays with externally managed metadata,
NeilBrown8bb93aa2006-01-06 00:20:50 -08004031 * or N.M for internally known formats
4032 */
4033static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004034metadata_show(struct mddev *mddev, char *page)
NeilBrown8bb93aa2006-01-06 00:20:50 -08004035{
4036 if (mddev->persistent)
4037 return sprintf(page, "%d.%d\n",
4038 mddev->major_version, mddev->minor_version);
NeilBrowne6910632008-02-06 01:39:51 -08004039 else if (mddev->external)
4040 return sprintf(page, "external:%s\n", mddev->metadata_type);
NeilBrown8bb93aa2006-01-06 00:20:50 -08004041 else
4042 return sprintf(page, "none\n");
4043}
4044
4045static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004046metadata_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown8bb93aa2006-01-06 00:20:50 -08004047{
4048 int major, minor;
4049 char *e;
NeilBrown67918752014-12-15 12:57:01 +11004050 int err;
NeilBrownea43ddd2008-10-13 11:55:11 +11004051 /* Changing the details of 'external' metadata is
4052 * always permitted. Otherwise there must be
4053 * no devices attached to the array.
4054 */
NeilBrown67918752014-12-15 12:57:01 +11004055
4056 err = mddev_lock(mddev);
4057 if (err)
4058 return err;
4059 err = -EBUSY;
NeilBrownea43ddd2008-10-13 11:55:11 +11004060 if (mddev->external && strncmp(buf, "external:", 9) == 0)
4061 ;
4062 else if (!list_empty(&mddev->disks))
NeilBrown67918752014-12-15 12:57:01 +11004063 goto out_unlock;
NeilBrown8bb93aa2006-01-06 00:20:50 -08004064
NeilBrown67918752014-12-15 12:57:01 +11004065 err = 0;
NeilBrown8bb93aa2006-01-06 00:20:50 -08004066 if (cmd_match(buf, "none")) {
4067 mddev->persistent = 0;
NeilBrowne6910632008-02-06 01:39:51 -08004068 mddev->external = 0;
4069 mddev->major_version = 0;
4070 mddev->minor_version = 90;
NeilBrown67918752014-12-15 12:57:01 +11004071 goto out_unlock;
NeilBrowne6910632008-02-06 01:39:51 -08004072 }
4073 if (strncmp(buf, "external:", 9) == 0) {
NeilBrown20a49ff2008-02-06 01:39:57 -08004074 size_t namelen = len-9;
NeilBrowne6910632008-02-06 01:39:51 -08004075 if (namelen >= sizeof(mddev->metadata_type))
4076 namelen = sizeof(mddev->metadata_type)-1;
4077 strncpy(mddev->metadata_type, buf+9, namelen);
4078 mddev->metadata_type[namelen] = 0;
4079 if (namelen && mddev->metadata_type[namelen-1] == '\n')
4080 mddev->metadata_type[--namelen] = 0;
4081 mddev->persistent = 0;
4082 mddev->external = 1;
NeilBrown8bb93aa2006-01-06 00:20:50 -08004083 mddev->major_version = 0;
4084 mddev->minor_version = 90;
NeilBrown67918752014-12-15 12:57:01 +11004085 goto out_unlock;
NeilBrown8bb93aa2006-01-06 00:20:50 -08004086 }
4087 major = simple_strtoul(buf, &e, 10);
NeilBrown67918752014-12-15 12:57:01 +11004088 err = -EINVAL;
NeilBrown8bb93aa2006-01-06 00:20:50 -08004089 if (e==buf || *e != '.')
NeilBrown67918752014-12-15 12:57:01 +11004090 goto out_unlock;
NeilBrown8bb93aa2006-01-06 00:20:50 -08004091 buf = e+1;
4092 minor = simple_strtoul(buf, &e, 10);
NeilBrown3f9d7b02006-12-22 01:11:41 -08004093 if (e==buf || (*e && *e != '\n') )
NeilBrown67918752014-12-15 12:57:01 +11004094 goto out_unlock;
4095 err = -ENOENT;
Ahmed S. Darwish50511da2007-05-09 02:35:34 -07004096 if (major >= ARRAY_SIZE(super_types) || super_types[major].name == NULL)
NeilBrown67918752014-12-15 12:57:01 +11004097 goto out_unlock;
NeilBrown8bb93aa2006-01-06 00:20:50 -08004098 mddev->major_version = major;
4099 mddev->minor_version = minor;
4100 mddev->persistent = 1;
NeilBrowne6910632008-02-06 01:39:51 -08004101 mddev->external = 0;
NeilBrown67918752014-12-15 12:57:01 +11004102 err = 0;
4103out_unlock:
4104 mddev_unlock(mddev);
4105 return err ?: len;
NeilBrown8bb93aa2006-01-06 00:20:50 -08004106}
4107
4108static struct md_sysfs_entry md_metadata =
NeilBrown750f1992014-09-30 08:53:05 +10004109__ATTR_PREALLOC(metadata_version, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
NeilBrown8bb93aa2006-01-06 00:20:50 -08004110
NeilBrowna35b0d692006-01-06 00:20:49 -08004111static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004112action_show(struct mddev *mddev, char *page)
NeilBrown24dd4692005-11-08 21:39:26 -08004113{
NeilBrown7eec3142005-11-08 21:39:44 -08004114 char *type = "idle";
NeilBrownb7b17c92014-12-15 12:56:59 +11004115 unsigned long recovery = mddev->recovery;
4116 if (test_bit(MD_RECOVERY_FROZEN, &recovery))
NeilBrownb6a9ce62009-05-26 09:41:17 +10004117 type = "frozen";
NeilBrownb7b17c92014-12-15 12:56:59 +11004118 else if (test_bit(MD_RECOVERY_RUNNING, &recovery) ||
4119 (!mddev->ro && test_bit(MD_RECOVERY_NEEDED, &recovery))) {
4120 if (test_bit(MD_RECOVERY_RESHAPE, &recovery))
NeilBrownccfcc3c2006-03-27 01:18:09 -08004121 type = "reshape";
NeilBrownb7b17c92014-12-15 12:56:59 +11004122 else if (test_bit(MD_RECOVERY_SYNC, &recovery)) {
4123 if (!test_bit(MD_RECOVERY_REQUESTED, &recovery))
NeilBrown24dd4692005-11-08 21:39:26 -08004124 type = "resync";
NeilBrownb7b17c92014-12-15 12:56:59 +11004125 else if (test_bit(MD_RECOVERY_CHECK, &recovery))
NeilBrown24dd4692005-11-08 21:39:26 -08004126 type = "check";
4127 else
4128 type = "repair";
NeilBrownb7b17c92014-12-15 12:56:59 +11004129 } else if (test_bit(MD_RECOVERY_RECOVER, &recovery))
NeilBrown24dd4692005-11-08 21:39:26 -08004130 type = "recover";
4131 }
4132 return sprintf(page, "%s\n", type);
4133}
4134
4135static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004136action_store(struct mddev *mddev, const char *page, size_t len)
NeilBrown24dd4692005-11-08 21:39:26 -08004137{
NeilBrown7eec3142005-11-08 21:39:44 -08004138 if (!mddev->pers || !mddev->pers->sync_request)
4139 return -EINVAL;
4140
NeilBrownb6a9ce62009-05-26 09:41:17 +10004141 if (cmd_match(page, "frozen"))
4142 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4143 else
4144 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
4145
4146 if (cmd_match(page, "idle") || cmd_match(page, "frozen")) {
NeilBrownf851b602014-12-11 10:02:10 +11004147 flush_workqueue(md_misc_wq);
NeilBrown7eec3142005-11-08 21:39:44 -08004148 if (mddev->sync_thread) {
4149 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrown67918752014-12-15 12:57:01 +11004150 if (mddev_lock(mddev) == 0) {
4151 md_reap_sync_thread(mddev);
4152 mddev_unlock(mddev);
4153 }
NeilBrown7eec3142005-11-08 21:39:44 -08004154 }
NeilBrown03c902e2006-01-06 00:20:46 -08004155 } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
4156 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
NeilBrown24dd4692005-11-08 21:39:26 -08004157 return -EBUSY;
Neil Brown72a23c22008-06-28 08:31:41 +10004158 else if (cmd_match(page, "resync"))
NeilBrown7eec3142005-11-08 21:39:44 -08004159 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
Neil Brown72a23c22008-06-28 08:31:41 +10004160 else if (cmd_match(page, "recover")) {
4161 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
4162 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4163 } else if (cmd_match(page, "reshape")) {
NeilBrown16484bf2006-03-27 01:18:13 -08004164 int err;
4165 if (mddev->pers->start_reshape == NULL)
4166 return -EINVAL;
NeilBrown67918752014-12-15 12:57:01 +11004167 err = mddev_lock(mddev);
4168 if (!err) {
4169 err = mddev->pers->start_reshape(mddev);
4170 mddev_unlock(mddev);
4171 }
NeilBrown16484bf2006-03-27 01:18:13 -08004172 if (err)
4173 return err;
Neil Browna99ac972008-06-28 08:31:43 +10004174 sysfs_notify(&mddev->kobj, NULL, "degraded");
NeilBrown16484bf2006-03-27 01:18:13 -08004175 } else {
NeilBrownbce74da2006-01-06 00:20:41 -08004176 if (cmd_match(page, "check"))
NeilBrown7eec3142005-11-08 21:39:44 -08004177 set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
NeilBrown2adc7d42006-05-20 14:59:57 -07004178 else if (!cmd_match(page, "repair"))
NeilBrown7eec3142005-11-08 21:39:44 -08004179 return -EINVAL;
4180 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
4181 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
NeilBrown7eec3142005-11-08 21:39:44 -08004182 }
NeilBrown48c26dd2012-10-11 14:19:39 +11004183 if (mddev->ro == 2) {
4184 /* A write to sync_action is enough to justify
4185 * canceling read-auto mode
4186 */
4187 mddev->ro = 0;
4188 md_wakeup_thread(mddev->sync_thread);
4189 }
NeilBrown03c902e2006-01-06 00:20:46 -08004190 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrown24dd4692005-11-08 21:39:26 -08004191 md_wakeup_thread(mddev->thread);
NeilBrown00bcb4a2010-06-01 19:37:23 +10004192 sysfs_notify_dirent_safe(mddev->sysfs_action);
NeilBrown24dd4692005-11-08 21:39:26 -08004193 return len;
4194}
4195
Jonathan Brassowc4a39552013-06-25 01:23:59 -05004196static struct md_sysfs_entry md_scan_mode =
NeilBrown750f1992014-09-30 08:53:05 +10004197__ATTR_PREALLOC(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
Jonathan Brassowc4a39552013-06-25 01:23:59 -05004198
4199static ssize_t
4200last_sync_action_show(struct mddev *mddev, char *page)
4201{
4202 return sprintf(page, "%s\n", mddev->last_sync_action);
4203}
4204
4205static struct md_sysfs_entry md_last_scan_mode = __ATTR_RO(last_sync_action);
4206
NeilBrown9d888832005-11-08 21:39:26 -08004207static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004208mismatch_cnt_show(struct mddev *mddev, char *page)
NeilBrown9d888832005-11-08 21:39:26 -08004209{
4210 return sprintf(page, "%llu\n",
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11004211 (unsigned long long)
4212 atomic64_read(&mddev->resync_mismatches));
NeilBrown9d888832005-11-08 21:39:26 -08004213}
4214
NeilBrown80ca3a42006-07-10 04:44:18 -07004215static struct md_sysfs_entry md_mismatches = __ATTR_RO(mismatch_cnt);
NeilBrown9d888832005-11-08 21:39:26 -08004216
NeilBrown88202a02006-01-06 00:21:36 -08004217static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004218sync_min_show(struct mddev *mddev, char *page)
NeilBrown88202a02006-01-06 00:21:36 -08004219{
4220 return sprintf(page, "%d (%s)\n", speed_min(mddev),
4221 mddev->sync_speed_min ? "local": "system");
4222}
4223
4224static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004225sync_min_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown88202a02006-01-06 00:21:36 -08004226{
4227 int min;
4228 char *e;
4229 if (strncmp(buf, "system", 6)==0) {
4230 mddev->sync_speed_min = 0;
4231 return len;
4232 }
4233 min = simple_strtoul(buf, &e, 10);
4234 if (buf == e || (*e && *e != '\n') || min <= 0)
4235 return -EINVAL;
4236 mddev->sync_speed_min = min;
4237 return len;
4238}
4239
4240static struct md_sysfs_entry md_sync_min =
4241__ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
4242
4243static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004244sync_max_show(struct mddev *mddev, char *page)
NeilBrown88202a02006-01-06 00:21:36 -08004245{
4246 return sprintf(page, "%d (%s)\n", speed_max(mddev),
4247 mddev->sync_speed_max ? "local": "system");
4248}
4249
4250static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004251sync_max_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown88202a02006-01-06 00:21:36 -08004252{
4253 int max;
4254 char *e;
4255 if (strncmp(buf, "system", 6)==0) {
4256 mddev->sync_speed_max = 0;
4257 return len;
4258 }
4259 max = simple_strtoul(buf, &e, 10);
4260 if (buf == e || (*e && *e != '\n') || max <= 0)
4261 return -EINVAL;
4262 mddev->sync_speed_max = max;
4263 return len;
4264}
4265
4266static struct md_sysfs_entry md_sync_max =
4267__ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
4268
Iustin Popd7f3d292007-10-16 23:30:54 -07004269static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004270degraded_show(struct mddev *mddev, char *page)
Iustin Popd7f3d292007-10-16 23:30:54 -07004271{
4272 return sprintf(page, "%d\n", mddev->degraded);
4273}
4274static struct md_sysfs_entry md_degraded = __ATTR_RO(degraded);
NeilBrown88202a02006-01-06 00:21:36 -08004275
4276static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004277sync_force_parallel_show(struct mddev *mddev, char *page)
Bernd Schubert90b08712008-05-23 13:04:38 -07004278{
4279 return sprintf(page, "%d\n", mddev->parallel_resync);
4280}
4281
4282static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004283sync_force_parallel_store(struct mddev *mddev, const char *buf, size_t len)
Bernd Schubert90b08712008-05-23 13:04:38 -07004284{
4285 long n;
4286
Jingoo Hanb29bebd2013-06-01 16:15:16 +09004287 if (kstrtol(buf, 10, &n))
Bernd Schubert90b08712008-05-23 13:04:38 -07004288 return -EINVAL;
4289
4290 if (n != 0 && n != 1)
4291 return -EINVAL;
4292
4293 mddev->parallel_resync = n;
4294
4295 if (mddev->sync_thread)
4296 wake_up(&resync_wait);
4297
4298 return len;
4299}
4300
4301/* force parallel resync, even with shared block devices */
4302static struct md_sysfs_entry md_sync_force_parallel =
4303__ATTR(sync_force_parallel, S_IRUGO|S_IWUSR,
4304 sync_force_parallel_show, sync_force_parallel_store);
4305
4306static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004307sync_speed_show(struct mddev *mddev, char *page)
NeilBrown88202a02006-01-06 00:21:36 -08004308{
4309 unsigned long resync, dt, db;
NeilBrownd1a7c502009-03-31 15:24:32 +11004310 if (mddev->curr_resync == 0)
4311 return sprintf(page, "none\n");
Andre Noll9687a602008-03-25 22:24:09 +01004312 resync = mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active);
4313 dt = (jiffies - mddev->resync_mark) / HZ;
NeilBrown88202a02006-01-06 00:21:36 -08004314 if (!dt) dt++;
Andre Noll9687a602008-03-25 22:24:09 +01004315 db = resync - mddev->resync_mark_cnt;
4316 return sprintf(page, "%lu\n", db/dt/2); /* K/sec */
NeilBrown88202a02006-01-06 00:21:36 -08004317}
4318
NeilBrown80ca3a42006-07-10 04:44:18 -07004319static struct md_sysfs_entry md_sync_speed = __ATTR_RO(sync_speed);
NeilBrown88202a02006-01-06 00:21:36 -08004320
4321static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004322sync_completed_show(struct mddev *mddev, char *page)
NeilBrown88202a02006-01-06 00:21:36 -08004323{
Rémi Rérolle13ae8642011-01-14 09:14:34 +11004324 unsigned long long max_sectors, resync;
NeilBrown88202a02006-01-06 00:21:36 -08004325
NeilBrownacb180b2009-04-14 16:28:34 +10004326 if (!test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
4327 return sprintf(page, "none\n");
4328
NeilBrown72f36d52012-10-11 14:25:57 +11004329 if (mddev->curr_resync == 1 ||
4330 mddev->curr_resync == 2)
4331 return sprintf(page, "delayed\n");
4332
NeilBrownc804cde2012-05-21 09:28:33 +10004333 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
4334 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
Andre Noll58c0fed2009-03-31 14:33:13 +11004335 max_sectors = mddev->resync_max_sectors;
NeilBrown88202a02006-01-06 00:21:36 -08004336 else
Andre Noll58c0fed2009-03-31 14:33:13 +11004337 max_sectors = mddev->dev_sectors;
NeilBrown88202a02006-01-06 00:21:36 -08004338
NeilBrownacb180b2009-04-14 16:28:34 +10004339 resync = mddev->curr_resync_completed;
Rémi Rérolle13ae8642011-01-14 09:14:34 +11004340 return sprintf(page, "%llu / %llu\n", resync, max_sectors);
NeilBrown88202a02006-01-06 00:21:36 -08004341}
4342
NeilBrown750f1992014-09-30 08:53:05 +10004343static struct md_sysfs_entry md_sync_completed =
4344 __ATTR_PREALLOC(sync_completed, S_IRUGO, sync_completed_show, NULL);
NeilBrown88202a02006-01-06 00:21:36 -08004345
NeilBrowne464eaf2006-03-27 01:18:14 -08004346static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004347min_sync_show(struct mddev *mddev, char *page)
Neil Brown5e96ee62008-06-28 08:31:24 +10004348{
4349 return sprintf(page, "%llu\n",
4350 (unsigned long long)mddev->resync_min);
4351}
4352static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004353min_sync_store(struct mddev *mddev, const char *buf, size_t len)
Neil Brown5e96ee62008-06-28 08:31:24 +10004354{
4355 unsigned long long min;
NeilBrown23da4222014-12-15 12:57:01 +11004356 int err;
4357 int chunk;
4358
Jingoo Hanb29bebd2013-06-01 16:15:16 +09004359 if (kstrtoull(buf, 10, &min))
Neil Brown5e96ee62008-06-28 08:31:24 +10004360 return -EINVAL;
NeilBrown23da4222014-12-15 12:57:01 +11004361
4362 spin_lock(&mddev->lock);
4363 err = -EINVAL;
Neil Brown5e96ee62008-06-28 08:31:24 +10004364 if (min > mddev->resync_max)
NeilBrown23da4222014-12-15 12:57:01 +11004365 goto out_unlock;
4366
4367 err = -EBUSY;
Neil Brown5e96ee62008-06-28 08:31:24 +10004368 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown23da4222014-12-15 12:57:01 +11004369 goto out_unlock;
Neil Brown5e96ee62008-06-28 08:31:24 +10004370
4371 /* Must be a multiple of chunk_size */
NeilBrown23da4222014-12-15 12:57:01 +11004372 chunk = mddev->chunk_sectors;
4373 if (chunk) {
raz ben yehuda2ac06c32009-06-16 17:01:42 +10004374 sector_t temp = min;
NeilBrown23da4222014-12-15 12:57:01 +11004375
4376 err = -EINVAL;
4377 if (sector_div(temp, chunk))
4378 goto out_unlock;
Neil Brown5e96ee62008-06-28 08:31:24 +10004379 }
4380 mddev->resync_min = min;
NeilBrown23da4222014-12-15 12:57:01 +11004381 err = 0;
Neil Brown5e96ee62008-06-28 08:31:24 +10004382
NeilBrown23da4222014-12-15 12:57:01 +11004383out_unlock:
4384 spin_unlock(&mddev->lock);
4385 return err ?: len;
Neil Brown5e96ee62008-06-28 08:31:24 +10004386}
4387
4388static struct md_sysfs_entry md_min_sync =
4389__ATTR(sync_min, S_IRUGO|S_IWUSR, min_sync_show, min_sync_store);
4390
4391static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004392max_sync_show(struct mddev *mddev, char *page)
NeilBrownc6207272008-02-06 01:39:52 -08004393{
4394 if (mddev->resync_max == MaxSector)
4395 return sprintf(page, "max\n");
4396 else
4397 return sprintf(page, "%llu\n",
4398 (unsigned long long)mddev->resync_max);
4399}
4400static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004401max_sync_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrownc6207272008-02-06 01:39:52 -08004402{
NeilBrown23da4222014-12-15 12:57:01 +11004403 int err;
4404 spin_lock(&mddev->lock);
NeilBrownc6207272008-02-06 01:39:52 -08004405 if (strncmp(buf, "max", 3) == 0)
4406 mddev->resync_max = MaxSector;
4407 else {
Neil Brown5e96ee62008-06-28 08:31:24 +10004408 unsigned long long max;
NeilBrown23da4222014-12-15 12:57:01 +11004409 int chunk;
4410
4411 err = -EINVAL;
Jingoo Hanb29bebd2013-06-01 16:15:16 +09004412 if (kstrtoull(buf, 10, &max))
NeilBrown23da4222014-12-15 12:57:01 +11004413 goto out_unlock;
Neil Brown5e96ee62008-06-28 08:31:24 +10004414 if (max < mddev->resync_min)
NeilBrown23da4222014-12-15 12:57:01 +11004415 goto out_unlock;
4416
4417 err = -EBUSY;
NeilBrownc6207272008-02-06 01:39:52 -08004418 if (max < mddev->resync_max &&
NeilBrown4d484a42009-08-13 10:41:50 +10004419 mddev->ro == 0 &&
NeilBrownc6207272008-02-06 01:39:52 -08004420 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown23da4222014-12-15 12:57:01 +11004421 goto out_unlock;
NeilBrownc6207272008-02-06 01:39:52 -08004422
4423 /* Must be a multiple of chunk_size */
NeilBrown23da4222014-12-15 12:57:01 +11004424 chunk = mddev->chunk_sectors;
4425 if (chunk) {
raz ben yehuda2ac06c32009-06-16 17:01:42 +10004426 sector_t temp = max;
NeilBrown23da4222014-12-15 12:57:01 +11004427
4428 err = -EINVAL;
4429 if (sector_div(temp, chunk))
4430 goto out_unlock;
NeilBrownc6207272008-02-06 01:39:52 -08004431 }
4432 mddev->resync_max = max;
4433 }
4434 wake_up(&mddev->recovery_wait);
NeilBrown23da4222014-12-15 12:57:01 +11004435 err = 0;
4436out_unlock:
4437 spin_unlock(&mddev->lock);
4438 return err ?: len;
NeilBrownc6207272008-02-06 01:39:52 -08004439}
4440
4441static struct md_sysfs_entry md_max_sync =
4442__ATTR(sync_max, S_IRUGO|S_IWUSR, max_sync_show, max_sync_store);
4443
4444static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004445suspend_lo_show(struct mddev *mddev, char *page)
NeilBrowne464eaf2006-03-27 01:18:14 -08004446{
4447 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_lo);
4448}
4449
4450static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004451suspend_lo_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrowne464eaf2006-03-27 01:18:14 -08004452{
4453 char *e;
4454 unsigned long long new = simple_strtoull(buf, &e, 10);
NeilBrown67918752014-12-15 12:57:01 +11004455 unsigned long long old;
4456 int err;
NeilBrowne464eaf2006-03-27 01:18:14 -08004457
NeilBrowne464eaf2006-03-27 01:18:14 -08004458 if (buf == e || (*e && *e != '\n'))
4459 return -EINVAL;
NeilBrown23ddff32011-01-14 09:14:34 +11004460
NeilBrown67918752014-12-15 12:57:01 +11004461 err = mddev_lock(mddev);
4462 if (err)
4463 return err;
4464 err = -EINVAL;
4465 if (mddev->pers == NULL ||
4466 mddev->pers->quiesce == NULL)
4467 goto unlock;
4468 old = mddev->suspend_lo;
NeilBrown23ddff32011-01-14 09:14:34 +11004469 mddev->suspend_lo = new;
4470 if (new >= old)
4471 /* Shrinking suspended region */
NeilBrowne464eaf2006-03-27 01:18:14 -08004472 mddev->pers->quiesce(mddev, 2);
NeilBrown23ddff32011-01-14 09:14:34 +11004473 else {
4474 /* Expanding suspended region - need to wait */
4475 mddev->pers->quiesce(mddev, 1);
4476 mddev->pers->quiesce(mddev, 0);
4477 }
NeilBrown67918752014-12-15 12:57:01 +11004478 err = 0;
4479unlock:
4480 mddev_unlock(mddev);
4481 return err ?: len;
NeilBrowne464eaf2006-03-27 01:18:14 -08004482}
4483static struct md_sysfs_entry md_suspend_lo =
4484__ATTR(suspend_lo, S_IRUGO|S_IWUSR, suspend_lo_show, suspend_lo_store);
4485
NeilBrowne464eaf2006-03-27 01:18:14 -08004486static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004487suspend_hi_show(struct mddev *mddev, char *page)
NeilBrowne464eaf2006-03-27 01:18:14 -08004488{
4489 return sprintf(page, "%llu\n", (unsigned long long)mddev->suspend_hi);
4490}
4491
4492static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004493suspend_hi_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrowne464eaf2006-03-27 01:18:14 -08004494{
4495 char *e;
4496 unsigned long long new = simple_strtoull(buf, &e, 10);
NeilBrown67918752014-12-15 12:57:01 +11004497 unsigned long long old;
4498 int err;
NeilBrowne464eaf2006-03-27 01:18:14 -08004499
NeilBrowne464eaf2006-03-27 01:18:14 -08004500 if (buf == e || (*e && *e != '\n'))
4501 return -EINVAL;
NeilBrown23ddff32011-01-14 09:14:34 +11004502
NeilBrown67918752014-12-15 12:57:01 +11004503 err = mddev_lock(mddev);
4504 if (err)
4505 return err;
4506 err = -EINVAL;
4507 if (mddev->pers == NULL ||
4508 mddev->pers->quiesce == NULL)
4509 goto unlock;
4510 old = mddev->suspend_hi;
NeilBrown23ddff32011-01-14 09:14:34 +11004511 mddev->suspend_hi = new;
4512 if (new <= old)
4513 /* Shrinking suspended region */
4514 mddev->pers->quiesce(mddev, 2);
4515 else {
4516 /* Expanding suspended region - need to wait */
NeilBrowne464eaf2006-03-27 01:18:14 -08004517 mddev->pers->quiesce(mddev, 1);
4518 mddev->pers->quiesce(mddev, 0);
NeilBrown23ddff32011-01-14 09:14:34 +11004519 }
NeilBrown67918752014-12-15 12:57:01 +11004520 err = 0;
4521unlock:
4522 mddev_unlock(mddev);
4523 return err ?: len;
NeilBrowne464eaf2006-03-27 01:18:14 -08004524}
4525static struct md_sysfs_entry md_suspend_hi =
4526__ATTR(suspend_hi, S_IRUGO|S_IWUSR, suspend_hi_show, suspend_hi_store);
4527
NeilBrown08a02ec2007-05-09 02:35:38 -07004528static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004529reshape_position_show(struct mddev *mddev, char *page)
NeilBrown08a02ec2007-05-09 02:35:38 -07004530{
4531 if (mddev->reshape_position != MaxSector)
4532 return sprintf(page, "%llu\n",
4533 (unsigned long long)mddev->reshape_position);
4534 strcpy(page, "none\n");
4535 return 5;
4536}
4537
4538static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004539reshape_position_store(struct mddev *mddev, const char *buf, size_t len)
NeilBrown08a02ec2007-05-09 02:35:38 -07004540{
NeilBrownc6563a82012-05-21 09:27:00 +10004541 struct md_rdev *rdev;
NeilBrown08a02ec2007-05-09 02:35:38 -07004542 char *e;
NeilBrown67918752014-12-15 12:57:01 +11004543 int err;
NeilBrown08a02ec2007-05-09 02:35:38 -07004544 unsigned long long new = simple_strtoull(buf, &e, 10);
NeilBrown67918752014-12-15 12:57:01 +11004545
NeilBrown08a02ec2007-05-09 02:35:38 -07004546 if (buf == e || (*e && *e != '\n'))
4547 return -EINVAL;
NeilBrown67918752014-12-15 12:57:01 +11004548 err = mddev_lock(mddev);
4549 if (err)
4550 return err;
4551 err = -EBUSY;
4552 if (mddev->pers)
4553 goto unlock;
NeilBrown08a02ec2007-05-09 02:35:38 -07004554 mddev->reshape_position = new;
4555 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10004556 mddev->reshape_backwards = 0;
NeilBrown08a02ec2007-05-09 02:35:38 -07004557 mddev->new_level = mddev->level;
4558 mddev->new_layout = mddev->layout;
Andre Noll664e7c42009-06-18 08:45:27 +10004559 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrownc6563a82012-05-21 09:27:00 +10004560 rdev_for_each(rdev, mddev)
4561 rdev->new_data_offset = rdev->data_offset;
NeilBrown67918752014-12-15 12:57:01 +11004562 err = 0;
4563unlock:
4564 mddev_unlock(mddev);
4565 return err ?: len;
NeilBrown08a02ec2007-05-09 02:35:38 -07004566}
4567
4568static struct md_sysfs_entry md_reshape_position =
4569__ATTR(reshape_position, S_IRUGO|S_IWUSR, reshape_position_show,
4570 reshape_position_store);
4571
Dan Williamsb522adc2009-03-31 15:00:31 +11004572static ssize_t
NeilBrown2c810cd2012-05-21 09:27:00 +10004573reshape_direction_show(struct mddev *mddev, char *page)
4574{
4575 return sprintf(page, "%s\n",
4576 mddev->reshape_backwards ? "backwards" : "forwards");
4577}
4578
4579static ssize_t
4580reshape_direction_store(struct mddev *mddev, const char *buf, size_t len)
4581{
4582 int backwards = 0;
NeilBrown67918752014-12-15 12:57:01 +11004583 int err;
4584
NeilBrown2c810cd2012-05-21 09:27:00 +10004585 if (cmd_match(buf, "forwards"))
4586 backwards = 0;
4587 else if (cmd_match(buf, "backwards"))
4588 backwards = 1;
4589 else
4590 return -EINVAL;
4591 if (mddev->reshape_backwards == backwards)
4592 return len;
4593
NeilBrown67918752014-12-15 12:57:01 +11004594 err = mddev_lock(mddev);
4595 if (err)
4596 return err;
NeilBrown2c810cd2012-05-21 09:27:00 +10004597 /* check if we are allowed to change */
4598 if (mddev->delta_disks)
NeilBrown67918752014-12-15 12:57:01 +11004599 err = -EBUSY;
4600 else if (mddev->persistent &&
NeilBrown2c810cd2012-05-21 09:27:00 +10004601 mddev->major_version == 0)
NeilBrown67918752014-12-15 12:57:01 +11004602 err = -EINVAL;
4603 else
4604 mddev->reshape_backwards = backwards;
4605 mddev_unlock(mddev);
4606 return err ?: len;
NeilBrown2c810cd2012-05-21 09:27:00 +10004607}
4608
4609static struct md_sysfs_entry md_reshape_direction =
4610__ATTR(reshape_direction, S_IRUGO|S_IWUSR, reshape_direction_show,
4611 reshape_direction_store);
4612
4613static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004614array_size_show(struct mddev *mddev, char *page)
Dan Williamsb522adc2009-03-31 15:00:31 +11004615{
4616 if (mddev->external_size)
4617 return sprintf(page, "%llu\n",
4618 (unsigned long long)mddev->array_sectors/2);
4619 else
4620 return sprintf(page, "default\n");
4621}
4622
4623static ssize_t
NeilBrownfd01b882011-10-11 16:47:53 +11004624array_size_store(struct mddev *mddev, const char *buf, size_t len)
Dan Williamsb522adc2009-03-31 15:00:31 +11004625{
4626 sector_t sectors;
NeilBrown67918752014-12-15 12:57:01 +11004627 int err;
4628
4629 err = mddev_lock(mddev);
4630 if (err)
4631 return err;
Dan Williamsb522adc2009-03-31 15:00:31 +11004632
4633 if (strncmp(buf, "default", 7) == 0) {
4634 if (mddev->pers)
4635 sectors = mddev->pers->size(mddev, 0, 0);
4636 else
4637 sectors = mddev->array_sectors;
4638
4639 mddev->external_size = 0;
4640 } else {
4641 if (strict_blocks_to_sectors(buf, &sectors) < 0)
NeilBrown67918752014-12-15 12:57:01 +11004642 err = -EINVAL;
4643 else if (mddev->pers && mddev->pers->size(mddev, 0, 0) < sectors)
4644 err = -E2BIG;
4645 else
4646 mddev->external_size = 1;
Dan Williamsb522adc2009-03-31 15:00:31 +11004647 }
4648
NeilBrown67918752014-12-15 12:57:01 +11004649 if (!err) {
4650 mddev->array_sectors = sectors;
4651 if (mddev->pers) {
4652 set_capacity(mddev->gendisk, mddev->array_sectors);
4653 revalidate_disk(mddev->gendisk);
4654 }
NeilBrowncbe6ef12011-02-16 13:58:38 +11004655 }
NeilBrown67918752014-12-15 12:57:01 +11004656 mddev_unlock(mddev);
4657 return err ?: len;
Dan Williamsb522adc2009-03-31 15:00:31 +11004658}
4659
4660static struct md_sysfs_entry md_array_size =
4661__ATTR(array_size, S_IRUGO|S_IWUSR, array_size_show,
4662 array_size_store);
NeilBrowne464eaf2006-03-27 01:18:14 -08004663
NeilBrowneae17012005-11-08 21:39:23 -08004664static struct attribute *md_default_attrs[] = {
4665 &md_level.attr,
NeilBrownd4dbd022006-06-26 00:27:59 -07004666 &md_layout.attr,
NeilBrowneae17012005-11-08 21:39:23 -08004667 &md_raid_disks.attr,
NeilBrown3b343802006-01-06 00:20:47 -08004668 &md_chunk_size.attr,
NeilBrowna35b0d692006-01-06 00:20:49 -08004669 &md_size.attr,
NeilBrowna94213b2006-06-26 00:28:00 -07004670 &md_resync_start.attr,
NeilBrown8bb93aa2006-01-06 00:20:50 -08004671 &md_metadata.attr,
NeilBrown6d7ff732006-01-06 00:21:16 -08004672 &md_new_device.attr,
NeilBrown16f17b32006-06-26 00:27:37 -07004673 &md_safe_delay.attr,
NeilBrown9e653b62006-06-26 00:27:58 -07004674 &md_array_state.attr,
NeilBrown08a02ec2007-05-09 02:35:38 -07004675 &md_reshape_position.attr,
NeilBrown2c810cd2012-05-21 09:27:00 +10004676 &md_reshape_direction.attr,
Dan Williamsb522adc2009-03-31 15:00:31 +11004677 &md_array_size.attr,
Robert Becker1e509152009-12-14 12:49:58 +11004678 &max_corr_read_errors.attr,
NeilBrown411036f2005-11-08 21:39:40 -08004679 NULL,
4680};
4681
4682static struct attribute *md_redundancy_attrs[] = {
NeilBrown24dd4692005-11-08 21:39:26 -08004683 &md_scan_mode.attr,
Jonathan Brassowc4a39552013-06-25 01:23:59 -05004684 &md_last_scan_mode.attr,
NeilBrown9d888832005-11-08 21:39:26 -08004685 &md_mismatches.attr,
NeilBrown88202a02006-01-06 00:21:36 -08004686 &md_sync_min.attr,
4687 &md_sync_max.attr,
4688 &md_sync_speed.attr,
Bernd Schubert90b08712008-05-23 13:04:38 -07004689 &md_sync_force_parallel.attr,
NeilBrown88202a02006-01-06 00:21:36 -08004690 &md_sync_completed.attr,
Neil Brown5e96ee62008-06-28 08:31:24 +10004691 &md_min_sync.attr,
NeilBrownc6207272008-02-06 01:39:52 -08004692 &md_max_sync.attr,
NeilBrowne464eaf2006-03-27 01:18:14 -08004693 &md_suspend_lo.attr,
4694 &md_suspend_hi.attr,
Paul Clements9b1d1da2006-10-03 01:15:49 -07004695 &md_bitmap.attr,
Iustin Popd7f3d292007-10-16 23:30:54 -07004696 &md_degraded.attr,
NeilBrowneae17012005-11-08 21:39:23 -08004697 NULL,
4698};
NeilBrown411036f2005-11-08 21:39:40 -08004699static struct attribute_group md_redundancy_group = {
4700 .name = NULL,
4701 .attrs = md_redundancy_attrs,
4702};
4703
NeilBrowneae17012005-11-08 21:39:23 -08004704static ssize_t
4705md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
4706{
4707 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
NeilBrownfd01b882011-10-11 16:47:53 +11004708 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
NeilBrown96de1e62005-11-08 21:39:39 -08004709 ssize_t rv;
NeilBrowneae17012005-11-08 21:39:23 -08004710
4711 if (!entry->show)
4712 return -EIO;
NeilBrownaf8a2432011-12-08 15:49:46 +11004713 spin_lock(&all_mddevs_lock);
4714 if (list_empty(&mddev->all_mddevs)) {
4715 spin_unlock(&all_mddevs_lock);
4716 return -EBUSY;
4717 }
4718 mddev_get(mddev);
4719 spin_unlock(&all_mddevs_lock);
4720
NeilBrownb7b17c92014-12-15 12:56:59 +11004721 rv = entry->show(mddev, page);
NeilBrownaf8a2432011-12-08 15:49:46 +11004722 mddev_put(mddev);
NeilBrown96de1e62005-11-08 21:39:39 -08004723 return rv;
NeilBrowneae17012005-11-08 21:39:23 -08004724}
4725
4726static ssize_t
4727md_attr_store(struct kobject *kobj, struct attribute *attr,
4728 const char *page, size_t length)
4729{
4730 struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
NeilBrownfd01b882011-10-11 16:47:53 +11004731 struct mddev *mddev = container_of(kobj, struct mddev, kobj);
NeilBrown96de1e62005-11-08 21:39:39 -08004732 ssize_t rv;
NeilBrowneae17012005-11-08 21:39:23 -08004733
4734 if (!entry->store)
4735 return -EIO;
NeilBrown67463ac2006-07-10 04:44:19 -07004736 if (!capable(CAP_SYS_ADMIN))
4737 return -EACCES;
NeilBrownaf8a2432011-12-08 15:49:46 +11004738 spin_lock(&all_mddevs_lock);
4739 if (list_empty(&mddev->all_mddevs)) {
4740 spin_unlock(&all_mddevs_lock);
4741 return -EBUSY;
4742 }
4743 mddev_get(mddev);
4744 spin_unlock(&all_mddevs_lock);
NeilBrown67918752014-12-15 12:57:01 +11004745 rv = entry->store(mddev, page, length);
NeilBrownaf8a2432011-12-08 15:49:46 +11004746 mddev_put(mddev);
NeilBrown96de1e62005-11-08 21:39:39 -08004747 return rv;
NeilBrowneae17012005-11-08 21:39:23 -08004748}
4749
4750static void md_free(struct kobject *ko)
4751{
NeilBrownfd01b882011-10-11 16:47:53 +11004752 struct mddev *mddev = container_of(ko, struct mddev, kobj);
NeilBrowna21d1502009-01-09 08:31:09 +11004753
4754 if (mddev->sysfs_state)
4755 sysfs_put(mddev->sysfs_state);
4756
4757 if (mddev->gendisk) {
4758 del_gendisk(mddev->gendisk);
4759 put_disk(mddev->gendisk);
4760 }
4761 if (mddev->queue)
4762 blk_cleanup_queue(mddev->queue);
4763
NeilBrowneae17012005-11-08 21:39:23 -08004764 kfree(mddev);
4765}
4766
Emese Revfy52cf25d2010-01-19 02:58:23 +01004767static const struct sysfs_ops md_sysfs_ops = {
NeilBrowneae17012005-11-08 21:39:23 -08004768 .show = md_attr_show,
4769 .store = md_attr_store,
4770};
4771static struct kobj_type md_ktype = {
4772 .release = md_free,
4773 .sysfs_ops = &md_sysfs_ops,
4774 .default_attrs = md_default_attrs,
4775};
4776
Linus Torvalds1da177e2005-04-16 15:20:36 -07004777int mdp_major = 0;
4778
Dan Williams5fd3a172009-03-04 00:57:25 -07004779static void mddev_delayed_delete(struct work_struct *ws)
4780{
NeilBrownfd01b882011-10-11 16:47:53 +11004781 struct mddev *mddev = container_of(ws, struct mddev, del_work);
Dan Williams5fd3a172009-03-04 00:57:25 -07004782
NeilBrown43a70502009-12-14 12:49:55 +11004783 sysfs_remove_group(&mddev->kobj, &md_bitmap_group);
Dan Williams5fd3a172009-03-04 00:57:25 -07004784 kobject_del(&mddev->kobj);
4785 kobject_put(&mddev->kobj);
4786}
4787
NeilBrownefeb53c2009-01-09 08:31:10 +11004788static int md_alloc(dev_t dev, char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004789{
Arjan van de Ven48c9c272006-03-27 01:18:20 -08004790 static DEFINE_MUTEX(disks_mutex);
NeilBrownfd01b882011-10-11 16:47:53 +11004791 struct mddev *mddev = mddev_find(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004792 struct gendisk *disk;
NeilBrownefeb53c2009-01-09 08:31:10 +11004793 int partitioned;
4794 int shift;
4795 int unit;
Greg Kroah-Hartman3830c622007-12-17 15:54:39 -04004796 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004797
4798 if (!mddev)
NeilBrownefeb53c2009-01-09 08:31:10 +11004799 return -ENODEV;
4800
4801 partitioned = (MAJOR(mddev->unit) != MD_MAJOR);
4802 shift = partitioned ? MdpMinorShift : 0;
4803 unit = MINOR(mddev->unit) >> shift;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004804
Tejun Heoe804ac72010-10-15 15:36:08 +02004805 /* wait for any previous instance of this device to be
4806 * completely removed (mddev_delayed_delete).
NeilBrownd3374822009-01-09 08:31:10 +11004807 */
Tejun Heoe804ac72010-10-15 15:36:08 +02004808 flush_workqueue(md_misc_wq);
NeilBrownd3374822009-01-09 08:31:10 +11004809
Arjan van de Ven48c9c272006-03-27 01:18:20 -08004810 mutex_lock(&disks_mutex);
NeilBrown0909dc42009-07-01 12:27:21 +10004811 error = -EEXIST;
4812 if (mddev->gendisk)
4813 goto abort;
NeilBrownefeb53c2009-01-09 08:31:10 +11004814
4815 if (name) {
4816 /* Need to ensure that 'name' is not a duplicate.
4817 */
NeilBrownfd01b882011-10-11 16:47:53 +11004818 struct mddev *mddev2;
NeilBrownefeb53c2009-01-09 08:31:10 +11004819 spin_lock(&all_mddevs_lock);
4820
4821 list_for_each_entry(mddev2, &all_mddevs, all_mddevs)
4822 if (mddev2->gendisk &&
4823 strcmp(mddev2->gendisk->disk_name, name) == 0) {
4824 spin_unlock(&all_mddevs_lock);
NeilBrown0909dc42009-07-01 12:27:21 +10004825 goto abort;
NeilBrownefeb53c2009-01-09 08:31:10 +11004826 }
4827 spin_unlock(&all_mddevs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004828 }
NeilBrown8b765392009-01-09 08:31:08 +11004829
NeilBrown0909dc42009-07-01 12:27:21 +10004830 error = -ENOMEM;
NeilBrown8b765392009-01-09 08:31:08 +11004831 mddev->queue = blk_alloc_queue(GFP_KERNEL);
NeilBrown0909dc42009-07-01 12:27:21 +10004832 if (!mddev->queue)
4833 goto abort;
NeilBrown409c57f2009-03-31 14:39:39 +11004834 mddev->queue->queuedata = mddev;
4835
NeilBrown409c57f2009-03-31 14:39:39 +11004836 blk_queue_make_request(mddev->queue, md_make_request);
Martin K. Petersenb1bd0552012-01-11 16:27:11 +01004837 blk_set_stacking_limits(&mddev->queue->limits);
NeilBrown8b765392009-01-09 08:31:08 +11004838
Linus Torvalds1da177e2005-04-16 15:20:36 -07004839 disk = alloc_disk(1 << shift);
4840 if (!disk) {
NeilBrown8b765392009-01-09 08:31:08 +11004841 blk_cleanup_queue(mddev->queue);
4842 mddev->queue = NULL;
NeilBrown0909dc42009-07-01 12:27:21 +10004843 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004844 }
NeilBrownefeb53c2009-01-09 08:31:10 +11004845 disk->major = MAJOR(mddev->unit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004846 disk->first_minor = unit << shift;
NeilBrownefeb53c2009-01-09 08:31:10 +11004847 if (name)
4848 strcpy(disk->disk_name, name);
4849 else if (partitioned)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004850 sprintf(disk->disk_name, "md_d%d", unit);
Greg Kroah-Hartmance7b0f42005-06-20 21:15:16 -07004851 else
Linus Torvalds1da177e2005-04-16 15:20:36 -07004852 sprintf(disk->disk_name, "md%d", unit);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004853 disk->fops = &md_fops;
4854 disk->private_data = mddev;
4855 disk->queue = mddev->queue;
NeilBrownb0140892011-05-10 17:49:01 +10004856 blk_queue_flush(mddev->queue, REQ_FLUSH | REQ_FUA);
NeilBrown92850bb2008-10-21 13:25:32 +11004857 /* Allow extended partitions. This makes the
NeilBrownd3374822009-01-09 08:31:10 +11004858 * 'mdp' device redundant, but we can't really
NeilBrown92850bb2008-10-21 13:25:32 +11004859 * remove it now.
4860 */
4861 disk->flags |= GENHD_FL_EXT_DEVT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004862 mddev->gendisk = disk;
NeilBrownb0140892011-05-10 17:49:01 +10004863 /* As soon as we call add_disk(), another thread could get
4864 * through to md_open, so make sure it doesn't get too far
4865 */
4866 mutex_lock(&mddev->open_mutex);
4867 add_disk(disk);
4868
Tejun Heoed9e1982008-08-25 19:56:05 +09004869 error = kobject_init_and_add(&mddev->kobj, &md_ktype,
4870 &disk_to_dev(disk)->kobj, "%s", "md");
NeilBrown0909dc42009-07-01 12:27:21 +10004871 if (error) {
4872 /* This isn't possible, but as kobject_init_and_add is marked
4873 * __must_check, we must do something with the result
4874 */
NeilBrown5e55e2f2007-03-26 21:32:14 -08004875 printk(KERN_WARNING "md: cannot register %s/md - name in use\n",
4876 disk->disk_name);
NeilBrown0909dc42009-07-01 12:27:21 +10004877 error = 0;
4878 }
NeilBrown00bcb4a2010-06-01 19:37:23 +10004879 if (mddev->kobj.sd &&
4880 sysfs_create_group(&mddev->kobj, &md_bitmap_group))
NeilBrown43a70502009-12-14 12:49:55 +11004881 printk(KERN_DEBUG "pointless warning\n");
NeilBrownb0140892011-05-10 17:49:01 +10004882 mutex_unlock(&mddev->open_mutex);
NeilBrown0909dc42009-07-01 12:27:21 +10004883 abort:
4884 mutex_unlock(&disks_mutex);
NeilBrown00bcb4a2010-06-01 19:37:23 +10004885 if (!error && mddev->kobj.sd) {
Greg Kroah-Hartman3830c622007-12-17 15:54:39 -04004886 kobject_uevent(&mddev->kobj, KOBJ_ADD);
NeilBrown00bcb4a2010-06-01 19:37:23 +10004887 mddev->sysfs_state = sysfs_get_dirent_safe(mddev->kobj.sd, "array_state");
NeilBrownb62b7592008-10-21 13:25:21 +11004888 }
NeilBrownd3374822009-01-09 08:31:10 +11004889 mddev_put(mddev);
NeilBrown0909dc42009-07-01 12:27:21 +10004890 return error;
NeilBrownefeb53c2009-01-09 08:31:10 +11004891}
4892
4893static struct kobject *md_probe(dev_t dev, int *part, void *data)
4894{
4895 md_alloc(dev, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004896 return NULL;
4897}
4898
NeilBrownefeb53c2009-01-09 08:31:10 +11004899static int add_named_array(const char *val, struct kernel_param *kp)
4900{
4901 /* val must be "md_*" where * is not all digits.
4902 * We allocate an array with a large free minor number, and
4903 * set the name to val. val must not already be an active name.
4904 */
4905 int len = strlen(val);
4906 char buf[DISK_NAME_LEN];
4907
4908 while (len && val[len-1] == '\n')
4909 len--;
4910 if (len >= DISK_NAME_LEN)
4911 return -E2BIG;
4912 strlcpy(buf, val, len+1);
4913 if (strncmp(buf, "md_", 3) != 0)
4914 return -EINVAL;
4915 return md_alloc(0, buf);
4916}
4917
Linus Torvalds1da177e2005-04-16 15:20:36 -07004918static void md_safemode_timeout(unsigned long data)
4919{
NeilBrownfd01b882011-10-11 16:47:53 +11004920 struct mddev *mddev = (struct mddev *) data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004921
Neil Brown0fd62b82008-06-28 08:31:36 +10004922 if (!atomic_read(&mddev->writes_pending)) {
4923 mddev->safemode = 1;
4924 if (mddev->external)
NeilBrown00bcb4a2010-06-01 19:37:23 +10004925 sysfs_notify_dirent_safe(mddev->sysfs_state);
Neil Brown0fd62b82008-06-28 08:31:36 +10004926 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004927 md_wakeup_thread(mddev->thread);
4928}
4929
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08004930static int start_dirty_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004931
NeilBrownfd01b882011-10-11 16:47:53 +11004932int md_run(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004933{
NeilBrown2604b702006-01-06 00:20:36 -08004934 int err;
NeilBrown3cb03002011-10-11 16:45:26 +11004935 struct md_rdev *rdev;
NeilBrown84fc4b52011-10-11 16:49:58 +11004936 struct md_personality *pers;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004937
NeilBrowna757e642005-04-16 15:26:42 -07004938 if (list_empty(&mddev->disks))
4939 /* cannot run an array with no devices.. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004940 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004941
4942 if (mddev->pers)
4943 return -EBUSY;
NeilBrownbb4f1e92010-08-08 21:18:03 +10004944 /* Cannot run until previous stop completes properly */
4945 if (mddev->sysfs_active)
4946 return -EBUSY;
NeilBrownb6eb1272010-04-15 10:13:47 +10004947
Linus Torvalds1da177e2005-04-16 15:20:36 -07004948 /*
4949 * Analyze all RAID superblock(s)
4950 */
NeilBrown1ec4a932008-02-06 01:39:53 -08004951 if (!mddev->raid_disks) {
4952 if (!mddev->persistent)
4953 return -EINVAL;
NeilBrowna757e642005-04-16 15:26:42 -07004954 analyze_sbs(mddev);
NeilBrown1ec4a932008-02-06 01:39:53 -08004955 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07004956
NeilBrownd9d166c2006-01-06 00:20:51 -08004957 if (mddev->level != LEVEL_NONE)
4958 request_module("md-level-%d", mddev->level);
4959 else if (mddev->clevel[0])
4960 request_module("md-%s", mddev->clevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004961
4962 /*
4963 * Drop all container device buffers, from now on
4964 * the only valid external interface is through the md
4965 * device.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004966 */
NeilBrowndafb20f2012-03-19 12:46:39 +11004967 rdev_for_each(rdev, mddev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08004968 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07004969 continue;
4970 sync_blockdev(rdev->bdev);
Peter Zijlstraf98393a2007-05-06 14:49:54 -07004971 invalidate_bdev(rdev->bdev);
NeilBrownf0d76d72007-07-17 04:06:12 -07004972
4973 /* perform some consistency tests on the device.
4974 * We don't want the data to overlap the metadata,
Andre Noll58c0fed2009-03-31 14:33:13 +11004975 * Internal Bitmap issues have been handled elsewhere.
NeilBrownf0d76d72007-07-17 04:06:12 -07004976 */
Jonathan Brassowa6ff7e02011-01-14 09:14:34 +11004977 if (rdev->meta_bdev) {
4978 /* Nothing to check */;
4979 } else if (rdev->data_offset < rdev->sb_start) {
Andre Noll58c0fed2009-03-31 14:33:13 +11004980 if (mddev->dev_sectors &&
4981 rdev->data_offset + mddev->dev_sectors
Andre Noll0f420352008-07-11 22:02:23 +10004982 > rdev->sb_start) {
NeilBrownf0d76d72007-07-17 04:06:12 -07004983 printk("md: %s: data overlaps metadata\n",
4984 mdname(mddev));
4985 return -EINVAL;
4986 }
4987 } else {
Andre Noll0f420352008-07-11 22:02:23 +10004988 if (rdev->sb_start + rdev->sb_size/512
NeilBrownf0d76d72007-07-17 04:06:12 -07004989 > rdev->data_offset) {
4990 printk("md: %s: metadata overlaps data\n",
4991 mdname(mddev));
4992 return -EINVAL;
4993 }
4994 }
NeilBrown00bcb4a2010-06-01 19:37:23 +10004995 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004996 }
4997
NeilBrowna167f662010-10-26 18:31:13 +11004998 if (mddev->bio_set == NULL)
Kent Overstreet395c72a2012-09-06 15:34:55 -07004999 mddev->bio_set = bioset_create(BIO_POOL_SIZE, 0);
NeilBrowna167f662010-10-26 18:31:13 +11005000
Linus Torvalds1da177e2005-04-16 15:20:36 -07005001 spin_lock(&pers_lock);
NeilBrownd9d166c2006-01-06 00:20:51 -08005002 pers = find_pers(mddev->level, mddev->clevel);
NeilBrown2604b702006-01-06 00:20:36 -08005003 if (!pers || !try_module_get(pers->owner)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005004 spin_unlock(&pers_lock);
NeilBrownd9d166c2006-01-06 00:20:51 -08005005 if (mddev->level != LEVEL_NONE)
5006 printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
5007 mddev->level);
5008 else
5009 printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
5010 mddev->clevel);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005011 return -EINVAL;
5012 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005013 spin_unlock(&pers_lock);
NeilBrown34817e82009-03-31 14:39:38 +11005014 if (mddev->level != pers->level) {
5015 mddev->level = pers->level;
5016 mddev->new_level = pers->level;
5017 }
NeilBrownd9d166c2006-01-06 00:20:51 -08005018 strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
Linus Torvalds1da177e2005-04-16 15:20:36 -07005019
NeilBrownf6705572006-03-27 01:18:11 -08005020 if (mddev->reshape_position != MaxSector &&
NeilBrown63c70c42006-03-27 01:18:13 -08005021 pers->start_reshape == NULL) {
NeilBrownf6705572006-03-27 01:18:11 -08005022 /* This personality cannot handle reshaping... */
NeilBrownf6705572006-03-27 01:18:11 -08005023 module_put(pers->owner);
5024 return -EINVAL;
5025 }
5026
NeilBrown7dd5e7c32007-02-28 20:11:35 -08005027 if (pers->sync_request) {
5028 /* Warn if this is a potentially silly
5029 * configuration.
5030 */
5031 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11005032 struct md_rdev *rdev2;
NeilBrown7dd5e7c32007-02-28 20:11:35 -08005033 int warned = 0;
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005034
NeilBrowndafb20f2012-03-19 12:46:39 +11005035 rdev_for_each(rdev, mddev)
5036 rdev_for_each(rdev2, mddev) {
NeilBrown7dd5e7c32007-02-28 20:11:35 -08005037 if (rdev < rdev2 &&
5038 rdev->bdev->bd_contains ==
5039 rdev2->bdev->bd_contains) {
5040 printk(KERN_WARNING
5041 "%s: WARNING: %s appears to be"
5042 " on the same physical disk as"
5043 " %s.\n",
5044 mdname(mddev),
5045 bdevname(rdev->bdev,b),
5046 bdevname(rdev2->bdev,b2));
5047 warned = 1;
5048 }
5049 }
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005050
NeilBrown7dd5e7c32007-02-28 20:11:35 -08005051 if (warned)
5052 printk(KERN_WARNING
5053 "True protection against single-disk"
5054 " failure might be compromised.\n");
5055 }
5056
NeilBrown657390d2005-08-26 18:34:16 -07005057 mddev->recovery = 0;
Andre Noll58c0fed2009-03-31 14:33:13 +11005058 /* may be over-ridden by personality */
5059 mddev->resync_max_sectors = mddev->dev_sectors;
5060
NeilBrown6ff8d8ec2006-01-06 00:20:15 -08005061 mddev->ok_start_degraded = start_dirty_degraded;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005062
NeilBrown0f9552b2009-12-30 12:08:50 +11005063 if (start_readonly && mddev->ro == 0)
NeilBrownf91de922005-11-08 21:39:36 -08005064 mddev->ro = 2; /* read-only, but switch on first write */
5065
NeilBrown36d091f2014-12-15 12:56:58 +11005066 err = pers->run(mddev);
Andre Noll13e53df2008-03-26 00:07:03 +01005067 if (err)
5068 printk(KERN_ERR "md: pers->run() failed ...\n");
NeilBrown36d091f2014-12-15 12:56:58 +11005069 else if (pers->size(mddev, 0, 0) < mddev->array_sectors) {
Dan Williamsb522adc2009-03-31 15:00:31 +11005070 WARN_ONCE(!mddev->external_size, "%s: default size too small,"
5071 " but 'external_size' not in effect?\n", __func__);
5072 printk(KERN_ERR
5073 "md: invalid array_size %llu > default size %llu\n",
5074 (unsigned long long)mddev->array_sectors / 2,
NeilBrown36d091f2014-12-15 12:56:58 +11005075 (unsigned long long)pers->size(mddev, 0, 0) / 2);
Dan Williamsb522adc2009-03-31 15:00:31 +11005076 err = -EINVAL;
Dan Williamsb522adc2009-03-31 15:00:31 +11005077 }
NeilBrown36d091f2014-12-15 12:56:58 +11005078 if (err == 0 && pers->sync_request &&
NeilBrownef99bf42012-05-22 13:55:08 +10005079 (mddev->bitmap_info.file || mddev->bitmap_info.offset)) {
NeilBrownb15c2e52006-01-06 00:20:16 -08005080 err = bitmap_create(mddev);
NeilBrown5aa61f42014-12-15 12:56:57 +11005081 if (err)
NeilBrownb15c2e52006-01-06 00:20:16 -08005082 printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
5083 mdname(mddev), err);
NeilBrownb15c2e52006-01-06 00:20:16 -08005084 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005085 if (err) {
NeilBrown5aa61f42014-12-15 12:56:57 +11005086 mddev_detach(mddev);
NeilBrown0c35bd42015-03-13 11:51:18 +11005087 if (mddev->private)
5088 pers->free(mddev, mddev->private);
NeilBrown36d091f2014-12-15 12:56:58 +11005089 module_put(pers->owner);
NeilBrown32a76272005-06-21 17:17:14 -07005090 bitmap_destroy(mddev);
5091 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005092 }
NeilBrown5c675f82014-12-15 12:56:56 +11005093 if (mddev->queue) {
5094 mddev->queue->backing_dev_info.congested_data = mddev;
5095 mddev->queue->backing_dev_info.congested_fn = md_congested;
NeilBrown64590f42014-12-15 12:56:57 +11005096 blk_queue_merge_bvec(mddev->queue, md_mergeable_bvec);
NeilBrown5c675f82014-12-15 12:56:56 +11005097 }
NeilBrown36d091f2014-12-15 12:56:58 +11005098 if (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
NeilBrownf72ffdd2014-09-30 14:23:59 +10005108 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;
NeilBrown0ca698862011-01-14 09:14:33 +11005116 smp_wmb();
NeilBrown36d091f2014-12-15 12:56:58 +11005117 spin_lock(&mddev->lock);
5118 mddev->pers = pers;
NeilBrown0ca698862011-01-14 09:14:33 +11005119 mddev->ready = 1;
NeilBrown36d091f2014-12-15 12:56:58 +11005120 spin_unlock(&mddev->lock);
NeilBrowndafb20f2012-03-19 12:46:39 +11005121 rdev_for_each(rdev, mddev)
Namhyung Kim36fad852011-07-27 11:00:36 +10005122 if (rdev->raid_disk >= 0)
5123 if (sysfs_link_rdev(mddev, rdev))
NeilBrown00bcb4a2010-06-01 19:37:23 +10005124 /* failure here is OK */;
NeilBrownf72ffdd2014-09-30 14:23:59 +10005125
Linus Torvalds1da177e2005-04-16 15:20:36 -07005126 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrownf72ffdd2014-09-30 14:23:59 +10005127
NeilBrown7a0a5352013-08-27 16:28:23 +10005128 if (mddev->flags & MD_UPDATE_SB_FLAGS)
NeilBrown850b2b42006-10-03 01:15:46 -07005129 md_update_sb(mddev, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005130
NeilBrownd7603b72006-01-06 00:20:30 -08005131 md_new_event(mddev);
NeilBrown00bcb4a2010-06-01 19:37:23 +10005132 sysfs_notify_dirent_safe(mddev->sysfs_state);
5133 sysfs_notify_dirent_safe(mddev->sysfs_action);
Neil Browna99ac972008-06-28 08:31:43 +10005134 sysfs_notify(&mddev->kobj, NULL, "degraded");
Linus Torvalds1da177e2005-04-16 15:20:36 -07005135 return 0;
5136}
NeilBrown390ee602010-06-01 19:37:27 +10005137EXPORT_SYMBOL_GPL(md_run);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005138
NeilBrownfd01b882011-10-11 16:47:53 +11005139static int do_md_run(struct mddev *mddev)
NeilBrownfe60b012010-03-29 11:10:42 +11005140{
5141 int err;
5142
5143 err = md_run(mddev);
5144 if (err)
5145 goto out;
NeilBrown69e51b42010-06-01 19:37:35 +10005146 err = bitmap_load(mddev);
5147 if (err) {
5148 bitmap_destroy(mddev);
5149 goto out;
5150 }
Jonathan Brassow0fd018a2011-06-07 17:49:36 -05005151
5152 md_wakeup_thread(mddev->thread);
5153 md_wakeup_thread(mddev->sync_thread); /* possibly kick off a reshape */
5154
NeilBrownfe60b012010-03-29 11:10:42 +11005155 set_capacity(mddev->gendisk, mddev->array_sectors);
5156 revalidate_disk(mddev->gendisk);
NeilBrownf0b4f7e2011-02-24 17:26:41 +11005157 mddev->changed = 1;
NeilBrownfe60b012010-03-29 11:10:42 +11005158 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
5159out:
5160 return err;
5161}
5162
NeilBrownfd01b882011-10-11 16:47:53 +11005163static int restart_array(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005164{
5165 struct gendisk *disk = mddev->gendisk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005166
Andre Noll80fab1d2008-07-11 22:02:21 +10005167 /* Complain if it has no devices */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005168 if (list_empty(&mddev->disks))
Andre Noll80fab1d2008-07-11 22:02:21 +10005169 return -ENXIO;
5170 if (!mddev->pers)
5171 return -EINVAL;
5172 if (!mddev->ro)
5173 return -EBUSY;
5174 mddev->safemode = 0;
5175 mddev->ro = 0;
5176 set_disk_ro(disk, 0);
5177 printk(KERN_INFO "md: %s switched to read-write mode.\n",
5178 mdname(mddev));
5179 /* Kick recovery or resync if necessary */
5180 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5181 md_wakeup_thread(mddev->thread);
5182 md_wakeup_thread(mddev->sync_thread);
NeilBrown00bcb4a2010-06-01 19:37:23 +10005183 sysfs_notify_dirent_safe(mddev->sysfs_state);
Andre Noll80fab1d2008-07-11 22:02:21 +10005184 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005185}
5186
NeilBrownfd01b882011-10-11 16:47:53 +11005187static void md_clean(struct mddev *mddev)
NeilBrown6177b472010-03-29 11:37:13 +11005188{
5189 mddev->array_sectors = 0;
5190 mddev->external_size = 0;
5191 mddev->dev_sectors = 0;
5192 mddev->raid_disks = 0;
5193 mddev->recovery_cp = 0;
5194 mddev->resync_min = 0;
5195 mddev->resync_max = MaxSector;
5196 mddev->reshape_position = MaxSector;
5197 mddev->external = 0;
5198 mddev->persistent = 0;
5199 mddev->level = LEVEL_NONE;
5200 mddev->clevel[0] = 0;
5201 mddev->flags = 0;
5202 mddev->ro = 0;
5203 mddev->metadata_type[0] = 0;
5204 mddev->chunk_sectors = 0;
5205 mddev->ctime = mddev->utime = 0;
5206 mddev->layout = 0;
5207 mddev->max_disks = 0;
5208 mddev->events = 0;
NeilBrowna8707c02010-05-18 09:28:43 +10005209 mddev->can_decrease_events = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005210 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10005211 mddev->reshape_backwards = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005212 mddev->new_level = LEVEL_NONE;
5213 mddev->new_layout = 0;
5214 mddev->new_chunk_sectors = 0;
5215 mddev->curr_resync = 0;
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11005216 atomic64_set(&mddev->resync_mismatches, 0);
NeilBrown6177b472010-03-29 11:37:13 +11005217 mddev->suspend_lo = mddev->suspend_hi = 0;
5218 mddev->sync_speed_min = mddev->sync_speed_max = 0;
5219 mddev->recovery = 0;
5220 mddev->in_sync = 0;
NeilBrownf0b4f7e2011-02-24 17:26:41 +11005221 mddev->changed = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005222 mddev->degraded = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005223 mddev->safemode = 0;
NeilBrown050b6612012-03-19 12:46:39 +11005224 mddev->merge_check_needed = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005225 mddev->bitmap_info.offset = 0;
5226 mddev->bitmap_info.default_offset = 0;
NeilBrown6409bb02012-05-22 13:55:07 +10005227 mddev->bitmap_info.default_space = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005228 mddev->bitmap_info.chunksize = 0;
5229 mddev->bitmap_info.daemon_sleep = 0;
5230 mddev->bitmap_info.max_write_behind = 0;
5231}
5232
NeilBrownfd01b882011-10-11 16:47:53 +11005233static void __md_stop_writes(struct mddev *mddev)
NeilBrowna047e122010-03-29 12:07:53 +11005234{
NeilBrown6b6204e2013-05-09 09:48:30 +10005235 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
NeilBrownf851b602014-12-11 10:02:10 +11005236 flush_workqueue(md_misc_wq);
NeilBrowna047e122010-03-29 12:07:53 +11005237 if (mddev->sync_thread) {
NeilBrowna047e122010-03-29 12:07:53 +11005238 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
Jonathan Brassowa91d5ac2013-04-24 11:42:43 +10005239 md_reap_sync_thread(mddev);
NeilBrowna047e122010-03-29 12:07:53 +11005240 }
5241
5242 del_timer_sync(&mddev->safemode_timer);
5243
5244 bitmap_flush(mddev);
5245 md_super_wait(mddev);
5246
NeilBrownb6d428c2013-04-24 11:42:42 +10005247 if (mddev->ro == 0 &&
NeilBrown7a0a5352013-08-27 16:28:23 +10005248 (!mddev->in_sync || (mddev->flags & MD_UPDATE_SB_FLAGS))) {
NeilBrowna047e122010-03-29 12:07:53 +11005249 /* mark array as shutdown cleanly */
5250 mddev->in_sync = 1;
5251 md_update_sb(mddev, 1);
5252 }
5253}
NeilBrowndefad612011-01-14 09:14:33 +11005254
NeilBrownfd01b882011-10-11 16:47:53 +11005255void md_stop_writes(struct mddev *mddev)
NeilBrowndefad612011-01-14 09:14:33 +11005256{
NeilBrown29f097c2013-11-14 17:54:51 +11005257 mddev_lock_nointr(mddev);
NeilBrowndefad612011-01-14 09:14:33 +11005258 __md_stop_writes(mddev);
5259 mddev_unlock(mddev);
5260}
NeilBrown390ee602010-06-01 19:37:27 +10005261EXPORT_SYMBOL_GPL(md_stop_writes);
NeilBrowna047e122010-03-29 12:07:53 +11005262
NeilBrown5aa61f42014-12-15 12:56:57 +11005263static void mddev_detach(struct mddev *mddev)
5264{
5265 struct bitmap *bitmap = mddev->bitmap;
5266 /* wait for behind writes to complete */
5267 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
5268 printk(KERN_INFO "md:%s: behind writes in progress - waiting to stop.\n",
5269 mdname(mddev));
5270 /* need to kick something here to make sure I/O goes? */
5271 wait_event(bitmap->behind_wait,
5272 atomic_read(&bitmap->behind_writes) == 0);
5273 }
NeilBrown36d091f2014-12-15 12:56:58 +11005274 if (mddev->pers && mddev->pers->quiesce) {
NeilBrown5aa61f42014-12-15 12:56:57 +11005275 mddev->pers->quiesce(mddev, 1);
5276 mddev->pers->quiesce(mddev, 0);
5277 }
5278 md_unregister_thread(&mddev->thread);
5279 if (mddev->queue)
5280 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
5281}
5282
NeilBrown5eff3c42012-11-19 10:47:48 +11005283static void __md_stop(struct mddev *mddev)
NeilBrown6177b472010-03-29 11:37:13 +11005284{
NeilBrown36d091f2014-12-15 12:56:58 +11005285 struct md_personality *pers = mddev->pers;
NeilBrown5aa61f42014-12-15 12:56:57 +11005286 mddev_detach(mddev);
NeilBrown36d091f2014-12-15 12:56:58 +11005287 spin_lock(&mddev->lock);
5288 mddev->ready = 0;
NeilBrown6177b472010-03-29 11:37:13 +11005289 mddev->pers = NULL;
NeilBrown36d091f2014-12-15 12:56:58 +11005290 spin_unlock(&mddev->lock);
5291 pers->free(mddev, mddev->private);
5292 if (pers->sync_request && mddev->to_remove == NULL)
5293 mddev->to_remove = &md_redundancy_group;
5294 module_put(pers->owner);
NeilBrowncca9cf92010-04-01 12:08:16 +11005295 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
NeilBrown6177b472010-03-29 11:37:13 +11005296}
NeilBrown5eff3c42012-11-19 10:47:48 +11005297
5298void md_stop(struct mddev *mddev)
5299{
5300 /* stop the array and free an attached data structures.
5301 * This is called from dm-raid
5302 */
5303 __md_stop(mddev);
5304 bitmap_destroy(mddev);
5305 if (mddev->bio_set)
5306 bioset_free(mddev->bio_set);
5307}
5308
NeilBrown390ee602010-06-01 19:37:27 +10005309EXPORT_SYMBOL_GPL(md_stop);
NeilBrown6177b472010-03-29 11:37:13 +11005310
NeilBrowna05b7ea2012-07-19 15:59:18 +10005311static int md_set_readonly(struct mddev *mddev, struct block_device *bdev)
NeilBrowna4bd82d2010-03-29 13:23:10 +11005312{
5313 int err = 0;
NeilBrown30b8feb2013-11-14 15:16:17 +11005314 int did_freeze = 0;
5315
5316 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5317 did_freeze = 1;
5318 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5319 md_wakeup_thread(mddev->thread);
5320 }
NeilBrownf851b602014-12-11 10:02:10 +11005321 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown30b8feb2013-11-14 15:16:17 +11005322 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrownf851b602014-12-11 10:02:10 +11005323 if (mddev->sync_thread)
NeilBrown30b8feb2013-11-14 15:16:17 +11005324 /* Thread might be blocked waiting for metadata update
5325 * which will now never happen */
5326 wake_up_process(mddev->sync_thread->tsk);
NeilBrownf851b602014-12-11 10:02:10 +11005327
NeilBrown30b8feb2013-11-14 15:16:17 +11005328 mddev_unlock(mddev);
NeilBrownf851b602014-12-11 10:02:10 +11005329 wait_event(resync_wait, !test_bit(MD_RECOVERY_RUNNING,
5330 &mddev->recovery));
NeilBrown30b8feb2013-11-14 15:16:17 +11005331 mddev_lock_nointr(mddev);
5332
NeilBrowna4bd82d2010-03-29 13:23:10 +11005333 mutex_lock(&mddev->open_mutex);
NeilBrown9ba3b7f2014-09-09 14:00:15 +10005334 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
NeilBrown30b8feb2013-11-14 15:16:17 +11005335 mddev->sync_thread ||
NeilBrownf851b602014-12-11 10:02:10 +11005336 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
NeilBrown30b8feb2013-11-14 15:16:17 +11005337 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
NeilBrowna4bd82d2010-03-29 13:23:10 +11005338 printk("md: %s still in use.\n",mdname(mddev));
NeilBrown30b8feb2013-11-14 15:16:17 +11005339 if (did_freeze) {
5340 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
NeilBrown45eaf452014-10-29 08:49:50 +11005341 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrown30b8feb2013-11-14 15:16:17 +11005342 md_wakeup_thread(mddev->thread);
5343 }
NeilBrowna4bd82d2010-03-29 13:23:10 +11005344 err = -EBUSY;
5345 goto out;
5346 }
5347 if (mddev->pers) {
NeilBrowndefad612011-01-14 09:14:33 +11005348 __md_stop_writes(mddev);
NeilBrowna4bd82d2010-03-29 13:23:10 +11005349
5350 err = -ENXIO;
5351 if (mddev->ro==1)
5352 goto out;
5353 mddev->ro = 1;
5354 set_disk_ro(mddev->gendisk, 1);
5355 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
NeilBrown45eaf452014-10-29 08:49:50 +11005356 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5357 md_wakeup_thread(mddev->thread);
NeilBrown00bcb4a2010-06-01 19:37:23 +10005358 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrown30b8feb2013-11-14 15:16:17 +11005359 err = 0;
NeilBrowna4bd82d2010-03-29 13:23:10 +11005360 }
5361out:
5362 mutex_unlock(&mddev->open_mutex);
5363 return err;
5364}
5365
NeilBrown9e653b62006-06-26 00:27:58 -07005366/* mode:
5367 * 0 - completely stop and dis-assemble array
NeilBrown9e653b62006-06-26 00:27:58 -07005368 * 2 - stop but do not disassemble array
5369 */
NeilBrownf72ffdd2014-09-30 14:23:59 +10005370static int do_md_stop(struct mddev *mddev, int mode,
NeilBrowna05b7ea2012-07-19 15:59:18 +10005371 struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005372{
Linus Torvalds1da177e2005-04-16 15:20:36 -07005373 struct gendisk *disk = mddev->gendisk;
NeilBrown3cb03002011-10-11 16:45:26 +11005374 struct md_rdev *rdev;
NeilBrown30b8feb2013-11-14 15:16:17 +11005375 int did_freeze = 0;
5376
5377 if (!test_bit(MD_RECOVERY_FROZEN, &mddev->recovery)) {
5378 did_freeze = 1;
5379 set_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
5380 md_wakeup_thread(mddev->thread);
5381 }
NeilBrownf851b602014-12-11 10:02:10 +11005382 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery))
NeilBrown30b8feb2013-11-14 15:16:17 +11005383 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrownf851b602014-12-11 10:02:10 +11005384 if (mddev->sync_thread)
NeilBrown30b8feb2013-11-14 15:16:17 +11005385 /* Thread might be blocked waiting for metadata update
5386 * which will now never happen */
5387 wake_up_process(mddev->sync_thread->tsk);
NeilBrownf851b602014-12-11 10:02:10 +11005388
NeilBrown30b8feb2013-11-14 15:16:17 +11005389 mddev_unlock(mddev);
NeilBrownf851b602014-12-11 10:02:10 +11005390 wait_event(resync_wait, (mddev->sync_thread == NULL &&
5391 !test_bit(MD_RECOVERY_RUNNING,
5392 &mddev->recovery)));
NeilBrown30b8feb2013-11-14 15:16:17 +11005393 mddev_lock_nointr(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005394
NeilBrownc8c00a62009-08-10 12:50:52 +10005395 mutex_lock(&mddev->open_mutex);
NeilBrown9ba3b7f2014-09-09 14:00:15 +10005396 if ((mddev->pers && atomic_read(&mddev->openers) > !!bdev) ||
NeilBrown30b8feb2013-11-14 15:16:17 +11005397 mddev->sysfs_active ||
5398 mddev->sync_thread ||
NeilBrownf851b602014-12-11 10:02:10 +11005399 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
NeilBrown30b8feb2013-11-14 15:16:17 +11005400 (bdev && !test_bit(MD_STILL_CLOSED, &mddev->flags))) {
Neil Browndf5b20c2008-07-11 22:02:22 +10005401 printk("md: %s still in use.\n",mdname(mddev));
NeilBrown6e17b022010-08-07 21:41:19 +10005402 mutex_unlock(&mddev->open_mutex);
NeilBrown30b8feb2013-11-14 15:16:17 +11005403 if (did_freeze) {
5404 clear_bit(MD_RECOVERY_FROZEN, &mddev->recovery);
NeilBrown45eaf452014-10-29 08:49:50 +11005405 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrown30b8feb2013-11-14 15:16:17 +11005406 md_wakeup_thread(mddev->thread);
5407 }
NeilBrown260fa032013-08-27 16:44:13 +10005408 return -EBUSY;
5409 }
NeilBrown6e17b022010-08-07 21:41:19 +10005410 if (mddev->pers) {
NeilBrowna4bd82d2010-03-29 13:23:10 +11005411 if (mddev->ro)
5412 set_disk_ro(disk, 0);
NeilBrown409c57f2009-03-31 14:39:39 +11005413
NeilBrowndefad612011-01-14 09:14:33 +11005414 __md_stop_writes(mddev);
NeilBrown5eff3c42012-11-19 10:47:48 +11005415 __md_stop(mddev);
NeilBrowna4bd82d2010-03-29 13:23:10 +11005416 mddev->queue->merge_bvec_fn = NULL;
NeilBrowna4bd82d2010-03-29 13:23:10 +11005417 mddev->queue->backing_dev_info.congested_fn = NULL;
NeilBrown6177b472010-03-29 11:37:13 +11005418
NeilBrowna4bd82d2010-03-29 13:23:10 +11005419 /* tell userspace to handle 'inactive' */
NeilBrown00bcb4a2010-06-01 19:37:23 +10005420 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrown0d4ca602006-12-10 02:20:44 -08005421
NeilBrowndafb20f2012-03-19 12:46:39 +11005422 rdev_for_each(rdev, mddev)
Namhyung Kim36fad852011-07-27 11:00:36 +10005423 if (rdev->raid_disk >= 0)
5424 sysfs_unlink_rdev(mddev, rdev);
NeilBrownc4647292009-05-07 12:51:06 +10005425
NeilBrowna4bd82d2010-03-29 13:23:10 +11005426 set_capacity(disk, 0);
NeilBrown6e17b022010-08-07 21:41:19 +10005427 mutex_unlock(&mddev->open_mutex);
NeilBrownf0b4f7e2011-02-24 17:26:41 +11005428 mddev->changed = 1;
NeilBrowna4bd82d2010-03-29 13:23:10 +11005429 revalidate_disk(disk);
NeilBrown0d4ca602006-12-10 02:20:44 -08005430
NeilBrowna4bd82d2010-03-29 13:23:10 +11005431 if (mddev->ro)
5432 mddev->ro = 0;
NeilBrown6e17b022010-08-07 21:41:19 +10005433 } else
5434 mutex_unlock(&mddev->open_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005435 /*
5436 * Free resources if final stop
5437 */
NeilBrown9e653b62006-06-26 00:27:58 -07005438 if (mode == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005439 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
5440
NeilBrown978f9462006-02-02 14:28:05 -08005441 bitmap_destroy(mddev);
NeilBrownc3d97142009-12-14 12:49:52 +11005442 if (mddev->bitmap_info.file) {
NeilBrown4af1a042014-12-15 12:57:00 +11005443 struct file *f = mddev->bitmap_info.file;
5444 spin_lock(&mddev->lock);
NeilBrownc3d97142009-12-14 12:49:52 +11005445 mddev->bitmap_info.file = NULL;
NeilBrown4af1a042014-12-15 12:57:00 +11005446 spin_unlock(&mddev->lock);
5447 fput(f);
NeilBrown978f9462006-02-02 14:28:05 -08005448 }
NeilBrownc3d97142009-12-14 12:49:52 +11005449 mddev->bitmap_info.offset = 0;
NeilBrown978f9462006-02-02 14:28:05 -08005450
Linus Torvalds1da177e2005-04-16 15:20:36 -07005451 export_array(mddev);
5452
NeilBrown6177b472010-03-29 11:37:13 +11005453 md_clean(mddev);
NeilBrown934d9c22008-10-28 17:01:23 +11005454 kobject_uevent(&disk_to_dev(mddev->gendisk)->kobj, KOBJ_CHANGE);
NeilBrownefeb53c2009-01-09 08:31:10 +11005455 if (mddev->hold_active == UNTIL_STOP)
5456 mddev->hold_active = 0;
NeilBrowna4bd82d2010-03-29 13:23:10 +11005457 }
Martin K. Petersen3f9d99c2009-03-31 14:27:02 +11005458 blk_integrity_unregister(disk);
NeilBrownd7603b72006-01-06 00:20:30 -08005459 md_new_event(mddev);
NeilBrown00bcb4a2010-06-01 19:37:23 +10005460 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrown6e17b022010-08-07 21:41:19 +10005461 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005462}
5463
Jeff Garzikfdee8ae2006-12-10 02:20:50 -08005464#ifndef MODULE
NeilBrownfd01b882011-10-11 16:47:53 +11005465static void autorun_array(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005466{
NeilBrown3cb03002011-10-11 16:45:26 +11005467 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005468 int err;
5469
NeilBrowna757e642005-04-16 15:26:42 -07005470 if (list_empty(&mddev->disks))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005471 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005472
5473 printk(KERN_INFO "md: running: ");
5474
NeilBrowndafb20f2012-03-19 12:46:39 +11005475 rdev_for_each(rdev, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005476 char b[BDEVNAME_SIZE];
5477 printk("<%s>", bdevname(rdev->bdev,b));
5478 }
5479 printk("\n");
5480
NeilBrownd710e132008-10-13 11:55:12 +11005481 err = do_md_run(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005482 if (err) {
5483 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
NeilBrowna05b7ea2012-07-19 15:59:18 +10005484 do_md_stop(mddev, 0, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005485 }
5486}
5487
5488/*
5489 * lets try to run arrays based on all disks that have arrived
5490 * until now. (those are in pending_raid_disks)
5491 *
5492 * the method: pick the first pending disk, collect all disks with
5493 * the same UUID, remove all from the pending list and put them into
5494 * the 'same_array' list. Then order this list based on superblock
5495 * update time (freshest comes first), kick out 'old' disks and
5496 * compare superblocks. If everything's fine then run it.
5497 *
5498 * If "unit" is allocated, then bump its reference count
5499 */
5500static void autorun_devices(int part)
5501{
NeilBrown3cb03002011-10-11 16:45:26 +11005502 struct md_rdev *rdev0, *rdev, *tmp;
NeilBrownfd01b882011-10-11 16:47:53 +11005503 struct mddev *mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005504 char b[BDEVNAME_SIZE];
5505
5506 printk(KERN_INFO "md: autorun ...\n");
5507 while (!list_empty(&pending_raid_disks)) {
NeilBrowne8703fe2006-10-03 01:15:59 -07005508 int unit;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005509 dev_t dev;
NeilBrownad01c9e2006-03-27 01:18:07 -08005510 LIST_HEAD(candidates);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005511 rdev0 = list_entry(pending_raid_disks.next,
NeilBrown3cb03002011-10-11 16:45:26 +11005512 struct md_rdev, same_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005513
5514 printk(KERN_INFO "md: considering %s ...\n",
5515 bdevname(rdev0->bdev,b));
5516 INIT_LIST_HEAD(&candidates);
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005517 rdev_for_each_list(rdev, tmp, &pending_raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005518 if (super_90_load(rdev, rdev0, 0) >= 0) {
5519 printk(KERN_INFO "md: adding %s ...\n",
5520 bdevname(rdev->bdev,b));
5521 list_move(&rdev->same_set, &candidates);
5522 }
5523 /*
5524 * now we have a set of devices, with all of them having
5525 * mostly sane superblocks. It's time to allocate the
5526 * mddev.
5527 */
NeilBrowne8703fe2006-10-03 01:15:59 -07005528 if (part) {
5529 dev = MKDEV(mdp_major,
5530 rdev0->preferred_minor << MdpMinorShift);
5531 unit = MINOR(dev) >> MdpMinorShift;
5532 } else {
5533 dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
5534 unit = MINOR(dev);
5535 }
5536 if (rdev0->preferred_minor != unit) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005537 printk(KERN_INFO "md: unit number in %s is bad: %d\n",
5538 bdevname(rdev0->bdev, b), rdev0->preferred_minor);
5539 break;
5540 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005541
5542 md_probe(dev, NULL, NULL);
5543 mddev = mddev_find(dev);
Neil Brown9bbbca32008-06-28 08:31:17 +10005544 if (!mddev || !mddev->gendisk) {
5545 if (mddev)
5546 mddev_put(mddev);
5547 printk(KERN_ERR
Linus Torvalds1da177e2005-04-16 15:20:36 -07005548 "md: cannot allocate memory for md drive.\n");
5549 break;
5550 }
NeilBrownf72ffdd2014-09-30 14:23:59 +10005551 if (mddev_lock(mddev))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005552 printk(KERN_WARNING "md: %s locked, cannot run\n",
5553 mdname(mddev));
5554 else if (mddev->raid_disks || mddev->major_version
5555 || !list_empty(&mddev->disks)) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10005556 printk(KERN_WARNING
Linus Torvalds1da177e2005-04-16 15:20:36 -07005557 "md: %s already running, cannot run %s\n",
5558 mdname(mddev), bdevname(rdev0->bdev,b));
5559 mddev_unlock(mddev);
5560 } else {
5561 printk(KERN_INFO "md: created %s\n", mdname(mddev));
NeilBrown1ec4a932008-02-06 01:39:53 -08005562 mddev->persistent = 1;
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005563 rdev_for_each_list(rdev, tmp, &candidates) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005564 list_del_init(&rdev->same_set);
5565 if (bind_rdev_to_array(rdev, mddev))
5566 export_rdev(rdev);
5567 }
5568 autorun_array(mddev);
5569 mddev_unlock(mddev);
5570 }
5571 /* on success, candidates will be empty, on error
5572 * it won't...
5573 */
Cheng Renquan159ec1f2009-01-09 08:31:08 +11005574 rdev_for_each_list(rdev, tmp, &candidates) {
NeilBrown4b809912008-07-21 17:05:25 +10005575 list_del_init(&rdev->same_set);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005576 export_rdev(rdev);
NeilBrown4b809912008-07-21 17:05:25 +10005577 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005578 mddev_put(mddev);
5579 }
5580 printk(KERN_INFO "md: ... autorun DONE.\n");
5581}
Jeff Garzikfdee8ae2006-12-10 02:20:50 -08005582#endif /* !MODULE */
Linus Torvalds1da177e2005-04-16 15:20:36 -07005583
NeilBrownf72ffdd2014-09-30 14:23:59 +10005584static int get_version(void __user *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005585{
5586 mdu_version_t ver;
5587
5588 ver.major = MD_MAJOR_VERSION;
5589 ver.minor = MD_MINOR_VERSION;
5590 ver.patchlevel = MD_PATCHLEVEL_VERSION;
5591
5592 if (copy_to_user(arg, &ver, sizeof(ver)))
5593 return -EFAULT;
5594
5595 return 0;
5596}
5597
NeilBrownf72ffdd2014-09-30 14:23:59 +10005598static int get_array_info(struct mddev *mddev, void __user *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005599{
5600 mdu_array_info_t info;
NeilBrowna9f326e2009-09-23 18:06:41 +10005601 int nr,working,insync,failed,spare;
NeilBrown3cb03002011-10-11 16:45:26 +11005602 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005603
NeilBrown1ca69c42012-10-11 13:37:33 +11005604 nr = working = insync = failed = spare = 0;
5605 rcu_read_lock();
5606 rdev_for_each_rcu(rdev, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005607 nr++;
NeilBrownb2d444d2005-11-08 21:39:31 -08005608 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005609 failed++;
5610 else {
5611 working++;
NeilBrownb2d444d2005-11-08 21:39:31 -08005612 if (test_bit(In_sync, &rdev->flags))
NeilBrownf72ffdd2014-09-30 14:23:59 +10005613 insync++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005614 else
5615 spare++;
5616 }
5617 }
NeilBrown1ca69c42012-10-11 13:37:33 +11005618 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005619
5620 info.major_version = mddev->major_version;
5621 info.minor_version = mddev->minor_version;
5622 info.patch_version = MD_PATCHLEVEL_VERSION;
5623 info.ctime = mddev->ctime;
5624 info.level = mddev->level;
Andre Noll58c0fed2009-03-31 14:33:13 +11005625 info.size = mddev->dev_sectors / 2;
5626 if (info.size != mddev->dev_sectors / 2) /* overflow */
NeilBrown284ae7c2006-02-03 03:03:40 -08005627 info.size = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005628 info.nr_disks = nr;
5629 info.raid_disks = mddev->raid_disks;
5630 info.md_minor = mddev->md_minor;
5631 info.not_persistent= !mddev->persistent;
5632
5633 info.utime = mddev->utime;
5634 info.state = 0;
5635 if (mddev->in_sync)
5636 info.state = (1<<MD_SB_CLEAN);
NeilBrownc3d97142009-12-14 12:49:52 +11005637 if (mddev->bitmap && mddev->bitmap_info.offset)
NeilBrown9bd35922014-07-02 11:35:06 +10005638 info.state |= (1<<MD_SB_BITMAP_PRESENT);
NeilBrowna9f326e2009-09-23 18:06:41 +10005639 info.active_disks = insync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005640 info.working_disks = working;
5641 info.failed_disks = failed;
5642 info.spare_disks = spare;
5643
5644 info.layout = mddev->layout;
Andre Noll9d8f0362009-06-18 08:45:01 +10005645 info.chunk_size = mddev->chunk_sectors << 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005646
5647 if (copy_to_user(arg, &info, sizeof(info)))
5648 return -EFAULT;
5649
5650 return 0;
5651}
5652
NeilBrownf72ffdd2014-09-30 14:23:59 +10005653static int get_bitmap_file(struct mddev *mddev, void __user * arg)
NeilBrown32a76272005-06-21 17:17:14 -07005654{
5655 mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
NeilBrownf4ad3d32014-12-15 12:57:00 +11005656 char *ptr;
NeilBrown4af1a042014-12-15 12:57:00 +11005657 int err;
NeilBrown32a76272005-06-21 17:17:14 -07005658
NeilBrown60559da2013-07-16 16:44:29 +10005659 file = kmalloc(sizeof(*file), GFP_NOIO);
NeilBrown32a76272005-06-21 17:17:14 -07005660 if (!file)
NeilBrown4af1a042014-12-15 12:57:00 +11005661 return -ENOMEM;
NeilBrown32a76272005-06-21 17:17:14 -07005662
NeilBrown32a76272005-06-21 17:17:14 -07005663 err = 0;
NeilBrown4af1a042014-12-15 12:57:00 +11005664 spin_lock(&mddev->lock);
5665 /* bitmap disabled, zero the first byte and copy out */
5666 if (!mddev->bitmap_info.file)
5667 file->pathname[0] = '\0';
5668 else if ((ptr = d_path(&mddev->bitmap_info.file->f_path,
5669 file->pathname, sizeof(file->pathname))),
5670 IS_ERR(ptr))
5671 err = PTR_ERR(ptr);
5672 else
5673 memmove(file->pathname, ptr,
5674 sizeof(file->pathname)-(ptr-file->pathname));
5675 spin_unlock(&mddev->lock);
5676
5677 if (err == 0 &&
5678 copy_to_user(arg, file, sizeof(*file)))
NeilBrown32a76272005-06-21 17:17:14 -07005679 err = -EFAULT;
NeilBrown4af1a042014-12-15 12:57:00 +11005680
NeilBrown32a76272005-06-21 17:17:14 -07005681 kfree(file);
5682 return err;
5683}
5684
NeilBrownf72ffdd2014-09-30 14:23:59 +10005685static int get_disk_info(struct mddev *mddev, void __user * arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005686{
5687 mdu_disk_info_t info;
NeilBrown3cb03002011-10-11 16:45:26 +11005688 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005689
5690 if (copy_from_user(&info, arg, sizeof(info)))
5691 return -EFAULT;
5692
NeilBrown1ca69c42012-10-11 13:37:33 +11005693 rcu_read_lock();
5694 rdev = find_rdev_nr_rcu(mddev, info.number);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005695 if (rdev) {
5696 info.major = MAJOR(rdev->bdev->bd_dev);
5697 info.minor = MINOR(rdev->bdev->bd_dev);
5698 info.raid_disk = rdev->raid_disk;
5699 info.state = 0;
NeilBrownb2d444d2005-11-08 21:39:31 -08005700 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07005701 info.state |= (1<<MD_DISK_FAULTY);
NeilBrownb2d444d2005-11-08 21:39:31 -08005702 else if (test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07005703 info.state |= (1<<MD_DISK_ACTIVE);
5704 info.state |= (1<<MD_DISK_SYNC);
5705 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -07005706 if (test_bit(WriteMostly, &rdev->flags))
5707 info.state |= (1<<MD_DISK_WRITEMOSTLY);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005708 } else {
5709 info.major = info.minor = 0;
5710 info.raid_disk = -1;
5711 info.state = (1<<MD_DISK_REMOVED);
5712 }
NeilBrown1ca69c42012-10-11 13:37:33 +11005713 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07005714
5715 if (copy_to_user(arg, &info, sizeof(info)))
5716 return -EFAULT;
5717
5718 return 0;
5719}
5720
NeilBrownf72ffdd2014-09-30 14:23:59 +10005721static int add_new_disk(struct mddev *mddev, mdu_disk_info_t *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005722{
5723 char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11005724 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005725 dev_t dev = MKDEV(info->major,info->minor);
5726
5727 if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
5728 return -EOVERFLOW;
5729
5730 if (!mddev->raid_disks) {
5731 int err;
5732 /* expecting a device which has a superblock */
5733 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
5734 if (IS_ERR(rdev)) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10005735 printk(KERN_WARNING
Linus Torvalds1da177e2005-04-16 15:20:36 -07005736 "md: md_import_device returned %ld\n",
5737 PTR_ERR(rdev));
5738 return PTR_ERR(rdev);
5739 }
5740 if (!list_empty(&mddev->disks)) {
NeilBrown3cb03002011-10-11 16:45:26 +11005741 struct md_rdev *rdev0
5742 = list_entry(mddev->disks.next,
5743 struct md_rdev, same_set);
NeilBrowna9f326e2009-09-23 18:06:41 +10005744 err = super_types[mddev->major_version]
Linus Torvalds1da177e2005-04-16 15:20:36 -07005745 .load_super(rdev, rdev0, mddev->minor_version);
5746 if (err < 0) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10005747 printk(KERN_WARNING
Linus Torvalds1da177e2005-04-16 15:20:36 -07005748 "md: %s has different UUID to %s\n",
NeilBrownf72ffdd2014-09-30 14:23:59 +10005749 bdevname(rdev->bdev,b),
Linus Torvalds1da177e2005-04-16 15:20:36 -07005750 bdevname(rdev0->bdev,b2));
5751 export_rdev(rdev);
5752 return -EINVAL;
5753 }
5754 }
5755 err = bind_rdev_to_array(rdev, mddev);
5756 if (err)
5757 export_rdev(rdev);
5758 return err;
5759 }
5760
5761 /*
5762 * add_new_disk can be used once the array is assembled
5763 * to add "hot spares". They must already have a superblock
5764 * written
5765 */
5766 if (mddev->pers) {
5767 int err;
5768 if (!mddev->pers->hot_add_disk) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10005769 printk(KERN_WARNING
Linus Torvalds1da177e2005-04-16 15:20:36 -07005770 "%s: personality does not support diskops!\n",
5771 mdname(mddev));
5772 return -EINVAL;
5773 }
NeilBrown7b1e35f2005-09-09 16:23:50 -07005774 if (mddev->persistent)
5775 rdev = md_import_device(dev, mddev->major_version,
5776 mddev->minor_version);
5777 else
5778 rdev = md_import_device(dev, -1, -1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005779 if (IS_ERR(rdev)) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10005780 printk(KERN_WARNING
Linus Torvalds1da177e2005-04-16 15:20:36 -07005781 "md: md_import_device returned %ld\n",
5782 PTR_ERR(rdev));
5783 return PTR_ERR(rdev);
5784 }
NeilBrown1a855a02010-12-09 16:36:28 +11005785 /* set saved_raid_disk if appropriate */
NeilBrown41158c72005-06-21 17:17:25 -07005786 if (!mddev->persistent) {
5787 if (info->state & (1<<MD_DISK_SYNC) &&
NeilBrownbf572542011-01-12 09:03:35 +11005788 info->raid_disk < mddev->raid_disks) {
NeilBrown41158c72005-06-21 17:17:25 -07005789 rdev->raid_disk = info->raid_disk;
NeilBrownbf572542011-01-12 09:03:35 +11005790 set_bit(In_sync, &rdev->flags);
NeilBrown8313b8e2013-12-12 10:13:33 +11005791 clear_bit(Bitmap_sync, &rdev->flags);
NeilBrownbf572542011-01-12 09:03:35 +11005792 } else
NeilBrown41158c72005-06-21 17:17:25 -07005793 rdev->raid_disk = -1;
NeilBrownf4667222013-12-09 12:04:56 +11005794 rdev->saved_raid_disk = rdev->raid_disk;
NeilBrown41158c72005-06-21 17:17:25 -07005795 } else
5796 super_types[mddev->major_version].
5797 validate_super(mddev, rdev);
NeilBrownbedd86b2011-05-11 14:26:20 +10005798 if ((info->state & (1<<MD_DISK_SYNC)) &&
NeilBrownf4563092012-07-03 15:59:06 +10005799 rdev->raid_disk != info->raid_disk) {
NeilBrownbedd86b2011-05-11 14:26:20 +10005800 /* This was a hot-add request, but events doesn't
5801 * match, so reject it.
5802 */
5803 export_rdev(rdev);
5804 return -EINVAL;
5805 }
5806
NeilBrownb2d444d2005-11-08 21:39:31 -08005807 clear_bit(In_sync, &rdev->flags); /* just to be sure */
NeilBrown8ddf9ef2005-09-09 16:23:45 -07005808 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5809 set_bit(WriteMostly, &rdev->flags);
NeilBrown575a80f2009-03-31 14:33:13 +11005810 else
5811 clear_bit(WriteMostly, &rdev->flags);
NeilBrown8ddf9ef2005-09-09 16:23:45 -07005812
Linus Torvalds1da177e2005-04-16 15:20:36 -07005813 rdev->raid_disk = -1;
5814 err = bind_rdev_to_array(rdev, mddev);
NeilBrown7c7546c2006-06-26 00:27:41 -07005815 if (!err && !mddev->pers->hot_remove_disk) {
5816 /* If there is hot_add_disk but no hot_remove_disk
5817 * then added disks for geometry changes,
5818 * and should be added immediately.
5819 */
5820 super_types[mddev->major_version].
5821 validate_super(mddev, rdev);
5822 err = mddev->pers->hot_add_disk(mddev, rdev);
5823 if (err)
5824 unbind_rdev_from_array(rdev);
5825 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005826 if (err)
5827 export_rdev(rdev);
Neil Brown52664732008-06-28 08:31:44 +10005828 else
NeilBrown00bcb4a2010-06-01 19:37:23 +10005829 sysfs_notify_dirent_safe(rdev->sysfs_state);
NeilBrownc3617772005-06-21 17:17:10 -07005830
NeilBrown7ceb17e2013-04-24 11:42:42 +10005831 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Neil Brown72a23c22008-06-28 08:31:41 +10005832 if (mddev->degraded)
5833 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
NeilBrownc3617772005-06-21 17:17:10 -07005834 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
马建朋9864c002011-06-09 11:42:48 +10005835 if (!err)
5836 md_new_event(mddev);
NeilBrown005eca52005-08-22 13:11:08 -07005837 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005838 return err;
5839 }
5840
5841 /* otherwise, add_new_disk is only allowed
5842 * for major_version==0 superblocks
5843 */
5844 if (mddev->major_version != 0) {
5845 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
5846 mdname(mddev));
5847 return -EINVAL;
5848 }
5849
5850 if (!(info->state & (1<<MD_DISK_FAULTY))) {
5851 int err;
NeilBrownd710e132008-10-13 11:55:12 +11005852 rdev = md_import_device(dev, -1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005853 if (IS_ERR(rdev)) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10005854 printk(KERN_WARNING
Linus Torvalds1da177e2005-04-16 15:20:36 -07005855 "md: error, md_import_device() returned %ld\n",
5856 PTR_ERR(rdev));
5857 return PTR_ERR(rdev);
5858 }
5859 rdev->desc_nr = info->number;
5860 if (info->raid_disk < mddev->raid_disks)
5861 rdev->raid_disk = info->raid_disk;
5862 else
5863 rdev->raid_disk = -1;
5864
Linus Torvalds1da177e2005-04-16 15:20:36 -07005865 if (rdev->raid_disk < mddev->raid_disks)
NeilBrownb2d444d2005-11-08 21:39:31 -08005866 if (info->state & (1<<MD_DISK_SYNC))
5867 set_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005868
NeilBrown8ddf9ef2005-09-09 16:23:45 -07005869 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
5870 set_bit(WriteMostly, &rdev->flags);
5871
Linus Torvalds1da177e2005-04-16 15:20:36 -07005872 if (!mddev->persistent) {
5873 printk(KERN_INFO "md: nonpersistent superblock ...\n");
Mike Snitzer77304d22010-11-08 14:39:12 +01005874 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
5875 } else
Jonathan Brassow57b2caa2011-01-14 09:14:33 +11005876 rdev->sb_start = calc_dev_sboffset(rdev);
NeilBrown8190e752009-06-18 08:48:58 +10005877 rdev->sectors = rdev->sb_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005878
NeilBrown2bf071b2006-01-06 00:20:55 -08005879 err = bind_rdev_to_array(rdev, mddev);
5880 if (err) {
5881 export_rdev(rdev);
5882 return err;
5883 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07005884 }
5885
5886 return 0;
5887}
5888
NeilBrownf72ffdd2014-09-30 14:23:59 +10005889static int hot_remove_disk(struct mddev *mddev, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005890{
5891 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11005892 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005893
Linus Torvalds1da177e2005-04-16 15:20:36 -07005894 rdev = find_rdev(mddev, dev);
5895 if (!rdev)
5896 return -ENXIO;
5897
NeilBrown3ea8929d2013-04-24 11:42:41 +10005898 clear_bit(Blocked, &rdev->flags);
5899 remove_and_add_spares(mddev, rdev);
5900
Linus Torvalds1da177e2005-04-16 15:20:36 -07005901 if (rdev->raid_disk >= 0)
5902 goto busy;
5903
5904 kick_rdev_from_array(rdev);
NeilBrown850b2b42006-10-03 01:15:46 -07005905 md_update_sb(mddev, 1);
NeilBrownd7603b72006-01-06 00:20:30 -08005906 md_new_event(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005907
5908 return 0;
5909busy:
Nick Andrewfdefa4d2008-04-21 22:42:58 +00005910 printk(KERN_WARNING "md: cannot remove active disk %s from %s ...\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07005911 bdevname(rdev->bdev,b), mdname(mddev));
5912 return -EBUSY;
5913}
5914
NeilBrownf72ffdd2014-09-30 14:23:59 +10005915static int hot_add_disk(struct mddev *mddev, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07005916{
5917 char b[BDEVNAME_SIZE];
5918 int err;
NeilBrown3cb03002011-10-11 16:45:26 +11005919 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005920
5921 if (!mddev->pers)
5922 return -ENODEV;
5923
5924 if (mddev->major_version != 0) {
5925 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
5926 " version-0 superblocks.\n",
5927 mdname(mddev));
5928 return -EINVAL;
5929 }
5930 if (!mddev->pers->hot_add_disk) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10005931 printk(KERN_WARNING
Linus Torvalds1da177e2005-04-16 15:20:36 -07005932 "%s: personality does not support diskops!\n",
5933 mdname(mddev));
5934 return -EINVAL;
5935 }
5936
NeilBrownd710e132008-10-13 11:55:12 +11005937 rdev = md_import_device(dev, -1, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005938 if (IS_ERR(rdev)) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10005939 printk(KERN_WARNING
Linus Torvalds1da177e2005-04-16 15:20:36 -07005940 "md: error, md_import_device() returned %ld\n",
5941 PTR_ERR(rdev));
5942 return -EINVAL;
5943 }
5944
5945 if (mddev->persistent)
Jonathan Brassow57b2caa2011-01-14 09:14:33 +11005946 rdev->sb_start = calc_dev_sboffset(rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005947 else
Mike Snitzer77304d22010-11-08 14:39:12 +01005948 rdev->sb_start = i_size_read(rdev->bdev->bd_inode) / 512;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005949
NeilBrown8190e752009-06-18 08:48:58 +10005950 rdev->sectors = rdev->sb_start;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005951
NeilBrownb2d444d2005-11-08 21:39:31 -08005952 if (test_bit(Faulty, &rdev->flags)) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10005953 printk(KERN_WARNING
Linus Torvalds1da177e2005-04-16 15:20:36 -07005954 "md: can not hot-add faulty %s disk to %s!\n",
5955 bdevname(rdev->bdev,b), mdname(mddev));
5956 err = -EINVAL;
5957 goto abort_export;
5958 }
NeilBrownb2d444d2005-11-08 21:39:31 -08005959 clear_bit(In_sync, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005960 rdev->desc_nr = -1;
NeilBrown58427302006-10-06 00:44:04 -07005961 rdev->saved_raid_disk = -1;
NeilBrown2bf071b2006-01-06 00:20:55 -08005962 err = bind_rdev_to_array(rdev, mddev);
5963 if (err)
5964 goto abort_export;
Linus Torvalds1da177e2005-04-16 15:20:36 -07005965
5966 /*
5967 * The rest should better be atomic, we can have disk failures
5968 * noticed in interrupt contexts ...
5969 */
5970
Linus Torvalds1da177e2005-04-16 15:20:36 -07005971 rdev->raid_disk = -1;
5972
NeilBrown850b2b42006-10-03 01:15:46 -07005973 md_update_sb(mddev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005974
5975 /*
5976 * Kick recovery, maybe this spare has to be added to the
5977 * array immediately.
5978 */
5979 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
5980 md_wakeup_thread(mddev->thread);
NeilBrownd7603b72006-01-06 00:20:30 -08005981 md_new_event(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07005982 return 0;
5983
Linus Torvalds1da177e2005-04-16 15:20:36 -07005984abort_export:
5985 export_rdev(rdev);
5986 return err;
5987}
5988
NeilBrownfd01b882011-10-11 16:47:53 +11005989static int set_bitmap_file(struct mddev *mddev, int fd)
NeilBrown32a76272005-06-21 17:17:14 -07005990{
NeilBrown035328c2014-04-09 12:25:40 +10005991 int err = 0;
NeilBrown32a76272005-06-21 17:17:14 -07005992
NeilBrown36fa3062005-09-09 16:23:45 -07005993 if (mddev->pers) {
NeilBrownd66b1b32014-08-08 15:40:24 +10005994 if (!mddev->pers->quiesce || !mddev->thread)
NeilBrown36fa3062005-09-09 16:23:45 -07005995 return -EBUSY;
5996 if (mddev->recovery || mddev->sync_thread)
5997 return -EBUSY;
5998 /* we should be able to change the bitmap.. */
NeilBrown32a76272005-06-21 17:17:14 -07005999 }
6000
NeilBrown36fa3062005-09-09 16:23:45 -07006001 if (fd >= 0) {
NeilBrown035328c2014-04-09 12:25:40 +10006002 struct inode *inode;
NeilBrown1e594bb2014-12-15 12:57:00 +11006003 struct file *f;
NeilBrown36fa3062005-09-09 16:23:45 -07006004
NeilBrown1e594bb2014-12-15 12:57:00 +11006005 if (mddev->bitmap || mddev->bitmap_info.file)
6006 return -EEXIST; /* cannot add when bitmap is present */
6007 f = fget(fd);
6008
6009 if (f == NULL) {
NeilBrown36fa3062005-09-09 16:23:45 -07006010 printk(KERN_ERR "%s: error: failed to get bitmap file\n",
6011 mdname(mddev));
6012 return -EBADF;
6013 }
6014
NeilBrown1e594bb2014-12-15 12:57:00 +11006015 inode = f->f_mapping->host;
NeilBrown035328c2014-04-09 12:25:40 +10006016 if (!S_ISREG(inode->i_mode)) {
6017 printk(KERN_ERR "%s: error: bitmap file must be a regular file\n",
6018 mdname(mddev));
6019 err = -EBADF;
NeilBrown1e594bb2014-12-15 12:57:00 +11006020 } else if (!(f->f_mode & FMODE_WRITE)) {
NeilBrown035328c2014-04-09 12:25:40 +10006021 printk(KERN_ERR "%s: error: bitmap file must open for write\n",
6022 mdname(mddev));
6023 err = -EBADF;
6024 } else if (atomic_read(&inode->i_writecount) != 1) {
NeilBrown36fa3062005-09-09 16:23:45 -07006025 printk(KERN_ERR "%s: error: bitmap file is already in use\n",
6026 mdname(mddev));
NeilBrown035328c2014-04-09 12:25:40 +10006027 err = -EBUSY;
6028 }
6029 if (err) {
NeilBrown1e594bb2014-12-15 12:57:00 +11006030 fput(f);
NeilBrown36fa3062005-09-09 16:23:45 -07006031 return err;
6032 }
NeilBrown1e594bb2014-12-15 12:57:00 +11006033 mddev->bitmap_info.file = f;
NeilBrownc3d97142009-12-14 12:49:52 +11006034 mddev->bitmap_info.offset = 0; /* file overrides offset */
NeilBrown36fa3062005-09-09 16:23:45 -07006035 } else if (mddev->bitmap == NULL)
6036 return -ENOENT; /* cannot remove what isn't there */
6037 err = 0;
6038 if (mddev->pers) {
6039 mddev->pers->quiesce(mddev, 1);
NeilBrown69e51b42010-06-01 19:37:35 +10006040 if (fd >= 0) {
NeilBrown36fa3062005-09-09 16:23:45 -07006041 err = bitmap_create(mddev);
NeilBrown69e51b42010-06-01 19:37:35 +10006042 if (!err)
6043 err = bitmap_load(mddev);
6044 }
NeilBrownd7375ab2006-06-26 00:27:43 -07006045 if (fd < 0 || err) {
NeilBrown36fa3062005-09-09 16:23:45 -07006046 bitmap_destroy(mddev);
NeilBrownd7375ab2006-06-26 00:27:43 -07006047 fd = -1; /* make sure to put the file */
6048 }
NeilBrown36fa3062005-09-09 16:23:45 -07006049 mddev->pers->quiesce(mddev, 0);
NeilBrownd7375ab2006-06-26 00:27:43 -07006050 }
6051 if (fd < 0) {
NeilBrown4af1a042014-12-15 12:57:00 +11006052 struct file *f = mddev->bitmap_info.file;
6053 if (f) {
6054 spin_lock(&mddev->lock);
6055 mddev->bitmap_info.file = NULL;
6056 spin_unlock(&mddev->lock);
6057 fput(f);
6058 }
NeilBrown36fa3062005-09-09 16:23:45 -07006059 }
6060
NeilBrown32a76272005-06-21 17:17:14 -07006061 return err;
6062}
6063
Linus Torvalds1da177e2005-04-16 15:20:36 -07006064/*
6065 * set_array_info is used two different ways
6066 * The original usage is when creating a new array.
6067 * In this usage, raid_disks is > 0 and it together with
6068 * level, size, not_persistent,layout,chunksize determine the
6069 * shape of the array.
6070 * This will always create an array with a type-0.90.0 superblock.
6071 * The newer usage is when assembling an array.
6072 * In this case raid_disks will be 0, and the major_version field is
6073 * use to determine which style super-blocks are to be found on the devices.
6074 * The minor and patch _version numbers are also kept incase the
6075 * super_block handler wishes to interpret them.
6076 */
NeilBrownf72ffdd2014-09-30 14:23:59 +10006077static int set_array_info(struct mddev *mddev, mdu_array_info_t *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006078{
6079
6080 if (info->raid_disks == 0) {
6081 /* just setting version number for superblock loading */
6082 if (info->major_version < 0 ||
Ahmed S. Darwish50511da2007-05-09 02:35:34 -07006083 info->major_version >= ARRAY_SIZE(super_types) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07006084 super_types[info->major_version].name == NULL) {
6085 /* maybe try to auto-load a module? */
NeilBrownf72ffdd2014-09-30 14:23:59 +10006086 printk(KERN_INFO
Linus Torvalds1da177e2005-04-16 15:20:36 -07006087 "md: superblock version %d not known\n",
6088 info->major_version);
6089 return -EINVAL;
6090 }
6091 mddev->major_version = info->major_version;
6092 mddev->minor_version = info->minor_version;
6093 mddev->patch_version = info->patch_version;
NeilBrown3f9d7b02006-12-22 01:11:41 -08006094 mddev->persistent = !info->not_persistent;
NeilBrowncbd19982009-12-30 12:08:49 +11006095 /* ensure mddev_put doesn't delete this now that there
6096 * is some minimal configuration.
6097 */
6098 mddev->ctime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07006099 return 0;
6100 }
6101 mddev->major_version = MD_MAJOR_VERSION;
6102 mddev->minor_version = MD_MINOR_VERSION;
6103 mddev->patch_version = MD_PATCHLEVEL_VERSION;
6104 mddev->ctime = get_seconds();
6105
6106 mddev->level = info->level;
NeilBrown17115e02006-01-16 22:14:57 -08006107 mddev->clevel[0] = 0;
Andre Noll58c0fed2009-03-31 14:33:13 +11006108 mddev->dev_sectors = 2 * (sector_t)info->size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006109 mddev->raid_disks = info->raid_disks;
6110 /* don't set md_minor, it is determined by which /dev/md* was
6111 * openned
6112 */
6113 if (info->state & (1<<MD_SB_CLEAN))
6114 mddev->recovery_cp = MaxSector;
6115 else
6116 mddev->recovery_cp = 0;
6117 mddev->persistent = ! info->not_persistent;
NeilBrowne6910632008-02-06 01:39:51 -08006118 mddev->external = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006119
6120 mddev->layout = info->layout;
Andre Noll9d8f0362009-06-18 08:45:01 +10006121 mddev->chunk_sectors = info->chunk_size >> 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006122
6123 mddev->max_disks = MD_SB_DISKS;
6124
NeilBrowne6910632008-02-06 01:39:51 -08006125 if (mddev->persistent)
6126 mddev->flags = 0;
NeilBrown850b2b42006-10-03 01:15:46 -07006127 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006128
NeilBrownc3d97142009-12-14 12:49:52 +11006129 mddev->bitmap_info.default_offset = MD_SB_BYTES >> 9;
NeilBrown6409bb02012-05-22 13:55:07 +10006130 mddev->bitmap_info.default_space = 64*2 - (MD_SB_BYTES >> 9);
NeilBrownc3d97142009-12-14 12:49:52 +11006131 mddev->bitmap_info.offset = 0;
NeilBrownb2a27032005-11-28 13:44:12 -08006132
NeilBrownf6705572006-03-27 01:18:11 -08006133 mddev->reshape_position = MaxSector;
6134
Linus Torvalds1da177e2005-04-16 15:20:36 -07006135 /*
6136 * Generate a 128 bit UUID
6137 */
6138 get_random_bytes(mddev->uuid, 16);
6139
NeilBrownf6705572006-03-27 01:18:11 -08006140 mddev->new_level = mddev->level;
Andre Noll664e7c42009-06-18 08:45:27 +10006141 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrownf6705572006-03-27 01:18:11 -08006142 mddev->new_layout = mddev->layout;
6143 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10006144 mddev->reshape_backwards = 0;
NeilBrownf6705572006-03-27 01:18:11 -08006145
Linus Torvalds1da177e2005-04-16 15:20:36 -07006146 return 0;
6147}
6148
NeilBrownfd01b882011-10-11 16:47:53 +11006149void md_set_array_sectors(struct mddev *mddev, sector_t array_sectors)
Dan Williams1f403622009-03-31 14:59:03 +11006150{
Dan Williamsb522adc2009-03-31 15:00:31 +11006151 WARN(!mddev_is_locked(mddev), "%s: unlocked mddev!\n", __func__);
6152
6153 if (mddev->external_size)
6154 return;
6155
Dan Williams1f403622009-03-31 14:59:03 +11006156 mddev->array_sectors = array_sectors;
6157}
6158EXPORT_SYMBOL(md_set_array_sectors);
6159
NeilBrownfd01b882011-10-11 16:47:53 +11006160static int update_size(struct mddev *mddev, sector_t num_sectors)
NeilBrowna35b0d692006-01-06 00:20:49 -08006161{
NeilBrown3cb03002011-10-11 16:45:26 +11006162 struct md_rdev *rdev;
NeilBrowna35b0d692006-01-06 00:20:49 -08006163 int rv;
Andre Nolld71f9f82008-07-11 22:02:22 +10006164 int fit = (num_sectors == 0);
NeilBrowna35b0d692006-01-06 00:20:49 -08006165
6166 if (mddev->pers->resize == NULL)
6167 return -EINVAL;
Andre Nolld71f9f82008-07-11 22:02:22 +10006168 /* The "num_sectors" is the number of sectors of each device that
6169 * is used. This can only make sense for arrays with redundancy.
6170 * linear and raid0 always use whatever space is available. We can only
6171 * consider changing this number if no resync or reconstruction is
6172 * happening, and if the new size is acceptable. It must fit before the
Andre Noll0f420352008-07-11 22:02:23 +10006173 * sb_start or, if that is <data_offset, it must fit before the size
Andre Nolld71f9f82008-07-11 22:02:22 +10006174 * of each device. If num_sectors is zero, we find the largest size
6175 * that fits.
NeilBrowna35b0d692006-01-06 00:20:49 -08006176 */
NeilBrownf851b602014-12-11 10:02:10 +11006177 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6178 mddev->sync_thread)
NeilBrowna35b0d692006-01-06 00:20:49 -08006179 return -EBUSY;
NeilBrownbd8839e2014-05-28 13:39:21 +10006180 if (mddev->ro)
6181 return -EROFS;
NeilBrowna4a61252012-05-22 13:55:27 +10006182
NeilBrowndafb20f2012-03-19 12:46:39 +11006183 rdev_for_each(rdev, mddev) {
Andre Nolldd8ac332009-03-31 14:33:13 +11006184 sector_t avail = rdev->sectors;
NeilBrown01ab5662006-10-28 10:38:30 -07006185
Andre Nolld71f9f82008-07-11 22:02:22 +10006186 if (fit && (num_sectors == 0 || num_sectors > avail))
6187 num_sectors = avail;
6188 if (avail < num_sectors)
NeilBrowna35b0d692006-01-06 00:20:49 -08006189 return -ENOSPC;
6190 }
Andre Nolld71f9f82008-07-11 22:02:22 +10006191 rv = mddev->pers->resize(mddev, num_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10006192 if (!rv)
6193 revalidate_disk(mddev->gendisk);
NeilBrowna35b0d692006-01-06 00:20:49 -08006194 return rv;
6195}
6196
NeilBrownfd01b882011-10-11 16:47:53 +11006197static int update_raid_disks(struct mddev *mddev, int raid_disks)
NeilBrownda943b992006-01-06 00:20:54 -08006198{
6199 int rv;
NeilBrownc6563a82012-05-21 09:27:00 +10006200 struct md_rdev *rdev;
NeilBrownda943b992006-01-06 00:20:54 -08006201 /* change the number of raid disks */
NeilBrown63c70c42006-03-27 01:18:13 -08006202 if (mddev->pers->check_reshape == NULL)
NeilBrownda943b992006-01-06 00:20:54 -08006203 return -EINVAL;
NeilBrownbd8839e2014-05-28 13:39:21 +10006204 if (mddev->ro)
6205 return -EROFS;
NeilBrownda943b992006-01-06 00:20:54 -08006206 if (raid_disks <= 0 ||
NeilBrown233fca32010-04-14 17:02:09 +10006207 (mddev->max_disks && raid_disks >= mddev->max_disks))
NeilBrownda943b992006-01-06 00:20:54 -08006208 return -EINVAL;
NeilBrownf851b602014-12-11 10:02:10 +11006209 if (mddev->sync_thread ||
6210 test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
6211 mddev->reshape_position != MaxSector)
NeilBrownda943b992006-01-06 00:20:54 -08006212 return -EBUSY;
NeilBrownc6563a82012-05-21 09:27:00 +10006213
6214 rdev_for_each(rdev, mddev) {
6215 if (mddev->raid_disks < raid_disks &&
6216 rdev->data_offset < rdev->new_data_offset)
6217 return -EINVAL;
6218 if (mddev->raid_disks > raid_disks &&
6219 rdev->data_offset > rdev->new_data_offset)
6220 return -EINVAL;
6221 }
6222
NeilBrown63c70c42006-03-27 01:18:13 -08006223 mddev->delta_disks = raid_disks - mddev->raid_disks;
NeilBrown2c810cd2012-05-21 09:27:00 +10006224 if (mddev->delta_disks < 0)
6225 mddev->reshape_backwards = 1;
6226 else if (mddev->delta_disks > 0)
6227 mddev->reshape_backwards = 0;
NeilBrown63c70c42006-03-27 01:18:13 -08006228
6229 rv = mddev->pers->check_reshape(mddev);
NeilBrown2c810cd2012-05-21 09:27:00 +10006230 if (rv < 0) {
NeilBrownde171cb2011-01-31 11:57:42 +11006231 mddev->delta_disks = 0;
NeilBrown2c810cd2012-05-21 09:27:00 +10006232 mddev->reshape_backwards = 0;
6233 }
NeilBrownda943b992006-01-06 00:20:54 -08006234 return rv;
6235}
6236
Linus Torvalds1da177e2005-04-16 15:20:36 -07006237/*
6238 * update_array_info is used to change the configuration of an
6239 * on-line array.
6240 * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
6241 * fields in the info are checked against the array.
6242 * Any differences that cannot be handled will cause an error.
6243 * Normally, only one change can be managed at a time.
6244 */
NeilBrownfd01b882011-10-11 16:47:53 +11006245static int update_array_info(struct mddev *mddev, mdu_array_info_t *info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006246{
6247 int rv = 0;
6248 int cnt = 0;
NeilBrown36fa3062005-09-09 16:23:45 -07006249 int state = 0;
6250
6251 /* calculate expected state,ignoring low bits */
NeilBrownc3d97142009-12-14 12:49:52 +11006252 if (mddev->bitmap && mddev->bitmap_info.offset)
NeilBrown36fa3062005-09-09 16:23:45 -07006253 state |= (1 << MD_SB_BITMAP_PRESENT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006254
6255 if (mddev->major_version != info->major_version ||
6256 mddev->minor_version != info->minor_version ||
6257/* mddev->patch_version != info->patch_version || */
6258 mddev->ctime != info->ctime ||
6259 mddev->level != info->level ||
6260/* mddev->layout != info->layout || */
6261 !mddev->persistent != info->not_persistent||
Andre Noll9d8f0362009-06-18 08:45:01 +10006262 mddev->chunk_sectors != info->chunk_size >> 9 ||
NeilBrown36fa3062005-09-09 16:23:45 -07006263 /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
6264 ((state^info->state) & 0xfffffe00)
6265 )
Linus Torvalds1da177e2005-04-16 15:20:36 -07006266 return -EINVAL;
6267 /* Check there is only one change */
Andre Noll58c0fed2009-03-31 14:33:13 +11006268 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
6269 cnt++;
6270 if (mddev->raid_disks != info->raid_disks)
6271 cnt++;
6272 if (mddev->layout != info->layout)
6273 cnt++;
6274 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT))
6275 cnt++;
6276 if (cnt == 0)
6277 return 0;
6278 if (cnt > 1)
6279 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006280
6281 if (mddev->layout != info->layout) {
6282 /* Change layout
6283 * we don't need to do anything at the md level, the
6284 * personality will take care of it all.
6285 */
NeilBrown50ac1682009-06-18 08:47:55 +10006286 if (mddev->pers->check_reshape == NULL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006287 return -EINVAL;
NeilBrown597a7112009-06-18 08:47:42 +10006288 else {
6289 mddev->new_layout = info->layout;
NeilBrown50ac1682009-06-18 08:47:55 +10006290 rv = mddev->pers->check_reshape(mddev);
NeilBrown597a7112009-06-18 08:47:42 +10006291 if (rv)
6292 mddev->new_layout = mddev->layout;
6293 return rv;
6294 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006295 }
Andre Noll58c0fed2009-03-31 14:33:13 +11006296 if (info->size >= 0 && mddev->dev_sectors / 2 != info->size)
Andre Nolld71f9f82008-07-11 22:02:22 +10006297 rv = update_size(mddev, (sector_t)info->size * 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006298
NeilBrownda943b992006-01-06 00:20:54 -08006299 if (mddev->raid_disks != info->raid_disks)
6300 rv = update_raid_disks(mddev, info->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006301
NeilBrown36fa3062005-09-09 16:23:45 -07006302 if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
NeilBrownd66b1b32014-08-08 15:40:24 +10006303 if (mddev->pers->quiesce == NULL || mddev->thread == NULL)
NeilBrown36fa3062005-09-09 16:23:45 -07006304 return -EINVAL;
6305 if (mddev->recovery || mddev->sync_thread)
6306 return -EBUSY;
6307 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
6308 /* add the bitmap */
6309 if (mddev->bitmap)
6310 return -EEXIST;
NeilBrownc3d97142009-12-14 12:49:52 +11006311 if (mddev->bitmap_info.default_offset == 0)
NeilBrown36fa3062005-09-09 16:23:45 -07006312 return -EINVAL;
NeilBrownc3d97142009-12-14 12:49:52 +11006313 mddev->bitmap_info.offset =
6314 mddev->bitmap_info.default_offset;
NeilBrown6409bb02012-05-22 13:55:07 +10006315 mddev->bitmap_info.space =
6316 mddev->bitmap_info.default_space;
NeilBrown36fa3062005-09-09 16:23:45 -07006317 mddev->pers->quiesce(mddev, 1);
6318 rv = bitmap_create(mddev);
NeilBrown69e51b42010-06-01 19:37:35 +10006319 if (!rv)
6320 rv = bitmap_load(mddev);
NeilBrown36fa3062005-09-09 16:23:45 -07006321 if (rv)
6322 bitmap_destroy(mddev);
6323 mddev->pers->quiesce(mddev, 0);
6324 } else {
6325 /* remove the bitmap */
6326 if (!mddev->bitmap)
6327 return -ENOENT;
NeilBrown1ec885c2012-05-22 13:55:10 +10006328 if (mddev->bitmap->storage.file)
NeilBrown36fa3062005-09-09 16:23:45 -07006329 return -EINVAL;
6330 mddev->pers->quiesce(mddev, 1);
6331 bitmap_destroy(mddev);
6332 mddev->pers->quiesce(mddev, 0);
NeilBrownc3d97142009-12-14 12:49:52 +11006333 mddev->bitmap_info.offset = 0;
NeilBrown36fa3062005-09-09 16:23:45 -07006334 }
6335 }
NeilBrown850b2b42006-10-03 01:15:46 -07006336 md_update_sb(mddev, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006337 return rv;
6338}
6339
NeilBrownfd01b882011-10-11 16:47:53 +11006340static int set_disk_faulty(struct mddev *mddev, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006341{
NeilBrown3cb03002011-10-11 16:45:26 +11006342 struct md_rdev *rdev;
NeilBrown1ca69c42012-10-11 13:37:33 +11006343 int err = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006344
6345 if (mddev->pers == NULL)
6346 return -ENODEV;
6347
NeilBrown1ca69c42012-10-11 13:37:33 +11006348 rcu_read_lock();
6349 rdev = find_rdev_rcu(mddev, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006350 if (!rdev)
NeilBrown1ca69c42012-10-11 13:37:33 +11006351 err = -ENODEV;
6352 else {
6353 md_error(mddev, rdev);
6354 if (!test_bit(Faulty, &rdev->flags))
6355 err = -EBUSY;
6356 }
6357 rcu_read_unlock();
6358 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006359}
6360
Andre Noll2f9618c2008-04-25 18:57:58 +02006361/*
6362 * We have a problem here : there is no easy way to give a CHS
6363 * virtual geometry. We currently pretend that we have a 2 heads
6364 * 4 sectors (with a BIG number of cylinders...). This drives
6365 * dosfs just mad... ;-)
6366 */
Christoph Hellwiga885c8c2006-01-08 01:02:50 -08006367static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
6368{
NeilBrownfd01b882011-10-11 16:47:53 +11006369 struct mddev *mddev = bdev->bd_disk->private_data;
Christoph Hellwiga885c8c2006-01-08 01:02:50 -08006370
6371 geo->heads = 2;
6372 geo->sectors = 4;
NeilBrown49ce6ce2010-03-29 10:51:42 +11006373 geo->cylinders = mddev->array_sectors / 8;
Christoph Hellwiga885c8c2006-01-08 01:02:50 -08006374 return 0;
6375}
6376
Nicolas Schichancb335f82014-01-15 16:58:52 +01006377static inline bool md_ioctl_valid(unsigned int cmd)
6378{
6379 switch (cmd) {
6380 case ADD_NEW_DISK:
6381 case BLKROSET:
6382 case GET_ARRAY_INFO:
6383 case GET_BITMAP_FILE:
6384 case GET_DISK_INFO:
6385 case HOT_ADD_DISK:
6386 case HOT_REMOVE_DISK:
Nicolas Schichancb335f82014-01-15 16:58:52 +01006387 case RAID_AUTORUN:
6388 case RAID_VERSION:
6389 case RESTART_ARRAY_RW:
6390 case RUN_ARRAY:
6391 case SET_ARRAY_INFO:
6392 case SET_BITMAP_FILE:
6393 case SET_DISK_FAULTY:
6394 case STOP_ARRAY:
6395 case STOP_ARRAY_RO:
6396 return true;
6397 default:
6398 return false;
6399 }
6400}
6401
Al Viroa39907f2008-03-02 10:31:15 -05006402static int md_ioctl(struct block_device *bdev, fmode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006403 unsigned int cmd, unsigned long arg)
6404{
6405 int err = 0;
6406 void __user *argp = (void __user *)arg;
NeilBrownfd01b882011-10-11 16:47:53 +11006407 struct mddev *mddev = NULL;
Dan Williamse2218352010-05-12 08:25:37 +10006408 int ro;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006409
Nicolas Schichancb335f82014-01-15 16:58:52 +01006410 if (!md_ioctl_valid(cmd))
6411 return -ENOTTY;
6412
NeilBrown506c9e42011-12-23 10:17:26 +11006413 switch (cmd) {
6414 case RAID_VERSION:
6415 case GET_ARRAY_INFO:
6416 case GET_DISK_INFO:
6417 break;
6418 default:
6419 if (!capable(CAP_SYS_ADMIN))
6420 return -EACCES;
6421 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006422
6423 /*
6424 * Commands dealing with the RAID driver but not any
6425 * particular array:
6426 */
NeilBrownc02c0ae2012-12-11 13:39:21 +11006427 switch (cmd) {
6428 case RAID_VERSION:
6429 err = get_version(argp);
NeilBrown3adc28d2014-09-30 15:46:41 +10006430 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006431
Linus Torvalds1da177e2005-04-16 15:20:36 -07006432#ifndef MODULE
NeilBrownc02c0ae2012-12-11 13:39:21 +11006433 case RAID_AUTORUN:
6434 err = 0;
6435 autostart_arrays(arg);
NeilBrown3adc28d2014-09-30 15:46:41 +10006436 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006437#endif
NeilBrownc02c0ae2012-12-11 13:39:21 +11006438 default:;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006439 }
6440
6441 /*
6442 * Commands creating/starting a new array:
6443 */
6444
Al Viroa39907f2008-03-02 10:31:15 -05006445 mddev = bdev->bd_disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006446
6447 if (!mddev) {
6448 BUG();
NeilBrown3adc28d2014-09-30 15:46:41 +10006449 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006450 }
6451
NeilBrown1ca69c42012-10-11 13:37:33 +11006452 /* Some actions do not requires the mutex */
6453 switch (cmd) {
6454 case GET_ARRAY_INFO:
6455 if (!mddev->raid_disks && !mddev->external)
6456 err = -ENODEV;
6457 else
6458 err = get_array_info(mddev, argp);
NeilBrown3adc28d2014-09-30 15:46:41 +10006459 goto out;
NeilBrown1ca69c42012-10-11 13:37:33 +11006460
6461 case GET_DISK_INFO:
6462 if (!mddev->raid_disks && !mddev->external)
6463 err = -ENODEV;
6464 else
6465 err = get_disk_info(mddev, argp);
NeilBrown3adc28d2014-09-30 15:46:41 +10006466 goto out;
NeilBrown1ca69c42012-10-11 13:37:33 +11006467
6468 case SET_DISK_FAULTY:
6469 err = set_disk_faulty(mddev, new_decode_dev(arg));
NeilBrown3adc28d2014-09-30 15:46:41 +10006470 goto out;
NeilBrown4af1a042014-12-15 12:57:00 +11006471
6472 case GET_BITMAP_FILE:
6473 err = get_bitmap_file(mddev, argp);
6474 goto out;
6475
NeilBrown1ca69c42012-10-11 13:37:33 +11006476 }
6477
NeilBrowna7a3f082012-12-11 13:35:54 +11006478 if (cmd == ADD_NEW_DISK)
6479 /* need to ensure md_delayed_delete() has completed */
6480 flush_workqueue(md_misc_wq);
6481
Hannes Reinecke90f5f7a2013-04-02 08:38:55 +02006482 if (cmd == HOT_REMOVE_DISK)
6483 /* need to ensure recovery thread has run */
6484 wait_event_interruptible_timeout(mddev->sb_wait,
6485 !test_bit(MD_RECOVERY_NEEDED,
6486 &mddev->flags),
6487 msecs_to_jiffies(5000));
NeilBrown260fa032013-08-27 16:44:13 +10006488 if (cmd == STOP_ARRAY || cmd == STOP_ARRAY_RO) {
6489 /* Need to flush page cache, and ensure no-one else opens
6490 * and writes
6491 */
6492 mutex_lock(&mddev->open_mutex);
NeilBrown9ba3b7f2014-09-09 14:00:15 +10006493 if (mddev->pers && atomic_read(&mddev->openers) > 1) {
NeilBrown260fa032013-08-27 16:44:13 +10006494 mutex_unlock(&mddev->open_mutex);
6495 err = -EBUSY;
NeilBrown3adc28d2014-09-30 15:46:41 +10006496 goto out;
NeilBrown260fa032013-08-27 16:44:13 +10006497 }
6498 set_bit(MD_STILL_CLOSED, &mddev->flags);
6499 mutex_unlock(&mddev->open_mutex);
6500 sync_blockdev(bdev);
6501 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006502 err = mddev_lock(mddev);
6503 if (err) {
NeilBrownf72ffdd2014-09-30 14:23:59 +10006504 printk(KERN_INFO
Linus Torvalds1da177e2005-04-16 15:20:36 -07006505 "md: ioctl lock interrupted, reason %d, cmd %d\n",
6506 err, cmd);
NeilBrown3adc28d2014-09-30 15:46:41 +10006507 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006508 }
6509
NeilBrownc02c0ae2012-12-11 13:39:21 +11006510 if (cmd == SET_ARRAY_INFO) {
6511 mdu_array_info_t info;
6512 if (!arg)
6513 memset(&info, 0, sizeof(info));
6514 else if (copy_from_user(&info, argp, sizeof(info))) {
6515 err = -EFAULT;
NeilBrown3adc28d2014-09-30 15:46:41 +10006516 goto unlock;
NeilBrownc02c0ae2012-12-11 13:39:21 +11006517 }
6518 if (mddev->pers) {
6519 err = update_array_info(mddev, &info);
6520 if (err) {
6521 printk(KERN_WARNING "md: couldn't update"
6522 " array info. %d\n", err);
NeilBrown3adc28d2014-09-30 15:46:41 +10006523 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006524 }
NeilBrown3adc28d2014-09-30 15:46:41 +10006525 goto unlock;
NeilBrownc02c0ae2012-12-11 13:39:21 +11006526 }
6527 if (!list_empty(&mddev->disks)) {
6528 printk(KERN_WARNING
6529 "md: array %s already has disks!\n",
6530 mdname(mddev));
6531 err = -EBUSY;
NeilBrown3adc28d2014-09-30 15:46:41 +10006532 goto unlock;
NeilBrownc02c0ae2012-12-11 13:39:21 +11006533 }
6534 if (mddev->raid_disks) {
6535 printk(KERN_WARNING
6536 "md: array %s already initialised!\n",
6537 mdname(mddev));
6538 err = -EBUSY;
NeilBrown3adc28d2014-09-30 15:46:41 +10006539 goto unlock;
NeilBrownc02c0ae2012-12-11 13:39:21 +11006540 }
6541 err = set_array_info(mddev, &info);
6542 if (err) {
6543 printk(KERN_WARNING "md: couldn't set"
6544 " array info. %d\n", err);
NeilBrown3adc28d2014-09-30 15:46:41 +10006545 goto unlock;
NeilBrownc02c0ae2012-12-11 13:39:21 +11006546 }
NeilBrown3adc28d2014-09-30 15:46:41 +10006547 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006548 }
6549
6550 /*
6551 * Commands querying/configuring an existing array:
6552 */
NeilBrown32a76272005-06-21 17:17:14 -07006553 /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
NeilBrown3f9d7b02006-12-22 01:11:41 -08006554 * RUN_ARRAY, and GET_ and SET_BITMAP_FILE are allowed */
NeilBrowna17184a2008-02-06 01:39:55 -08006555 if ((!mddev->raid_disks && !mddev->external)
6556 && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
6557 && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE
6558 && cmd != GET_BITMAP_FILE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006559 err = -ENODEV;
NeilBrown3adc28d2014-09-30 15:46:41 +10006560 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006561 }
6562
6563 /*
6564 * Commands even a read-only array can execute:
6565 */
NeilBrownc02c0ae2012-12-11 13:39:21 +11006566 switch (cmd) {
NeilBrownc02c0ae2012-12-11 13:39:21 +11006567 case RESTART_ARRAY_RW:
6568 err = restart_array(mddev);
NeilBrown3adc28d2014-09-30 15:46:41 +10006569 goto unlock;
NeilBrownc02c0ae2012-12-11 13:39:21 +11006570
6571 case STOP_ARRAY:
6572 err = do_md_stop(mddev, 0, bdev);
NeilBrown3adc28d2014-09-30 15:46:41 +10006573 goto unlock;
NeilBrownc02c0ae2012-12-11 13:39:21 +11006574
6575 case STOP_ARRAY_RO:
6576 err = md_set_readonly(mddev, bdev);
NeilBrown3adc28d2014-09-30 15:46:41 +10006577 goto unlock;
NeilBrownc02c0ae2012-12-11 13:39:21 +11006578
NeilBrown3ea8929d2013-04-24 11:42:41 +10006579 case HOT_REMOVE_DISK:
6580 err = hot_remove_disk(mddev, new_decode_dev(arg));
NeilBrown3adc28d2014-09-30 15:46:41 +10006581 goto unlock;
NeilBrown3ea8929d2013-04-24 11:42:41 +10006582
NeilBrown7ceb17e2013-04-24 11:42:42 +10006583 case ADD_NEW_DISK:
6584 /* We can support ADD_NEW_DISK on read-only arrays
6585 * on if we are re-adding a preexisting device.
6586 * So require mddev->pers and MD_DISK_SYNC.
6587 */
6588 if (mddev->pers) {
6589 mdu_disk_info_t info;
6590 if (copy_from_user(&info, argp, sizeof(info)))
6591 err = -EFAULT;
6592 else if (!(info.state & (1<<MD_DISK_SYNC)))
6593 /* Need to clear read-only for this */
6594 break;
6595 else
6596 err = add_new_disk(mddev, &info);
NeilBrown3adc28d2014-09-30 15:46:41 +10006597 goto unlock;
NeilBrown7ceb17e2013-04-24 11:42:42 +10006598 }
6599 break;
6600
NeilBrownc02c0ae2012-12-11 13:39:21 +11006601 case BLKROSET:
6602 if (get_user(ro, (int __user *)(arg))) {
6603 err = -EFAULT;
NeilBrown3adc28d2014-09-30 15:46:41 +10006604 goto unlock;
NeilBrownc02c0ae2012-12-11 13:39:21 +11006605 }
6606 err = -EINVAL;
6607
6608 /* if the bdev is going readonly the value of mddev->ro
6609 * does not matter, no writes are coming
6610 */
6611 if (ro)
NeilBrown3adc28d2014-09-30 15:46:41 +10006612 goto unlock;
NeilBrown32a76272005-06-21 17:17:14 -07006613
NeilBrownc02c0ae2012-12-11 13:39:21 +11006614 /* are we are already prepared for writes? */
6615 if (mddev->ro != 1)
NeilBrown3adc28d2014-09-30 15:46:41 +10006616 goto unlock;
NeilBrownc02c0ae2012-12-11 13:39:21 +11006617
6618 /* transitioning to readauto need only happen for
6619 * arrays that call md_write_start
6620 */
6621 if (mddev->pers) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006622 err = restart_array(mddev);
NeilBrownc02c0ae2012-12-11 13:39:21 +11006623 if (err == 0) {
6624 mddev->ro = 2;
6625 set_disk_ro(mddev->gendisk, 0);
Dan Williamse2218352010-05-12 08:25:37 +10006626 }
NeilBrownc02c0ae2012-12-11 13:39:21 +11006627 }
NeilBrown3adc28d2014-09-30 15:46:41 +10006628 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006629 }
6630
6631 /*
6632 * The remaining ioctls are changing the state of the
NeilBrownf91de922005-11-08 21:39:36 -08006633 * superblock, so we do not allow them on read-only arrays.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006634 */
NeilBrown326eb172014-09-30 15:36:28 +10006635 if (mddev->ro && mddev->pers) {
NeilBrownf91de922005-11-08 21:39:36 -08006636 if (mddev->ro == 2) {
6637 mddev->ro = 0;
NeilBrown00bcb4a2010-06-01 19:37:23 +10006638 sysfs_notify_dirent_safe(mddev->sysfs_state);
Neil Brown0fd62b82008-06-28 08:31:36 +10006639 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrownf3378b42013-02-28 11:59:03 +11006640 /* mddev_unlock will wake thread */
6641 /* If a device failed while we were read-only, we
6642 * need to make sure the metadata is updated now.
6643 */
6644 if (test_bit(MD_CHANGE_DEVS, &mddev->flags)) {
6645 mddev_unlock(mddev);
6646 wait_event(mddev->sb_wait,
6647 !test_bit(MD_CHANGE_DEVS, &mddev->flags) &&
6648 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
NeilBrown29f097c2013-11-14 17:54:51 +11006649 mddev_lock_nointr(mddev);
NeilBrownf3378b42013-02-28 11:59:03 +11006650 }
NeilBrownf91de922005-11-08 21:39:36 -08006651 } else {
6652 err = -EROFS;
NeilBrown3adc28d2014-09-30 15:46:41 +10006653 goto unlock;
NeilBrownf91de922005-11-08 21:39:36 -08006654 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006655 }
6656
NeilBrownc02c0ae2012-12-11 13:39:21 +11006657 switch (cmd) {
6658 case ADD_NEW_DISK:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006659 {
NeilBrownc02c0ae2012-12-11 13:39:21 +11006660 mdu_disk_info_t info;
6661 if (copy_from_user(&info, argp, sizeof(info)))
6662 err = -EFAULT;
6663 else
6664 err = add_new_disk(mddev, &info);
NeilBrown3adc28d2014-09-30 15:46:41 +10006665 goto unlock;
NeilBrownc02c0ae2012-12-11 13:39:21 +11006666 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006667
NeilBrownc02c0ae2012-12-11 13:39:21 +11006668 case HOT_ADD_DISK:
6669 err = hot_add_disk(mddev, new_decode_dev(arg));
NeilBrown3adc28d2014-09-30 15:46:41 +10006670 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006671
NeilBrownc02c0ae2012-12-11 13:39:21 +11006672 case RUN_ARRAY:
6673 err = do_md_run(mddev);
NeilBrown3adc28d2014-09-30 15:46:41 +10006674 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006675
NeilBrownc02c0ae2012-12-11 13:39:21 +11006676 case SET_BITMAP_FILE:
6677 err = set_bitmap_file(mddev, (int)arg);
NeilBrown3adc28d2014-09-30 15:46:41 +10006678 goto unlock;
NeilBrown32a76272005-06-21 17:17:14 -07006679
NeilBrownc02c0ae2012-12-11 13:39:21 +11006680 default:
6681 err = -EINVAL;
NeilBrown3adc28d2014-09-30 15:46:41 +10006682 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006683 }
6684
NeilBrown3adc28d2014-09-30 15:46:41 +10006685unlock:
NeilBrownd3374822009-01-09 08:31:10 +11006686 if (mddev->hold_active == UNTIL_IOCTL &&
6687 err != -EINVAL)
6688 mddev->hold_active = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006689 mddev_unlock(mddev);
NeilBrown3adc28d2014-09-30 15:46:41 +10006690out:
Linus Torvalds1da177e2005-04-16 15:20:36 -07006691 return err;
6692}
Arnd Bergmannaa98aa32009-12-14 12:50:05 +11006693#ifdef CONFIG_COMPAT
6694static int md_compat_ioctl(struct block_device *bdev, fmode_t mode,
6695 unsigned int cmd, unsigned long arg)
6696{
6697 switch (cmd) {
6698 case HOT_REMOVE_DISK:
6699 case HOT_ADD_DISK:
6700 case SET_DISK_FAULTY:
6701 case SET_BITMAP_FILE:
6702 /* These take in integer arg, do not convert */
6703 break;
6704 default:
6705 arg = (unsigned long)compat_ptr(arg);
6706 break;
6707 }
6708
6709 return md_ioctl(bdev, mode, cmd, arg);
6710}
6711#endif /* CONFIG_COMPAT */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006712
Al Viroa39907f2008-03-02 10:31:15 -05006713static int md_open(struct block_device *bdev, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006714{
6715 /*
6716 * Succeed if we can lock the mddev, which confirms that
6717 * it isn't being stopped right now.
6718 */
NeilBrownfd01b882011-10-11 16:47:53 +11006719 struct mddev *mddev = mddev_find(bdev->bd_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006720 int err;
6721
Yuanhan Liu0c098222012-05-22 13:55:32 +10006722 if (!mddev)
6723 return -ENODEV;
6724
NeilBrownd3374822009-01-09 08:31:10 +11006725 if (mddev->gendisk != bdev->bd_disk) {
6726 /* we are racing with mddev_put which is discarding this
6727 * bd_disk.
6728 */
6729 mddev_put(mddev);
6730 /* Wait until bdev->bd_disk is definitely gone */
Tejun Heoe804ac72010-10-15 15:36:08 +02006731 flush_workqueue(md_misc_wq);
NeilBrownd3374822009-01-09 08:31:10 +11006732 /* Then retry the open from the top */
6733 return -ERESTARTSYS;
6734 }
6735 BUG_ON(mddev != bdev->bd_disk->private_data);
6736
NeilBrownc8c00a62009-08-10 12:50:52 +10006737 if ((err = mutex_lock_interruptible(&mddev->open_mutex)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006738 goto out;
6739
6740 err = 0;
NeilBrownf2ea68c2008-07-21 17:05:25 +10006741 atomic_inc(&mddev->openers);
NeilBrown260fa032013-08-27 16:44:13 +10006742 clear_bit(MD_STILL_CLOSED, &mddev->flags);
NeilBrownc8c00a62009-08-10 12:50:52 +10006743 mutex_unlock(&mddev->open_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006744
NeilBrownf0b4f7e2011-02-24 17:26:41 +11006745 check_disk_change(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006746 out:
6747 return err;
6748}
6749
Al Virodb2a1442013-05-05 21:52:57 -04006750static void md_release(struct gendisk *disk, fmode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006751{
NeilBrownf72ffdd2014-09-30 14:23:59 +10006752 struct mddev *mddev = disk->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006753
Eric Sesterhenn52e5f9d2006-10-03 23:33:23 +02006754 BUG_ON(!mddev);
NeilBrownf2ea68c2008-07-21 17:05:25 +10006755 atomic_dec(&mddev->openers);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006756 mddev_put(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006757}
NeilBrownf0b4f7e2011-02-24 17:26:41 +11006758
6759static int md_media_changed(struct gendisk *disk)
6760{
NeilBrownfd01b882011-10-11 16:47:53 +11006761 struct mddev *mddev = disk->private_data;
NeilBrownf0b4f7e2011-02-24 17:26:41 +11006762
6763 return mddev->changed;
6764}
6765
6766static int md_revalidate(struct gendisk *disk)
6767{
NeilBrownfd01b882011-10-11 16:47:53 +11006768 struct mddev *mddev = disk->private_data;
NeilBrownf0b4f7e2011-02-24 17:26:41 +11006769
6770 mddev->changed = 0;
6771 return 0;
6772}
Alexey Dobriyan83d5cde2009-09-21 17:01:13 -07006773static const struct block_device_operations md_fops =
Linus Torvalds1da177e2005-04-16 15:20:36 -07006774{
6775 .owner = THIS_MODULE,
Al Viroa39907f2008-03-02 10:31:15 -05006776 .open = md_open,
6777 .release = md_release,
NeilBrownb492b852009-05-26 12:57:36 +10006778 .ioctl = md_ioctl,
Arnd Bergmannaa98aa32009-12-14 12:50:05 +11006779#ifdef CONFIG_COMPAT
6780 .compat_ioctl = md_compat_ioctl,
6781#endif
Christoph Hellwiga885c8c2006-01-08 01:02:50 -08006782 .getgeo = md_getgeo,
NeilBrownf0b4f7e2011-02-24 17:26:41 +11006783 .media_changed = md_media_changed,
6784 .revalidate_disk= md_revalidate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07006785};
6786
NeilBrownf72ffdd2014-09-30 14:23:59 +10006787static int md_thread(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006788{
NeilBrown2b8bf342011-10-11 16:48:23 +11006789 struct md_thread *thread = arg;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006790
Linus Torvalds1da177e2005-04-16 15:20:36 -07006791 /*
6792 * md_thread is a 'system-thread', it's priority should be very
6793 * high. We avoid resource deadlocks individually in each
6794 * raid personality. (RAID5 does preallocation) We also use RR and
6795 * the very same RT priority as kswapd, thus we will never get
6796 * into a priority inversion deadlock.
6797 *
6798 * we definitely have to have equal or higher priority than
6799 * bdflush, otherwise bdflush will deadlock if there are too
6800 * many dirty RAID5 blocks.
6801 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07006802
NeilBrown6985c432005-10-19 21:23:47 -07006803 allow_signal(SIGKILL);
NeilBrowna6fb0932005-09-09 16:23:56 -07006804 while (!kthread_should_stop()) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006805
NeilBrown93588e22005-11-15 00:09:12 -08006806 /* We need to wait INTERRUPTIBLE so that
6807 * we don't add to the load-average.
6808 * That means we need to be sure no signals are
6809 * pending
6810 */
6811 if (signal_pending(current))
6812 flush_signals(current);
6813
6814 wait_event_interruptible_timeout
6815 (thread->wqueue,
6816 test_bit(THREAD_WAKEUP, &thread->flags)
6817 || kthread_should_stop(),
6818 thread->timeout);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006819
NeilBrown6c987912011-01-14 09:13:53 +11006820 clear_bit(THREAD_WAKEUP, &thread->flags);
6821 if (!kthread_should_stop())
Shaohua Li4ed87312012-10-11 13:34:00 +11006822 thread->run(thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006823 }
NeilBrowna6fb0932005-09-09 16:23:56 -07006824
Linus Torvalds1da177e2005-04-16 15:20:36 -07006825 return 0;
6826}
6827
NeilBrown2b8bf342011-10-11 16:48:23 +11006828void md_wakeup_thread(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006829{
6830 if (thread) {
NeilBrown36a4e1f2011-10-07 14:23:17 +11006831 pr_debug("md: waking up MD thread %s.\n", thread->tsk->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006832 set_bit(THREAD_WAKEUP, &thread->flags);
6833 wake_up(&thread->wqueue);
6834 }
6835}
NeilBrown6c144d32014-09-30 16:15:38 +10006836EXPORT_SYMBOL(md_wakeup_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006837
Shaohua Li4ed87312012-10-11 13:34:00 +11006838struct md_thread *md_register_thread(void (*run) (struct md_thread *),
6839 struct mddev *mddev, const char *name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006840{
NeilBrown2b8bf342011-10-11 16:48:23 +11006841 struct md_thread *thread;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006842
NeilBrown2b8bf342011-10-11 16:48:23 +11006843 thread = kzalloc(sizeof(struct md_thread), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006844 if (!thread)
6845 return NULL;
6846
Linus Torvalds1da177e2005-04-16 15:20:36 -07006847 init_waitqueue_head(&thread->wqueue);
6848
Linus Torvalds1da177e2005-04-16 15:20:36 -07006849 thread->run = run;
6850 thread->mddev = mddev;
NeilBrown32a76272005-06-21 17:17:14 -07006851 thread->timeout = MAX_SCHEDULE_TIMEOUT;
NeilBrown0da3c612009-09-23 18:09:45 +10006852 thread->tsk = kthread_run(md_thread, thread,
6853 "%s_%s",
6854 mdname(thread->mddev),
NeilBrown02326052012-07-03 15:56:52 +10006855 name);
NeilBrowna6fb0932005-09-09 16:23:56 -07006856 if (IS_ERR(thread->tsk)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006857 kfree(thread);
6858 return NULL;
6859 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07006860 return thread;
6861}
NeilBrown6c144d32014-09-30 16:15:38 +10006862EXPORT_SYMBOL(md_register_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006863
NeilBrown2b8bf342011-10-11 16:48:23 +11006864void md_unregister_thread(struct md_thread **threadp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006865{
NeilBrown2b8bf342011-10-11 16:48:23 +11006866 struct md_thread *thread = *threadp;
NeilBrowne0cf8f02009-03-31 14:39:39 +11006867 if (!thread)
6868 return;
NeilBrown36a4e1f2011-10-07 14:23:17 +11006869 pr_debug("interrupting MD-thread pid %d\n", task_pid_nr(thread->tsk));
NeilBrown01f96c02011-09-21 15:30:20 +10006870 /* Locking ensures that mddev_unlock does not wake_up a
6871 * non-existent thread
6872 */
6873 spin_lock(&pers_lock);
6874 *threadp = NULL;
6875 spin_unlock(&pers_lock);
NeilBrowna6fb0932005-09-09 16:23:56 -07006876
6877 kthread_stop(thread->tsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006878 kfree(thread);
6879}
NeilBrown6c144d32014-09-30 16:15:38 +10006880EXPORT_SYMBOL(md_unregister_thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006881
NeilBrownfd01b882011-10-11 16:47:53 +11006882void md_error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006883{
NeilBrownb2d444d2005-11-08 21:39:31 -08006884 if (!rdev || test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07006885 return;
Dan Williams6bfe0b42008-04-30 00:52:32 -07006886
NeilBrownde393cd2011-07-28 11:31:48 +10006887 if (!mddev->pers || !mddev->pers->error_handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006888 return;
6889 mddev->pers->error_handler(mddev,rdev);
Neil Brown72a23c22008-06-28 08:31:41 +10006890 if (mddev->degraded)
6891 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
NeilBrown00bcb4a2010-06-01 19:37:23 +10006892 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006893 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
6894 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
6895 md_wakeup_thread(mddev->thread);
NeilBrown768a4182010-07-26 11:49:55 +10006896 if (mddev->event_work.func)
Tejun Heoe804ac72010-10-15 15:36:08 +02006897 queue_work(md_misc_wq, &mddev->event_work);
NeilBrownc331eb02006-05-30 21:27:13 -07006898 md_new_event_inintr(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006899}
NeilBrown6c144d32014-09-30 16:15:38 +10006900EXPORT_SYMBOL(md_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006901
6902/* seq_file implementation /proc/mdstat */
6903
6904static void status_unused(struct seq_file *seq)
6905{
6906 int i = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11006907 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006908
6909 seq_printf(seq, "unused devices: ");
6910
Cheng Renquan159ec1f2009-01-09 08:31:08 +11006911 list_for_each_entry(rdev, &pending_raid_disks, same_set) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07006912 char b[BDEVNAME_SIZE];
6913 i++;
6914 seq_printf(seq, "%s ",
6915 bdevname(rdev->bdev,b));
6916 }
6917 if (!i)
6918 seq_printf(seq, "<none>");
6919
6920 seq_printf(seq, "\n");
6921}
6922
NeilBrownf72ffdd2014-09-30 14:23:59 +10006923static void status_resync(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07006924{
NeilBrowndd71cf62009-05-07 12:49:35 +10006925 sector_t max_sectors, resync, res;
6926 unsigned long dt, db;
6927 sector_t rt;
NeilBrown4588b422006-03-27 01:18:04 -08006928 int scale;
6929 unsigned int per_milli;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006930
NeilBrown72f36d52012-10-11 14:25:57 +11006931 if (mddev->curr_resync <= 3)
6932 resync = 0;
6933 else
6934 resync = mddev->curr_resync
6935 - atomic_read(&mddev->recovery_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006936
NeilBrownc804cde2012-05-21 09:28:33 +10006937 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ||
6938 test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
NeilBrowndd71cf62009-05-07 12:49:35 +10006939 max_sectors = mddev->resync_max_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006940 else
NeilBrowndd71cf62009-05-07 12:49:35 +10006941 max_sectors = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006942
NeilBrown403df472014-09-30 15:52:29 +10006943 WARN_ON(max_sectors == 0);
NeilBrown4588b422006-03-27 01:18:04 -08006944 /* Pick 'scale' such that (resync>>scale)*1000 will fit
NeilBrowndd71cf62009-05-07 12:49:35 +10006945 * in a sector_t, and (max_sectors>>scale) will fit in a
NeilBrown4588b422006-03-27 01:18:04 -08006946 * u32, as those are the requirements for sector_div.
6947 * Thus 'scale' must be at least 10
6948 */
6949 scale = 10;
6950 if (sizeof(sector_t) > sizeof(unsigned long)) {
NeilBrowndd71cf62009-05-07 12:49:35 +10006951 while ( max_sectors/2 > (1ULL<<(scale+32)))
NeilBrown4588b422006-03-27 01:18:04 -08006952 scale++;
6953 }
6954 res = (resync>>scale)*1000;
NeilBrowndd71cf62009-05-07 12:49:35 +10006955 sector_div(res, (u32)((max_sectors>>scale)+1));
NeilBrown4588b422006-03-27 01:18:04 -08006956
6957 per_milli = res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006958 {
NeilBrown4588b422006-03-27 01:18:04 -08006959 int i, x = per_milli/50, y = 20-x;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006960 seq_printf(seq, "[");
6961 for (i = 0; i < x; i++)
6962 seq_printf(seq, "=");
6963 seq_printf(seq, ">");
6964 for (i = 0; i < y; i++)
6965 seq_printf(seq, ".");
6966 seq_printf(seq, "] ");
6967 }
NeilBrown4588b422006-03-27 01:18:04 -08006968 seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
NeilBrownccfcc3c2006-03-27 01:18:09 -08006969 (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
6970 "reshape" :
NeilBrown61df9d92006-10-03 01:15:57 -07006971 (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)?
6972 "check" :
6973 (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
6974 "resync" : "recovery"))),
6975 per_milli/10, per_milli % 10,
NeilBrowndd71cf62009-05-07 12:49:35 +10006976 (unsigned long long) resync/2,
6977 (unsigned long long) max_sectors/2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07006978
6979 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07006980 * dt: time from mark until now
6981 * db: blocks written from mark until now
6982 * rt: remaining time
NeilBrowndd71cf62009-05-07 12:49:35 +10006983 *
6984 * rt is a sector_t, so could be 32bit or 64bit.
6985 * So we divide before multiply in case it is 32bit and close
6986 * to the limit.
Lucas De Marchi25985ed2011-03-30 22:57:33 -03006987 * We scale the divisor (db) by 32 to avoid losing precision
NeilBrowndd71cf62009-05-07 12:49:35 +10006988 * near the end of resync when the number of remaining sectors
6989 * is close to 'db'.
6990 * We then divide rt by 32 after multiplying by db to compensate.
6991 * The '+1' avoids division by zero if db is very small.
Linus Torvalds1da177e2005-04-16 15:20:36 -07006992 */
6993 dt = ((jiffies - mddev->resync_mark) / HZ);
6994 if (!dt) dt++;
NeilBrownff4e8d92006-07-10 04:44:16 -07006995 db = (mddev->curr_mark_cnt - atomic_read(&mddev->recovery_active))
6996 - mddev->resync_mark_cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07006997
NeilBrowndd71cf62009-05-07 12:49:35 +10006998 rt = max_sectors - resync; /* number of remaining sectors */
6999 sector_div(rt, db/32+1);
7000 rt *= dt;
7001 rt >>= 5;
7002
7003 seq_printf(seq, " finish=%lu.%lumin", (unsigned long)rt / 60,
7004 ((unsigned long)rt % 60)/6);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007005
NeilBrownff4e8d92006-07-10 04:44:16 -07007006 seq_printf(seq, " speed=%ldK/sec", db/2/dt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007007}
7008
7009static void *md_seq_start(struct seq_file *seq, loff_t *pos)
7010{
7011 struct list_head *tmp;
7012 loff_t l = *pos;
NeilBrownfd01b882011-10-11 16:47:53 +11007013 struct mddev *mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007014
7015 if (l >= 0x10000)
7016 return NULL;
7017 if (!l--)
7018 /* header */
7019 return (void*)1;
7020
7021 spin_lock(&all_mddevs_lock);
7022 list_for_each(tmp,&all_mddevs)
7023 if (!l--) {
NeilBrownfd01b882011-10-11 16:47:53 +11007024 mddev = list_entry(tmp, struct mddev, all_mddevs);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007025 mddev_get(mddev);
7026 spin_unlock(&all_mddevs_lock);
7027 return mddev;
7028 }
7029 spin_unlock(&all_mddevs_lock);
7030 if (!l--)
7031 return (void*)2;/* tail */
7032 return NULL;
7033}
7034
7035static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
7036{
7037 struct list_head *tmp;
NeilBrownfd01b882011-10-11 16:47:53 +11007038 struct mddev *next_mddev, *mddev = v;
NeilBrownf72ffdd2014-09-30 14:23:59 +10007039
Linus Torvalds1da177e2005-04-16 15:20:36 -07007040 ++*pos;
7041 if (v == (void*)2)
7042 return NULL;
7043
7044 spin_lock(&all_mddevs_lock);
7045 if (v == (void*)1)
7046 tmp = all_mddevs.next;
7047 else
7048 tmp = mddev->all_mddevs.next;
7049 if (tmp != &all_mddevs)
NeilBrownfd01b882011-10-11 16:47:53 +11007050 next_mddev = mddev_get(list_entry(tmp,struct mddev,all_mddevs));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007051 else {
7052 next_mddev = (void*)2;
7053 *pos = 0x10000;
NeilBrownf72ffdd2014-09-30 14:23:59 +10007054 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007055 spin_unlock(&all_mddevs_lock);
7056
7057 if (v != (void*)1)
7058 mddev_put(mddev);
7059 return next_mddev;
7060
7061}
7062
7063static void md_seq_stop(struct seq_file *seq, void *v)
7064{
NeilBrownfd01b882011-10-11 16:47:53 +11007065 struct mddev *mddev = v;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007066
7067 if (mddev && v != (void*)1 && v != (void*)2)
7068 mddev_put(mddev);
7069}
7070
7071static int md_seq_show(struct seq_file *seq, void *v)
7072{
NeilBrownfd01b882011-10-11 16:47:53 +11007073 struct mddev *mddev = v;
Andre Nolldd8ac332009-03-31 14:33:13 +11007074 sector_t sectors;
NeilBrown3cb03002011-10-11 16:45:26 +11007075 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007076
7077 if (v == (void*)1) {
NeilBrown84fc4b52011-10-11 16:49:58 +11007078 struct md_personality *pers;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007079 seq_printf(seq, "Personalities : ");
7080 spin_lock(&pers_lock);
NeilBrown2604b702006-01-06 00:20:36 -08007081 list_for_each_entry(pers, &pers_list, list)
7082 seq_printf(seq, "[%s] ", pers->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007083
7084 spin_unlock(&pers_lock);
7085 seq_printf(seq, "\n");
Kay Sieversf1514632011-07-12 20:48:39 +02007086 seq->poll_event = atomic_read(&md_event_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007087 return 0;
7088 }
7089 if (v == (void*)2) {
7090 status_unused(seq);
7091 return 0;
7092 }
7093
NeilBrown36d091f2014-12-15 12:56:58 +11007094 spin_lock(&mddev->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007095 if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
7096 seq_printf(seq, "%s : %sactive", mdname(mddev),
7097 mddev->pers ? "" : "in");
7098 if (mddev->pers) {
NeilBrownf91de922005-11-08 21:39:36 -08007099 if (mddev->ro==1)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007100 seq_printf(seq, " (read-only)");
NeilBrownf91de922005-11-08 21:39:36 -08007101 if (mddev->ro==2)
NeilBrown52720ae2008-03-10 11:43:47 -07007102 seq_printf(seq, " (auto-read-only)");
Linus Torvalds1da177e2005-04-16 15:20:36 -07007103 seq_printf(seq, " %s", mddev->pers->name);
7104 }
7105
Andre Nolldd8ac332009-03-31 14:33:13 +11007106 sectors = 0;
NeilBrownf97fcad2014-12-15 12:56:59 +11007107 rcu_read_lock();
7108 rdev_for_each_rcu(rdev, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007109 char b[BDEVNAME_SIZE];
7110 seq_printf(seq, " %s[%d]",
7111 bdevname(rdev->bdev,b), rdev->desc_nr);
NeilBrown8ddf9ef2005-09-09 16:23:45 -07007112 if (test_bit(WriteMostly, &rdev->flags))
7113 seq_printf(seq, "(W)");
NeilBrownb2d444d2005-11-08 21:39:31 -08007114 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007115 seq_printf(seq, "(F)");
7116 continue;
NeilBrown2d78f8c2011-12-23 10:17:51 +11007117 }
7118 if (rdev->raid_disk < 0)
NeilBrownb325a322005-09-09 16:24:00 -07007119 seq_printf(seq, "(S)"); /* spare */
NeilBrown2d78f8c2011-12-23 10:17:51 +11007120 if (test_bit(Replacement, &rdev->flags))
7121 seq_printf(seq, "(R)");
Andre Nolldd8ac332009-03-31 14:33:13 +11007122 sectors += rdev->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007123 }
NeilBrownf97fcad2014-12-15 12:56:59 +11007124 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007125
7126 if (!list_empty(&mddev->disks)) {
7127 if (mddev->pers)
7128 seq_printf(seq, "\n %llu blocks",
Andre Nollf233ea52008-07-21 17:05:22 +10007129 (unsigned long long)
7130 mddev->array_sectors / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007131 else
7132 seq_printf(seq, "\n %llu blocks",
Andre Nolldd8ac332009-03-31 14:33:13 +11007133 (unsigned long long)sectors / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007134 }
NeilBrown1cd6bf12005-09-09 16:24:00 -07007135 if (mddev->persistent) {
7136 if (mddev->major_version != 0 ||
7137 mddev->minor_version != 90) {
7138 seq_printf(seq," super %d.%d",
7139 mddev->major_version,
7140 mddev->minor_version);
7141 }
NeilBrowne6910632008-02-06 01:39:51 -08007142 } else if (mddev->external)
7143 seq_printf(seq, " super external:%s",
7144 mddev->metadata_type);
7145 else
NeilBrown1cd6bf12005-09-09 16:24:00 -07007146 seq_printf(seq, " super non-persistent");
Linus Torvalds1da177e2005-04-16 15:20:36 -07007147
7148 if (mddev->pers) {
NeilBrownd710e132008-10-13 11:55:12 +11007149 mddev->pers->status(seq, mddev);
NeilBrownf72ffdd2014-09-30 14:23:59 +10007150 seq_printf(seq, "\n ");
NeilBrown8e1b39d2005-11-08 21:39:41 -08007151 if (mddev->pers->sync_request) {
7152 if (mddev->curr_resync > 2) {
NeilBrownd710e132008-10-13 11:55:12 +11007153 status_resync(seq, mddev);
NeilBrown8e1b39d2005-11-08 21:39:41 -08007154 seq_printf(seq, "\n ");
NeilBrown72f36d52012-10-11 14:25:57 +11007155 } else if (mddev->curr_resync >= 1)
NeilBrown8e1b39d2005-11-08 21:39:41 -08007156 seq_printf(seq, "\tresync=DELAYED\n ");
7157 else if (mddev->recovery_cp < MaxSector)
7158 seq_printf(seq, "\tresync=PENDING\n ");
7159 }
NeilBrown32a76272005-06-21 17:17:14 -07007160 } else
7161 seq_printf(seq, "\n ");
7162
NeilBrown57148962012-03-19 12:46:40 +11007163 bitmap_status(seq, mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007164
7165 seq_printf(seq, "\n");
7166 }
NeilBrown36d091f2014-12-15 12:56:58 +11007167 spin_unlock(&mddev->lock);
NeilBrownf72ffdd2014-09-30 14:23:59 +10007168
Linus Torvalds1da177e2005-04-16 15:20:36 -07007169 return 0;
7170}
7171
Jan Engelhardt110518b2009-05-07 12:49:37 +10007172static const struct seq_operations md_seq_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007173 .start = md_seq_start,
7174 .next = md_seq_next,
7175 .stop = md_seq_stop,
7176 .show = md_seq_show,
7177};
7178
7179static int md_seq_open(struct inode *inode, struct file *file)
7180{
Kay Sieversf1514632011-07-12 20:48:39 +02007181 struct seq_file *seq;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007182 int error;
7183
7184 error = seq_open(file, &md_seq_ops);
NeilBrownd7603b72006-01-06 00:20:30 -08007185 if (error)
Kay Sieversf1514632011-07-12 20:48:39 +02007186 return error;
7187
7188 seq = file->private_data;
7189 seq->poll_event = atomic_read(&md_event_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007190 return error;
7191}
7192
NeilBrowne2f23b62014-04-09 14:33:51 +10007193static int md_unloading;
NeilBrownd7603b72006-01-06 00:20:30 -08007194static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
7195{
Kay Sieversf1514632011-07-12 20:48:39 +02007196 struct seq_file *seq = filp->private_data;
NeilBrownd7603b72006-01-06 00:20:30 -08007197 int mask;
7198
NeilBrowne2f23b62014-04-09 14:33:51 +10007199 if (md_unloading)
kbuild test robot7d7e64f2014-12-03 16:07:59 +11007200 return POLLIN|POLLRDNORM|POLLERR|POLLPRI;
NeilBrownd7603b72006-01-06 00:20:30 -08007201 poll_wait(filp, &md_event_waiters, wait);
7202
7203 /* always allow read */
7204 mask = POLLIN | POLLRDNORM;
7205
Kay Sieversf1514632011-07-12 20:48:39 +02007206 if (seq->poll_event != atomic_read(&md_event_count))
NeilBrownd7603b72006-01-06 00:20:30 -08007207 mask |= POLLERR | POLLPRI;
7208 return mask;
7209}
7210
Arjan van de Venfa027c22007-02-12 00:55:33 -08007211static const struct file_operations md_seq_fops = {
Akinobu Mitae24650c2006-10-17 00:09:38 -07007212 .owner = THIS_MODULE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007213 .open = md_seq_open,
7214 .read = seq_read,
7215 .llseek = seq_lseek,
Martin Peschkec3f94b402007-05-09 02:35:35 -07007216 .release = seq_release_private,
NeilBrownd7603b72006-01-06 00:20:30 -08007217 .poll = mdstat_poll,
Linus Torvalds1da177e2005-04-16 15:20:36 -07007218};
7219
NeilBrown84fc4b52011-10-11 16:49:58 +11007220int register_md_personality(struct md_personality *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007221{
Chao Yu50bd3772014-09-25 15:28:34 +08007222 printk(KERN_INFO "md: %s personality registered for level %d\n",
7223 p->name, p->level);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007224 spin_lock(&pers_lock);
NeilBrown2604b702006-01-06 00:20:36 -08007225 list_add_tail(&p->list, &pers_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007226 spin_unlock(&pers_lock);
7227 return 0;
7228}
NeilBrown6c144d32014-09-30 16:15:38 +10007229EXPORT_SYMBOL(register_md_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007230
NeilBrown84fc4b52011-10-11 16:49:58 +11007231int unregister_md_personality(struct md_personality *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007232{
NeilBrown2604b702006-01-06 00:20:36 -08007233 printk(KERN_INFO "md: %s personality unregistered\n", p->name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007234 spin_lock(&pers_lock);
NeilBrown2604b702006-01-06 00:20:36 -08007235 list_del_init(&p->list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007236 spin_unlock(&pers_lock);
7237 return 0;
7238}
NeilBrown6c144d32014-09-30 16:15:38 +10007239EXPORT_SYMBOL(unregister_md_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007240
NeilBrownfd01b882011-10-11 16:47:53 +11007241static int is_mddev_idle(struct mddev *mddev, int init)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007242{
NeilBrownf72ffdd2014-09-30 14:23:59 +10007243 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007244 int idle;
NeilBrowneea1bf32009-03-31 14:27:02 +11007245 int curr_events;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007246
7247 idle = 1;
NeilBrown4b809912008-07-21 17:05:25 +10007248 rcu_read_lock();
7249 rdev_for_each_rcu(rdev, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007250 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
NeilBrowneea1bf32009-03-31 14:27:02 +11007251 curr_events = (int)part_stat_read(&disk->part0, sectors[0]) +
7252 (int)part_stat_read(&disk->part0, sectors[1]) -
7253 atomic_read(&disk->sync_io);
NeilBrown713f6ab2007-07-17 04:06:12 -07007254 /* sync IO will cause sync_io to increase before the disk_stats
7255 * as sync_io is counted when a request starts, and
7256 * disk_stats is counted when it completes.
7257 * So resync activity will cause curr_events to be smaller than
7258 * when there was no such activity.
7259 * non-sync IO will cause disk_stat to increase without
7260 * increasing sync_io so curr_events will (eventually)
7261 * be larger than it was before. Once it becomes
7262 * substantially larger, the test below will cause
7263 * the array to appear non-idle, and resync will slow
7264 * down.
7265 * If there is a lot of outstanding resync activity when
7266 * we set last_event to curr_events, then all that activity
7267 * completing might cause the array to appear non-idle
7268 * and resync will be slowed down even though there might
7269 * not have been non-resync activity. This will only
7270 * happen once though. 'last_events' will soon reflect
7271 * the state where there is little or no outstanding
7272 * resync requests, and further resync activity will
7273 * always make curr_events less than last_events.
NeilBrownc0e48522005-11-18 01:11:01 -08007274 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07007275 */
NeilBrowneea1bf32009-03-31 14:27:02 +11007276 if (init || curr_events - rdev->last_events > 64) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007277 rdev->last_events = curr_events;
7278 idle = 0;
7279 }
7280 }
NeilBrown4b809912008-07-21 17:05:25 +10007281 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07007282 return idle;
7283}
7284
NeilBrownfd01b882011-10-11 16:47:53 +11007285void md_done_sync(struct mddev *mddev, int blocks, int ok)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007286{
7287 /* another "blocks" (512byte) blocks have been synced */
7288 atomic_sub(blocks, &mddev->recovery_active);
7289 wake_up(&mddev->recovery_wait);
7290 if (!ok) {
NeilBrowndfc70642008-05-23 13:04:39 -07007291 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
majianpeng0a19caa2012-11-19 19:57:34 +08007292 set_bit(MD_RECOVERY_ERROR, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007293 md_wakeup_thread(mddev->thread);
7294 // stop recovery, signal do_sync ....
7295 }
7296}
NeilBrown6c144d32014-09-30 16:15:38 +10007297EXPORT_SYMBOL(md_done_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007298
NeilBrown06d91a52005-06-21 17:17:12 -07007299/* md_write_start(mddev, bi)
7300 * If we need to update some array metadata (e.g. 'active' flag
NeilBrown3d310eb2005-06-21 17:17:26 -07007301 * in superblock) before writing, schedule a superblock update
7302 * and wait for it to complete.
NeilBrown06d91a52005-06-21 17:17:12 -07007303 */
NeilBrownfd01b882011-10-11 16:47:53 +11007304void md_write_start(struct mddev *mddev, struct bio *bi)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007305{
Neil Brown0fd62b82008-06-28 08:31:36 +10007306 int did_change = 0;
NeilBrown06d91a52005-06-21 17:17:12 -07007307 if (bio_data_dir(bi) != WRITE)
NeilBrown3d310eb2005-06-21 17:17:26 -07007308 return;
NeilBrown06d91a52005-06-21 17:17:12 -07007309
NeilBrownf91de922005-11-08 21:39:36 -08007310 BUG_ON(mddev->ro == 1);
7311 if (mddev->ro == 2) {
7312 /* need to switch to read/write */
7313 mddev->ro = 0;
7314 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7315 md_wakeup_thread(mddev->thread);
NeilBrown25156192008-03-04 14:29:32 -08007316 md_wakeup_thread(mddev->sync_thread);
Neil Brown0fd62b82008-06-28 08:31:36 +10007317 did_change = 1;
NeilBrownf91de922005-11-08 21:39:36 -08007318 }
NeilBrown06d91a52005-06-21 17:17:12 -07007319 atomic_inc(&mddev->writes_pending);
NeilBrown31a59e32008-04-30 00:52:30 -07007320 if (mddev->safemode == 1)
7321 mddev->safemode = 0;
NeilBrown06d91a52005-06-21 17:17:12 -07007322 if (mddev->in_sync) {
NeilBrown85572d72014-12-15 12:56:56 +11007323 spin_lock(&mddev->lock);
NeilBrown3d310eb2005-06-21 17:17:26 -07007324 if (mddev->in_sync) {
7325 mddev->in_sync = 0;
NeilBrown850b2b42006-10-03 01:15:46 -07007326 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrown070dc6d2010-08-30 17:33:34 +10007327 set_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrown3d310eb2005-06-21 17:17:26 -07007328 md_wakeup_thread(mddev->thread);
Neil Brown0fd62b82008-06-28 08:31:36 +10007329 did_change = 1;
NeilBrown3d310eb2005-06-21 17:17:26 -07007330 }
NeilBrown85572d72014-12-15 12:56:56 +11007331 spin_unlock(&mddev->lock);
NeilBrown06d91a52005-06-21 17:17:12 -07007332 }
Neil Brown0fd62b82008-06-28 08:31:36 +10007333 if (did_change)
NeilBrown00bcb4a2010-06-01 19:37:23 +10007334 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrown09a44cc2008-05-23 13:04:36 -07007335 wait_event(mddev->sb_wait,
NeilBrown09a44cc2008-05-23 13:04:36 -07007336 !test_bit(MD_CHANGE_PENDING, &mddev->flags));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007337}
NeilBrown6c144d32014-09-30 16:15:38 +10007338EXPORT_SYMBOL(md_write_start);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007339
NeilBrownfd01b882011-10-11 16:47:53 +11007340void md_write_end(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007341{
7342 if (atomic_dec_and_test(&mddev->writes_pending)) {
7343 if (mddev->safemode == 2)
7344 md_wakeup_thread(mddev->thread);
NeilBrown16f17b32006-06-26 00:27:37 -07007345 else if (mddev->safemode_delay)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007346 mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
7347 }
7348}
NeilBrown6c144d32014-09-30 16:15:38 +10007349EXPORT_SYMBOL(md_write_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007350
NeilBrown2a2275d2007-01-26 00:57:11 -08007351/* md_allow_write(mddev)
7352 * Calling this ensures that the array is marked 'active' so that writes
7353 * may proceed without blocking. It is important to call this before
7354 * attempting a GFP_KERNEL allocation while holding the mddev lock.
7355 * Must be called with mddev_lock held.
Dan Williamsb5470dc2008-06-27 21:44:04 -07007356 *
7357 * In the ->external case MD_CHANGE_CLEAN can not be cleared until mddev->lock
7358 * is dropped, so return -EAGAIN after notifying userspace.
NeilBrown2a2275d2007-01-26 00:57:11 -08007359 */
NeilBrownfd01b882011-10-11 16:47:53 +11007360int md_allow_write(struct mddev *mddev)
NeilBrown2a2275d2007-01-26 00:57:11 -08007361{
7362 if (!mddev->pers)
Dan Williamsb5470dc2008-06-27 21:44:04 -07007363 return 0;
NeilBrown2a2275d2007-01-26 00:57:11 -08007364 if (mddev->ro)
Dan Williamsb5470dc2008-06-27 21:44:04 -07007365 return 0;
Neil Brown1a0fd492008-06-28 08:31:27 +10007366 if (!mddev->pers->sync_request)
Dan Williamsb5470dc2008-06-27 21:44:04 -07007367 return 0;
NeilBrown2a2275d2007-01-26 00:57:11 -08007368
NeilBrown85572d72014-12-15 12:56:56 +11007369 spin_lock(&mddev->lock);
NeilBrown2a2275d2007-01-26 00:57:11 -08007370 if (mddev->in_sync) {
7371 mddev->in_sync = 0;
7372 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrown070dc6d2010-08-30 17:33:34 +10007373 set_bit(MD_CHANGE_PENDING, &mddev->flags);
NeilBrown2a2275d2007-01-26 00:57:11 -08007374 if (mddev->safemode_delay &&
7375 mddev->safemode == 0)
7376 mddev->safemode = 1;
NeilBrown85572d72014-12-15 12:56:56 +11007377 spin_unlock(&mddev->lock);
NeilBrown2a2275d2007-01-26 00:57:11 -08007378 md_update_sb(mddev, 0);
NeilBrown00bcb4a2010-06-01 19:37:23 +10007379 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrown2a2275d2007-01-26 00:57:11 -08007380 } else
NeilBrown85572d72014-12-15 12:56:56 +11007381 spin_unlock(&mddev->lock);
Dan Williamsb5470dc2008-06-27 21:44:04 -07007382
NeilBrown070dc6d2010-08-30 17:33:34 +10007383 if (test_bit(MD_CHANGE_PENDING, &mddev->flags))
Dan Williamsb5470dc2008-06-27 21:44:04 -07007384 return -EAGAIN;
7385 else
7386 return 0;
NeilBrown2a2275d2007-01-26 00:57:11 -08007387}
7388EXPORT_SYMBOL_GPL(md_allow_write);
7389
Linus Torvalds1da177e2005-04-16 15:20:36 -07007390#define SYNC_MARKS 10
7391#define SYNC_MARK_STEP (3*HZ)
majianpeng54f89342012-10-31 11:59:10 +11007392#define UPDATE_FREQUENCY (5*60*HZ)
Shaohua Li4ed87312012-10-11 13:34:00 +11007393void md_do_sync(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007394{
Shaohua Li4ed87312012-10-11 13:34:00 +11007395 struct mddev *mddev = thread->mddev;
NeilBrownfd01b882011-10-11 16:47:53 +11007396 struct mddev *mddev2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007397 unsigned int currspeed = 0,
7398 window;
Xiao Niac7e50a2014-08-07 09:37:41 -04007399 sector_t max_sectors,j, io_sectors, recovery_done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007400 unsigned long mark[SYNC_MARKS];
majianpeng54f89342012-10-31 11:59:10 +11007401 unsigned long update_time;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007402 sector_t mark_cnt[SYNC_MARKS];
7403 int last_mark,m;
7404 struct list_head *tmp;
7405 sector_t last_check;
NeilBrown57afd892005-06-21 17:17:13 -07007406 int skipped = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11007407 struct md_rdev *rdev;
Jonathan Brassowc4a39552013-06-25 01:23:59 -05007408 char *desc, *action = NULL;
majianpeng7c2c57c2012-07-03 12:12:26 +10007409 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007410
7411 /* just incase thread restarts... */
7412 if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
7413 return;
NeilBrown3991b312014-05-28 13:39:23 +10007414 if (mddev->ro) {/* never try to sync a read-only array */
7415 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007416 return;
NeilBrown3991b312014-05-28 13:39:23 +10007417 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007418
NeilBrown61df9d92006-10-03 01:15:57 -07007419 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Jonathan Brassowc4a39552013-06-25 01:23:59 -05007420 if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
NeilBrown61df9d92006-10-03 01:15:57 -07007421 desc = "data-check";
Jonathan Brassowc4a39552013-06-25 01:23:59 -05007422 action = "check";
7423 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown61df9d92006-10-03 01:15:57 -07007424 desc = "requested-resync";
Jonathan Brassowc4a39552013-06-25 01:23:59 -05007425 action = "repair";
7426 } else
NeilBrown61df9d92006-10-03 01:15:57 -07007427 desc = "resync";
7428 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
7429 desc = "reshape";
7430 else
7431 desc = "recovery";
7432
Jonathan Brassowc4a39552013-06-25 01:23:59 -05007433 mddev->last_sync_action = action ?: desc;
7434
Linus Torvalds1da177e2005-04-16 15:20:36 -07007435 /* we overload curr_resync somewhat here.
7436 * 0 == not engaged in resync at all
7437 * 2 == checking that there is no conflict with another sync
7438 * 1 == like 2, but have yielded to allow conflicting resync to
7439 * commense
7440 * other == active in resync - this many blocks
7441 *
7442 * Before starting a resync we must have set curr_resync to
7443 * 2, and then checked that every "conflicting" array has curr_resync
7444 * less than ours. When we find one that is the same or higher
7445 * we wait on resync_wait. To avoid deadlock, we reduce curr_resync
7446 * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
7447 * This will mean we have to start checking from the beginning again.
7448 *
7449 */
7450
7451 do {
7452 mddev->curr_resync = 2;
7453
7454 try_again:
NeilBrown404e4b42009-12-30 15:25:23 +11007455 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
Linus Torvalds1da177e2005-04-16 15:20:36 -07007456 goto skip;
NeilBrown29ac4aa2008-02-06 01:39:58 -08007457 for_each_mddev(mddev2, tmp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007458 if (mddev2 == mddev)
7459 continue;
Bernd Schubert90b08712008-05-23 13:04:38 -07007460 if (!mddev->parallel_resync
7461 && mddev2->curr_resync
7462 && match_mddev_units(mddev, mddev2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007463 DEFINE_WAIT(wq);
7464 if (mddev < mddev2 && mddev->curr_resync == 2) {
7465 /* arbitrarily yield */
7466 mddev->curr_resync = 1;
7467 wake_up(&resync_wait);
7468 }
7469 if (mddev > mddev2 && mddev->curr_resync == 1)
7470 /* no need to wait here, we can wait the next
7471 * time 'round when curr_resync == 2
7472 */
7473 continue;
NeilBrown97441972008-09-19 11:49:54 +10007474 /* We need to wait 'interruptible' so as not to
7475 * contribute to the load average, and not to
7476 * be caught by 'softlockup'
7477 */
7478 prepare_to_wait(&resync_wait, &wq, TASK_INTERRUPTIBLE);
NeilBrownc91abf52013-11-19 12:02:01 +11007479 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
NeilBrown8712e552005-10-26 01:58:58 -07007480 mddev2->curr_resync >= mddev->curr_resync) {
NeilBrown61df9d92006-10-03 01:15:57 -07007481 printk(KERN_INFO "md: delaying %s of %s"
7482 " until %s has finished (they"
Linus Torvalds1da177e2005-04-16 15:20:36 -07007483 " share one or more physical units)\n",
NeilBrown61df9d92006-10-03 01:15:57 -07007484 desc, mdname(mddev), mdname(mddev2));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007485 mddev_put(mddev2);
NeilBrown97441972008-09-19 11:49:54 +10007486 if (signal_pending(current))
7487 flush_signals(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007488 schedule();
7489 finish_wait(&resync_wait, &wq);
7490 goto try_again;
7491 }
7492 finish_wait(&resync_wait, &wq);
7493 }
7494 }
7495 } while (mddev->curr_resync < 2);
7496
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007497 j = 0;
NeilBrown9d888832005-11-08 21:39:26 -08007498 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007499 /* resync follows the size requested by the personality,
NeilBrown57afd892005-06-21 17:17:13 -07007500 * which defaults to physical size, but can be virtual size
Linus Torvalds1da177e2005-04-16 15:20:36 -07007501 */
7502 max_sectors = mddev->resync_max_sectors;
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11007503 atomic64_set(&mddev->resync_mismatches, 0);
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007504 /* we don't use the checkpoint if there's a bitmap */
Neil Brown5e96ee62008-06-28 08:31:24 +10007505 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7506 j = mddev->resync_min;
7507 else if (!mddev->bitmap)
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007508 j = mddev->recovery_cp;
Neil Brown5e96ee62008-06-28 08:31:24 +10007509
NeilBrownccfcc3c2006-03-27 01:18:09 -08007510 } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
NeilBrownc804cde2012-05-21 09:28:33 +10007511 max_sectors = mddev->resync_max_sectors;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007512 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07007513 /* recovery follows the physical size of devices */
Andre Noll58c0fed2009-03-31 14:33:13 +11007514 max_sectors = mddev->dev_sectors;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007515 j = MaxSector;
Dan Williams4e59ca72009-12-12 21:17:06 -07007516 rcu_read_lock();
NeilBrowndafb20f2012-03-19 12:46:39 +11007517 rdev_for_each_rcu(rdev, mddev)
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007518 if (rdev->raid_disk >= 0 &&
7519 !test_bit(Faulty, &rdev->flags) &&
7520 !test_bit(In_sync, &rdev->flags) &&
7521 rdev->recovery_offset < j)
7522 j = rdev->recovery_offset;
Dan Williams4e59ca72009-12-12 21:17:06 -07007523 rcu_read_unlock();
NeilBrown133d4522014-07-02 12:04:14 +10007524
7525 /* If there is a bitmap, we need to make sure all
7526 * writes that started before we added a spare
7527 * complete before we start doing a recovery.
7528 * Otherwise the write might complete and (via
7529 * bitmap_endwrite) set a bit in the bitmap after the
7530 * recovery has checked that bit and skipped that
7531 * region.
7532 */
7533 if (mddev->bitmap) {
7534 mddev->pers->quiesce(mddev, 1);
7535 mddev->pers->quiesce(mddev, 0);
7536 }
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007537 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007538
NeilBrown61df9d92006-10-03 01:15:57 -07007539 printk(KERN_INFO "md: %s of RAID array %s\n", desc, mdname(mddev));
7540 printk(KERN_INFO "md: minimum _guaranteed_ speed:"
7541 " %d KB/sec/disk.\n", speed_min(mddev));
Adrian Bunk338cec32005-09-10 00:26:54 -07007542 printk(KERN_INFO "md: using maximum available idle IO bandwidth "
NeilBrown61df9d92006-10-03 01:15:57 -07007543 "(but not more than %d KB/sec) for %s.\n",
7544 speed_max(mddev), desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007545
NeilBrowneea1bf32009-03-31 14:27:02 +11007546 is_mddev_idle(mddev, 1); /* this initializes IO event counters */
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007547
NeilBrown57afd892005-06-21 17:17:13 -07007548 io_sectors = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007549 for (m = 0; m < SYNC_MARKS; m++) {
7550 mark[m] = jiffies;
NeilBrown57afd892005-06-21 17:17:13 -07007551 mark_cnt[m] = io_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007552 }
7553 last_mark = 0;
7554 mddev->resync_mark = mark[last_mark];
7555 mddev->resync_mark_cnt = mark_cnt[last_mark];
7556
7557 /*
7558 * Tune reconstruction:
7559 */
7560 window = 32*(PAGE_SIZE/512);
Jonathan Brassowac424502011-06-07 17:48:35 -05007561 printk(KERN_INFO "md: using %dk window, over a total of %lluk.\n",
7562 window/2, (unsigned long long)max_sectors/2);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007563
7564 atomic_set(&mddev->recovery_active, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007565 last_check = 0;
7566
7567 if (j>2) {
NeilBrownc91abf52013-11-19 12:02:01 +11007568 printk(KERN_INFO
NeilBrown61df9d92006-10-03 01:15:57 -07007569 "md: resuming %s of %s from checkpoint.\n",
7570 desc, mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07007571 mddev->curr_resync = j;
NeilBrown72f36d52012-10-11 14:25:57 +11007572 } else
7573 mddev->curr_resync = 3; /* no longer delayed */
NeilBrown75d3da42011-01-14 09:14:34 +11007574 mddev->curr_resync_completed = j;
NeilBrown72f36d52012-10-11 14:25:57 +11007575 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
7576 md_new_event(mddev);
majianpeng54f89342012-10-31 11:59:10 +11007577 update_time = jiffies;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007578
majianpeng7c2c57c2012-07-03 12:12:26 +10007579 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007580 while (j < max_sectors) {
NeilBrown57afd892005-06-21 17:17:13 -07007581 sector_t sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007582
NeilBrown57afd892005-06-21 17:17:13 -07007583 skipped = 0;
NeilBrown97e4f422009-03-31 14:33:13 +11007584
NeilBrown7a91ee12009-05-26 12:57:21 +10007585 if (!test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
7586 ((mddev->curr_resync > mddev->curr_resync_completed &&
7587 (mddev->curr_resync - mddev->curr_resync_completed)
7588 > (max_sectors >> 4)) ||
majianpeng54f89342012-10-31 11:59:10 +11007589 time_after_eq(jiffies, update_time + UPDATE_FREQUENCY) ||
NeilBrown7a91ee12009-05-26 12:57:21 +10007590 (j - mddev->curr_resync_completed)*2
7591 >= mddev->resync_max - mddev->curr_resync_completed
7592 )) {
NeilBrown97e4f422009-03-31 14:33:13 +11007593 /* time to update curr_resync_completed */
NeilBrown97e4f422009-03-31 14:33:13 +11007594 wait_event(mddev->recovery_wait,
7595 atomic_read(&mddev->recovery_active) == 0);
NeilBrown75d3da42011-01-14 09:14:34 +11007596 mddev->curr_resync_completed = j;
kernelmail35d78c62012-10-31 11:59:10 +11007597 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
7598 j > mddev->recovery_cp)
7599 mddev->recovery_cp = j;
majianpeng54f89342012-10-31 11:59:10 +11007600 update_time = jiffies;
NeilBrown070dc6d2010-08-30 17:33:34 +10007601 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrownacb180b2009-04-14 16:28:34 +10007602 sysfs_notify(&mddev->kobj, NULL, "sync_completed");
NeilBrown97e4f422009-03-31 14:33:13 +11007603 }
NeilBrownacb180b2009-04-14 16:28:34 +10007604
NeilBrownc91abf52013-11-19 12:02:01 +11007605 while (j >= mddev->resync_max &&
7606 !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
NeilBrowne62e58a2009-07-01 13:15:35 +10007607 /* As this condition is controlled by user-space,
7608 * we can block indefinitely, so use '_interruptible'
7609 * to avoid triggering warnings.
7610 */
7611 flush_signals(current); /* just in case */
7612 wait_event_interruptible(mddev->recovery_wait,
7613 mddev->resync_max > j
NeilBrownc91abf52013-11-19 12:02:01 +11007614 || test_bit(MD_RECOVERY_INTR,
7615 &mddev->recovery));
NeilBrowne62e58a2009-07-01 13:15:35 +10007616 }
NeilBrownacb180b2009-04-14 16:28:34 +10007617
NeilBrownc91abf52013-11-19 12:02:01 +11007618 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7619 break;
NeilBrownacb180b2009-04-14 16:28:34 +10007620
NeilBrown57afd892005-06-21 17:17:13 -07007621 sectors = mddev->pers->sync_request(mddev, j, &skipped,
NeilBrownc6207272008-02-06 01:39:52 -08007622 currspeed < speed_min(mddev));
NeilBrown57afd892005-06-21 17:17:13 -07007623 if (sectors == 0) {
NeilBrowndfc70642008-05-23 13:04:39 -07007624 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrownc91abf52013-11-19 12:02:01 +11007625 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007626 }
NeilBrown57afd892005-06-21 17:17:13 -07007627
7628 if (!skipped) { /* actual IO requested */
7629 io_sectors += sectors;
7630 atomic_add(sectors, &mddev->recovery_active);
7631 }
7632
NeilBrowne875ece2011-07-28 11:39:24 +10007633 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7634 break;
7635
Linus Torvalds1da177e2005-04-16 15:20:36 -07007636 j += sectors;
NeilBrown72f36d52012-10-11 14:25:57 +11007637 if (j > 2)
7638 mddev->curr_resync = j;
NeilBrownff4e8d92006-07-10 04:44:16 -07007639 mddev->curr_mark_cnt = io_sectors;
NeilBrownd7603b72006-01-06 00:20:30 -08007640 if (last_check == 0)
NeilBrowne875ece2011-07-28 11:39:24 +10007641 /* this is the earliest that rebuild will be
NeilBrownd7603b72006-01-06 00:20:30 -08007642 * visible in /proc/mdstat
7643 */
7644 md_new_event(mddev);
NeilBrown57afd892005-06-21 17:17:13 -07007645
7646 if (last_check + window > io_sectors || j == max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007647 continue;
7648
NeilBrown57afd892005-06-21 17:17:13 -07007649 last_check = io_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007650 repeat:
7651 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
7652 /* step marks */
7653 int next = (last_mark+1) % SYNC_MARKS;
7654
7655 mddev->resync_mark = mark[next];
7656 mddev->resync_mark_cnt = mark_cnt[next];
7657 mark[next] = jiffies;
NeilBrown57afd892005-06-21 17:17:13 -07007658 mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007659 last_mark = next;
7660 }
7661
NeilBrownc91abf52013-11-19 12:02:01 +11007662 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7663 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007664
7665 /*
7666 * this loop exits only if either when we are slower than
7667 * the 'hard' speed limit, or the system was IO-idle for
7668 * a jiffy.
7669 * the system might be non-idle CPU-wise, but we only care
7670 * about not overloading the IO subsystem. (things like an
7671 * e2fsck being done on the RAID array should execute fast)
7672 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07007673 cond_resched();
7674
Xiao Niac7e50a2014-08-07 09:37:41 -04007675 recovery_done = io_sectors - atomic_read(&mddev->recovery_active);
7676 currspeed = ((unsigned long)(recovery_done - mddev->resync_mark_cnt))/2
NeilBrown57afd892005-06-21 17:17:13 -07007677 /((jiffies-mddev->resync_mark)/HZ +1) +1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007678
NeilBrown88202a02006-01-06 00:21:36 -08007679 if (currspeed > speed_min(mddev)) {
7680 if ((currspeed > speed_max(mddev)) ||
NeilBrowneea1bf32009-03-31 14:27:02 +11007681 !is_mddev_idle(mddev, 0)) {
NeilBrownc0e48522005-11-18 01:11:01 -08007682 msleep(500);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007683 goto repeat;
7684 }
7685 }
7686 }
NeilBrownc91abf52013-11-19 12:02:01 +11007687 printk(KERN_INFO "md: %s: %s %s.\n",mdname(mddev), desc,
7688 test_bit(MD_RECOVERY_INTR, &mddev->recovery)
7689 ? "interrupted" : "done");
Linus Torvalds1da177e2005-04-16 15:20:36 -07007690 /*
7691 * this also signals 'finished resyncing' to md_stop
7692 */
majianpeng7c2c57c2012-07-03 12:12:26 +10007693 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007694 wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
7695
7696 /* tell personality that we are finished */
NeilBrown57afd892005-06-21 17:17:13 -07007697 mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007698
NeilBrowndfc70642008-05-23 13:04:39 -07007699 if (!test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007700 mddev->curr_resync > 2) {
7701 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
7702 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7703 if (mddev->curr_resync >= mddev->recovery_cp) {
7704 printk(KERN_INFO
NeilBrown61df9d92006-10-03 01:15:57 -07007705 "md: checkpointing %s of %s.\n",
7706 desc, mdname(mddev));
majianpeng0a19caa2012-11-19 19:57:34 +08007707 if (test_bit(MD_RECOVERY_ERROR,
7708 &mddev->recovery))
7709 mddev->recovery_cp =
7710 mddev->curr_resync_completed;
7711 else
7712 mddev->recovery_cp =
7713 mddev->curr_resync;
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007714 }
7715 } else
7716 mddev->recovery_cp = MaxSector;
7717 } else {
7718 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery))
7719 mddev->curr_resync = MaxSector;
Dan Williams4e59ca72009-12-12 21:17:06 -07007720 rcu_read_lock();
NeilBrowndafb20f2012-03-19 12:46:39 +11007721 rdev_for_each_rcu(rdev, mddev)
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007722 if (rdev->raid_disk >= 0 &&
NeilBrown70fffd02010-06-16 17:01:25 +10007723 mddev->delta_disks >= 0 &&
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007724 !test_bit(Faulty, &rdev->flags) &&
7725 !test_bit(In_sync, &rdev->flags) &&
7726 rdev->recovery_offset < mddev->curr_resync)
7727 rdev->recovery_offset = mddev->curr_resync;
Dan Williams4e59ca72009-12-12 21:17:06 -07007728 rcu_read_unlock();
NeilBrown5fd6c1d2006-06-26 00:27:40 -07007729 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07007730 }
NeilBrowndb91ff552012-02-07 12:01:51 +11007731 skip:
NeilBrown17571282006-12-10 02:20:52 -08007732 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007733
NeilBrown23da4222014-12-15 12:57:01 +11007734 spin_lock(&mddev->lock);
NeilBrownc07b70a2009-12-14 12:49:48 +11007735 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
7736 /* We completed so min/max setting can be forgotten if used. */
7737 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7738 mddev->resync_min = 0;
7739 mddev->resync_max = MaxSector;
7740 } else if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
7741 mddev->resync_min = mddev->curr_resync_completed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007742 mddev->curr_resync = 0;
NeilBrown23da4222014-12-15 12:57:01 +11007743 spin_unlock(&mddev->lock);
7744
Linus Torvalds1da177e2005-04-16 15:20:36 -07007745 wake_up(&resync_wait);
7746 set_bit(MD_RECOVERY_DONE, &mddev->recovery);
7747 md_wakeup_thread(mddev->thread);
NeilBrownc6207272008-02-06 01:39:52 -08007748 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007749}
NeilBrown29269552006-03-27 01:18:10 -08007750EXPORT_SYMBOL_GPL(md_do_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007751
NeilBrown746d3202013-04-24 11:42:41 +10007752static int remove_and_add_spares(struct mddev *mddev,
7753 struct md_rdev *this)
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007754{
NeilBrown3cb03002011-10-11 16:45:26 +11007755 struct md_rdev *rdev;
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007756 int spares = 0;
NeilBrownf2a371c2012-01-09 00:46:41 +11007757 int removed = 0;
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007758
NeilBrowndafb20f2012-03-19 12:46:39 +11007759 rdev_for_each(rdev, mddev)
NeilBrown746d3202013-04-24 11:42:41 +10007760 if ((this == NULL || rdev == this) &&
7761 rdev->raid_disk >= 0 &&
Dan Williams6bfe0b42008-04-30 00:52:32 -07007762 !test_bit(Blocked, &rdev->flags) &&
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007763 (test_bit(Faulty, &rdev->flags) ||
7764 ! test_bit(In_sync, &rdev->flags)) &&
7765 atomic_read(&rdev->nr_pending)==0) {
7766 if (mddev->pers->hot_remove_disk(
NeilBrownb8321b62011-12-23 10:17:51 +11007767 mddev, rdev) == 0) {
Namhyung Kim36fad852011-07-27 11:00:36 +10007768 sysfs_unlink_rdev(mddev, rdev);
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007769 rdev->raid_disk = -1;
NeilBrownf2a371c2012-01-09 00:46:41 +11007770 removed++;
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007771 }
7772 }
Jonathan Brassow90584fc2013-03-07 16:24:26 -06007773 if (removed && mddev->kobj.sd)
7774 sysfs_notify(&mddev->kobj, NULL, "degraded");
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007775
NeilBrown746d3202013-04-24 11:42:41 +10007776 if (this)
7777 goto no_add;
7778
NeilBrowndafb20f2012-03-19 12:46:39 +11007779 rdev_for_each(rdev, mddev) {
NeilBrown7bfec5f2011-12-23 10:17:53 +11007780 if (rdev->raid_disk >= 0 &&
7781 !test_bit(In_sync, &rdev->flags) &&
7782 !test_bit(Faulty, &rdev->flags))
7783 spares++;
NeilBrown7ceb17e2013-04-24 11:42:42 +10007784 if (rdev->raid_disk >= 0)
7785 continue;
7786 if (test_bit(Faulty, &rdev->flags))
7787 continue;
7788 if (mddev->ro &&
NeilBrown8313b8e2013-12-12 10:13:33 +11007789 ! (rdev->saved_raid_disk >= 0 &&
7790 !test_bit(Bitmap_sync, &rdev->flags)))
NeilBrown7ceb17e2013-04-24 11:42:42 +10007791 continue;
7792
NeilBrown7eb41882014-01-14 15:55:14 +11007793 if (rdev->saved_raid_disk < 0)
7794 rdev->recovery_offset = 0;
NeilBrown7ceb17e2013-04-24 11:42:42 +10007795 if (mddev->pers->
7796 hot_add_disk(mddev, rdev) == 0) {
7797 if (sysfs_link_rdev(mddev, rdev))
7798 /* failure here is OK */;
7799 spares++;
7800 md_new_event(mddev);
7801 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrowndfc70642008-05-23 13:04:39 -07007802 }
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007803 }
NeilBrown746d3202013-04-24 11:42:41 +10007804no_add:
NeilBrown6dafab62012-09-19 12:54:22 +10007805 if (removed)
7806 set_bit(MD_CHANGE_DEVS, &mddev->flags);
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007807 return spares;
7808}
NeilBrown7ebc0be2011-01-14 09:14:33 +11007809
NeilBrownac05f252014-09-30 08:10:42 +10007810static void md_start_sync(struct work_struct *ws)
7811{
7812 struct mddev *mddev = container_of(ws, struct mddev, del_work);
7813
7814 mddev->sync_thread = md_register_thread(md_do_sync,
7815 mddev,
7816 "resync");
7817 if (!mddev->sync_thread) {
7818 printk(KERN_ERR "%s: could not start resync"
7819 " thread...\n",
7820 mdname(mddev));
7821 /* leave the spares where they are, it shouldn't hurt */
7822 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7823 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
7824 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
7825 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
7826 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
NeilBrownf851b602014-12-11 10:02:10 +11007827 wake_up(&resync_wait);
NeilBrownac05f252014-09-30 08:10:42 +10007828 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
7829 &mddev->recovery))
7830 if (mddev->sysfs_action)
7831 sysfs_notify_dirent_safe(mddev->sysfs_action);
7832 } else
7833 md_wakeup_thread(mddev->sync_thread);
7834 sysfs_notify_dirent_safe(mddev->sysfs_action);
7835 md_new_event(mddev);
7836}
7837
Linus Torvalds1da177e2005-04-16 15:20:36 -07007838/*
7839 * This routine is regularly called by all per-raid-array threads to
7840 * deal with generic issues like resync and super-block update.
7841 * Raid personalities that don't have a thread (linear/raid0) do not
7842 * need this as they never do any recovery or update the superblock.
7843 *
7844 * It does not do any resync itself, but rather "forks" off other threads
7845 * to do that as needed.
7846 * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
7847 * "->recovery" and create a thread at ->sync_thread.
NeilBrowndfc70642008-05-23 13:04:39 -07007848 * When the thread finishes it sets MD_RECOVERY_DONE
Linus Torvalds1da177e2005-04-16 15:20:36 -07007849 * and wakeups up this thread which will reap the thread and finish up.
7850 * This thread also removes any faulty devices (with nr_pending == 0).
7851 *
7852 * The overall approach is:
7853 * 1/ if the superblock needs updating, update it.
7854 * 2/ If a recovery thread is running, don't do anything else.
7855 * 3/ If recovery has finished, clean up, possibly marking spares active.
7856 * 4/ If there are any faulty devices, remove them.
7857 * 5/ If array is degraded, try to add spares devices
7858 * 6/ If array has spares or is not in-sync, start a resync thread.
7859 */
NeilBrownfd01b882011-10-11 16:47:53 +11007860void md_check_recovery(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007861{
Jonathan Brassow68866e42011-06-08 15:10:08 +10007862 if (mddev->suspended)
7863 return;
7864
NeilBrown5f404022005-06-21 17:17:16 -07007865 if (mddev->bitmap)
NeilBrownaa5cbd12009-12-14 12:49:46 +11007866 bitmap_daemon_work(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007867
NeilBrownfca4d842005-06-21 17:17:11 -07007868 if (signal_pending(current)) {
NeilBrown31a59e32008-04-30 00:52:30 -07007869 if (mddev->pers->sync_request && !mddev->external) {
NeilBrownfca4d842005-06-21 17:17:11 -07007870 printk(KERN_INFO "md: %s in immediate safe mode\n",
7871 mdname(mddev));
7872 mddev->safemode = 2;
7873 }
7874 flush_signals(current);
7875 }
7876
NeilBrownc89a8ee2008-08-05 15:54:13 +10007877 if (mddev->ro && !test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
7878 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007879 if ( ! (
NeilBrown142d44c2013-11-28 10:34:18 +11007880 (mddev->flags & MD_UPDATE_SB_FLAGS & ~ (1<<MD_CHANGE_PENDING)) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07007881 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
NeilBrownfca4d842005-06-21 17:17:11 -07007882 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
NeilBrown31a59e32008-04-30 00:52:30 -07007883 (mddev->external == 0 && mddev->safemode == 1) ||
NeilBrownfca4d842005-06-21 17:17:11 -07007884 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
7885 && !mddev->in_sync && mddev->recovery_cp == MaxSector)
Linus Torvalds1da177e2005-04-16 15:20:36 -07007886 ))
7887 return;
NeilBrownfca4d842005-06-21 17:17:11 -07007888
NeilBrowndf5b89b2006-03-27 01:18:20 -08007889 if (mddev_trylock(mddev)) {
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007890 int spares = 0;
NeilBrownfca4d842005-06-21 17:17:11 -07007891
NeilBrownc89a8ee2008-08-05 15:54:13 +10007892 if (mddev->ro) {
NeilBrown7ceb17e2013-04-24 11:42:42 +10007893 /* On a read-only array we can:
7894 * - remove failed devices
7895 * - add already-in_sync devices if the array itself
7896 * is in-sync.
7897 * As we only add devices that are already in-sync,
7898 * we can activate the spares immediately.
NeilBrownc89a8ee2008-08-05 15:54:13 +10007899 */
NeilBrown7ceb17e2013-04-24 11:42:42 +10007900 remove_and_add_spares(mddev, NULL);
NeilBrown8313b8e2013-12-12 10:13:33 +11007901 /* There is no thread, but we need to call
7902 * ->spare_active and clear saved_raid_disk
7903 */
NeilBrown2ac295a2014-05-29 11:40:03 +10007904 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrown8313b8e2013-12-12 10:13:33 +11007905 md_reap_sync_thread(mddev);
7906 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
NeilBrownc89a8ee2008-08-05 15:54:13 +10007907 goto unlock;
7908 }
7909
NeilBrown31a59e32008-04-30 00:52:30 -07007910 if (!mddev->external) {
Neil Brown0fd62b82008-06-28 08:31:36 +10007911 int did_change = 0;
NeilBrown85572d72014-12-15 12:56:56 +11007912 spin_lock(&mddev->lock);
NeilBrown31a59e32008-04-30 00:52:30 -07007913 if (mddev->safemode &&
7914 !atomic_read(&mddev->writes_pending) &&
7915 !mddev->in_sync &&
7916 mddev->recovery_cp == MaxSector) {
7917 mddev->in_sync = 1;
Neil Brown0fd62b82008-06-28 08:31:36 +10007918 did_change = 1;
NeilBrown070dc6d2010-08-30 17:33:34 +10007919 set_bit(MD_CHANGE_CLEAN, &mddev->flags);
NeilBrown31a59e32008-04-30 00:52:30 -07007920 }
7921 if (mddev->safemode == 1)
7922 mddev->safemode = 0;
NeilBrown85572d72014-12-15 12:56:56 +11007923 spin_unlock(&mddev->lock);
Neil Brown0fd62b82008-06-28 08:31:36 +10007924 if (did_change)
NeilBrown00bcb4a2010-06-01 19:37:23 +10007925 sysfs_notify_dirent_safe(mddev->sysfs_state);
NeilBrownfca4d842005-06-21 17:17:11 -07007926 }
NeilBrownfca4d842005-06-21 17:17:11 -07007927
NeilBrown7a0a5352013-08-27 16:28:23 +10007928 if (mddev->flags & MD_UPDATE_SB_FLAGS)
NeilBrown850b2b42006-10-03 01:15:46 -07007929 md_update_sb(mddev, 0);
NeilBrown06d91a52005-06-21 17:17:12 -07007930
Linus Torvalds1da177e2005-04-16 15:20:36 -07007931 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
7932 !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
7933 /* resync/recovery still happening */
7934 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
7935 goto unlock;
7936 }
7937 if (mddev->sync_thread) {
Jonathan Brassowa91d5ac2013-04-24 11:42:43 +10007938 md_reap_sync_thread(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007939 goto unlock;
7940 }
Neil Brown72a23c22008-06-28 08:31:41 +10007941 /* Set RUNNING before clearing NEEDED to avoid
7942 * any transients in the value of "sync_action".
7943 */
NeilBrown72f36d52012-10-11 14:25:57 +11007944 mddev->curr_resync_completed = 0;
NeilBrown23da4222014-12-15 12:57:01 +11007945 spin_lock(&mddev->lock);
Neil Brown72a23c22008-06-28 08:31:41 +10007946 set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
NeilBrown23da4222014-12-15 12:57:01 +11007947 spin_unlock(&mddev->lock);
NeilBrown24dd4692005-11-08 21:39:26 -08007948 /* Clear some bits that don't mean anything, but
7949 * might be left set
7950 */
NeilBrown24dd4692005-11-08 21:39:26 -08007951 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
7952 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07007953
NeilBrowned209582012-04-24 10:23:14 +10007954 if (!test_and_clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
7955 test_bit(MD_RECOVERY_FROZEN, &mddev->recovery))
NeilBrownac05f252014-09-30 08:10:42 +10007956 goto not_running;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007957 /* no recovery is running.
7958 * remove any failed drives, then
7959 * add spares if possible.
NeilBrown72f36d52012-10-11 14:25:57 +11007960 * Spares are also removed and re-added, to allow
Linus Torvalds1da177e2005-04-16 15:20:36 -07007961 * the personality to fail the re-add.
7962 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07007963
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007964 if (mddev->reshape_position != MaxSector) {
NeilBrown50ac1682009-06-18 08:47:55 +10007965 if (mddev->pers->check_reshape == NULL ||
7966 mddev->pers->check_reshape(mddev) != 0)
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007967 /* Cannot proceed */
NeilBrownac05f252014-09-30 08:10:42 +10007968 goto not_running;
NeilBrownb4c4c7b2007-02-28 20:11:48 -08007969 set_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
Neil Brown72a23c22008-06-28 08:31:41 +10007970 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
NeilBrown746d3202013-04-24 11:42:41 +10007971 } else if ((spares = remove_and_add_spares(mddev, NULL))) {
NeilBrown24dd4692005-11-08 21:39:26 -08007972 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
7973 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
Dan Williams56ac36d2008-08-07 10:02:47 -07007974 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
Neil Brown72a23c22008-06-28 08:31:41 +10007975 set_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
NeilBrown24dd4692005-11-08 21:39:26 -08007976 } else if (mddev->recovery_cp < MaxSector) {
7977 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
Neil Brown72a23c22008-06-28 08:31:41 +10007978 clear_bit(MD_RECOVERY_RECOVER, &mddev->recovery);
NeilBrown24dd4692005-11-08 21:39:26 -08007979 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
7980 /* nothing to be done ... */
NeilBrownac05f252014-09-30 08:10:42 +10007981 goto not_running;
NeilBrown24dd4692005-11-08 21:39:26 -08007982
Linus Torvalds1da177e2005-04-16 15:20:36 -07007983 if (mddev->pers->sync_request) {
NeilBrownef99bf42012-05-22 13:55:08 +10007984 if (spares) {
NeilBrowna654b9d82005-06-21 17:17:27 -07007985 /* We are adding a device or devices to an array
7986 * which has the bitmap stored on all devices.
7987 * So make sure all bitmap pages get written
7988 */
7989 bitmap_write_all(mddev->bitmap);
7990 }
NeilBrownac05f252014-09-30 08:10:42 +10007991 INIT_WORK(&mddev->del_work, md_start_sync);
7992 queue_work(md_misc_wq, &mddev->del_work);
7993 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07007994 }
NeilBrownac05f252014-09-30 08:10:42 +10007995 not_running:
Neil Brown72a23c22008-06-28 08:31:41 +10007996 if (!mddev->sync_thread) {
7997 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
NeilBrownf851b602014-12-11 10:02:10 +11007998 wake_up(&resync_wait);
Neil Brown72a23c22008-06-28 08:31:41 +10007999 if (test_and_clear_bit(MD_RECOVERY_RECOVER,
8000 &mddev->recovery))
NeilBrown0c3573f2009-01-09 08:31:05 +11008001 if (mddev->sysfs_action)
NeilBrown00bcb4a2010-06-01 19:37:23 +10008002 sysfs_notify_dirent_safe(mddev->sysfs_action);
Neil Brown72a23c22008-06-28 08:31:41 +10008003 }
NeilBrownac05f252014-09-30 08:10:42 +10008004 unlock:
8005 wake_up(&mddev->sb_wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008006 mddev_unlock(mddev);
8007 }
8008}
NeilBrown6c144d32014-09-30 16:15:38 +10008009EXPORT_SYMBOL(md_check_recovery);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008010
Jonathan Brassowa91d5ac2013-04-24 11:42:43 +10008011void md_reap_sync_thread(struct mddev *mddev)
8012{
8013 struct md_rdev *rdev;
8014
8015 /* resync has finished, collect result */
8016 md_unregister_thread(&mddev->sync_thread);
8017 if (!test_bit(MD_RECOVERY_INTR, &mddev->recovery) &&
8018 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
8019 /* success...*/
8020 /* activate any spares */
8021 if (mddev->pers->spare_active(mddev)) {
8022 sysfs_notify(&mddev->kobj, NULL,
8023 "degraded");
8024 set_bit(MD_CHANGE_DEVS, &mddev->flags);
8025 }
8026 }
8027 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery) &&
8028 mddev->pers->finish_reshape)
8029 mddev->pers->finish_reshape(mddev);
8030
8031 /* If array is no-longer degraded, then any saved_raid_disk
NeilBrownf4667222013-12-09 12:04:56 +11008032 * information must be scrapped.
Jonathan Brassowa91d5ac2013-04-24 11:42:43 +10008033 */
NeilBrownf4667222013-12-09 12:04:56 +11008034 if (!mddev->degraded)
8035 rdev_for_each(rdev, mddev)
Jonathan Brassowa91d5ac2013-04-24 11:42:43 +10008036 rdev->saved_raid_disk = -1;
8037
8038 md_update_sb(mddev, 1);
8039 clear_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
8040 clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
8041 clear_bit(MD_RECOVERY_RESHAPE, &mddev->recovery);
8042 clear_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
8043 clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
NeilBrownf851b602014-12-11 10:02:10 +11008044 wake_up(&resync_wait);
Jonathan Brassowa91d5ac2013-04-24 11:42:43 +10008045 /* flag recovery needed just to double check */
8046 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
8047 sysfs_notify_dirent_safe(mddev->sysfs_action);
8048 md_new_event(mddev);
8049 if (mddev->event_work.func)
8050 queue_work(md_misc_wq, &mddev->event_work);
8051}
NeilBrown6c144d32014-09-30 16:15:38 +10008052EXPORT_SYMBOL(md_reap_sync_thread);
Jonathan Brassowa91d5ac2013-04-24 11:42:43 +10008053
NeilBrownfd01b882011-10-11 16:47:53 +11008054void md_wait_for_blocked_rdev(struct md_rdev *rdev, struct mddev *mddev)
Dan Williams6bfe0b42008-04-30 00:52:32 -07008055{
NeilBrown00bcb4a2010-06-01 19:37:23 +10008056 sysfs_notify_dirent_safe(rdev->sysfs_state);
Dan Williams6bfe0b42008-04-30 00:52:32 -07008057 wait_event_timeout(rdev->blocked_wait,
NeilBrownde393cd2011-07-28 11:31:48 +10008058 !test_bit(Blocked, &rdev->flags) &&
8059 !test_bit(BlockedBadBlocks, &rdev->flags),
Dan Williams6bfe0b42008-04-30 00:52:32 -07008060 msecs_to_jiffies(5000));
8061 rdev_dec_pending(rdev, mddev);
8062}
8063EXPORT_SYMBOL(md_wait_for_blocked_rdev);
8064
NeilBrownc6563a82012-05-21 09:27:00 +10008065void md_finish_reshape(struct mddev *mddev)
8066{
8067 /* called be personality module when reshape completes. */
8068 struct md_rdev *rdev;
8069
8070 rdev_for_each(rdev, mddev) {
8071 if (rdev->data_offset > rdev->new_data_offset)
8072 rdev->sectors += rdev->data_offset - rdev->new_data_offset;
8073 else
8074 rdev->sectors -= rdev->new_data_offset - rdev->data_offset;
8075 rdev->data_offset = rdev->new_data_offset;
8076 }
8077}
8078EXPORT_SYMBOL(md_finish_reshape);
NeilBrown2230dfe2011-07-28 11:31:46 +10008079
8080/* Bad block management.
8081 * We can record which blocks on each device are 'bad' and so just
8082 * fail those blocks, or that stripe, rather than the whole device.
8083 * Entries in the bad-block table are 64bits wide. This comprises:
8084 * Length of bad-range, in sectors: 0-511 for lengths 1-512
8085 * Start of bad-range, sector offset, 54 bits (allows 8 exbibytes)
8086 * A 'shift' can be set so that larger blocks are tracked and
8087 * consequently larger devices can be covered.
8088 * 'Acknowledged' flag - 1 bit. - the most significant bit.
8089 *
8090 * Locking of the bad-block table uses a seqlock so md_is_badblock
8091 * might need to retry if it is very unlucky.
8092 * We will sometimes want to check for bad blocks in a bi_end_io function,
8093 * so we use the write_seqlock_irq variant.
8094 *
8095 * When looking for a bad block we specify a range and want to
8096 * know if any block in the range is bad. So we binary-search
8097 * to the last range that starts at-or-before the given endpoint,
8098 * (or "before the sector after the target range")
8099 * then see if it ends after the given start.
8100 * We return
8101 * 0 if there are no known bad blocks in the range
8102 * 1 if there are known bad block which are all acknowledged
8103 * -1 if there are bad blocks which have not yet been acknowledged in metadata.
8104 * plus the start/length of the first bad section we overlap.
8105 */
8106int md_is_badblock(struct badblocks *bb, sector_t s, int sectors,
8107 sector_t *first_bad, int *bad_sectors)
8108{
8109 int hi;
majianpengab056132012-11-06 17:13:44 +08008110 int lo;
NeilBrown2230dfe2011-07-28 11:31:46 +10008111 u64 *p = bb->page;
majianpengab056132012-11-06 17:13:44 +08008112 int rv;
NeilBrown2230dfe2011-07-28 11:31:46 +10008113 sector_t target = s + sectors;
8114 unsigned seq;
8115
8116 if (bb->shift > 0) {
8117 /* round the start down, and the end up */
8118 s >>= bb->shift;
8119 target += (1<<bb->shift) - 1;
8120 target >>= bb->shift;
8121 sectors = target - s;
8122 }
8123 /* 'target' is now the first block after the bad range */
8124
8125retry:
8126 seq = read_seqbegin(&bb->lock);
majianpengab056132012-11-06 17:13:44 +08008127 lo = 0;
8128 rv = 0;
NeilBrown2230dfe2011-07-28 11:31:46 +10008129 hi = bb->count;
8130
8131 /* Binary search between lo and hi for 'target'
8132 * i.e. for the last range that starts before 'target'
8133 */
8134 /* INVARIANT: ranges before 'lo' and at-or-after 'hi'
8135 * are known not to be the last range before target.
8136 * VARIANT: hi-lo is the number of possible
8137 * ranges, and decreases until it reaches 1
8138 */
8139 while (hi - lo > 1) {
8140 int mid = (lo + hi) / 2;
8141 sector_t a = BB_OFFSET(p[mid]);
8142 if (a < target)
8143 /* This could still be the one, earlier ranges
8144 * could not. */
8145 lo = mid;
8146 else
8147 /* This and later ranges are definitely out. */
8148 hi = mid;
8149 }
8150 /* 'lo' might be the last that started before target, but 'hi' isn't */
8151 if (hi > lo) {
8152 /* need to check all range that end after 's' to see if
8153 * any are unacknowledged.
8154 */
8155 while (lo >= 0 &&
8156 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8157 if (BB_OFFSET(p[lo]) < target) {
8158 /* starts before the end, and finishes after
8159 * the start, so they must overlap
8160 */
8161 if (rv != -1 && BB_ACK(p[lo]))
8162 rv = 1;
8163 else
8164 rv = -1;
8165 *first_bad = BB_OFFSET(p[lo]);
8166 *bad_sectors = BB_LEN(p[lo]);
8167 }
8168 lo--;
8169 }
8170 }
8171
8172 if (read_seqretry(&bb->lock, seq))
8173 goto retry;
8174
8175 return rv;
8176}
8177EXPORT_SYMBOL_GPL(md_is_badblock);
8178
8179/*
8180 * Add a range of bad blocks to the table.
8181 * This might extend the table, or might contract it
8182 * if two adjacent ranges can be merged.
8183 * We binary-search to find the 'insertion' point, then
8184 * decide how best to handle it.
8185 */
8186static int md_set_badblocks(struct badblocks *bb, sector_t s, int sectors,
8187 int acknowledged)
8188{
8189 u64 *p;
8190 int lo, hi;
8191 int rv = 1;
Bian Yu905b0292013-10-12 01:10:03 -04008192 unsigned long flags;
NeilBrown2230dfe2011-07-28 11:31:46 +10008193
8194 if (bb->shift < 0)
8195 /* badblocks are disabled */
8196 return 0;
8197
8198 if (bb->shift) {
8199 /* round the start down, and the end up */
8200 sector_t next = s + sectors;
8201 s >>= bb->shift;
8202 next += (1<<bb->shift) - 1;
8203 next >>= bb->shift;
8204 sectors = next - s;
8205 }
8206
Bian Yu905b0292013-10-12 01:10:03 -04008207 write_seqlock_irqsave(&bb->lock, flags);
NeilBrown2230dfe2011-07-28 11:31:46 +10008208
8209 p = bb->page;
8210 lo = 0;
8211 hi = bb->count;
8212 /* Find the last range that starts at-or-before 's' */
8213 while (hi - lo > 1) {
8214 int mid = (lo + hi) / 2;
8215 sector_t a = BB_OFFSET(p[mid]);
8216 if (a <= s)
8217 lo = mid;
8218 else
8219 hi = mid;
8220 }
8221 if (hi > lo && BB_OFFSET(p[lo]) > s)
8222 hi = lo;
8223
8224 if (hi > lo) {
8225 /* we found a range that might merge with the start
8226 * of our new range
8227 */
8228 sector_t a = BB_OFFSET(p[lo]);
8229 sector_t e = a + BB_LEN(p[lo]);
8230 int ack = BB_ACK(p[lo]);
8231 if (e >= s) {
8232 /* Yes, we can merge with a previous range */
8233 if (s == a && s + sectors >= e)
8234 /* new range covers old */
8235 ack = acknowledged;
8236 else
8237 ack = ack && acknowledged;
8238
8239 if (e < s + sectors)
8240 e = s + sectors;
8241 if (e - a <= BB_MAX_LEN) {
8242 p[lo] = BB_MAKE(a, e-a, ack);
8243 s = e;
8244 } else {
8245 /* does not all fit in one range,
8246 * make p[lo] maximal
8247 */
8248 if (BB_LEN(p[lo]) != BB_MAX_LEN)
8249 p[lo] = BB_MAKE(a, BB_MAX_LEN, ack);
8250 s = a + BB_MAX_LEN;
8251 }
8252 sectors = e - s;
8253 }
8254 }
8255 if (sectors && hi < bb->count) {
8256 /* 'hi' points to the first range that starts after 's'.
8257 * Maybe we can merge with the start of that range */
8258 sector_t a = BB_OFFSET(p[hi]);
8259 sector_t e = a + BB_LEN(p[hi]);
8260 int ack = BB_ACK(p[hi]);
8261 if (a <= s + sectors) {
8262 /* merging is possible */
8263 if (e <= s + sectors) {
8264 /* full overlap */
8265 e = s + sectors;
8266 ack = acknowledged;
8267 } else
8268 ack = ack && acknowledged;
8269
8270 a = s;
8271 if (e - a <= BB_MAX_LEN) {
8272 p[hi] = BB_MAKE(a, e-a, ack);
8273 s = e;
8274 } else {
8275 p[hi] = BB_MAKE(a, BB_MAX_LEN, ack);
8276 s = a + BB_MAX_LEN;
8277 }
8278 sectors = e - s;
8279 lo = hi;
8280 hi++;
8281 }
8282 }
8283 if (sectors == 0 && hi < bb->count) {
8284 /* we might be able to combine lo and hi */
8285 /* Note: 's' is at the end of 'lo' */
8286 sector_t a = BB_OFFSET(p[hi]);
8287 int lolen = BB_LEN(p[lo]);
8288 int hilen = BB_LEN(p[hi]);
8289 int newlen = lolen + hilen - (s - a);
8290 if (s >= a && newlen < BB_MAX_LEN) {
8291 /* yes, we can combine them */
8292 int ack = BB_ACK(p[lo]) && BB_ACK(p[hi]);
8293 p[lo] = BB_MAKE(BB_OFFSET(p[lo]), newlen, ack);
8294 memmove(p + hi, p + hi + 1,
8295 (bb->count - hi - 1) * 8);
8296 bb->count--;
8297 }
8298 }
8299 while (sectors) {
8300 /* didn't merge (it all).
8301 * Need to add a range just before 'hi' */
8302 if (bb->count >= MD_MAX_BADBLOCKS) {
8303 /* No room for more */
8304 rv = 0;
8305 break;
8306 } else {
8307 int this_sectors = sectors;
8308 memmove(p + hi + 1, p + hi,
8309 (bb->count - hi) * 8);
8310 bb->count++;
8311
8312 if (this_sectors > BB_MAX_LEN)
8313 this_sectors = BB_MAX_LEN;
8314 p[hi] = BB_MAKE(s, this_sectors, acknowledged);
8315 sectors -= this_sectors;
8316 s += this_sectors;
8317 }
8318 }
8319
8320 bb->changed = 1;
NeilBrownde393cd2011-07-28 11:31:48 +10008321 if (!acknowledged)
8322 bb->unacked_exist = 1;
Bian Yu905b0292013-10-12 01:10:03 -04008323 write_sequnlock_irqrestore(&bb->lock, flags);
NeilBrown2230dfe2011-07-28 11:31:46 +10008324
8325 return rv;
8326}
8327
NeilBrown3cb03002011-10-11 16:45:26 +11008328int rdev_set_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
NeilBrownc6563a82012-05-21 09:27:00 +10008329 int is_new)
NeilBrown2230dfe2011-07-28 11:31:46 +10008330{
NeilBrownc6563a82012-05-21 09:27:00 +10008331 int rv;
8332 if (is_new)
8333 s += rdev->new_data_offset;
8334 else
8335 s += rdev->data_offset;
8336 rv = md_set_badblocks(&rdev->badblocks,
8337 s, sectors, 0);
NeilBrown2230dfe2011-07-28 11:31:46 +10008338 if (rv) {
8339 /* Make sure they get written out promptly */
NeilBrown8bd2f0a2011-12-08 16:26:08 +11008340 sysfs_notify_dirent_safe(rdev->sysfs_state);
NeilBrown2230dfe2011-07-28 11:31:46 +10008341 set_bit(MD_CHANGE_CLEAN, &rdev->mddev->flags);
8342 md_wakeup_thread(rdev->mddev->thread);
8343 }
8344 return rv;
8345}
8346EXPORT_SYMBOL_GPL(rdev_set_badblocks);
8347
8348/*
8349 * Remove a range of bad blocks from the table.
8350 * This may involve extending the table if we spilt a region,
8351 * but it must not fail. So if the table becomes full, we just
8352 * drop the remove request.
8353 */
8354static int md_clear_badblocks(struct badblocks *bb, sector_t s, int sectors)
8355{
8356 u64 *p;
8357 int lo, hi;
8358 sector_t target = s + sectors;
8359 int rv = 0;
8360
8361 if (bb->shift > 0) {
8362 /* When clearing we round the start up and the end down.
8363 * This should not matter as the shift should align with
8364 * the block size and no rounding should ever be needed.
8365 * However it is better the think a block is bad when it
8366 * isn't than to think a block is not bad when it is.
8367 */
8368 s += (1<<bb->shift) - 1;
8369 s >>= bb->shift;
8370 target >>= bb->shift;
8371 sectors = target - s;
8372 }
8373
8374 write_seqlock_irq(&bb->lock);
8375
8376 p = bb->page;
8377 lo = 0;
8378 hi = bb->count;
8379 /* Find the last range that starts before 'target' */
8380 while (hi - lo > 1) {
8381 int mid = (lo + hi) / 2;
8382 sector_t a = BB_OFFSET(p[mid]);
8383 if (a < target)
8384 lo = mid;
8385 else
8386 hi = mid;
8387 }
8388 if (hi > lo) {
8389 /* p[lo] is the last range that could overlap the
8390 * current range. Earlier ranges could also overlap,
8391 * but only this one can overlap the end of the range.
8392 */
8393 if (BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > target) {
8394 /* Partial overlap, leave the tail of this range */
8395 int ack = BB_ACK(p[lo]);
8396 sector_t a = BB_OFFSET(p[lo]);
8397 sector_t end = a + BB_LEN(p[lo]);
8398
8399 if (a < s) {
8400 /* we need to split this range */
8401 if (bb->count >= MD_MAX_BADBLOCKS) {
NeilBrown8b32bf52014-05-28 13:39:22 +10008402 rv = -ENOSPC;
NeilBrown2230dfe2011-07-28 11:31:46 +10008403 goto out;
8404 }
8405 memmove(p+lo+1, p+lo, (bb->count - lo) * 8);
8406 bb->count++;
8407 p[lo] = BB_MAKE(a, s-a, ack);
8408 lo++;
8409 }
8410 p[lo] = BB_MAKE(target, end - target, ack);
8411 /* there is no longer an overlap */
8412 hi = lo;
8413 lo--;
8414 }
8415 while (lo >= 0 &&
8416 BB_OFFSET(p[lo]) + BB_LEN(p[lo]) > s) {
8417 /* This range does overlap */
8418 if (BB_OFFSET(p[lo]) < s) {
8419 /* Keep the early parts of this range. */
8420 int ack = BB_ACK(p[lo]);
8421 sector_t start = BB_OFFSET(p[lo]);
8422 p[lo] = BB_MAKE(start, s - start, ack);
8423 /* now low doesn't overlap, so.. */
8424 break;
8425 }
8426 lo--;
8427 }
8428 /* 'lo' is strictly before, 'hi' is strictly after,
8429 * anything between needs to be discarded
8430 */
8431 if (hi - lo > 1) {
8432 memmove(p+lo+1, p+hi, (bb->count - hi) * 8);
8433 bb->count -= (hi - lo - 1);
8434 }
8435 }
8436
8437 bb->changed = 1;
8438out:
8439 write_sequnlock_irq(&bb->lock);
8440 return rv;
8441}
8442
NeilBrownc6563a82012-05-21 09:27:00 +10008443int rdev_clear_badblocks(struct md_rdev *rdev, sector_t s, int sectors,
8444 int is_new)
NeilBrown2230dfe2011-07-28 11:31:46 +10008445{
NeilBrownc6563a82012-05-21 09:27:00 +10008446 if (is_new)
8447 s += rdev->new_data_offset;
8448 else
8449 s += rdev->data_offset;
NeilBrown2230dfe2011-07-28 11:31:46 +10008450 return md_clear_badblocks(&rdev->badblocks,
NeilBrownc6563a82012-05-21 09:27:00 +10008451 s, sectors);
NeilBrown2230dfe2011-07-28 11:31:46 +10008452}
8453EXPORT_SYMBOL_GPL(rdev_clear_badblocks);
8454
8455/*
8456 * Acknowledge all bad blocks in a list.
8457 * This only succeeds if ->changed is clear. It is used by
8458 * in-kernel metadata updates
8459 */
8460void md_ack_all_badblocks(struct badblocks *bb)
8461{
8462 if (bb->page == NULL || bb->changed)
8463 /* no point even trying */
8464 return;
8465 write_seqlock_irq(&bb->lock);
8466
majianpengecb178b2012-03-19 12:46:42 +11008467 if (bb->changed == 0 && bb->unacked_exist) {
NeilBrown2230dfe2011-07-28 11:31:46 +10008468 u64 *p = bb->page;
8469 int i;
8470 for (i = 0; i < bb->count ; i++) {
8471 if (!BB_ACK(p[i])) {
8472 sector_t start = BB_OFFSET(p[i]);
8473 int len = BB_LEN(p[i]);
8474 p[i] = BB_MAKE(start, len, 1);
8475 }
8476 }
NeilBrownde393cd2011-07-28 11:31:48 +10008477 bb->unacked_exist = 0;
NeilBrown2230dfe2011-07-28 11:31:46 +10008478 }
8479 write_sequnlock_irq(&bb->lock);
8480}
8481EXPORT_SYMBOL_GPL(md_ack_all_badblocks);
8482
NeilBrown16c791a2011-07-28 11:31:47 +10008483/* sysfs access to bad-blocks list.
8484 * We present two files.
8485 * 'bad-blocks' lists sector numbers and lengths of ranges that
8486 * are recorded as bad. The list is truncated to fit within
8487 * the one-page limit of sysfs.
8488 * Writing "sector length" to this file adds an acknowledged
8489 * bad block list.
8490 * 'unacknowledged-bad-blocks' lists bad blocks that have not yet
8491 * been acknowledged. Writing to this file adds bad blocks
8492 * without acknowledging them. This is largely for testing.
8493 */
8494
8495static ssize_t
8496badblocks_show(struct badblocks *bb, char *page, int unack)
8497{
8498 size_t len;
8499 int i;
8500 u64 *p = bb->page;
8501 unsigned seq;
8502
8503 if (bb->shift < 0)
8504 return 0;
8505
8506retry:
8507 seq = read_seqbegin(&bb->lock);
8508
8509 len = 0;
8510 i = 0;
8511
8512 while (len < PAGE_SIZE && i < bb->count) {
8513 sector_t s = BB_OFFSET(p[i]);
8514 unsigned int length = BB_LEN(p[i]);
8515 int ack = BB_ACK(p[i]);
8516 i++;
8517
8518 if (unack && ack)
8519 continue;
8520
8521 len += snprintf(page+len, PAGE_SIZE-len, "%llu %u\n",
8522 (unsigned long long)s << bb->shift,
8523 length << bb->shift);
8524 }
NeilBrownde393cd2011-07-28 11:31:48 +10008525 if (unack && len == 0)
8526 bb->unacked_exist = 0;
NeilBrown16c791a2011-07-28 11:31:47 +10008527
8528 if (read_seqretry(&bb->lock, seq))
8529 goto retry;
8530
8531 return len;
8532}
8533
8534#define DO_DEBUG 1
8535
8536static ssize_t
8537badblocks_store(struct badblocks *bb, const char *page, size_t len, int unack)
8538{
8539 unsigned long long sector;
8540 int length;
8541 char newline;
8542#ifdef DO_DEBUG
8543 /* Allow clearing via sysfs *only* for testing/debugging.
8544 * Normally only a successful write may clear a badblock
8545 */
8546 int clear = 0;
8547 if (page[0] == '-') {
8548 clear = 1;
8549 page++;
8550 }
8551#endif /* DO_DEBUG */
8552
8553 switch (sscanf(page, "%llu %d%c", &sector, &length, &newline)) {
8554 case 3:
8555 if (newline != '\n')
8556 return -EINVAL;
8557 case 2:
8558 if (length <= 0)
8559 return -EINVAL;
8560 break;
8561 default:
8562 return -EINVAL;
8563 }
8564
8565#ifdef DO_DEBUG
8566 if (clear) {
8567 md_clear_badblocks(bb, sector, length);
8568 return len;
8569 }
8570#endif /* DO_DEBUG */
8571 if (md_set_badblocks(bb, sector, length, !unack))
8572 return len;
8573 else
8574 return -ENOSPC;
8575}
8576
Adrian Bunk75c96f82005-05-05 16:16:09 -07008577static int md_notify_reboot(struct notifier_block *this,
8578 unsigned long code, void *x)
Linus Torvalds1da177e2005-04-16 15:20:36 -07008579{
8580 struct list_head *tmp;
NeilBrownfd01b882011-10-11 16:47:53 +11008581 struct mddev *mddev;
Daniel P. Berrange2dba6a92011-09-23 10:40:45 +01008582 int need_delay = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008583
NeilBrownc744a652012-03-19 12:46:37 +11008584 for_each_mddev(mddev, tmp) {
8585 if (mddev_trylock(mddev)) {
NeilBrown30b8aa92012-04-24 10:23:16 +10008586 if (mddev->pers)
8587 __md_stop_writes(mddev);
NeilBrown0f62fb22014-05-06 09:36:08 +10008588 if (mddev->persistent)
8589 mddev->safemode = 2;
NeilBrownc744a652012-03-19 12:46:37 +11008590 mddev_unlock(mddev);
Daniel P. Berrange2dba6a92011-09-23 10:40:45 +01008591 }
NeilBrownc744a652012-03-19 12:46:37 +11008592 need_delay = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008593 }
NeilBrownc744a652012-03-19 12:46:37 +11008594 /*
8595 * certain more exotic SCSI devices are known to be
8596 * volatile wrt too early system reboots. While the
8597 * right place to handle this issue is the given
8598 * driver, we do want to have a safe RAID driver ...
8599 */
8600 if (need_delay)
8601 mdelay(1000*1);
8602
Linus Torvalds1da177e2005-04-16 15:20:36 -07008603 return NOTIFY_DONE;
8604}
8605
Adrian Bunk75c96f82005-05-05 16:16:09 -07008606static struct notifier_block md_notifier = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07008607 .notifier_call = md_notify_reboot,
8608 .next = NULL,
8609 .priority = INT_MAX, /* before any real devices */
8610};
8611
8612static void md_geninit(void)
8613{
NeilBrown36a4e1f2011-10-07 14:23:17 +11008614 pr_debug("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
Linus Torvalds1da177e2005-04-16 15:20:36 -07008615
Denis V. Lunevc7705f3442008-04-29 01:02:35 -07008616 proc_create("mdstat", S_IRUGO, NULL, &md_seq_fops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008617}
8618
Adrian Bunk75c96f82005-05-05 16:16:09 -07008619static int __init md_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07008620{
Tejun Heoe804ac72010-10-15 15:36:08 +02008621 int ret = -ENOMEM;
8622
Tejun Heoada609e2011-01-25 14:35:54 +01008623 md_wq = alloc_workqueue("md", WQ_MEM_RECLAIM, 0);
Tejun Heoe804ac72010-10-15 15:36:08 +02008624 if (!md_wq)
8625 goto err_wq;
8626
8627 md_misc_wq = alloc_workqueue("md_misc", 0, 0);
8628 if (!md_misc_wq)
8629 goto err_misc_wq;
8630
8631 if ((ret = register_blkdev(MD_MAJOR, "md")) < 0)
8632 goto err_md;
8633
8634 if ((ret = register_blkdev(0, "mdp")) < 0)
8635 goto err_mdp;
8636 mdp_major = ret;
8637
NeilBrownaf5628f2014-07-31 13:54:54 +10008638 blk_register_region(MKDEV(MD_MAJOR, 0), 512, THIS_MODULE,
NeilBrowne8703fe2006-10-03 01:15:59 -07008639 md_probe, NULL, NULL);
8640 blk_register_region(MKDEV(mdp_major, 0), 1UL<<MINORBITS, THIS_MODULE,
Linus Torvalds1da177e2005-04-16 15:20:36 -07008641 md_probe, NULL, NULL);
8642
Linus Torvalds1da177e2005-04-16 15:20:36 -07008643 register_reboot_notifier(&md_notifier);
Eric W. Biederman0b4d4142007-02-14 00:34:09 -08008644 raid_table_header = register_sysctl_table(raid_root_table);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008645
8646 md_geninit();
NeilBrownd710e132008-10-13 11:55:12 +11008647 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008648
Tejun Heoe804ac72010-10-15 15:36:08 +02008649err_mdp:
8650 unregister_blkdev(MD_MAJOR, "md");
8651err_md:
8652 destroy_workqueue(md_misc_wq);
8653err_misc_wq:
8654 destroy_workqueue(md_wq);
8655err_wq:
8656 return ret;
8657}
Linus Torvalds1da177e2005-04-16 15:20:36 -07008658
8659#ifndef MODULE
8660
8661/*
8662 * Searches all registered partitions for autorun RAID arrays
8663 * at boot time.
8664 */
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008665
8666static LIST_HEAD(all_detected_devices);
8667struct detected_devices_node {
8668 struct list_head list;
8669 dev_t dev;
8670};
Linus Torvalds1da177e2005-04-16 15:20:36 -07008671
8672void md_autodetect_dev(dev_t dev)
8673{
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008674 struct detected_devices_node *node_detected_dev;
8675
8676 node_detected_dev = kzalloc(sizeof(*node_detected_dev), GFP_KERNEL);
8677 if (node_detected_dev) {
8678 node_detected_dev->dev = dev;
8679 list_add_tail(&node_detected_dev->list, &all_detected_devices);
8680 } else {
8681 printk(KERN_CRIT "md: md_autodetect_dev: kzalloc failed"
8682 ", skipping dev(%d,%d)\n", MAJOR(dev), MINOR(dev));
8683 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07008684}
8685
Linus Torvalds1da177e2005-04-16 15:20:36 -07008686static void autostart_arrays(int part)
8687{
NeilBrown3cb03002011-10-11 16:45:26 +11008688 struct md_rdev *rdev;
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008689 struct detected_devices_node *node_detected_dev;
8690 dev_t dev;
8691 int i_scanned, i_passed;
8692
8693 i_scanned = 0;
8694 i_passed = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008695
8696 printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
8697
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008698 while (!list_empty(&all_detected_devices) && i_scanned < INT_MAX) {
8699 i_scanned++;
8700 node_detected_dev = list_entry(all_detected_devices.next,
8701 struct detected_devices_node, list);
8702 list_del(&node_detected_dev->list);
8703 dev = node_detected_dev->dev;
8704 kfree(node_detected_dev);
NeilBrowndf968c42007-07-17 04:06:11 -07008705 rdev = md_import_device(dev,0, 90);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008706 if (IS_ERR(rdev))
8707 continue;
8708
NeilBrown403df472014-09-30 15:52:29 +10008709 if (test_bit(Faulty, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07008710 continue;
NeilBrown403df472014-09-30 15:52:29 +10008711
NeilBrownd0fae182008-03-04 14:29:31 -08008712 set_bit(AutoDetected, &rdev->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008713 list_add(&rdev->same_set, &pending_raid_disks);
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008714 i_passed++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008715 }
Michael J. Evans4d936ec2007-10-16 23:30:52 -07008716
8717 printk(KERN_INFO "md: Scanned %d and added %d devices.\n",
8718 i_scanned, i_passed);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008719
8720 autorun_devices(part);
8721}
8722
Jeff Garzikfdee8ae2006-12-10 02:20:50 -08008723#endif /* !MODULE */
Linus Torvalds1da177e2005-04-16 15:20:36 -07008724
8725static __exit void md_exit(void)
8726{
NeilBrownfd01b882011-10-11 16:47:53 +11008727 struct mddev *mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008728 struct list_head *tmp;
NeilBrowne2f23b62014-04-09 14:33:51 +10008729 int delay = 1;
Greg Kroah-Hartman8ab5e4c2005-06-20 21:15:16 -07008730
NeilBrownaf5628f2014-07-31 13:54:54 +10008731 blk_unregister_region(MKDEV(MD_MAJOR,0), 512);
NeilBrowne8703fe2006-10-03 01:15:59 -07008732 blk_unregister_region(MKDEV(mdp_major,0), 1U << MINORBITS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008733
Christoph Hellwig3dbd8c22009-03-31 14:27:02 +11008734 unregister_blkdev(MD_MAJOR,"md");
Linus Torvalds1da177e2005-04-16 15:20:36 -07008735 unregister_blkdev(mdp_major, "mdp");
8736 unregister_reboot_notifier(&md_notifier);
8737 unregister_sysctl_table(raid_table_header);
NeilBrowne2f23b62014-04-09 14:33:51 +10008738
8739 /* We cannot unload the modules while some process is
8740 * waiting for us in select() or poll() - wake them up
8741 */
8742 md_unloading = 1;
8743 while (waitqueue_active(&md_event_waiters)) {
8744 /* not safe to leave yet */
8745 wake_up(&md_event_waiters);
8746 msleep(delay);
8747 delay += delay;
8748 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07008749 remove_proc_entry("mdstat", NULL);
NeilBrowne2f23b62014-04-09 14:33:51 +10008750
NeilBrown29ac4aa2008-02-06 01:39:58 -08008751 for_each_mddev(mddev, tmp) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07008752 export_array(mddev);
NeilBrownd3374822009-01-09 08:31:10 +11008753 mddev->hold_active = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07008754 }
Tejun Heoe804ac72010-10-15 15:36:08 +02008755 destroy_workqueue(md_misc_wq);
8756 destroy_workqueue(md_wq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008757}
8758
Dan Williams685784a2007-07-09 11:56:42 -07008759subsys_initcall(md_init);
Linus Torvalds1da177e2005-04-16 15:20:36 -07008760module_exit(md_exit)
8761
NeilBrownf91de922005-11-08 21:39:36 -08008762static int get_ro(char *buffer, struct kernel_param *kp)
8763{
8764 return sprintf(buffer, "%d", start_readonly);
8765}
8766static int set_ro(const char *val, struct kernel_param *kp)
8767{
8768 char *e;
8769 int num = simple_strtoul(val, &e, 10);
8770 if (*val && (*e == '\0' || *e == '\n')) {
8771 start_readonly = num;
NeilBrown4dbcdc72006-01-06 00:20:52 -08008772 return 0;
NeilBrownf91de922005-11-08 21:39:36 -08008773 }
8774 return -EINVAL;
8775}
8776
NeilBrown80ca3a42006-07-10 04:44:18 -07008777module_param_call(start_ro, set_ro, get_ro, NULL, S_IRUSR|S_IWUSR);
8778module_param(start_dirty_degraded, int, S_IRUGO|S_IWUSR);
NeilBrownefeb53c2009-01-09 08:31:10 +11008779module_param_call(new_array, add_named_array, NULL, NULL, S_IWUSR);
NeilBrownf91de922005-11-08 21:39:36 -08008780
Linus Torvalds1da177e2005-04-16 15:20:36 -07008781MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11008782MODULE_DESCRIPTION("MD RAID framework");
NeilBrownaa1595e2005-08-04 12:53:32 -07008783MODULE_ALIAS("md");
NeilBrown72008652005-08-26 18:34:15 -07008784MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);