blob: 859bd3ffe4351bad9204aa04e906e751a788c64a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * raid1.c : Multiple Devices driver for Linux
3 *
4 * Copyright (C) 1999, 2000, 2001 Ingo Molnar, Red Hat
5 *
6 * Copyright (C) 1996, 1997, 1998 Ingo Molnar, Miguel de Icaza, Gadi Oxman
7 *
8 * RAID-1 management functions.
9 *
10 * Better read-balancing code written by Mika Kuoppala <miku@iki.fi>, 2000
11 *
Jan Engelhardt96de0e22007-10-19 23:21:04 +020012 * Fixes to reconstruction by Jakob Østergaard" <jakob@ostenfeld.dk>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 * Various fixes by Neil Brown <neilb@cse.unsw.edu.au>
14 *
NeilBrown191ea9b2005-06-21 17:17:23 -070015 * Changes by Peter T. Breuer <ptb@it.uc3m.es> 31/1/2003 to support
16 * bitmapped intelligence in resync:
17 *
18 * - bitmap marked during normal i/o
19 * - bitmap used to skip nondirty blocks during sync
20 *
21 * Additions to bitmap code, (C) 2003-2004 Paul Clements, SteelEye Technology:
22 * - persistent bitmap code
23 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070024 * This program is free software; you can redistribute it and/or modify
25 * it under the terms of the GNU General Public License as published by
26 * the Free Software Foundation; either version 2, or (at your option)
27 * any later version.
28 *
29 * You should have received a copy of the GNU General Public License
30 * (for example /usr/src/linux/COPYING); if not, write to the Free
31 * Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
32 */
33
Stephen Rothwell25570722008-10-15 09:09:21 +110034#include <linux/delay.h>
NeilBrownbff61972009-03-31 14:33:13 +110035#include <linux/blkdev.h>
NeilBrownbff61972009-03-31 14:33:13 +110036#include <linux/seq_file.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110037#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110038#include "raid1.h"
39#include "bitmap.h"
NeilBrown191ea9b2005-06-21 17:17:23 -070040
41#define DEBUG 0
42#if DEBUG
43#define PRINTK(x...) printk(x)
44#else
45#define PRINTK(x...)
46#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48/*
49 * Number of guaranteed r1bios in case of extreme VM load:
50 */
51#define NR_RAID1_BIOS 256
52
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54static void unplug_slaves(mddev_t *mddev);
55
NeilBrown17999be2006-01-06 00:20:12 -080056static void allow_barrier(conf_t *conf);
57static void lower_barrier(conf_t *conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Al Virodd0fc662005-10-07 07:46:04 +010059static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070060{
61 struct pool_info *pi = data;
62 r1bio_t *r1_bio;
63 int size = offsetof(r1bio_t, bios[pi->raid_disks]);
64
65 /* allocate a r1bio with room for raid_disks entries in the bios array */
NeilBrown9ffae0c2006-01-06 00:20:32 -080066 r1_bio = kzalloc(size, gfp_flags);
NeilBrowned9bfdf2009-10-16 15:55:44 +110067 if (!r1_bio && pi->mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -070068 unplug_slaves(pi->mddev);
69
70 return r1_bio;
71}
72
73static void r1bio_pool_free(void *r1_bio, void *data)
74{
75 kfree(r1_bio);
76}
77
78#define RESYNC_BLOCK_SIZE (64*1024)
79//#define RESYNC_BLOCK_SIZE PAGE_SIZE
80#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
81#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
82#define RESYNC_WINDOW (2048*1024)
83
Al Virodd0fc662005-10-07 07:46:04 +010084static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070085{
86 struct pool_info *pi = data;
87 struct page *page;
88 r1bio_t *r1_bio;
89 struct bio *bio;
90 int i, j;
91
92 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
93 if (!r1_bio) {
94 unplug_slaves(pi->mddev);
95 return NULL;
96 }
97
98 /*
99 * Allocate bios : 1 for reading, n-1 for writing
100 */
101 for (j = pi->raid_disks ; j-- ; ) {
102 bio = bio_alloc(gfp_flags, RESYNC_PAGES);
103 if (!bio)
104 goto out_free_bio;
105 r1_bio->bios[j] = bio;
106 }
107 /*
108 * Allocate RESYNC_PAGES data pages and attach them to
NeilBrownd11c1712006-01-06 00:20:26 -0800109 * the first bio.
110 * If this is a user-requested check/repair, allocate
111 * RESYNC_PAGES for each bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 */
NeilBrownd11c1712006-01-06 00:20:26 -0800113 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
114 j = pi->raid_disks;
115 else
116 j = 1;
117 while(j--) {
118 bio = r1_bio->bios[j];
119 for (i = 0; i < RESYNC_PAGES; i++) {
120 page = alloc_page(gfp_flags);
121 if (unlikely(!page))
122 goto out_free_pages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
NeilBrownd11c1712006-01-06 00:20:26 -0800124 bio->bi_io_vec[i].bv_page = page;
NeilBrown303a0e12009-04-06 14:40:38 +1000125 bio->bi_vcnt = i+1;
NeilBrownd11c1712006-01-06 00:20:26 -0800126 }
127 }
128 /* If not user-requests, copy the page pointers to all bios */
129 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
130 for (i=0; i<RESYNC_PAGES ; i++)
131 for (j=1; j<pi->raid_disks; j++)
132 r1_bio->bios[j]->bi_io_vec[i].bv_page =
133 r1_bio->bios[0]->bi_io_vec[i].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 }
135
136 r1_bio->master_bio = NULL;
137
138 return r1_bio;
139
140out_free_pages:
NeilBrown303a0e12009-04-06 14:40:38 +1000141 for (j=0 ; j < pi->raid_disks; j++)
142 for (i=0; i < r1_bio->bios[j]->bi_vcnt ; i++)
143 put_page(r1_bio->bios[j]->bi_io_vec[i].bv_page);
NeilBrownd11c1712006-01-06 00:20:26 -0800144 j = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145out_free_bio:
146 while ( ++j < pi->raid_disks )
147 bio_put(r1_bio->bios[j]);
148 r1bio_pool_free(r1_bio, data);
149 return NULL;
150}
151
152static void r1buf_pool_free(void *__r1_bio, void *data)
153{
154 struct pool_info *pi = data;
NeilBrownd11c1712006-01-06 00:20:26 -0800155 int i,j;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 r1bio_t *r1bio = __r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
NeilBrownd11c1712006-01-06 00:20:26 -0800158 for (i = 0; i < RESYNC_PAGES; i++)
159 for (j = pi->raid_disks; j-- ;) {
160 if (j == 0 ||
161 r1bio->bios[j]->bi_io_vec[i].bv_page !=
162 r1bio->bios[0]->bi_io_vec[i].bv_page)
NeilBrown1345b1d2006-01-06 00:20:40 -0800163 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
NeilBrownd11c1712006-01-06 00:20:26 -0800164 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 for (i=0 ; i < pi->raid_disks; i++)
166 bio_put(r1bio->bios[i]);
167
168 r1bio_pool_free(r1bio, data);
169}
170
171static void put_all_bios(conf_t *conf, r1bio_t *r1_bio)
172{
173 int i;
174
175 for (i = 0; i < conf->raid_disks; i++) {
176 struct bio **bio = r1_bio->bios + i;
NeilBrowncf30a472006-01-06 00:20:23 -0800177 if (*bio && *bio != IO_BLOCKED)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 bio_put(*bio);
179 *bio = NULL;
180 }
181}
182
Arjan van de Ven858119e2006-01-14 13:20:43 -0800183static void free_r1bio(r1bio_t *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
NeilBrown070ec552009-06-16 16:54:21 +1000185 conf_t *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
187 /*
188 * Wake up any possible resync thread that waits for the device
189 * to go idle.
190 */
NeilBrown17999be2006-01-06 00:20:12 -0800191 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
193 put_all_bios(conf, r1_bio);
194 mempool_free(r1_bio, conf->r1bio_pool);
195}
196
Arjan van de Ven858119e2006-01-14 13:20:43 -0800197static void put_buf(r1bio_t *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198{
NeilBrown070ec552009-06-16 16:54:21 +1000199 conf_t *conf = r1_bio->mddev->private;
NeilBrown3e198f72006-01-06 00:20:21 -0800200 int i;
201
202 for (i=0; i<conf->raid_disks; i++) {
203 struct bio *bio = r1_bio->bios[i];
204 if (bio->bi_end_io)
205 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
208 mempool_free(r1_bio, conf->r1buf_pool);
209
NeilBrown17999be2006-01-06 00:20:12 -0800210 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211}
212
213static void reschedule_retry(r1bio_t *r1_bio)
214{
215 unsigned long flags;
216 mddev_t *mddev = r1_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +1000217 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218
219 spin_lock_irqsave(&conf->device_lock, flags);
220 list_add(&r1_bio->retry_list, &conf->retry_list);
NeilBrownddaf22a2006-01-06 00:20:19 -0800221 conf->nr_queued ++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 spin_unlock_irqrestore(&conf->device_lock, flags);
223
NeilBrown17999be2006-01-06 00:20:12 -0800224 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 md_wakeup_thread(mddev->thread);
226}
227
228/*
229 * raid_end_bio_io() is called when we have finished servicing a mirrored
230 * operation and are ready to return a success/failure code to the buffer
231 * cache layer.
232 */
233static void raid_end_bio_io(r1bio_t *r1_bio)
234{
235 struct bio *bio = r1_bio->master_bio;
236
NeilBrown4b6d2872005-09-09 16:23:47 -0700237 /* if nobody has done the final endio yet, do it now */
238 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
239 PRINTK(KERN_DEBUG "raid1: sync end %s on sectors %llu-%llu\n",
240 (bio_data_dir(bio) == WRITE) ? "write" : "read",
241 (unsigned long long) bio->bi_sector,
242 (unsigned long long) bio->bi_sector +
243 (bio->bi_size >> 9) - 1);
244
NeilBrown6712ecf2007-09-27 12:47:43 +0200245 bio_endio(bio,
NeilBrown4b6d2872005-09-09 16:23:47 -0700246 test_bit(R1BIO_Uptodate, &r1_bio->state) ? 0 : -EIO);
247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 free_r1bio(r1_bio);
249}
250
251/*
252 * Update disk head position estimator based on IRQ completion info.
253 */
254static inline void update_head_pos(int disk, r1bio_t *r1_bio)
255{
NeilBrown070ec552009-06-16 16:54:21 +1000256 conf_t *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257
258 conf->mirrors[disk].head_position =
259 r1_bio->sector + (r1_bio->sectors);
260}
261
NeilBrown6712ecf2007-09-27 12:47:43 +0200262static void raid1_end_read_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263{
264 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
265 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
266 int mirror;
NeilBrown070ec552009-06-16 16:54:21 +1000267 conf_t *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 mirror = r1_bio->read_disk;
270 /*
271 * this branch is our 'one mirror IO has finished' event handler:
272 */
NeilBrownddaf22a2006-01-06 00:20:19 -0800273 update_head_pos(mirror, r1_bio);
274
NeilBrowndd00a992007-05-10 03:15:50 -0700275 if (uptodate)
276 set_bit(R1BIO_Uptodate, &r1_bio->state);
277 else {
278 /* If all other devices have failed, we want to return
279 * the error upwards rather than fail the last device.
280 * Here we redefine "uptodate" to mean "Don't want to retry"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 */
NeilBrowndd00a992007-05-10 03:15:50 -0700282 unsigned long flags;
283 spin_lock_irqsave(&conf->device_lock, flags);
284 if (r1_bio->mddev->degraded == conf->raid_disks ||
285 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
286 !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags)))
287 uptodate = 1;
288 spin_unlock_irqrestore(&conf->device_lock, flags);
289 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
NeilBrowndd00a992007-05-10 03:15:50 -0700291 if (uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 raid_end_bio_io(r1_bio);
NeilBrowndd00a992007-05-10 03:15:50 -0700293 else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 /*
295 * oops, read error:
296 */
297 char b[BDEVNAME_SIZE];
298 if (printk_ratelimit())
299 printk(KERN_ERR "raid1: %s: rescheduling sector %llu\n",
300 bdevname(conf->mirrors[mirror].rdev->bdev,b), (unsigned long long)r1_bio->sector);
301 reschedule_retry(r1_bio);
302 }
303
304 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305}
306
NeilBrown6712ecf2007-09-27 12:47:43 +0200307static void raid1_end_write_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308{
309 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
310 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
NeilBrowna9701a32005-11-08 21:39:34 -0800311 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
NeilBrown070ec552009-06-16 16:54:21 +1000312 conf_t *conf = r1_bio->mddev->private;
NeilBrown04b857f2006-03-09 17:33:46 -0800313 struct bio *to_put = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315
316 for (mirror = 0; mirror < conf->raid_disks; mirror++)
317 if (r1_bio->bios[mirror] == bio)
318 break;
319
NeilBrownbea27712006-05-01 12:15:46 -0700320 if (error == -EOPNOTSUPP && test_bit(R1BIO_Barrier, &r1_bio->state)) {
NeilBrowna9701a32005-11-08 21:39:34 -0800321 set_bit(BarriersNotsupp, &conf->mirrors[mirror].rdev->flags);
322 set_bit(R1BIO_BarrierRetry, &r1_bio->state);
323 r1_bio->mddev->barriers_work = 0;
NeilBrown5e7dd2a2006-05-01 12:15:47 -0700324 /* Don't rdev_dec_pending in this branch - keep it for the retry */
NeilBrowna9701a32005-11-08 21:39:34 -0800325 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 /*
NeilBrowna9701a32005-11-08 21:39:34 -0800327 * this branch is our 'one mirror IO has finished' event handler:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 */
NeilBrowna9701a32005-11-08 21:39:34 -0800329 r1_bio->bios[mirror] = NULL;
NeilBrown04b857f2006-03-09 17:33:46 -0800330 to_put = bio;
NeilBrowna9701a32005-11-08 21:39:34 -0800331 if (!uptodate) {
332 md_error(r1_bio->mddev, conf->mirrors[mirror].rdev);
333 /* an I/O failed, we can't clear the bitmap */
334 set_bit(R1BIO_Degraded, &r1_bio->state);
335 } else
336 /*
337 * Set R1BIO_Uptodate in our master bio, so that
338 * we will return a good error code for to the higher
339 * levels even if IO on some other mirrored buffer fails.
340 *
341 * The 'master' represents the composite IO operation to
342 * user-side. So if something waits for IO, then it will
343 * wait for the 'master' bio.
344 */
345 set_bit(R1BIO_Uptodate, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
NeilBrowna9701a32005-11-08 21:39:34 -0800347 update_head_pos(mirror, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348
NeilBrowna9701a32005-11-08 21:39:34 -0800349 if (behind) {
350 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
351 atomic_dec(&r1_bio->behind_remaining);
NeilBrown4b6d2872005-09-09 16:23:47 -0700352
NeilBrowna9701a32005-11-08 21:39:34 -0800353 /* In behind mode, we ACK the master bio once the I/O has safely
354 * reached all non-writemostly disks. Setting the Returned bit
355 * ensures that this gets done only once -- we don't ever want to
356 * return -EIO here, instead we'll wait */
NeilBrown4b6d2872005-09-09 16:23:47 -0700357
NeilBrowna9701a32005-11-08 21:39:34 -0800358 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
359 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
360 /* Maybe we can return now */
361 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
362 struct bio *mbio = r1_bio->master_bio;
363 PRINTK(KERN_DEBUG "raid1: behind end write sectors %llu-%llu\n",
364 (unsigned long long) mbio->bi_sector,
365 (unsigned long long) mbio->bi_sector +
366 (mbio->bi_size >> 9) - 1);
NeilBrown6712ecf2007-09-27 12:47:43 +0200367 bio_endio(mbio, 0);
NeilBrowna9701a32005-11-08 21:39:34 -0800368 }
NeilBrown4b6d2872005-09-09 16:23:47 -0700369 }
370 }
NeilBrown5e7dd2a2006-05-01 12:15:47 -0700371 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
NeilBrown4b6d2872005-09-09 16:23:47 -0700372 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 /*
374 *
375 * Let's see if all mirrored write operations have finished
376 * already.
377 */
378 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrownc70810b2006-06-26 00:27:35 -0700379 if (test_bit(R1BIO_BarrierRetry, &r1_bio->state))
NeilBrowna9701a32005-11-08 21:39:34 -0800380 reschedule_retry(r1_bio);
NeilBrownc70810b2006-06-26 00:27:35 -0700381 else {
382 /* it really is the end of this request */
383 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
384 /* free extra copy of the data pages */
385 int i = bio->bi_vcnt;
386 while (i--)
387 safe_put_page(bio->bi_io_vec[i].bv_page);
388 }
389 /* clear the bitmap if all writes complete successfully */
390 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
391 r1_bio->sectors,
392 !test_bit(R1BIO_Degraded, &r1_bio->state),
393 behind);
394 md_write_end(r1_bio->mddev);
395 raid_end_bio_io(r1_bio);
NeilBrowna9701a32005-11-08 21:39:34 -0800396 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 }
NeilBrownc70810b2006-06-26 00:27:35 -0700398
NeilBrown04b857f2006-03-09 17:33:46 -0800399 if (to_put)
400 bio_put(to_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401}
402
403
404/*
405 * This routine returns the disk from which the requested read should
406 * be done. There is a per-array 'next expected sequential IO' sector
407 * number - if this matches on the next IO then we use the last disk.
408 * There is also a per-disk 'last know head position' sector that is
409 * maintained from IRQ contexts, both the normal and the resync IO
410 * completion handlers update this position correctly. If there is no
411 * perfect sequential match then we pick the disk whose head is closest.
412 *
413 * If there are 2 mirrors in the same 2 devices, performance degrades
414 * because position is mirror, not device based.
415 *
416 * The rdev for the device selected will have nr_pending incremented.
417 */
418static int read_balance(conf_t *conf, r1bio_t *r1_bio)
419{
420 const unsigned long this_sector = r1_bio->sector;
421 int new_disk = conf->last_used, disk = new_disk;
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700422 int wonly_disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 const int sectors = r1_bio->sectors;
424 sector_t new_distance, current_distance;
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700425 mdk_rdev_t *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426
427 rcu_read_lock();
428 /*
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700429 * Check if we can balance. We can balance on the whole
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 * device if no resync is going on, or below the resync window.
431 * We take the first readable disk when above the resync window.
432 */
433 retry:
434 if (conf->mddev->recovery_cp < MaxSector &&
435 (this_sector + sectors >= conf->next_resync)) {
436 /* Choose the first operation device, for consistancy */
437 new_disk = 0;
438
Suzanne Woodd6065f72005-11-08 21:39:27 -0800439 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
NeilBrowncf30a472006-01-06 00:20:23 -0800440 r1_bio->bios[new_disk] == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800441 !rdev || !test_bit(In_sync, &rdev->flags)
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700442 || test_bit(WriteMostly, &rdev->flags);
Suzanne Woodd6065f72005-11-08 21:39:27 -0800443 rdev = rcu_dereference(conf->mirrors[++new_disk].rdev)) {
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700444
NeilBrowncf30a472006-01-06 00:20:23 -0800445 if (rdev && test_bit(In_sync, &rdev->flags) &&
446 r1_bio->bios[new_disk] != IO_BLOCKED)
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700447 wonly_disk = new_disk;
448
449 if (new_disk == conf->raid_disks - 1) {
450 new_disk = wonly_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 break;
452 }
453 }
454 goto rb_out;
455 }
456
457
458 /* make sure the disk is operational */
Suzanne Woodd6065f72005-11-08 21:39:27 -0800459 for (rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
NeilBrowncf30a472006-01-06 00:20:23 -0800460 r1_bio->bios[new_disk] == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800461 !rdev || !test_bit(In_sync, &rdev->flags) ||
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700462 test_bit(WriteMostly, &rdev->flags);
Suzanne Woodd6065f72005-11-08 21:39:27 -0800463 rdev = rcu_dereference(conf->mirrors[new_disk].rdev)) {
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700464
NeilBrowncf30a472006-01-06 00:20:23 -0800465 if (rdev && test_bit(In_sync, &rdev->flags) &&
466 r1_bio->bios[new_disk] != IO_BLOCKED)
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700467 wonly_disk = new_disk;
468
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469 if (new_disk <= 0)
470 new_disk = conf->raid_disks;
471 new_disk--;
472 if (new_disk == disk) {
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700473 new_disk = wonly_disk;
474 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 }
476 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700477
478 if (new_disk < 0)
479 goto rb_out;
480
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481 disk = new_disk;
482 /* now disk == new_disk == starting point for search */
483
484 /*
485 * Don't change to another disk for sequential reads:
486 */
487 if (conf->next_seq_sect == this_sector)
488 goto rb_out;
489 if (this_sector == conf->mirrors[new_disk].head_position)
490 goto rb_out;
491
492 current_distance = abs(this_sector - conf->mirrors[disk].head_position);
493
494 /* Find the disk whose head is closest */
495
496 do {
497 if (disk <= 0)
498 disk = conf->raid_disks;
499 disk--;
500
Suzanne Woodd6065f72005-11-08 21:39:27 -0800501 rdev = rcu_dereference(conf->mirrors[disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700502
NeilBrowncf30a472006-01-06 00:20:23 -0800503 if (!rdev || r1_bio->bios[disk] == IO_BLOCKED ||
NeilBrownb2d444d2005-11-08 21:39:31 -0800504 !test_bit(In_sync, &rdev->flags) ||
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700505 test_bit(WriteMostly, &rdev->flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 continue;
507
508 if (!atomic_read(&rdev->nr_pending)) {
509 new_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 break;
511 }
512 new_distance = abs(this_sector - conf->mirrors[disk].head_position);
513 if (new_distance < current_distance) {
514 current_distance = new_distance;
515 new_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 }
517 } while (disk != conf->last_used);
518
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700519 rb_out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520
521
522 if (new_disk >= 0) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800523 rdev = rcu_dereference(conf->mirrors[new_disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700524 if (!rdev)
525 goto retry;
526 atomic_inc(&rdev->nr_pending);
NeilBrownb2d444d2005-11-08 21:39:31 -0800527 if (!test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 /* cannot risk returning a device that failed
529 * before we inc'ed nr_pending
530 */
NeilBrown03c902e2006-01-06 00:20:46 -0800531 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 goto retry;
533 }
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700534 conf->next_seq_sect = this_sector + sectors;
535 conf->last_used = new_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 }
537 rcu_read_unlock();
538
539 return new_disk;
540}
541
542static void unplug_slaves(mddev_t *mddev)
543{
NeilBrown070ec552009-06-16 16:54:21 +1000544 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 int i;
546
547 rcu_read_lock();
548 for (i=0; i<mddev->raid_disks; i++) {
Suzanne Woodd6065f72005-11-08 21:39:27 -0800549 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownb2d444d2005-11-08 21:39:31 -0800550 if (rdev && !test_bit(Faulty, &rdev->flags) && atomic_read(&rdev->nr_pending)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200551 struct request_queue *r_queue = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
553 atomic_inc(&rdev->nr_pending);
554 rcu_read_unlock();
555
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -0500556 blk_unplug(r_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
558 rdev_dec_pending(rdev, mddev);
559 rcu_read_lock();
560 }
561 }
562 rcu_read_unlock();
563}
564
Jens Axboe165125e2007-07-24 09:28:11 +0200565static void raid1_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566{
NeilBrown191ea9b2005-06-21 17:17:23 -0700567 mddev_t *mddev = q->queuedata;
568
569 unplug_slaves(mddev);
570 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571}
572
NeilBrown0d129222006-10-03 01:15:54 -0700573static int raid1_congested(void *data, int bits)
574{
575 mddev_t *mddev = data;
NeilBrown070ec552009-06-16 16:54:21 +1000576 conf_t *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700577 int i, ret = 0;
578
NeilBrown3fa841d2009-09-23 18:10:29 +1000579 if (mddev_congested(mddev, bits))
580 return 1;
581
NeilBrown0d129222006-10-03 01:15:54 -0700582 rcu_read_lock();
583 for (i = 0; i < mddev->raid_disks; i++) {
584 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
585 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200586 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700587
588 /* Note the '|| 1' - when read_balance prefers
589 * non-congested targets, it can be removed
590 */
Alexander Beregalov91a9e992009-04-07 01:35:56 +0400591 if ((bits & (1<<BDI_async_congested)) || 1)
NeilBrown0d129222006-10-03 01:15:54 -0700592 ret |= bdi_congested(&q->backing_dev_info, bits);
593 else
594 ret &= bdi_congested(&q->backing_dev_info, bits);
595 }
596 }
597 rcu_read_unlock();
598 return ret;
599}
600
601
NeilBrowna35e63e2008-03-04 14:29:29 -0800602static int flush_pending_writes(conf_t *conf)
603{
604 /* Any writes that have been queued but are awaiting
605 * bitmap updates get flushed here.
606 * We return 1 if any requests were actually submitted.
607 */
608 int rv = 0;
609
610 spin_lock_irq(&conf->device_lock);
611
612 if (conf->pending_bio_list.head) {
613 struct bio *bio;
614 bio = bio_list_get(&conf->pending_bio_list);
615 blk_remove_plug(conf->mddev->queue);
616 spin_unlock_irq(&conf->device_lock);
617 /* flush any pending bitmap writes to
618 * disk before proceeding w/ I/O */
619 bitmap_unplug(conf->mddev->bitmap);
620
621 while (bio) { /* submit pending writes */
622 struct bio *next = bio->bi_next;
623 bio->bi_next = NULL;
624 generic_make_request(bio);
625 bio = next;
626 }
627 rv = 1;
628 } else
629 spin_unlock_irq(&conf->device_lock);
630 return rv;
631}
632
NeilBrown17999be2006-01-06 00:20:12 -0800633/* Barriers....
634 * Sometimes we need to suspend IO while we do something else,
635 * either some resync/recovery, or reconfigure the array.
636 * To do this we raise a 'barrier'.
637 * The 'barrier' is a counter that can be raised multiple times
638 * to count how many activities are happening which preclude
639 * normal IO.
640 * We can only raise the barrier if there is no pending IO.
641 * i.e. if nr_pending == 0.
642 * We choose only to raise the barrier if no-one is waiting for the
643 * barrier to go down. This means that as soon as an IO request
644 * is ready, no other operations which require a barrier will start
645 * until the IO request has had a chance.
646 *
647 * So: regular IO calls 'wait_barrier'. When that returns there
648 * is no backgroup IO happening, It must arrange to call
649 * allow_barrier when it has finished its IO.
650 * backgroup IO calls must call raise_barrier. Once that returns
651 * there is no normal IO happeing. It must arrange to call
652 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 */
654#define RESYNC_DEPTH 32
655
NeilBrown17999be2006-01-06 00:20:12 -0800656static void raise_barrier(conf_t *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
658 spin_lock_irq(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800659
660 /* Wait until no block IO is waiting */
661 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
662 conf->resync_lock,
663 raid1_unplug(conf->mddev->queue));
664
665 /* block any new IO from starting */
666 conf->barrier++;
667
668 /* No wait for all pending IO to complete */
669 wait_event_lock_irq(conf->wait_barrier,
670 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
671 conf->resync_lock,
672 raid1_unplug(conf->mddev->queue));
673
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674 spin_unlock_irq(&conf->resync_lock);
675}
676
NeilBrown17999be2006-01-06 00:20:12 -0800677static void lower_barrier(conf_t *conf)
678{
679 unsigned long flags;
NeilBrown709ae482009-12-14 12:49:51 +1100680 BUG_ON(conf->barrier <= 0);
NeilBrown17999be2006-01-06 00:20:12 -0800681 spin_lock_irqsave(&conf->resync_lock, flags);
682 conf->barrier--;
683 spin_unlock_irqrestore(&conf->resync_lock, flags);
684 wake_up(&conf->wait_barrier);
685}
686
687static void wait_barrier(conf_t *conf)
688{
689 spin_lock_irq(&conf->resync_lock);
690 if (conf->barrier) {
691 conf->nr_waiting++;
692 wait_event_lock_irq(conf->wait_barrier, !conf->barrier,
693 conf->resync_lock,
694 raid1_unplug(conf->mddev->queue));
695 conf->nr_waiting--;
696 }
697 conf->nr_pending++;
698 spin_unlock_irq(&conf->resync_lock);
699}
700
701static void allow_barrier(conf_t *conf)
702{
703 unsigned long flags;
704 spin_lock_irqsave(&conf->resync_lock, flags);
705 conf->nr_pending--;
706 spin_unlock_irqrestore(&conf->resync_lock, flags);
707 wake_up(&conf->wait_barrier);
708}
709
NeilBrownddaf22a2006-01-06 00:20:19 -0800710static void freeze_array(conf_t *conf)
711{
712 /* stop syncio and normal IO and wait for everything to
713 * go quite.
714 * We increment barrier and nr_waiting, and then
NeilBrown1c830532008-03-04 14:29:35 -0800715 * wait until nr_pending match nr_queued+1
716 * This is called in the context of one normal IO request
717 * that has failed. Thus any sync request that might be pending
718 * will be blocked by nr_pending, and we need to wait for
719 * pending IO requests to complete or be queued for re-try.
720 * Thus the number queued (nr_queued) plus this request (1)
721 * must match the number of pending IOs (nr_pending) before
722 * we continue.
NeilBrownddaf22a2006-01-06 00:20:19 -0800723 */
724 spin_lock_irq(&conf->resync_lock);
725 conf->barrier++;
726 conf->nr_waiting++;
727 wait_event_lock_irq(conf->wait_barrier,
NeilBrown1c830532008-03-04 14:29:35 -0800728 conf->nr_pending == conf->nr_queued+1,
NeilBrownddaf22a2006-01-06 00:20:19 -0800729 conf->resync_lock,
NeilBrowna35e63e2008-03-04 14:29:29 -0800730 ({ flush_pending_writes(conf);
731 raid1_unplug(conf->mddev->queue); }));
NeilBrownddaf22a2006-01-06 00:20:19 -0800732 spin_unlock_irq(&conf->resync_lock);
733}
734static void unfreeze_array(conf_t *conf)
735{
736 /* reverse the effect of the freeze */
737 spin_lock_irq(&conf->resync_lock);
738 conf->barrier--;
739 conf->nr_waiting--;
740 wake_up(&conf->wait_barrier);
741 spin_unlock_irq(&conf->resync_lock);
742}
743
NeilBrown17999be2006-01-06 00:20:12 -0800744
NeilBrown4b6d2872005-09-09 16:23:47 -0700745/* duplicate the data pages for behind I/O */
746static struct page **alloc_behind_pages(struct bio *bio)
747{
748 int i;
749 struct bio_vec *bvec;
NeilBrown9ffae0c2006-01-06 00:20:32 -0800750 struct page **pages = kzalloc(bio->bi_vcnt * sizeof(struct page *),
NeilBrown4b6d2872005-09-09 16:23:47 -0700751 GFP_NOIO);
752 if (unlikely(!pages))
753 goto do_sync_io;
754
NeilBrown4b6d2872005-09-09 16:23:47 -0700755 bio_for_each_segment(bvec, bio, i) {
756 pages[i] = alloc_page(GFP_NOIO);
757 if (unlikely(!pages[i]))
758 goto do_sync_io;
759 memcpy(kmap(pages[i]) + bvec->bv_offset,
760 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
761 kunmap(pages[i]);
762 kunmap(bvec->bv_page);
763 }
764
765 return pages;
766
767do_sync_io:
768 if (pages)
769 for (i = 0; i < bio->bi_vcnt && pages[i]; i++)
NeilBrown2d1f3b52006-01-06 00:20:31 -0800770 put_page(pages[i]);
NeilBrown4b6d2872005-09-09 16:23:47 -0700771 kfree(pages);
772 PRINTK("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
773 return NULL;
774}
775
Jens Axboe165125e2007-07-24 09:28:11 +0200776static int make_request(struct request_queue *q, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777{
778 mddev_t *mddev = q->queuedata;
NeilBrown070ec552009-06-16 16:54:21 +1000779 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 mirror_info_t *mirror;
781 r1bio_t *r1_bio;
782 struct bio *read_bio;
NeilBrown191ea9b2005-06-21 17:17:23 -0700783 int i, targets = 0, disks;
NeilBrown84255d12008-05-23 13:04:32 -0700784 struct bitmap *bitmap;
NeilBrown191ea9b2005-06-21 17:17:23 -0700785 unsigned long flags;
786 struct bio_list bl;
NeilBrown4b6d2872005-09-09 16:23:47 -0700787 struct page **behind_pages = NULL;
Jens Axboea3623572005-11-01 09:26:16 +0100788 const int rw = bio_data_dir(bio);
Jens Axboe1f98a132009-09-11 14:32:04 +0200789 const bool do_sync = bio_rw_flagged(bio, BIO_RW_SYNCIO);
790 int cpu;
791 bool do_barriers;
Dan Williams6bfe0b42008-04-30 00:52:32 -0700792 mdk_rdev_t *blocked_rdev;
NeilBrown191ea9b2005-06-21 17:17:23 -0700793
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 /*
795 * Register the new request and wait if the reconstruction
796 * thread has put up a bar for new requests.
797 * Continue immediately if no resync is active currently.
NeilBrown62de6082006-05-01 12:15:47 -0700798 * We test barriers_work *after* md_write_start as md_write_start
799 * may cause the first superblock write, and that will check out
800 * if barriers work.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 */
NeilBrown62de6082006-05-01 12:15:47 -0700802
NeilBrown3d310eb2005-06-21 17:17:26 -0700803 md_write_start(mddev, bio); /* wait on superblock update early */
804
NeilBrown6eef4b22009-12-14 12:49:51 +1100805 if (bio_data_dir(bio) == WRITE &&
806 bio->bi_sector + bio->bi_size/512 > mddev->suspend_lo &&
807 bio->bi_sector < mddev->suspend_hi) {
808 /* As the suspend_* range is controlled by
809 * userspace, we want an interruptible
810 * wait.
811 */
812 DEFINE_WAIT(w);
813 for (;;) {
814 flush_signals(current);
815 prepare_to_wait(&conf->wait_barrier,
816 &w, TASK_INTERRUPTIBLE);
817 if (bio->bi_sector + bio->bi_size/512 <= mddev->suspend_lo ||
818 bio->bi_sector >= mddev->suspend_hi)
819 break;
820 schedule();
821 }
822 finish_wait(&conf->wait_barrier, &w);
823 }
Jens Axboe1f98a132009-09-11 14:32:04 +0200824 if (unlikely(!mddev->barriers_work &&
825 bio_rw_flagged(bio, BIO_RW_BARRIER))) {
NeilBrown62de6082006-05-01 12:15:47 -0700826 if (rw == WRITE)
827 md_write_end(mddev);
NeilBrown6712ecf2007-09-27 12:47:43 +0200828 bio_endio(bio, -EOPNOTSUPP);
NeilBrown62de6082006-05-01 12:15:47 -0700829 return 0;
830 }
831
NeilBrown17999be2006-01-06 00:20:12 -0800832 wait_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
NeilBrown84255d12008-05-23 13:04:32 -0700834 bitmap = mddev->bitmap;
835
Tejun Heo074a7ac2008-08-25 19:56:14 +0900836 cpu = part_stat_lock();
837 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
838 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
839 bio_sectors(bio));
840 part_stat_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
842 /*
843 * make_request() can abort the operation when READA is being
844 * used and no empty request is available.
845 *
846 */
847 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
848
849 r1_bio->master_bio = bio;
850 r1_bio->sectors = bio->bi_size >> 9;
NeilBrown191ea9b2005-06-21 17:17:23 -0700851 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 r1_bio->mddev = mddev;
853 r1_bio->sector = bio->bi_sector;
854
Jens Axboea3623572005-11-01 09:26:16 +0100855 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 /*
857 * read balancing logic:
858 */
859 int rdisk = read_balance(conf, r1_bio);
860
861 if (rdisk < 0) {
862 /* couldn't find anywhere to read from */
863 raid_end_bio_io(r1_bio);
864 return 0;
865 }
866 mirror = conf->mirrors + rdisk;
867
868 r1_bio->read_disk = rdisk;
869
870 read_bio = bio_clone(bio, GFP_NOIO);
871
872 r1_bio->bios[rdisk] = read_bio;
873
874 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
875 read_bio->bi_bdev = mirror->rdev->bdev;
876 read_bio->bi_end_io = raid1_end_read_request;
Dmitry Monakhov1ef04fe2009-09-20 05:52:25 +0400877 read_bio->bi_rw = READ | (do_sync << BIO_RW_SYNCIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 read_bio->bi_private = r1_bio;
879
880 generic_make_request(read_bio);
881 return 0;
882 }
883
884 /*
885 * WRITE:
886 */
887 /* first select target devices under spinlock and
888 * inc refcount on their rdev. Record them by setting
889 * bios[x] to bio
890 */
891 disks = conf->raid_disks;
NeilBrown191ea9b2005-06-21 17:17:23 -0700892#if 0
893 { static int first=1;
894 if (first) printk("First Write sector %llu disks %d\n",
895 (unsigned long long)r1_bio->sector, disks);
896 first = 0;
897 }
898#endif
Dan Williams6bfe0b42008-04-30 00:52:32 -0700899 retry_write:
900 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901 rcu_read_lock();
902 for (i = 0; i < disks; i++) {
Dan Williams6bfe0b42008-04-30 00:52:32 -0700903 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
904 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
905 atomic_inc(&rdev->nr_pending);
906 blocked_rdev = rdev;
907 break;
908 }
909 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 atomic_inc(&rdev->nr_pending);
NeilBrownb2d444d2005-11-08 21:39:31 -0800911 if (test_bit(Faulty, &rdev->flags)) {
NeilBrown03c902e2006-01-06 00:20:46 -0800912 rdev_dec_pending(rdev, mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 r1_bio->bios[i] = NULL;
914 } else
915 r1_bio->bios[i] = bio;
NeilBrown191ea9b2005-06-21 17:17:23 -0700916 targets++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917 } else
918 r1_bio->bios[i] = NULL;
919 }
920 rcu_read_unlock();
921
Dan Williams6bfe0b42008-04-30 00:52:32 -0700922 if (unlikely(blocked_rdev)) {
923 /* Wait for this device to become unblocked */
924 int j;
925
926 for (j = 0; j < i; j++)
927 if (r1_bio->bios[j])
928 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
929
930 allow_barrier(conf);
931 md_wait_for_blocked_rdev(blocked_rdev, mddev);
932 wait_barrier(conf);
933 goto retry_write;
934 }
935
NeilBrown4b6d2872005-09-09 16:23:47 -0700936 BUG_ON(targets == 0); /* we never fail the last device */
937
NeilBrown191ea9b2005-06-21 17:17:23 -0700938 if (targets < conf->raid_disks) {
939 /* array is degraded, we will not clear the bitmap
940 * on I/O completion (see raid1_end_write_request) */
941 set_bit(R1BIO_Degraded, &r1_bio->state);
942 }
NeilBrown06d91a52005-06-21 17:17:12 -0700943
NeilBrown4b6d2872005-09-09 16:23:47 -0700944 /* do behind I/O ? */
945 if (bitmap &&
NeilBrown42a04b52009-12-14 12:49:53 +1100946 (atomic_read(&bitmap->behind_writes)
947 < mddev->bitmap_info.max_write_behind) &&
NeilBrown4b6d2872005-09-09 16:23:47 -0700948 (behind_pages = alloc_behind_pages(bio)) != NULL)
949 set_bit(R1BIO_BehindIO, &r1_bio->state);
950
NeilBrown191ea9b2005-06-21 17:17:23 -0700951 atomic_set(&r1_bio->remaining, 0);
NeilBrown4b6d2872005-09-09 16:23:47 -0700952 atomic_set(&r1_bio->behind_remaining, 0);
NeilBrown191ea9b2005-06-21 17:17:23 -0700953
Jens Axboe1f98a132009-09-11 14:32:04 +0200954 do_barriers = bio_rw_flagged(bio, BIO_RW_BARRIER);
NeilBrowna9701a32005-11-08 21:39:34 -0800955 if (do_barriers)
956 set_bit(R1BIO_Barrier, &r1_bio->state);
957
NeilBrown191ea9b2005-06-21 17:17:23 -0700958 bio_list_init(&bl);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 for (i = 0; i < disks; i++) {
960 struct bio *mbio;
961 if (!r1_bio->bios[i])
962 continue;
963
964 mbio = bio_clone(bio, GFP_NOIO);
965 r1_bio->bios[i] = mbio;
966
967 mbio->bi_sector = r1_bio->sector + conf->mirrors[i].rdev->data_offset;
968 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
969 mbio->bi_end_io = raid1_end_write_request;
Dmitry Monakhov1ef04fe2009-09-20 05:52:25 +0400970 mbio->bi_rw = WRITE | (do_barriers << BIO_RW_BARRIER) |
971 (do_sync << BIO_RW_SYNCIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 mbio->bi_private = r1_bio;
973
NeilBrown4b6d2872005-09-09 16:23:47 -0700974 if (behind_pages) {
975 struct bio_vec *bvec;
976 int j;
977
978 /* Yes, I really want the '__' version so that
979 * we clear any unused pointer in the io_vec, rather
980 * than leave them unchanged. This is important
981 * because when we come to free the pages, we won't
982 * know the originial bi_idx, so we just free
983 * them all
984 */
985 __bio_for_each_segment(bvec, mbio, j, 0)
986 bvec->bv_page = behind_pages[j];
987 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
988 atomic_inc(&r1_bio->behind_remaining);
989 }
990
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 atomic_inc(&r1_bio->remaining);
NeilBrown191ea9b2005-06-21 17:17:23 -0700992
993 bio_list_add(&bl, mbio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994 }
NeilBrown4b6d2872005-09-09 16:23:47 -0700995 kfree(behind_pages); /* the behind pages are attached to the bios now */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
NeilBrown4b6d2872005-09-09 16:23:47 -0700997 bitmap_startwrite(bitmap, bio->bi_sector, r1_bio->sectors,
998 test_bit(R1BIO_BehindIO, &r1_bio->state));
NeilBrown191ea9b2005-06-21 17:17:23 -0700999 spin_lock_irqsave(&conf->device_lock, flags);
1000 bio_list_merge(&conf->pending_bio_list, &bl);
1001 bio_list_init(&bl);
1002
1003 blk_plug_device(mddev->queue);
1004 spin_unlock_irqrestore(&conf->device_lock, flags);
1005
NeilBrowna35e63e2008-03-04 14:29:29 -08001006 /* In case raid1d snuck into freeze_array */
1007 wake_up(&conf->wait_barrier);
1008
Lars Ellenberge3881a62007-01-10 23:15:37 -08001009 if (do_sync)
1010 md_wakeup_thread(mddev->thread);
NeilBrown191ea9b2005-06-21 17:17:23 -07001011#if 0
1012 while ((bio = bio_list_pop(&bl)) != NULL)
1013 generic_make_request(bio);
1014#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
1016 return 0;
1017}
1018
1019static void status(struct seq_file *seq, mddev_t *mddev)
1020{
NeilBrown070ec552009-06-16 16:54:21 +10001021 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 int i;
1023
1024 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown11ce99e2006-10-03 01:15:52 -07001025 conf->raid_disks - mddev->degraded);
NeilBrownddac7c72006-08-31 21:27:36 -07001026 rcu_read_lock();
1027 for (i = 0; i < conf->raid_disks; i++) {
1028 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 seq_printf(seq, "%s",
NeilBrownddac7c72006-08-31 21:27:36 -07001030 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1031 }
1032 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 seq_printf(seq, "]");
1034}
1035
1036
1037static void error(mddev_t *mddev, mdk_rdev_t *rdev)
1038{
1039 char b[BDEVNAME_SIZE];
NeilBrown070ec552009-06-16 16:54:21 +10001040 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041
1042 /*
1043 * If it is not operational, then we have already marked it as dead
1044 * else if it is the last working disks, ignore the error, let the
1045 * next level up know.
1046 * else mark the drive as failed
1047 */
NeilBrownb2d444d2005-11-08 21:39:31 -08001048 if (test_bit(In_sync, &rdev->flags)
NeilBrown4044ba52009-01-09 08:31:11 +11001049 && (conf->raid_disks - mddev->degraded) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 /*
1051 * Don't fail the drive, act as though we were just a
NeilBrown4044ba52009-01-09 08:31:11 +11001052 * normal single drive.
1053 * However don't try a recovery from this drive as
1054 * it is very likely to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 */
NeilBrown4044ba52009-01-09 08:31:11 +11001056 mddev->recovery_disabled = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 return;
NeilBrown4044ba52009-01-09 08:31:11 +11001058 }
NeilBrownc04be0a2006-10-03 01:15:53 -07001059 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1060 unsigned long flags;
1061 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 mddev->degraded++;
NeilBrowndd00a992007-05-10 03:15:50 -07001063 set_bit(Faulty, &rdev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001064 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 /*
1066 * if recovery is running, make sure it aborts.
1067 */
NeilBrowndfc70642008-05-23 13:04:39 -07001068 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrowndd00a992007-05-10 03:15:50 -07001069 } else
1070 set_bit(Faulty, &rdev->flags);
NeilBrown850b2b42006-10-03 01:15:46 -07001071 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Nick Andrewd7a420c2008-04-28 02:15:55 -07001072 printk(KERN_ALERT "raid1: Disk failure on %s, disabling device.\n"
1073 "raid1: Operation continuing on %d devices.\n",
NeilBrown11ce99e2006-10-03 01:15:52 -07001074 bdevname(rdev->bdev,b), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001075}
1076
1077static void print_conf(conf_t *conf)
1078{
1079 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 printk("RAID1 conf printout:\n");
1082 if (!conf) {
1083 printk("(!conf)\n");
1084 return;
1085 }
NeilBrown11ce99e2006-10-03 01:15:52 -07001086 printk(" --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 conf->raid_disks);
1088
NeilBrownddac7c72006-08-31 21:27:36 -07001089 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 for (i = 0; i < conf->raid_disks; i++) {
1091 char b[BDEVNAME_SIZE];
NeilBrownddac7c72006-08-31 21:27:36 -07001092 mdk_rdev_t *rdev = rcu_dereference(conf->mirrors[i].rdev);
1093 if (rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 printk(" disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownddac7c72006-08-31 21:27:36 -07001095 i, !test_bit(In_sync, &rdev->flags),
1096 !test_bit(Faulty, &rdev->flags),
1097 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 }
NeilBrownddac7c72006-08-31 21:27:36 -07001099 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100}
1101
1102static void close_sync(conf_t *conf)
1103{
NeilBrown17999be2006-01-06 00:20:12 -08001104 wait_barrier(conf);
1105 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
1107 mempool_destroy(conf->r1buf_pool);
1108 conf->r1buf_pool = NULL;
1109}
1110
1111static int raid1_spare_active(mddev_t *mddev)
1112{
1113 int i;
1114 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115
1116 /*
1117 * Find all failed disks within the RAID1 configuration
NeilBrownddac7c72006-08-31 21:27:36 -07001118 * and mark them readable.
1119 * Called under mddev lock, so rcu protection not needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 */
1121 for (i = 0; i < conf->raid_disks; i++) {
NeilBrownddac7c72006-08-31 21:27:36 -07001122 mdk_rdev_t *rdev = conf->mirrors[i].rdev;
1123 if (rdev
1124 && !test_bit(Faulty, &rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07001125 && !test_and_set_bit(In_sync, &rdev->flags)) {
1126 unsigned long flags;
1127 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 mddev->degraded--;
NeilBrownc04be0a2006-10-03 01:15:53 -07001129 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 }
1131 }
1132
1133 print_conf(conf);
1134 return 0;
1135}
1136
1137
1138static int raid1_add_disk(mddev_t *mddev, mdk_rdev_t *rdev)
1139{
1140 conf_t *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001141 int err = -EEXIST;
NeilBrown41158c72005-06-21 17:17:25 -07001142 int mirror = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 mirror_info_t *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10001144 int first = 0;
1145 int last = mddev->raid_disks - 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146
Neil Brown6c2fce22008-06-28 08:31:31 +10001147 if (rdev->raid_disk >= 0)
1148 first = last = rdev->raid_disk;
1149
1150 for (mirror = first; mirror <= last; mirror++)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151 if ( !(p=conf->mirrors+mirror)->rdev) {
1152
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +10001153 disk_stack_limits(mddev->gendisk, rdev->bdev,
1154 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 /* as we don't honour merge_bvec_fn, we must never risk
1156 * violating it, so limit ->max_sector to one PAGE, as
1157 * a one page request is never in violation.
1158 */
1159 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
Martin K. Petersenae03bf62009-05-22 17:17:50 -04001160 queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
1162
1163 p->head_position = 0;
1164 rdev->raid_disk = mirror;
Neil Brown199050e2008-06-28 08:31:33 +10001165 err = 0;
NeilBrown6aea114a2005-11-28 13:44:13 -08001166 /* As all devices are equivalent, we don't need a full recovery
1167 * if this was recently any drive of the array
1168 */
1169 if (rdev->saved_raid_disk < 0)
NeilBrown41158c72005-06-21 17:17:25 -07001170 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001171 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 break;
1173 }
Andre Nollac5e7112009-08-03 10:59:47 +10001174 md_integrity_add_rdev(rdev, mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001176 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177}
1178
1179static int raid1_remove_disk(mddev_t *mddev, int number)
1180{
1181 conf_t *conf = mddev->private;
1182 int err = 0;
1183 mdk_rdev_t *rdev;
1184 mirror_info_t *p = conf->mirrors+ number;
1185
1186 print_conf(conf);
1187 rdev = p->rdev;
1188 if (rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001189 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190 atomic_read(&rdev->nr_pending)) {
1191 err = -EBUSY;
1192 goto abort;
1193 }
NeilBrowndfc70642008-05-23 13:04:39 -07001194 /* Only remove non-faulty devices is recovery
1195 * is not possible.
1196 */
1197 if (!test_bit(Faulty, &rdev->flags) &&
1198 mddev->degraded < conf->raid_disks) {
1199 err = -EBUSY;
1200 goto abort;
1201 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07001203 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 if (atomic_read(&rdev->nr_pending)) {
1205 /* lost the race, try later */
1206 err = -EBUSY;
1207 p->rdev = rdev;
Andre Nollac5e7112009-08-03 10:59:47 +10001208 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 }
Andre Nollac5e7112009-08-03 10:59:47 +10001210 md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 }
1212abort:
1213
1214 print_conf(conf);
1215 return err;
1216}
1217
1218
NeilBrown6712ecf2007-09-27 12:47:43 +02001219static void end_sync_read(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001220{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
NeilBrownd11c1712006-01-06 00:20:26 -08001222 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223
NeilBrownd11c1712006-01-06 00:20:26 -08001224 for (i=r1_bio->mddev->raid_disks; i--; )
1225 if (r1_bio->bios[i] == bio)
1226 break;
1227 BUG_ON(i < 0);
1228 update_head_pos(i, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 /*
1230 * we have read a block, now it needs to be re-written,
1231 * or re-read if the read failed.
1232 * We don't do much here, just schedule handling by raid1d
1233 */
NeilBrown69382e82006-01-06 00:20:22 -08001234 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrownd11c1712006-01-06 00:20:26 -08001236
1237 if (atomic_dec_and_test(&r1_bio->remaining))
1238 reschedule_retry(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239}
1240
NeilBrown6712ecf2007-09-27 12:47:43 +02001241static void end_sync_write(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242{
1243 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
1244 r1bio_t * r1_bio = (r1bio_t *)(bio->bi_private);
1245 mddev_t *mddev = r1_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10001246 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247 int i;
1248 int mirror=0;
1249
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 for (i = 0; i < conf->raid_disks; i++)
1251 if (r1_bio->bios[i] == bio) {
1252 mirror = i;
1253 break;
1254 }
NeilBrown6b1117d2006-03-31 02:31:57 -08001255 if (!uptodate) {
1256 int sync_blocks = 0;
1257 sector_t s = r1_bio->sector;
1258 long sectors_to_go = r1_bio->sectors;
1259 /* make sure these bits doesn't get cleared. */
1260 do {
NeilBrown5e3db642006-07-10 04:44:18 -07001261 bitmap_end_sync(mddev->bitmap, s,
NeilBrown6b1117d2006-03-31 02:31:57 -08001262 &sync_blocks, 1);
1263 s += sync_blocks;
1264 sectors_to_go -= sync_blocks;
1265 } while (sectors_to_go > 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001266 md_error(mddev, conf->mirrors[mirror].rdev);
NeilBrown6b1117d2006-03-31 02:31:57 -08001267 }
NeilBrowne3b97032005-08-04 12:53:34 -07001268
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 update_head_pos(mirror, r1_bio);
1270
1271 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown73d5c382009-02-25 13:18:47 +11001272 sector_t s = r1_bio->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 put_buf(r1_bio);
NeilBrown73d5c382009-02-25 13:18:47 +11001274 md_done_sync(mddev, s, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276}
1277
1278static void sync_request_write(mddev_t *mddev, r1bio_t *r1_bio)
1279{
NeilBrown070ec552009-06-16 16:54:21 +10001280 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 int i;
1282 int disks = conf->raid_disks;
1283 struct bio *bio, *wbio;
1284
1285 bio = r1_bio->bios[r1_bio->read_disk];
1286
NeilBrown69382e82006-01-06 00:20:22 -08001287
NeilBrownd11c1712006-01-06 00:20:26 -08001288 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1289 /* We have read all readable devices. If we haven't
1290 * got the block, then there is no hope left.
1291 * If we have, then we want to do a comparison
1292 * and skip the write if everything is the same.
1293 * If any blocks failed to read, then we need to
1294 * attempt an over-write
1295 */
1296 int primary;
1297 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
1298 for (i=0; i<mddev->raid_disks; i++)
1299 if (r1_bio->bios[i]->bi_end_io == end_sync_read)
1300 md_error(mddev, conf->mirrors[i].rdev);
1301
1302 md_done_sync(mddev, r1_bio->sectors, 1);
1303 put_buf(r1_bio);
1304 return;
1305 }
1306 for (primary=0; primary<mddev->raid_disks; primary++)
1307 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1308 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1309 r1_bio->bios[primary]->bi_end_io = NULL;
NeilBrown03c902e2006-01-06 00:20:46 -08001310 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
NeilBrownd11c1712006-01-06 00:20:26 -08001311 break;
1312 }
1313 r1_bio->read_disk = primary;
1314 for (i=0; i<mddev->raid_disks; i++)
Mike Accettaed456662007-06-16 10:16:07 -07001315 if (r1_bio->bios[i]->bi_end_io == end_sync_read) {
NeilBrownd11c1712006-01-06 00:20:26 -08001316 int j;
1317 int vcnt = r1_bio->sectors >> (PAGE_SHIFT- 9);
1318 struct bio *pbio = r1_bio->bios[primary];
1319 struct bio *sbio = r1_bio->bios[i];
Mike Accettaed456662007-06-16 10:16:07 -07001320
1321 if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) {
1322 for (j = vcnt; j-- ; ) {
1323 struct page *p, *s;
1324 p = pbio->bi_io_vec[j].bv_page;
1325 s = sbio->bi_io_vec[j].bv_page;
1326 if (memcmp(page_address(p),
1327 page_address(s),
1328 PAGE_SIZE))
1329 break;
1330 }
1331 } else
1332 j = 0;
NeilBrownd11c1712006-01-06 00:20:26 -08001333 if (j >= 0)
1334 mddev->resync_mismatches += r1_bio->sectors;
NeilBrowncf7a4412007-10-16 23:30:55 -07001335 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
1336 && test_bit(BIO_UPTODATE, &sbio->bi_flags))) {
NeilBrownd11c1712006-01-06 00:20:26 -08001337 sbio->bi_end_io = NULL;
NeilBrown03c902e2006-01-06 00:20:46 -08001338 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1339 } else {
NeilBrownd11c1712006-01-06 00:20:26 -08001340 /* fixup the bio for reuse */
NeilBrown698b18c2008-05-23 13:04:35 -07001341 int size;
NeilBrownd11c1712006-01-06 00:20:26 -08001342 sbio->bi_vcnt = vcnt;
1343 sbio->bi_size = r1_bio->sectors << 9;
1344 sbio->bi_idx = 0;
1345 sbio->bi_phys_segments = 0;
NeilBrownd11c1712006-01-06 00:20:26 -08001346 sbio->bi_flags &= ~(BIO_POOL_MASK - 1);
1347 sbio->bi_flags |= 1 << BIO_UPTODATE;
1348 sbio->bi_next = NULL;
1349 sbio->bi_sector = r1_bio->sector +
1350 conf->mirrors[i].rdev->data_offset;
1351 sbio->bi_bdev = conf->mirrors[i].rdev->bdev;
NeilBrown698b18c2008-05-23 13:04:35 -07001352 size = sbio->bi_size;
1353 for (j = 0; j < vcnt ; j++) {
1354 struct bio_vec *bi;
1355 bi = &sbio->bi_io_vec[j];
1356 bi->bv_offset = 0;
1357 if (size > PAGE_SIZE)
1358 bi->bv_len = PAGE_SIZE;
1359 else
1360 bi->bv_len = size;
1361 size -= PAGE_SIZE;
1362 memcpy(page_address(bi->bv_page),
NeilBrown3eda22d2007-01-26 00:57:01 -08001363 page_address(pbio->bi_io_vec[j].bv_page),
1364 PAGE_SIZE);
NeilBrown698b18c2008-05-23 13:04:35 -07001365 }
NeilBrown3eda22d2007-01-26 00:57:01 -08001366
NeilBrownd11c1712006-01-06 00:20:26 -08001367 }
1368 }
1369 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 if (!test_bit(R1BIO_Uptodate, &r1_bio->state)) {
NeilBrown69382e82006-01-06 00:20:22 -08001371 /* ouch - failed to read all of that.
1372 * Try some synchronous reads of other devices to get
1373 * good data, much like with normal read errors. Only
NeilBrownddac7c72006-08-31 21:27:36 -07001374 * read into the pages we already have so we don't
NeilBrown69382e82006-01-06 00:20:22 -08001375 * need to re-issue the read request.
1376 * We don't need to freeze the array, because being in an
1377 * active sync request, there is no normal IO, and
1378 * no overlapping syncs.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 */
NeilBrown69382e82006-01-06 00:20:22 -08001380 sector_t sect = r1_bio->sector;
1381 int sectors = r1_bio->sectors;
1382 int idx = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383
NeilBrown69382e82006-01-06 00:20:22 -08001384 while(sectors) {
1385 int s = sectors;
1386 int d = r1_bio->read_disk;
1387 int success = 0;
1388 mdk_rdev_t *rdev;
1389
1390 if (s > (PAGE_SIZE>>9))
1391 s = PAGE_SIZE >> 9;
1392 do {
1393 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
NeilBrownddac7c72006-08-31 21:27:36 -07001394 /* No rcu protection needed here devices
1395 * can only be removed when no resync is
1396 * active, and resync is currently active
1397 */
NeilBrown69382e82006-01-06 00:20:22 -08001398 rdev = conf->mirrors[d].rdev;
1399 if (sync_page_io(rdev->bdev,
1400 sect + rdev->data_offset,
1401 s<<9,
1402 bio->bi_io_vec[idx].bv_page,
1403 READ)) {
1404 success = 1;
1405 break;
1406 }
1407 }
1408 d++;
1409 if (d == conf->raid_disks)
1410 d = 0;
1411 } while (!success && d != r1_bio->read_disk);
1412
1413 if (success) {
NeilBrown097426f2006-01-06 00:20:37 -08001414 int start = d;
NeilBrown69382e82006-01-06 00:20:22 -08001415 /* write it back and re-read */
1416 set_bit(R1BIO_Uptodate, &r1_bio->state);
1417 while (d != r1_bio->read_disk) {
1418 if (d == 0)
1419 d = conf->raid_disks;
1420 d--;
1421 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1422 continue;
1423 rdev = conf->mirrors[d].rdev;
NeilBrown4dbcdc72006-01-06 00:20:52 -08001424 atomic_add(s, &rdev->corrected_errors);
NeilBrown69382e82006-01-06 00:20:22 -08001425 if (sync_page_io(rdev->bdev,
1426 sect + rdev->data_offset,
1427 s<<9,
1428 bio->bi_io_vec[idx].bv_page,
NeilBrown097426f2006-01-06 00:20:37 -08001429 WRITE) == 0)
1430 md_error(mddev, rdev);
1431 }
1432 d = start;
1433 while (d != r1_bio->read_disk) {
1434 if (d == 0)
1435 d = conf->raid_disks;
1436 d--;
1437 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1438 continue;
1439 rdev = conf->mirrors[d].rdev;
1440 if (sync_page_io(rdev->bdev,
NeilBrown69382e82006-01-06 00:20:22 -08001441 sect + rdev->data_offset,
1442 s<<9,
1443 bio->bi_io_vec[idx].bv_page,
NeilBrown097426f2006-01-06 00:20:37 -08001444 READ) == 0)
NeilBrown69382e82006-01-06 00:20:22 -08001445 md_error(mddev, rdev);
NeilBrown69382e82006-01-06 00:20:22 -08001446 }
1447 } else {
1448 char b[BDEVNAME_SIZE];
1449 /* Cannot read from anywhere, array is toast */
1450 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
1451 printk(KERN_ALERT "raid1: %s: unrecoverable I/O read error"
1452 " for block %llu\n",
1453 bdevname(bio->bi_bdev,b),
1454 (unsigned long long)r1_bio->sector);
1455 md_done_sync(mddev, r1_bio->sectors, 0);
1456 put_buf(r1_bio);
1457 return;
1458 }
1459 sectors -= s;
1460 sect += s;
1461 idx ++;
1462 }
1463 }
NeilBrownd11c1712006-01-06 00:20:26 -08001464
1465 /*
1466 * schedule writes
1467 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468 atomic_set(&r1_bio->remaining, 1);
1469 for (i = 0; i < disks ; i++) {
1470 wbio = r1_bio->bios[i];
NeilBrown3e198f72006-01-06 00:20:21 -08001471 if (wbio->bi_end_io == NULL ||
1472 (wbio->bi_end_io == end_sync_read &&
1473 (i == r1_bio->read_disk ||
1474 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 continue;
1476
NeilBrown3e198f72006-01-06 00:20:21 -08001477 wbio->bi_rw = WRITE;
1478 wbio->bi_end_io = end_sync_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479 atomic_inc(&r1_bio->remaining);
1480 md_sync_acct(conf->mirrors[i].rdev->bdev, wbio->bi_size >> 9);
NeilBrown191ea9b2005-06-21 17:17:23 -07001481
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 generic_make_request(wbio);
1483 }
1484
1485 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001486 /* if we're here, all write(s) have completed, so clean up */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 md_done_sync(mddev, r1_bio->sectors, 1);
1488 put_buf(r1_bio);
1489 }
1490}
1491
1492/*
1493 * This is a kernel thread which:
1494 *
1495 * 1. Retries failed read operations on working mirrors.
1496 * 2. Updates the raid superblock when problems encounter.
1497 * 3. Performs writes following reads for array syncronising.
1498 */
1499
NeilBrown867868f2006-10-03 01:15:51 -07001500static void fix_read_error(conf_t *conf, int read_disk,
1501 sector_t sect, int sectors)
1502{
1503 mddev_t *mddev = conf->mddev;
1504 while(sectors) {
1505 int s = sectors;
1506 int d = read_disk;
1507 int success = 0;
1508 int start;
1509 mdk_rdev_t *rdev;
1510
1511 if (s > (PAGE_SIZE>>9))
1512 s = PAGE_SIZE >> 9;
1513
1514 do {
1515 /* Note: no rcu protection needed here
1516 * as this is synchronous in the raid1d thread
1517 * which is the thread that might remove
1518 * a device. If raid1d ever becomes multi-threaded....
1519 */
1520 rdev = conf->mirrors[d].rdev;
1521 if (rdev &&
1522 test_bit(In_sync, &rdev->flags) &&
1523 sync_page_io(rdev->bdev,
1524 sect + rdev->data_offset,
1525 s<<9,
1526 conf->tmppage, READ))
1527 success = 1;
1528 else {
1529 d++;
1530 if (d == conf->raid_disks)
1531 d = 0;
1532 }
1533 } while (!success && d != read_disk);
1534
1535 if (!success) {
1536 /* Cannot read from anywhere -- bye bye array */
1537 md_error(mddev, conf->mirrors[read_disk].rdev);
1538 break;
1539 }
1540 /* write it back and re-read */
1541 start = d;
1542 while (d != read_disk) {
1543 if (d==0)
1544 d = conf->raid_disks;
1545 d--;
1546 rdev = conf->mirrors[d].rdev;
1547 if (rdev &&
1548 test_bit(In_sync, &rdev->flags)) {
1549 if (sync_page_io(rdev->bdev,
1550 sect + rdev->data_offset,
1551 s<<9, conf->tmppage, WRITE)
1552 == 0)
1553 /* Well, this device is dead */
1554 md_error(mddev, rdev);
1555 }
1556 }
1557 d = start;
1558 while (d != read_disk) {
1559 char b[BDEVNAME_SIZE];
1560 if (d==0)
1561 d = conf->raid_disks;
1562 d--;
1563 rdev = conf->mirrors[d].rdev;
1564 if (rdev &&
1565 test_bit(In_sync, &rdev->flags)) {
1566 if (sync_page_io(rdev->bdev,
1567 sect + rdev->data_offset,
1568 s<<9, conf->tmppage, READ)
1569 == 0)
1570 /* Well, this device is dead */
1571 md_error(mddev, rdev);
1572 else {
1573 atomic_add(s, &rdev->corrected_errors);
1574 printk(KERN_INFO
1575 "raid1:%s: read error corrected "
1576 "(%d sectors at %llu on %s)\n",
1577 mdname(mddev), s,
Randy Dunlap969b7552006-10-28 10:38:32 -07001578 (unsigned long long)(sect +
1579 rdev->data_offset),
NeilBrown867868f2006-10-03 01:15:51 -07001580 bdevname(rdev->bdev, b));
1581 }
1582 }
1583 }
1584 sectors -= s;
1585 sect += s;
1586 }
1587}
1588
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589static void raid1d(mddev_t *mddev)
1590{
1591 r1bio_t *r1_bio;
1592 struct bio *bio;
1593 unsigned long flags;
NeilBrown070ec552009-06-16 16:54:21 +10001594 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 struct list_head *head = &conf->retry_list;
1596 int unplug=0;
1597 mdk_rdev_t *rdev;
1598
1599 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600
1601 for (;;) {
1602 char b[BDEVNAME_SIZE];
NeilBrowna35e63e2008-03-04 14:29:29 -08001603
1604 unplug += flush_pending_writes(conf);
1605
Linus Torvalds1da177e2005-04-16 15:20:36 -07001606 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08001607 if (list_empty(head)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001608 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08001610 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 r1_bio = list_entry(head->prev, r1bio_t, retry_list);
1612 list_del(head->prev);
NeilBrownddaf22a2006-01-06 00:20:19 -08001613 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 spin_unlock_irqrestore(&conf->device_lock, flags);
1615
1616 mddev = r1_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10001617 conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
1619 sync_request_write(mddev, r1_bio);
1620 unplug = 1;
NeilBrowna9701a32005-11-08 21:39:34 -08001621 } else if (test_bit(R1BIO_BarrierRetry, &r1_bio->state)) {
1622 /* some requests in the r1bio were BIO_RW_BARRIER
NeilBrownbea27712006-05-01 12:15:46 -07001623 * requests which failed with -EOPNOTSUPP. Hohumm..
NeilBrowna9701a32005-11-08 21:39:34 -08001624 * Better resubmit without the barrier.
1625 * We know which devices to resubmit for, because
1626 * all others have had their bios[] entry cleared.
NeilBrown5e7dd2a2006-05-01 12:15:47 -07001627 * We already have a nr_pending reference on these rdevs.
NeilBrowna9701a32005-11-08 21:39:34 -08001628 */
1629 int i;
Jens Axboe1f98a132009-09-11 14:32:04 +02001630 const bool do_sync = bio_rw_flagged(r1_bio->master_bio, BIO_RW_SYNCIO);
NeilBrowna9701a32005-11-08 21:39:34 -08001631 clear_bit(R1BIO_BarrierRetry, &r1_bio->state);
1632 clear_bit(R1BIO_Barrier, &r1_bio->state);
1633 for (i=0; i < conf->raid_disks; i++)
NeilBrown2f889122006-03-27 01:18:19 -08001634 if (r1_bio->bios[i])
1635 atomic_inc(&r1_bio->remaining);
1636 for (i=0; i < conf->raid_disks; i++)
NeilBrowna9701a32005-11-08 21:39:34 -08001637 if (r1_bio->bios[i]) {
1638 struct bio_vec *bvec;
1639 int j;
1640
1641 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1642 /* copy pages from the failed bio, as
1643 * this might be a write-behind device */
1644 __bio_for_each_segment(bvec, bio, j, 0)
1645 bvec->bv_page = bio_iovec_idx(r1_bio->bios[i], j)->bv_page;
1646 bio_put(r1_bio->bios[i]);
1647 bio->bi_sector = r1_bio->sector +
1648 conf->mirrors[i].rdev->data_offset;
1649 bio->bi_bdev = conf->mirrors[i].rdev->bdev;
1650 bio->bi_end_io = raid1_end_write_request;
Dmitry Monakhov1ef04fe2009-09-20 05:52:25 +04001651 bio->bi_rw = WRITE |
1652 (do_sync << BIO_RW_SYNCIO);
NeilBrowna9701a32005-11-08 21:39:34 -08001653 bio->bi_private = r1_bio;
1654 r1_bio->bios[i] = bio;
1655 generic_make_request(bio);
1656 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 } else {
1658 int disk;
NeilBrownddaf22a2006-01-06 00:20:19 -08001659
1660 /* we got a read error. Maybe the drive is bad. Maybe just
1661 * the block and we can fix it.
1662 * We freeze all other IO, and try reading the block from
1663 * other devices. When we find one, we re-write
1664 * and check it that fixes the read error.
1665 * This is all done synchronously while the array is
1666 * frozen
1667 */
NeilBrown867868f2006-10-03 01:15:51 -07001668 if (mddev->ro == 0) {
1669 freeze_array(conf);
1670 fix_read_error(conf, r1_bio->read_disk,
1671 r1_bio->sector,
1672 r1_bio->sectors);
1673 unfreeze_array(conf);
NeilBrownd0e26072009-12-01 17:30:59 +11001674 } else
1675 md_error(mddev,
1676 conf->mirrors[r1_bio->read_disk].rdev);
NeilBrownddaf22a2006-01-06 00:20:19 -08001677
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 bio = r1_bio->bios[r1_bio->read_disk];
NeilBrownd0e26072009-12-01 17:30:59 +11001679 if ((disk=read_balance(conf, r1_bio)) == -1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 printk(KERN_ALERT "raid1: %s: unrecoverable I/O"
1681 " read error for block %llu\n",
1682 bdevname(bio->bi_bdev,b),
1683 (unsigned long long)r1_bio->sector);
1684 raid_end_bio_io(r1_bio);
1685 } else {
Jens Axboe1f98a132009-09-11 14:32:04 +02001686 const bool do_sync = bio_rw_flagged(r1_bio->master_bio, BIO_RW_SYNCIO);
NeilBrowncf30a472006-01-06 00:20:23 -08001687 r1_bio->bios[r1_bio->read_disk] =
1688 mddev->ro ? IO_BLOCKED : NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 r1_bio->read_disk = disk;
1690 bio_put(bio);
1691 bio = bio_clone(r1_bio->master_bio, GFP_NOIO);
1692 r1_bio->bios[r1_bio->read_disk] = bio;
1693 rdev = conf->mirrors[disk].rdev;
1694 if (printk_ratelimit())
1695 printk(KERN_ERR "raid1: %s: redirecting sector %llu to"
1696 " another mirror\n",
1697 bdevname(rdev->bdev,b),
1698 (unsigned long long)r1_bio->sector);
1699 bio->bi_sector = r1_bio->sector + rdev->data_offset;
1700 bio->bi_bdev = rdev->bdev;
1701 bio->bi_end_io = raid1_end_read_request;
Dmitry Monakhov1ef04fe2009-09-20 05:52:25 +04001702 bio->bi_rw = READ | (do_sync << BIO_RW_SYNCIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 bio->bi_private = r1_bio;
1704 unplug = 1;
1705 generic_make_request(bio);
1706 }
1707 }
NeilBrown1d9d5242009-10-16 15:55:32 +11001708 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 if (unplug)
1711 unplug_slaves(mddev);
1712}
1713
1714
1715static int init_resync(conf_t *conf)
1716{
1717 int buffs;
1718
1719 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02001720 BUG_ON(conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
1722 conf->poolinfo);
1723 if (!conf->r1buf_pool)
1724 return -ENOMEM;
1725 conf->next_resync = 0;
1726 return 0;
1727}
1728
1729/*
1730 * perform a "sync" on one "block"
1731 *
1732 * We need to make sure that no normal I/O request - particularly write
1733 * requests - conflict with active sync requests.
1734 *
1735 * This is achieved by tracking pending requests and a 'barrier' concept
1736 * that can be installed to exclude normal IO requests.
1737 */
1738
NeilBrown57afd892005-06-21 17:17:13 -07001739static sector_t sync_request(mddev_t *mddev, sector_t sector_nr, int *skipped, int go_faster)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740{
NeilBrown070ec552009-06-16 16:54:21 +10001741 conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 r1bio_t *r1_bio;
1743 struct bio *bio;
1744 sector_t max_sector, nr_sectors;
NeilBrown3e198f72006-01-06 00:20:21 -08001745 int disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 int i;
NeilBrown3e198f72006-01-06 00:20:21 -08001747 int wonly = -1;
1748 int write_targets = 0, read_targets = 0;
NeilBrown191ea9b2005-06-21 17:17:23 -07001749 int sync_blocks;
NeilBrowne3b97032005-08-04 12:53:34 -07001750 int still_degraded = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751
1752 if (!conf->r1buf_pool)
NeilBrown191ea9b2005-06-21 17:17:23 -07001753 {
1754/*
1755 printk("sync start - bitmap %p\n", mddev->bitmap);
1756*/
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07001758 return 0;
NeilBrown191ea9b2005-06-21 17:17:23 -07001759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760
Andre Noll58c0fed2009-03-31 14:33:13 +11001761 max_sector = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 if (sector_nr >= max_sector) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001763 /* If we aborted, we need to abort the
1764 * sync on the 'current' bitmap chunk (there will
1765 * only be one in raid1 resync.
1766 * We can find the current addess in mddev->curr_resync
1767 */
NeilBrown6a806c52005-07-15 03:56:35 -07001768 if (mddev->curr_resync < max_sector) /* aborted */
1769 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
NeilBrown191ea9b2005-06-21 17:17:23 -07001770 &sync_blocks, 1);
NeilBrown6a806c52005-07-15 03:56:35 -07001771 else /* completed sync */
NeilBrown191ea9b2005-06-21 17:17:23 -07001772 conf->fullsync = 0;
NeilBrown6a806c52005-07-15 03:56:35 -07001773
1774 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 close_sync(conf);
1776 return 0;
1777 }
1778
NeilBrown07d84d102006-06-26 00:27:56 -07001779 if (mddev->bitmap == NULL &&
1780 mddev->recovery_cp == MaxSector &&
NeilBrown6394cca2006-08-27 01:23:50 -07001781 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown07d84d102006-06-26 00:27:56 -07001782 conf->fullsync == 0) {
1783 *skipped = 1;
1784 return max_sector - sector_nr;
1785 }
NeilBrown6394cca2006-08-27 01:23:50 -07001786 /* before building a request, check if we can skip these blocks..
1787 * This call the bitmap_start_sync doesn't actually record anything
1788 */
NeilBrowne3b97032005-08-04 12:53:34 -07001789 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrowne5de4852005-11-08 21:39:38 -08001790 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001791 /* We can skip this block, and probably several more */
1792 *skipped = 1;
1793 return sync_blocks;
1794 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 /*
NeilBrown17999be2006-01-06 00:20:12 -08001796 * If there is non-resync activity waiting for a turn,
1797 * and resync is going fast enough,
1798 * then let it though before starting on this new sync request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799 */
NeilBrown17999be2006-01-06 00:20:12 -08001800 if (!go_faster && conf->nr_waiting)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 msleep_interruptible(1000);
NeilBrown17999be2006-01-06 00:20:12 -08001802
NeilBrownb47490c2008-02-06 01:39:50 -08001803 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
NeilBrown17999be2006-01-06 00:20:12 -08001804 raise_barrier(conf);
1805
1806 conf->next_resync = sector_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
NeilBrown3e198f72006-01-06 00:20:21 -08001809 rcu_read_lock();
1810 /*
1811 * If we get a correctably read error during resync or recovery,
1812 * we might want to read from a different device. So we
1813 * flag all drives that could conceivably be read from for READ,
1814 * and any others (which will be non-In_sync devices) for WRITE.
1815 * If a read fails, we try reading from something else for which READ
1816 * is OK.
1817 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 r1_bio->mddev = mddev;
1820 r1_bio->sector = sector_nr;
NeilBrown191ea9b2005-06-21 17:17:23 -07001821 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 set_bit(R1BIO_IsSync, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
1824 for (i=0; i < conf->raid_disks; i++) {
NeilBrown3e198f72006-01-06 00:20:21 -08001825 mdk_rdev_t *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 bio = r1_bio->bios[i];
1827
1828 /* take from bio_init */
1829 bio->bi_next = NULL;
1830 bio->bi_flags |= 1 << BIO_UPTODATE;
NeilBrown802ba062006-12-13 00:34:13 -08001831 bio->bi_rw = READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 bio->bi_vcnt = 0;
1833 bio->bi_idx = 0;
1834 bio->bi_phys_segments = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 bio->bi_size = 0;
1836 bio->bi_end_io = NULL;
1837 bio->bi_private = NULL;
1838
NeilBrown3e198f72006-01-06 00:20:21 -08001839 rdev = rcu_dereference(conf->mirrors[i].rdev);
1840 if (rdev == NULL ||
1841 test_bit(Faulty, &rdev->flags)) {
NeilBrowne3b97032005-08-04 12:53:34 -07001842 still_degraded = 1;
1843 continue;
NeilBrown3e198f72006-01-06 00:20:21 -08001844 } else if (!test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 bio->bi_rw = WRITE;
1846 bio->bi_end_io = end_sync_write;
1847 write_targets ++;
NeilBrown3e198f72006-01-06 00:20:21 -08001848 } else {
1849 /* may need to read from here */
1850 bio->bi_rw = READ;
1851 bio->bi_end_io = end_sync_read;
1852 if (test_bit(WriteMostly, &rdev->flags)) {
1853 if (wonly < 0)
1854 wonly = i;
1855 } else {
1856 if (disk < 0)
1857 disk = i;
1858 }
1859 read_targets++;
1860 }
1861 atomic_inc(&rdev->nr_pending);
1862 bio->bi_sector = sector_nr + rdev->data_offset;
1863 bio->bi_bdev = rdev->bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 bio->bi_private = r1_bio;
1865 }
NeilBrown3e198f72006-01-06 00:20:21 -08001866 rcu_read_unlock();
1867 if (disk < 0)
1868 disk = wonly;
1869 r1_bio->read_disk = disk;
NeilBrown191ea9b2005-06-21 17:17:23 -07001870
NeilBrown3e198f72006-01-06 00:20:21 -08001871 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
1872 /* extra read targets are also write targets */
1873 write_targets += read_targets-1;
1874
1875 if (write_targets == 0 || read_targets == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 /* There is nowhere to write, so all non-sync
1877 * drives must be failed - so we are finished
1878 */
NeilBrown57afd892005-06-21 17:17:13 -07001879 sector_t rv = max_sector - sector_nr;
1880 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 put_buf(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 return rv;
1883 }
1884
NeilBrownc6207272008-02-06 01:39:52 -08001885 if (max_sector > mddev->resync_max)
1886 max_sector = mddev->resync_max; /* Don't do IO beyond here */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 nr_sectors = 0;
NeilBrown289e99e2005-06-21 17:17:24 -07001888 sync_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 do {
1890 struct page *page;
1891 int len = PAGE_SIZE;
1892 if (sector_nr + (len>>9) > max_sector)
1893 len = (max_sector - sector_nr) << 9;
1894 if (len == 0)
1895 break;
NeilBrown6a806c52005-07-15 03:56:35 -07001896 if (sync_blocks == 0) {
1897 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
NeilBrowne5de4852005-11-08 21:39:38 -08001898 &sync_blocks, still_degraded) &&
1899 !conf->fullsync &&
1900 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrown6a806c52005-07-15 03:56:35 -07001901 break;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02001902 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
NeilBrown6a806c52005-07-15 03:56:35 -07001903 if (len > (sync_blocks<<9))
1904 len = sync_blocks<<9;
NeilBrownab7a30c2005-06-21 17:17:23 -07001905 }
NeilBrown191ea9b2005-06-21 17:17:23 -07001906
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 for (i=0 ; i < conf->raid_disks; i++) {
1908 bio = r1_bio->bios[i];
1909 if (bio->bi_end_io) {
NeilBrownd11c1712006-01-06 00:20:26 -08001910 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 if (bio_add_page(bio, page, len, 0) == 0) {
1912 /* stop here */
NeilBrownd11c1712006-01-06 00:20:26 -08001913 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 while (i > 0) {
1915 i--;
1916 bio = r1_bio->bios[i];
NeilBrown6a806c52005-07-15 03:56:35 -07001917 if (bio->bi_end_io==NULL)
1918 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 /* remove last page from this bio */
1920 bio->bi_vcnt--;
1921 bio->bi_size -= len;
1922 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
1923 }
1924 goto bio_full;
1925 }
1926 }
1927 }
1928 nr_sectors += len>>9;
1929 sector_nr += len>>9;
NeilBrown191ea9b2005-06-21 17:17:23 -07001930 sync_blocks -= (len>>9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
1932 bio_full:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 r1_bio->sectors = nr_sectors;
1934
NeilBrownd11c1712006-01-06 00:20:26 -08001935 /* For a user-requested sync, we read all readable devices and do a
1936 * compare
1937 */
1938 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
1939 atomic_set(&r1_bio->remaining, read_targets);
1940 for (i=0; i<conf->raid_disks; i++) {
1941 bio = r1_bio->bios[i];
1942 if (bio->bi_end_io == end_sync_read) {
NeilBrownddac7c72006-08-31 21:27:36 -07001943 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08001944 generic_make_request(bio);
1945 }
1946 }
1947 } else {
1948 atomic_set(&r1_bio->remaining, 1);
1949 bio = r1_bio->bios[r1_bio->read_disk];
NeilBrownddac7c72006-08-31 21:27:36 -07001950 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08001951 generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952
NeilBrownd11c1712006-01-06 00:20:26 -08001953 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 return nr_sectors;
1955}
1956
Dan Williams80c3a6c2009-03-17 18:10:40 -07001957static sector_t raid1_size(mddev_t *mddev, sector_t sectors, int raid_disks)
1958{
1959 if (sectors)
1960 return sectors;
1961
1962 return mddev->dev_sectors;
1963}
1964
NeilBrown709ae482009-12-14 12:49:51 +11001965static conf_t *setup_conf(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966{
1967 conf_t *conf;
NeilBrown709ae482009-12-14 12:49:51 +11001968 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 mirror_info_t *disk;
1970 mdk_rdev_t *rdev;
NeilBrown709ae482009-12-14 12:49:51 +11001971 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972
NeilBrown9ffae0c2006-01-06 00:20:32 -08001973 conf = kzalloc(sizeof(conf_t), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 if (!conf)
NeilBrown709ae482009-12-14 12:49:51 +11001975 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976
NeilBrown9ffae0c2006-01-06 00:20:32 -08001977 conf->mirrors = kzalloc(sizeof(struct mirror_info)*mddev->raid_disks,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001978 GFP_KERNEL);
1979 if (!conf->mirrors)
NeilBrown709ae482009-12-14 12:49:51 +11001980 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981
NeilBrownddaf22a2006-01-06 00:20:19 -08001982 conf->tmppage = alloc_page(GFP_KERNEL);
1983 if (!conf->tmppage)
NeilBrown709ae482009-12-14 12:49:51 +11001984 goto abort;
NeilBrownddaf22a2006-01-06 00:20:19 -08001985
NeilBrown709ae482009-12-14 12:49:51 +11001986 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987 if (!conf->poolinfo)
NeilBrown709ae482009-12-14 12:49:51 +11001988 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 conf->poolinfo->raid_disks = mddev->raid_disks;
1990 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
1991 r1bio_pool_free,
1992 conf->poolinfo);
1993 if (!conf->r1bio_pool)
NeilBrown709ae482009-12-14 12:49:51 +11001994 goto abort;
1995
NeilBrowned9bfdf2009-10-16 15:55:44 +11001996 conf->poolinfo->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997
Neil Browne7e72bf2008-05-14 16:05:54 -07001998 spin_lock_init(&conf->device_lock);
Cheng Renquan159ec1f2009-01-09 08:31:08 +11001999 list_for_each_entry(rdev, &mddev->disks, same_set) {
NeilBrown709ae482009-12-14 12:49:51 +11002000 int disk_idx = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002001 if (disk_idx >= mddev->raid_disks
2002 || disk_idx < 0)
2003 continue;
2004 disk = conf->mirrors + disk_idx;
2005
2006 disk->rdev = rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007
2008 disk->head_position = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 }
2010 conf->raid_disks = mddev->raid_disks;
2011 conf->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 INIT_LIST_HEAD(&conf->retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013
2014 spin_lock_init(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -08002015 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016
NeilBrown191ea9b2005-06-21 17:17:23 -07002017 bio_list_init(&conf->pending_bio_list);
2018 bio_list_init(&conf->flushing_bio_list);
2019
NeilBrown709ae482009-12-14 12:49:51 +11002020 conf->last_used = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002021 for (i = 0; i < conf->raid_disks; i++) {
2022
2023 disk = conf->mirrors + i;
2024
NeilBrown5fd6c1d2006-06-26 00:27:40 -07002025 if (!disk->rdev ||
2026 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 disk->head_position = 0;
NeilBrown918f0232007-08-22 14:01:52 -07002028 if (disk->rdev)
2029 conf->fullsync = 1;
NeilBrown709ae482009-12-14 12:49:51 +11002030 } else if (conf->last_used < 0)
2031 /*
2032 * The first working device is used as a
2033 * starting point to read balancing.
2034 */
2035 conf->last_used = i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 }
NeilBrown709ae482009-12-14 12:49:51 +11002037
2038 err = -EIO;
2039 if (conf->last_used < 0) {
NeilBrown11ce99e2006-10-03 01:15:52 -07002040 printk(KERN_ERR "raid1: no operational mirrors for %s\n",
NeilBrown709ae482009-12-14 12:49:51 +11002041 mdname(mddev));
2042 goto abort;
NeilBrown11ce99e2006-10-03 01:15:52 -07002043 }
NeilBrown709ae482009-12-14 12:49:51 +11002044 err = -ENOMEM;
2045 conf->thread = md_register_thread(raid1d, mddev, NULL);
2046 if (!conf->thread) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002047 printk(KERN_ERR
2048 "raid1: couldn't allocate thread for %s\n",
2049 mdname(mddev));
NeilBrown709ae482009-12-14 12:49:51 +11002050 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002051 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002052
NeilBrown709ae482009-12-14 12:49:51 +11002053 return conf;
2054
2055 abort:
2056 if (conf) {
2057 if (conf->r1bio_pool)
2058 mempool_destroy(conf->r1bio_pool);
2059 kfree(conf->mirrors);
2060 safe_put_page(conf->tmppage);
2061 kfree(conf->poolinfo);
2062 kfree(conf);
2063 }
2064 return ERR_PTR(err);
2065}
2066
2067static int run(mddev_t *mddev)
2068{
2069 conf_t *conf;
2070 int i;
2071 mdk_rdev_t *rdev;
2072
2073 if (mddev->level != 1) {
2074 printk("raid1: %s: raid level not set to mirroring (%d)\n",
2075 mdname(mddev), mddev->level);
2076 return -EIO;
2077 }
2078 if (mddev->reshape_position != MaxSector) {
2079 printk("raid1: %s: reshape_position set but not supported\n",
2080 mdname(mddev));
2081 return -EIO;
2082 }
2083 /*
2084 * copy the already verified devices into our private RAID1
2085 * bookkeeping area. [whatever we allocate in run(),
2086 * should be freed in stop()]
2087 */
2088 if (mddev->private == NULL)
2089 conf = setup_conf(mddev);
2090 else
2091 conf = mddev->private;
2092
2093 if (IS_ERR(conf))
2094 return PTR_ERR(conf);
2095
2096 mddev->queue->queue_lock = &conf->device_lock;
2097 list_for_each_entry(rdev, &mddev->disks, same_set) {
2098 disk_stack_limits(mddev->gendisk, rdev->bdev,
2099 rdev->data_offset << 9);
2100 /* as we don't honour merge_bvec_fn, we must never risk
2101 * violating it, so limit ->max_sector to one PAGE, as
2102 * a one page request is never in violation.
2103 */
2104 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
2105 queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
2106 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
2107 }
2108
2109 mddev->degraded = 0;
2110 for (i=0; i < conf->raid_disks; i++)
2111 if (conf->mirrors[i].rdev == NULL ||
2112 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
2113 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
2114 mddev->degraded++;
2115
2116 if (conf->raid_disks - mddev->degraded == 1)
2117 mddev->recovery_cp = MaxSector;
2118
Andre Noll8c6ac862009-06-18 08:48:06 +10002119 if (mddev->recovery_cp != MaxSector)
2120 printk(KERN_NOTICE "raid1: %s is not clean"
2121 " -- starting background reconstruction\n",
2122 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002123 printk(KERN_INFO
2124 "raid1: raid set %s active with %d out of %d mirrors\n",
2125 mdname(mddev), mddev->raid_disks - mddev->degraded,
2126 mddev->raid_disks);
NeilBrown709ae482009-12-14 12:49:51 +11002127
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128 /*
2129 * Ok, everything is just fine now
2130 */
NeilBrown709ae482009-12-14 12:49:51 +11002131 mddev->thread = conf->thread;
2132 conf->thread = NULL;
2133 mddev->private = conf;
2134
Dan Williams1f403622009-03-31 14:59:03 +11002135 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136
NeilBrown7a5febe2005-05-16 21:53:16 -07002137 mddev->queue->unplug_fn = raid1_unplug;
NeilBrown0d129222006-10-03 01:15:54 -07002138 mddev->queue->backing_dev_info.congested_fn = raid1_congested;
2139 mddev->queue->backing_dev_info.congested_data = mddev;
Andre Nollac5e7112009-08-03 10:59:47 +10002140 md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002141 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002142}
2143
2144static int stop(mddev_t *mddev)
2145{
NeilBrown070ec552009-06-16 16:54:21 +10002146 conf_t *conf = mddev->private;
NeilBrown4b6d2872005-09-09 16:23:47 -07002147 struct bitmap *bitmap = mddev->bitmap;
2148 int behind_wait = 0;
2149
2150 /* wait for behind writes to complete */
2151 while (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
2152 behind_wait++;
2153 printk(KERN_INFO "raid1: behind writes in progress on device %s, waiting to stop (%d)\n", mdname(mddev), behind_wait);
2154 set_current_state(TASK_UNINTERRUPTIBLE);
2155 schedule_timeout(HZ); /* wait a second */
2156 /* need to kick something here to make sure I/O goes? */
2157 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158
NeilBrown409c57f2009-03-31 14:39:39 +11002159 raise_barrier(conf);
2160 lower_barrier(conf);
2161
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162 md_unregister_thread(mddev->thread);
2163 mddev->thread = NULL;
2164 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
2165 if (conf->r1bio_pool)
2166 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002167 kfree(conf->mirrors);
2168 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002169 kfree(conf);
2170 mddev->private = NULL;
2171 return 0;
2172}
2173
2174static int raid1_resize(mddev_t *mddev, sector_t sectors)
2175{
2176 /* no resync is happening, and there is enough space
2177 * on all devices, so we can resize.
2178 * We need to make sure resync covers any new space.
2179 * If the array is shrinking we should possibly wait until
2180 * any io in the removed space completes, but it hardly seems
2181 * worth it.
2182 */
Dan Williams1f403622009-03-31 14:59:03 +11002183 md_set_array_sectors(mddev, raid1_size(mddev, sectors, 0));
Dan Williamsb522adc2009-03-31 15:00:31 +11002184 if (mddev->array_sectors > raid1_size(mddev, sectors, 0))
2185 return -EINVAL;
Andre Nollf233ea52008-07-21 17:05:22 +10002186 set_capacity(mddev->gendisk, mddev->array_sectors);
Linus Torvalds44ce6292007-05-09 18:51:36 -07002187 mddev->changed = 1;
NeilBrown449aad32009-08-03 10:59:58 +10002188 revalidate_disk(mddev->gendisk);
Dan Williamsb522adc2009-03-31 15:00:31 +11002189 if (sectors > mddev->dev_sectors &&
Andre Nollf233ea52008-07-21 17:05:22 +10002190 mddev->recovery_cp == MaxSector) {
Andre Noll58c0fed2009-03-31 14:33:13 +11002191 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2193 }
Dan Williamsb522adc2009-03-31 15:00:31 +11002194 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07002195 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 return 0;
2197}
2198
NeilBrown63c70c42006-03-27 01:18:13 -08002199static int raid1_reshape(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200{
2201 /* We need to:
2202 * 1/ resize the r1bio_pool
2203 * 2/ resize conf->mirrors
2204 *
2205 * We allocate a new r1bio_pool if we can.
2206 * Then raise a device barrier and wait until all IO stops.
2207 * Then resize conf->mirrors and swap in the new r1bio pool.
NeilBrown6ea9c072005-06-21 17:17:09 -07002208 *
2209 * At the same time, we "pack" the devices so that all the missing
2210 * devices have the higher raid_disk numbers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 */
2212 mempool_t *newpool, *oldpool;
2213 struct pool_info *newpoolinfo;
2214 mirror_info_t *newmirrors;
NeilBrown070ec552009-06-16 16:54:21 +10002215 conf_t *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08002216 int cnt, raid_disks;
NeilBrownc04be0a2006-10-03 01:15:53 -07002217 unsigned long flags;
Dan Williamsb5470dc2008-06-27 21:44:04 -07002218 int d, d2, err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219
NeilBrown63c70c42006-03-27 01:18:13 -08002220 /* Cannot change chunk_size, layout, or level */
Andre Noll664e7c42009-06-18 08:45:27 +10002221 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
NeilBrown63c70c42006-03-27 01:18:13 -08002222 mddev->layout != mddev->new_layout ||
2223 mddev->level != mddev->new_level) {
Andre Noll664e7c42009-06-18 08:45:27 +10002224 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown63c70c42006-03-27 01:18:13 -08002225 mddev->new_layout = mddev->layout;
2226 mddev->new_level = mddev->level;
2227 return -EINVAL;
2228 }
2229
Dan Williamsb5470dc2008-06-27 21:44:04 -07002230 err = md_allow_write(mddev);
2231 if (err)
2232 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08002233
NeilBrown63c70c42006-03-27 01:18:13 -08002234 raid_disks = mddev->raid_disks + mddev->delta_disks;
2235
NeilBrown6ea9c072005-06-21 17:17:09 -07002236 if (raid_disks < conf->raid_disks) {
2237 cnt=0;
2238 for (d= 0; d < conf->raid_disks; d++)
2239 if (conf->mirrors[d].rdev)
2240 cnt++;
2241 if (cnt > raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 return -EBUSY;
NeilBrown6ea9c072005-06-21 17:17:09 -07002243 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244
2245 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
2246 if (!newpoolinfo)
2247 return -ENOMEM;
2248 newpoolinfo->mddev = mddev;
2249 newpoolinfo->raid_disks = raid_disks;
2250
2251 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2252 r1bio_pool_free, newpoolinfo);
2253 if (!newpool) {
2254 kfree(newpoolinfo);
2255 return -ENOMEM;
2256 }
NeilBrown9ffae0c2006-01-06 00:20:32 -08002257 newmirrors = kzalloc(sizeof(struct mirror_info) * raid_disks, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 if (!newmirrors) {
2259 kfree(newpoolinfo);
2260 mempool_destroy(newpool);
2261 return -ENOMEM;
2262 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263
NeilBrown17999be2006-01-06 00:20:12 -08002264 raise_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
2266 /* ok, everything is stopped */
2267 oldpool = conf->r1bio_pool;
2268 conf->r1bio_pool = newpool;
NeilBrown6ea9c072005-06-21 17:17:09 -07002269
NeilBrowna88aa782007-08-22 14:01:53 -07002270 for (d = d2 = 0; d < conf->raid_disks; d++) {
2271 mdk_rdev_t *rdev = conf->mirrors[d].rdev;
2272 if (rdev && rdev->raid_disk != d2) {
2273 char nm[20];
2274 sprintf(nm, "rd%d", rdev->raid_disk);
2275 sysfs_remove_link(&mddev->kobj, nm);
2276 rdev->raid_disk = d2;
2277 sprintf(nm, "rd%d", rdev->raid_disk);
2278 sysfs_remove_link(&mddev->kobj, nm);
2279 if (sysfs_create_link(&mddev->kobj,
2280 &rdev->kobj, nm))
2281 printk(KERN_WARNING
2282 "md/raid1: cannot register "
2283 "%s for %s\n",
2284 nm, mdname(mddev));
NeilBrown6ea9c072005-06-21 17:17:09 -07002285 }
NeilBrowna88aa782007-08-22 14:01:53 -07002286 if (rdev)
2287 newmirrors[d2++].rdev = rdev;
2288 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 kfree(conf->mirrors);
2290 conf->mirrors = newmirrors;
2291 kfree(conf->poolinfo);
2292 conf->poolinfo = newpoolinfo;
2293
NeilBrownc04be0a2006-10-03 01:15:53 -07002294 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 mddev->degraded += (raid_disks - conf->raid_disks);
NeilBrownc04be0a2006-10-03 01:15:53 -07002296 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 conf->raid_disks = mddev->raid_disks = raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08002298 mddev->delta_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299
NeilBrown6ea9c072005-06-21 17:17:09 -07002300 conf->last_used = 0; /* just make sure it is in-range */
NeilBrown17999be2006-01-06 00:20:12 -08002301 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302
2303 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2304 md_wakeup_thread(mddev->thread);
2305
2306 mempool_destroy(oldpool);
2307 return 0;
2308}
2309
NeilBrown500af872005-09-09 16:23:58 -07002310static void raid1_quiesce(mddev_t *mddev, int state)
NeilBrown36fa3062005-09-09 16:23:45 -07002311{
NeilBrown070ec552009-06-16 16:54:21 +10002312 conf_t *conf = mddev->private;
NeilBrown36fa3062005-09-09 16:23:45 -07002313
2314 switch(state) {
NeilBrown6eef4b22009-12-14 12:49:51 +11002315 case 2: /* wake for suspend */
2316 wake_up(&conf->wait_barrier);
2317 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07002318 case 1:
NeilBrown17999be2006-01-06 00:20:12 -08002319 raise_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002320 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07002321 case 0:
NeilBrown17999be2006-01-06 00:20:12 -08002322 lower_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07002323 break;
2324 }
NeilBrown36fa3062005-09-09 16:23:45 -07002325}
2326
NeilBrown709ae482009-12-14 12:49:51 +11002327static void *raid1_takeover(mddev_t *mddev)
2328{
2329 /* raid1 can take over:
2330 * raid5 with 2 devices, any layout or chunk size
2331 */
2332 if (mddev->level == 5 && mddev->raid_disks == 2) {
2333 conf_t *conf;
2334 mddev->new_level = 1;
2335 mddev->new_layout = 0;
2336 mddev->new_chunk_sectors = 0;
2337 conf = setup_conf(mddev);
2338 if (!IS_ERR(conf))
2339 conf->barrier = 1;
2340 return conf;
2341 }
2342 return ERR_PTR(-EINVAL);
2343}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344
NeilBrown2604b702006-01-06 00:20:36 -08002345static struct mdk_personality raid1_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346{
2347 .name = "raid1",
NeilBrown2604b702006-01-06 00:20:36 -08002348 .level = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 .owner = THIS_MODULE,
2350 .make_request = make_request,
2351 .run = run,
2352 .stop = stop,
2353 .status = status,
2354 .error_handler = error,
2355 .hot_add_disk = raid1_add_disk,
2356 .hot_remove_disk= raid1_remove_disk,
2357 .spare_active = raid1_spare_active,
2358 .sync_request = sync_request,
2359 .resize = raid1_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07002360 .size = raid1_size,
NeilBrown63c70c42006-03-27 01:18:13 -08002361 .check_reshape = raid1_reshape,
NeilBrown36fa3062005-09-09 16:23:45 -07002362 .quiesce = raid1_quiesce,
NeilBrown709ae482009-12-14 12:49:51 +11002363 .takeover = raid1_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364};
2365
2366static int __init raid_init(void)
2367{
NeilBrown2604b702006-01-06 00:20:36 -08002368 return register_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369}
2370
2371static void raid_exit(void)
2372{
NeilBrown2604b702006-01-06 00:20:36 -08002373 unregister_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374}
2375
2376module_init(raid_init);
2377module_exit(raid_exit);
2378MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11002379MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380MODULE_ALIAS("md-personality-3"); /* RAID1 */
NeilBrownd9d166c2006-01-06 00:20:51 -08002381MODULE_ALIAS("md-raid1");
NeilBrown2604b702006-01-06 00:20:36 -08002382MODULE_ALIAS("md-level-1");