blob: aacf6bf352d87978a15633b05a0d51d3579ce16d [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
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090034#include <linux/slab.h>
Stephen Rothwell25570722008-10-15 09:09:21 +110035#include <linux/delay.h>
NeilBrownbff61972009-03-31 14:33:13 +110036#include <linux/blkdev.h>
Paul Gortmaker056075c2011-07-03 13:58:33 -040037#include <linux/module.h>
NeilBrownbff61972009-03-31 14:33:13 +110038#include <linux/seq_file.h>
Christian Dietrich8bda4702011-07-27 11:00:36 +100039#include <linux/ratelimit.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110040#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110041#include "raid1.h"
42#include "bitmap.h"
NeilBrown191ea9b2005-06-21 17:17:23 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/*
45 * Number of guaranteed r1bios in case of extreme VM load:
46 */
47#define NR_RAID1_BIOS 256
48
Jonathan Brassow473e87c2012-07-31 10:03:52 +100049/* when we get a read error on a read-only array, we redirect to another
50 * device without failing the first device, or trying to over-write to
51 * correct the read error. To keep track of bad blocks on a per-bio
52 * level, we store IO_BLOCKED in the appropriate 'bios' pointer
53 */
54#define IO_BLOCKED ((struct bio *)1)
55/* When we successfully write to a known bad-block, we need to remove the
56 * bad-block marking which must be done from process context. So we record
57 * the success by setting devs[n].bio to IO_MADE_GOOD
58 */
59#define IO_MADE_GOOD ((struct bio *)2)
60
61#define BIO_SPECIAL(bio) ((unsigned long)bio <= 2)
62
NeilBrown34db0cd2011-10-11 16:50:01 +110063/* When there are this many requests queue to be written by
64 * the raid1 thread, we become 'congested' to provide back-pressure
65 * for writeback.
66 */
67static int max_queued_requests = 1024;
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
NeilBrowne8096362011-10-11 16:49:05 +110069static void allow_barrier(struct r1conf *conf);
70static void lower_barrier(struct r1conf *conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071
Al Virodd0fc662005-10-07 07:46:04 +010072static void * r1bio_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073{
74 struct pool_info *pi = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +110075 int size = offsetof(struct r1bio, bios[pi->raid_disks]);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
77 /* allocate a r1bio with room for raid_disks entries in the bios array */
Jens Axboe7eaceac2011-03-10 08:52:07 +010078 return kzalloc(size, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -070079}
80
81static void r1bio_pool_free(void *r1_bio, void *data)
82{
83 kfree(r1_bio);
84}
85
86#define RESYNC_BLOCK_SIZE (64*1024)
87//#define RESYNC_BLOCK_SIZE PAGE_SIZE
88#define RESYNC_SECTORS (RESYNC_BLOCK_SIZE >> 9)
89#define RESYNC_PAGES ((RESYNC_BLOCK_SIZE + PAGE_SIZE-1) / PAGE_SIZE)
90#define RESYNC_WINDOW (2048*1024)
91
Al Virodd0fc662005-10-07 07:46:04 +010092static void * r1buf_pool_alloc(gfp_t gfp_flags, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -070093{
94 struct pool_info *pi = data;
NeilBrown9f2c9d12011-10-11 16:48:43 +110095 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 struct bio *bio;
97 int i, j;
98
99 r1_bio = r1bio_pool_alloc(gfp_flags, pi);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100100 if (!r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
103 /*
104 * Allocate bios : 1 for reading, n-1 for writing
105 */
106 for (j = pi->raid_disks ; j-- ; ) {
NeilBrown67465572010-10-26 17:33:54 +1100107 bio = bio_kmalloc(gfp_flags, RESYNC_PAGES);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 if (!bio)
109 goto out_free_bio;
110 r1_bio->bios[j] = bio;
111 }
112 /*
113 * Allocate RESYNC_PAGES data pages and attach them to
NeilBrownd11c1712006-01-06 00:20:26 -0800114 * the first bio.
115 * If this is a user-requested check/repair, allocate
116 * RESYNC_PAGES for each bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 */
NeilBrownd11c1712006-01-06 00:20:26 -0800118 if (test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery))
119 j = pi->raid_disks;
120 else
121 j = 1;
122 while(j--) {
123 bio = r1_bio->bios[j];
Kent Overstreeta0787602012-09-10 14:03:28 -0700124 bio->bi_vcnt = RESYNC_PAGES;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
Kent Overstreeta0787602012-09-10 14:03:28 -0700126 if (bio_alloc_pages(bio, gfp_flags))
127 goto out_free_bio;
NeilBrownd11c1712006-01-06 00:20:26 -0800128 }
129 /* If not user-requests, copy the page pointers to all bios */
130 if (!test_bit(MD_RECOVERY_REQUESTED, &pi->mddev->recovery)) {
131 for (i=0; i<RESYNC_PAGES ; i++)
132 for (j=1; j<pi->raid_disks; j++)
133 r1_bio->bios[j]->bi_io_vec[i].bv_page =
134 r1_bio->bios[0]->bi_io_vec[i].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 }
136
137 r1_bio->master_bio = NULL;
138
139 return r1_bio;
140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141out_free_bio:
NeilBrown8f19ccb2011-12-23 10:17:56 +1100142 while (++j < pi->raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 bio_put(r1_bio->bios[j]);
144 r1bio_pool_free(r1_bio, data);
145 return NULL;
146}
147
148static void r1buf_pool_free(void *__r1_bio, void *data)
149{
150 struct pool_info *pi = data;
NeilBrownd11c1712006-01-06 00:20:26 -0800151 int i,j;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100152 struct r1bio *r1bio = __r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153
NeilBrownd11c1712006-01-06 00:20:26 -0800154 for (i = 0; i < RESYNC_PAGES; i++)
155 for (j = pi->raid_disks; j-- ;) {
156 if (j == 0 ||
157 r1bio->bios[j]->bi_io_vec[i].bv_page !=
158 r1bio->bios[0]->bi_io_vec[i].bv_page)
NeilBrown1345b1d2006-01-06 00:20:40 -0800159 safe_put_page(r1bio->bios[j]->bi_io_vec[i].bv_page);
NeilBrownd11c1712006-01-06 00:20:26 -0800160 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 for (i=0 ; i < pi->raid_disks; i++)
162 bio_put(r1bio->bios[i]);
163
164 r1bio_pool_free(r1bio, data);
165}
166
NeilBrowne8096362011-10-11 16:49:05 +1100167static void put_all_bios(struct r1conf *conf, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168{
169 int i;
170
NeilBrown8f19ccb2011-12-23 10:17:56 +1100171 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 struct bio **bio = r1_bio->bios + i;
NeilBrown4367af52011-07-28 11:31:49 +1000173 if (!BIO_SPECIAL(*bio))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 bio_put(*bio);
175 *bio = NULL;
176 }
177}
178
NeilBrown9f2c9d12011-10-11 16:48:43 +1100179static void free_r1bio(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180{
NeilBrowne8096362011-10-11 16:49:05 +1100181 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 put_all_bios(conf, r1_bio);
184 mempool_free(r1_bio, conf->r1bio_pool);
185}
186
NeilBrown9f2c9d12011-10-11 16:48:43 +1100187static void put_buf(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188{
NeilBrowne8096362011-10-11 16:49:05 +1100189 struct r1conf *conf = r1_bio->mddev->private;
NeilBrown3e198f72006-01-06 00:20:21 -0800190 int i;
191
NeilBrown8f19ccb2011-12-23 10:17:56 +1100192 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3e198f72006-01-06 00:20:21 -0800193 struct bio *bio = r1_bio->bios[i];
194 if (bio->bi_end_io)
195 rdev_dec_pending(conf->mirrors[i].rdev, r1_bio->mddev);
196 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
198 mempool_free(r1_bio, conf->r1buf_pool);
199
NeilBrown17999be2006-01-06 00:20:12 -0800200 lower_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201}
202
NeilBrown9f2c9d12011-10-11 16:48:43 +1100203static void reschedule_retry(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
205 unsigned long flags;
NeilBrownfd01b882011-10-11 16:47:53 +1100206 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +1100207 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
209 spin_lock_irqsave(&conf->device_lock, flags);
210 list_add(&r1_bio->retry_list, &conf->retry_list);
NeilBrownddaf22a2006-01-06 00:20:19 -0800211 conf->nr_queued ++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 spin_unlock_irqrestore(&conf->device_lock, flags);
213
NeilBrown17999be2006-01-06 00:20:12 -0800214 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 md_wakeup_thread(mddev->thread);
216}
217
218/*
219 * raid_end_bio_io() is called when we have finished servicing a mirrored
220 * operation and are ready to return a success/failure code to the buffer
221 * cache layer.
222 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100223static void call_bio_endio(struct r1bio *r1_bio)
NeilBrownd2eb35a2011-07-28 11:31:48 +1000224{
225 struct bio *bio = r1_bio->master_bio;
226 int done;
NeilBrowne8096362011-10-11 16:49:05 +1100227 struct r1conf *conf = r1_bio->mddev->private;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000228
229 if (bio->bi_phys_segments) {
230 unsigned long flags;
231 spin_lock_irqsave(&conf->device_lock, flags);
232 bio->bi_phys_segments--;
233 done = (bio->bi_phys_segments == 0);
234 spin_unlock_irqrestore(&conf->device_lock, flags);
235 } else
236 done = 1;
237
238 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
239 clear_bit(BIO_UPTODATE, &bio->bi_flags);
240 if (done) {
241 bio_endio(bio, 0);
242 /*
243 * Wake up any possible resync thread that waits for the device
244 * to go idle.
245 */
246 allow_barrier(conf);
247 }
248}
249
NeilBrown9f2c9d12011-10-11 16:48:43 +1100250static void raid_end_bio_io(struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251{
252 struct bio *bio = r1_bio->master_bio;
253
NeilBrown4b6d2872005-09-09 16:23:47 -0700254 /* if nobody has done the final endio yet, do it now */
255 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
NeilBrown36a4e1f2011-10-07 14:23:17 +1100256 pr_debug("raid1: sync end %s on sectors %llu-%llu\n",
257 (bio_data_dir(bio) == WRITE) ? "write" : "read",
258 (unsigned long long) bio->bi_sector,
259 (unsigned long long) bio->bi_sector +
Kent Overstreetaa8b57a2013-02-05 15:19:29 -0800260 bio_sectors(bio) - 1);
NeilBrown4b6d2872005-09-09 16:23:47 -0700261
NeilBrownd2eb35a2011-07-28 11:31:48 +1000262 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700263 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 free_r1bio(r1_bio);
265}
266
267/*
268 * Update disk head position estimator based on IRQ completion info.
269 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100270static inline void update_head_pos(int disk, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
NeilBrowne8096362011-10-11 16:49:05 +1100272 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273
274 conf->mirrors[disk].head_position =
275 r1_bio->sector + (r1_bio->sectors);
276}
277
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100278/*
279 * Find the disk number which triggered given bio
280 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100281static int find_bio_disk(struct r1bio *r1_bio, struct bio *bio)
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100282{
283 int mirror;
NeilBrown30194632011-12-23 10:17:56 +1100284 struct r1conf *conf = r1_bio->mddev->private;
285 int raid_disks = conf->raid_disks;
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100286
NeilBrown8f19ccb2011-12-23 10:17:56 +1100287 for (mirror = 0; mirror < raid_disks * 2; mirror++)
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100288 if (r1_bio->bios[mirror] == bio)
289 break;
290
NeilBrown8f19ccb2011-12-23 10:17:56 +1100291 BUG_ON(mirror == raid_disks * 2);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100292 update_head_pos(mirror, r1_bio);
293
294 return mirror;
295}
296
NeilBrown6712ecf2007-09-27 12:47:43 +0200297static void raid1_end_read_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
299 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +1100300 struct r1bio *r1_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 int mirror;
NeilBrowne8096362011-10-11 16:49:05 +1100302 struct r1conf *conf = r1_bio->mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304 mirror = r1_bio->read_disk;
305 /*
306 * this branch is our 'one mirror IO has finished' event handler:
307 */
NeilBrownddaf22a2006-01-06 00:20:19 -0800308 update_head_pos(mirror, r1_bio);
309
NeilBrowndd00a992007-05-10 03:15:50 -0700310 if (uptodate)
311 set_bit(R1BIO_Uptodate, &r1_bio->state);
312 else {
313 /* If all other devices have failed, we want to return
314 * the error upwards rather than fail the last device.
315 * Here we redefine "uptodate" to mean "Don't want to retry"
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 */
NeilBrowndd00a992007-05-10 03:15:50 -0700317 unsigned long flags;
318 spin_lock_irqsave(&conf->device_lock, flags);
319 if (r1_bio->mddev->degraded == conf->raid_disks ||
320 (r1_bio->mddev->degraded == conf->raid_disks-1 &&
321 !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags)))
322 uptodate = 1;
323 spin_unlock_irqrestore(&conf->device_lock, flags);
324 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100326 if (uptodate) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 raid_end_bio_io(r1_bio);
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100328 rdev_dec_pending(conf->mirrors[mirror].rdev, conf->mddev);
329 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 /*
331 * oops, read error:
332 */
333 char b[BDEVNAME_SIZE];
Christian Dietrich8bda4702011-07-27 11:00:36 +1000334 printk_ratelimited(
335 KERN_ERR "md/raid1:%s: %s: "
336 "rescheduling sector %llu\n",
337 mdname(conf->mddev),
338 bdevname(conf->mirrors[mirror].rdev->bdev,
339 b),
340 (unsigned long long)r1_bio->sector);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000341 set_bit(R1BIO_ReadError, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 reschedule_retry(r1_bio);
NeilBrown7ad4d4a2012-10-11 13:44:30 +1100343 /* don't drop the reference on read_disk yet */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345}
346
NeilBrown9f2c9d12011-10-11 16:48:43 +1100347static void close_write(struct r1bio *r1_bio)
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000348{
349 /* it really is the end of this request */
350 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
351 /* free extra copy of the data pages */
352 int i = r1_bio->behind_page_count;
353 while (i--)
354 safe_put_page(r1_bio->behind_bvecs[i].bv_page);
355 kfree(r1_bio->behind_bvecs);
356 r1_bio->behind_bvecs = NULL;
357 }
358 /* clear the bitmap if all writes complete successfully */
359 bitmap_endwrite(r1_bio->mddev->bitmap, r1_bio->sector,
360 r1_bio->sectors,
361 !test_bit(R1BIO_Degraded, &r1_bio->state),
362 test_bit(R1BIO_BehindIO, &r1_bio->state));
363 md_write_end(r1_bio->mddev);
364}
365
NeilBrown9f2c9d12011-10-11 16:48:43 +1100366static void r1_bio_write_done(struct r1bio *r1_bio)
NeilBrown4e780642010-10-19 12:54:01 +1100367{
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000368 if (!atomic_dec_and_test(&r1_bio->remaining))
369 return;
370
371 if (test_bit(R1BIO_WriteError, &r1_bio->state))
372 reschedule_retry(r1_bio);
373 else {
374 close_write(r1_bio);
NeilBrown4367af52011-07-28 11:31:49 +1000375 if (test_bit(R1BIO_MadeGood, &r1_bio->state))
376 reschedule_retry(r1_bio);
377 else
378 raid_end_bio_io(r1_bio);
NeilBrown4e780642010-10-19 12:54:01 +1100379 }
380}
381
NeilBrown6712ecf2007-09-27 12:47:43 +0200382static void raid1_end_write_request(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383{
384 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +1100385 struct r1bio *r1_bio = bio->bi_private;
NeilBrowna9701a32005-11-08 21:39:34 -0800386 int mirror, behind = test_bit(R1BIO_BehindIO, &r1_bio->state);
NeilBrowne8096362011-10-11 16:49:05 +1100387 struct r1conf *conf = r1_bio->mddev->private;
NeilBrown04b857f2006-03-09 17:33:46 -0800388 struct bio *to_put = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
Namhyung Kimba3ae3b2011-10-07 14:22:53 +1100390 mirror = find_bio_disk(r1_bio, bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
Tejun Heoe9c74692010-09-03 11:56:18 +0200392 /*
393 * 'one mirror IO has finished' event handler:
394 */
Tejun Heoe9c74692010-09-03 11:56:18 +0200395 if (!uptodate) {
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000396 set_bit(WriteErrorSeen,
397 &conf->mirrors[mirror].rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +1100398 if (!test_and_set_bit(WantReplacement,
399 &conf->mirrors[mirror].rdev->flags))
400 set_bit(MD_RECOVERY_NEEDED, &
401 conf->mddev->recovery);
402
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000403 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +1000404 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405 /*
Tejun Heoe9c74692010-09-03 11:56:18 +0200406 * Set R1BIO_Uptodate in our master bio, so that we
407 * will return a good error code for to the higher
408 * levels even if IO on some other mirrored buffer
409 * fails.
410 *
411 * The 'master' represents the composite IO operation
412 * to user-side. So if something waits for IO, then it
413 * will wait for the 'master' bio.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 */
NeilBrown4367af52011-07-28 11:31:49 +1000415 sector_t first_bad;
416 int bad_sectors;
417
NeilBrowncd5ff9a12011-07-28 11:32:41 +1000418 r1_bio->bios[mirror] = NULL;
419 to_put = bio;
Alex Lyakas3056e3a2013-06-04 20:42:21 +0300420 /*
421 * Do not set R1BIO_Uptodate if the current device is
422 * rebuilding or Faulty. This is because we cannot use
423 * such device for properly reading the data back (we could
424 * potentially use it, if the current write would have felt
425 * before rdev->recovery_offset, but for simplicity we don't
426 * check this here.
427 */
428 if (test_bit(In_sync, &conf->mirrors[mirror].rdev->flags) &&
429 !test_bit(Faulty, &conf->mirrors[mirror].rdev->flags))
430 set_bit(R1BIO_Uptodate, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
NeilBrown4367af52011-07-28 11:31:49 +1000432 /* Maybe we can clear some bad blocks. */
433 if (is_badblock(conf->mirrors[mirror].rdev,
434 r1_bio->sector, r1_bio->sectors,
435 &first_bad, &bad_sectors)) {
436 r1_bio->bios[mirror] = IO_MADE_GOOD;
437 set_bit(R1BIO_MadeGood, &r1_bio->state);
438 }
439 }
440
Tejun Heoe9c74692010-09-03 11:56:18 +0200441 if (behind) {
442 if (test_bit(WriteMostly, &conf->mirrors[mirror].rdev->flags))
443 atomic_dec(&r1_bio->behind_remaining);
NeilBrown4b6d2872005-09-09 16:23:47 -0700444
Tejun Heoe9c74692010-09-03 11:56:18 +0200445 /*
446 * In behind mode, we ACK the master bio once the I/O
447 * has safely reached all non-writemostly
448 * disks. Setting the Returned bit ensures that this
449 * gets done only once -- we don't ever want to return
450 * -EIO here, instead we'll wait
451 */
452 if (atomic_read(&r1_bio->behind_remaining) >= (atomic_read(&r1_bio->remaining)-1) &&
453 test_bit(R1BIO_Uptodate, &r1_bio->state)) {
454 /* Maybe we can return now */
455 if (!test_and_set_bit(R1BIO_Returned, &r1_bio->state)) {
456 struct bio *mbio = r1_bio->master_bio;
NeilBrown36a4e1f2011-10-07 14:23:17 +1100457 pr_debug("raid1: behind end write sectors"
458 " %llu-%llu\n",
459 (unsigned long long) mbio->bi_sector,
460 (unsigned long long) mbio->bi_sector +
Kent Overstreetaa8b57a2013-02-05 15:19:29 -0800461 bio_sectors(mbio) - 1);
NeilBrownd2eb35a2011-07-28 11:31:48 +1000462 call_bio_endio(r1_bio);
NeilBrown4b6d2872005-09-09 16:23:47 -0700463 }
464 }
465 }
NeilBrown4367af52011-07-28 11:31:49 +1000466 if (r1_bio->bios[mirror] == NULL)
467 rdev_dec_pending(conf->mirrors[mirror].rdev,
468 conf->mddev);
Tejun Heoe9c74692010-09-03 11:56:18 +0200469
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471 * Let's see if all mirrored write operations have finished
472 * already.
473 */
NeilBrownaf6d7b72011-05-11 14:51:19 +1000474 r1_bio_write_done(r1_bio);
NeilBrownc70810b2006-06-26 00:27:35 -0700475
NeilBrown04b857f2006-03-09 17:33:46 -0800476 if (to_put)
477 bio_put(to_put);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478}
479
480
481/*
482 * This routine returns the disk from which the requested read should
483 * be done. There is a per-array 'next expected sequential IO' sector
484 * number - if this matches on the next IO then we use the last disk.
485 * There is also a per-disk 'last know head position' sector that is
486 * maintained from IRQ contexts, both the normal and the resync IO
487 * completion handlers update this position correctly. If there is no
488 * perfect sequential match then we pick the disk whose head is closest.
489 *
490 * If there are 2 mirrors in the same 2 devices, performance degrades
491 * because position is mirror, not device based.
492 *
493 * The rdev for the device selected will have nr_pending incremented.
494 */
NeilBrowne8096362011-10-11 16:49:05 +1100495static int read_balance(struct r1conf *conf, struct r1bio *r1_bio, int *max_sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
NeilBrownaf3a2cd2010-05-08 08:20:17 +1000497 const sector_t this_sector = r1_bio->sector;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000498 int sectors;
499 int best_good_sectors;
Shaohua Li9dedf602012-07-31 10:03:53 +1000500 int best_disk, best_dist_disk, best_pending_disk;
501 int has_nonrot_disk;
Shaohua Libe4d3282012-07-31 10:03:53 +1000502 int disk;
NeilBrown76073052011-05-11 14:34:56 +1000503 sector_t best_dist;
Shaohua Li9dedf602012-07-31 10:03:53 +1000504 unsigned int min_pending;
NeilBrown3cb03002011-10-11 16:45:26 +1100505 struct md_rdev *rdev;
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000506 int choose_first;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000507 int choose_next_idle;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
509 rcu_read_lock();
510 /*
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700511 * Check if we can balance. We can balance on the whole
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 * device if no resync is going on, or below the resync window.
513 * We take the first readable disk when above the resync window.
514 */
515 retry:
NeilBrownd2eb35a2011-07-28 11:31:48 +1000516 sectors = r1_bio->sectors;
NeilBrown76073052011-05-11 14:34:56 +1000517 best_disk = -1;
Shaohua Li9dedf602012-07-31 10:03:53 +1000518 best_dist_disk = -1;
NeilBrown76073052011-05-11 14:34:56 +1000519 best_dist = MaxSector;
Shaohua Li9dedf602012-07-31 10:03:53 +1000520 best_pending_disk = -1;
521 min_pending = UINT_MAX;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000522 best_good_sectors = 0;
Shaohua Li9dedf602012-07-31 10:03:53 +1000523 has_nonrot_disk = 0;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000524 choose_next_idle = 0;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 if (conf->mddev->recovery_cp < MaxSector &&
Shaohua Libe4d3282012-07-31 10:03:53 +1000527 (this_sector + sectors >= conf->next_resync))
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000528 choose_first = 1;
Shaohua Libe4d3282012-07-31 10:03:53 +1000529 else
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000530 choose_first = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
Shaohua Libe4d3282012-07-31 10:03:53 +1000532 for (disk = 0 ; disk < conf->raid_disks * 2 ; disk++) {
NeilBrown76073052011-05-11 14:34:56 +1000533 sector_t dist;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000534 sector_t first_bad;
535 int bad_sectors;
Shaohua Li9dedf602012-07-31 10:03:53 +1000536 unsigned int pending;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000537 bool nonrot;
NeilBrownd2eb35a2011-07-28 11:31:48 +1000538
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000539 rdev = rcu_dereference(conf->mirrors[disk].rdev);
540 if (r1_bio->bios[disk] == IO_BLOCKED
541 || rdev == NULL
NeilBrown6b740b82012-03-19 12:46:39 +1100542 || test_bit(Unmerged, &rdev->flags)
NeilBrown76073052011-05-11 14:34:56 +1000543 || test_bit(Faulty, &rdev->flags))
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000544 continue;
NeilBrown76073052011-05-11 14:34:56 +1000545 if (!test_bit(In_sync, &rdev->flags) &&
546 rdev->recovery_offset < this_sector + sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 continue;
NeilBrown76073052011-05-11 14:34:56 +1000548 if (test_bit(WriteMostly, &rdev->flags)) {
549 /* Don't balance among write-mostly, just
550 * use the first as a last resort */
NeilBrown307729c2012-01-09 01:41:51 +1100551 if (best_disk < 0) {
552 if (is_badblock(rdev, this_sector, sectors,
553 &first_bad, &bad_sectors)) {
554 if (first_bad < this_sector)
555 /* Cannot use this */
556 continue;
557 best_good_sectors = first_bad - this_sector;
558 } else
559 best_good_sectors = sectors;
NeilBrown76073052011-05-11 14:34:56 +1000560 best_disk = disk;
NeilBrown307729c2012-01-09 01:41:51 +1100561 }
NeilBrown76073052011-05-11 14:34:56 +1000562 continue;
563 }
564 /* This is a reasonable device to use. It might
565 * even be best.
566 */
NeilBrownd2eb35a2011-07-28 11:31:48 +1000567 if (is_badblock(rdev, this_sector, sectors,
568 &first_bad, &bad_sectors)) {
569 if (best_dist < MaxSector)
570 /* already have a better device */
571 continue;
572 if (first_bad <= this_sector) {
573 /* cannot read here. If this is the 'primary'
574 * device, then we must not read beyond
575 * bad_sectors from another device..
576 */
577 bad_sectors -= (this_sector - first_bad);
578 if (choose_first && sectors > bad_sectors)
579 sectors = bad_sectors;
580 if (best_good_sectors > sectors)
581 best_good_sectors = sectors;
582
583 } else {
584 sector_t good_sectors = first_bad - this_sector;
585 if (good_sectors > best_good_sectors) {
586 best_good_sectors = good_sectors;
587 best_disk = disk;
588 }
589 if (choose_first)
590 break;
591 }
592 continue;
593 } else
594 best_good_sectors = sectors;
595
Shaohua Li12cee5a2012-07-31 10:03:53 +1000596 nonrot = blk_queue_nonrot(bdev_get_queue(rdev->bdev));
597 has_nonrot_disk |= nonrot;
Shaohua Li9dedf602012-07-31 10:03:53 +1000598 pending = atomic_read(&rdev->nr_pending);
NeilBrown76073052011-05-11 14:34:56 +1000599 dist = abs(this_sector - conf->mirrors[disk].head_position);
Shaohua Li12cee5a2012-07-31 10:03:53 +1000600 if (choose_first) {
NeilBrown76073052011-05-11 14:34:56 +1000601 best_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700602 break;
603 }
Shaohua Li12cee5a2012-07-31 10:03:53 +1000604 /* Don't change to another disk for sequential reads */
605 if (conf->mirrors[disk].next_seq_sect == this_sector
606 || dist == 0) {
607 int opt_iosize = bdev_io_opt(rdev->bdev) >> 9;
608 struct raid1_info *mirror = &conf->mirrors[disk];
609
610 best_disk = disk;
611 /*
612 * If buffered sequential IO size exceeds optimal
613 * iosize, check if there is idle disk. If yes, choose
614 * the idle disk. read_balance could already choose an
615 * idle disk before noticing it's a sequential IO in
616 * this disk. This doesn't matter because this disk
617 * will idle, next time it will be utilized after the
618 * first disk has IO size exceeds optimal iosize. In
619 * this way, iosize of the first disk will be optimal
620 * iosize at least. iosize of the second disk might be
621 * small, but not a big deal since when the second disk
622 * starts IO, the first disk is likely still busy.
623 */
624 if (nonrot && opt_iosize > 0 &&
625 mirror->seq_start != MaxSector &&
626 mirror->next_seq_sect > opt_iosize &&
627 mirror->next_seq_sect - opt_iosize >=
628 mirror->seq_start) {
629 choose_next_idle = 1;
630 continue;
631 }
632 break;
633 }
634 /* If device is idle, use it */
635 if (pending == 0) {
636 best_disk = disk;
637 break;
638 }
639
640 if (choose_next_idle)
641 continue;
Shaohua Li9dedf602012-07-31 10:03:53 +1000642
643 if (min_pending > pending) {
644 min_pending = pending;
645 best_pending_disk = disk;
646 }
647
NeilBrown76073052011-05-11 14:34:56 +1000648 if (dist < best_dist) {
649 best_dist = dist;
Shaohua Li9dedf602012-07-31 10:03:53 +1000650 best_dist_disk = disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 }
NeilBrownf3ac8bf2010-09-06 14:10:08 +1000652 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653
Shaohua Li9dedf602012-07-31 10:03:53 +1000654 /*
655 * If all disks are rotational, choose the closest disk. If any disk is
656 * non-rotational, choose the disk with less pending request even the
657 * disk is rotational, which might/might not be optimal for raids with
658 * mixed ratation/non-rotational disks depending on workload.
659 */
660 if (best_disk == -1) {
661 if (has_nonrot_disk)
662 best_disk = best_pending_disk;
663 else
664 best_disk = best_dist_disk;
665 }
666
NeilBrown76073052011-05-11 14:34:56 +1000667 if (best_disk >= 0) {
668 rdev = rcu_dereference(conf->mirrors[best_disk].rdev);
NeilBrown8ddf9ef2005-09-09 16:23:45 -0700669 if (!rdev)
670 goto retry;
671 atomic_inc(&rdev->nr_pending);
NeilBrown76073052011-05-11 14:34:56 +1000672 if (test_bit(Faulty, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 /* cannot risk returning a device that failed
674 * before we inc'ed nr_pending
675 */
NeilBrown03c902e2006-01-06 00:20:46 -0800676 rdev_dec_pending(rdev, conf->mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 goto retry;
678 }
NeilBrownd2eb35a2011-07-28 11:31:48 +1000679 sectors = best_good_sectors;
Shaohua Li12cee5a2012-07-31 10:03:53 +1000680
681 if (conf->mirrors[best_disk].next_seq_sect != this_sector)
682 conf->mirrors[best_disk].seq_start = this_sector;
683
Shaohua Libe4d3282012-07-31 10:03:53 +1000684 conf->mirrors[best_disk].next_seq_sect = this_sector + sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 }
686 rcu_read_unlock();
NeilBrownd2eb35a2011-07-28 11:31:48 +1000687 *max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
NeilBrown76073052011-05-11 14:34:56 +1000689 return best_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690}
691
NeilBrown6b740b82012-03-19 12:46:39 +1100692static int raid1_mergeable_bvec(struct request_queue *q,
693 struct bvec_merge_data *bvm,
694 struct bio_vec *biovec)
695{
696 struct mddev *mddev = q->queuedata;
697 struct r1conf *conf = mddev->private;
698 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
699 int max = biovec->bv_len;
700
701 if (mddev->merge_check_needed) {
702 int disk;
703 rcu_read_lock();
704 for (disk = 0; disk < conf->raid_disks * 2; disk++) {
705 struct md_rdev *rdev = rcu_dereference(
706 conf->mirrors[disk].rdev);
707 if (rdev && !test_bit(Faulty, &rdev->flags)) {
708 struct request_queue *q =
709 bdev_get_queue(rdev->bdev);
710 if (q->merge_bvec_fn) {
711 bvm->bi_sector = sector +
712 rdev->data_offset;
713 bvm->bi_bdev = rdev->bdev;
714 max = min(max, q->merge_bvec_fn(
715 q, bvm, biovec));
716 }
717 }
718 }
719 rcu_read_unlock();
720 }
721 return max;
722
723}
724
NeilBrownfd01b882011-10-11 16:47:53 +1100725int md_raid1_congested(struct mddev *mddev, int bits)
NeilBrown0d129222006-10-03 01:15:54 -0700726{
NeilBrowne8096362011-10-11 16:49:05 +1100727 struct r1conf *conf = mddev->private;
NeilBrown0d129222006-10-03 01:15:54 -0700728 int i, ret = 0;
729
NeilBrown34db0cd2011-10-11 16:50:01 +1100730 if ((bits & (1 << BDI_async_congested)) &&
731 conf->pending_count >= max_queued_requests)
732 return 1;
733
NeilBrown0d129222006-10-03 01:15:54 -0700734 rcu_read_lock();
NeilBrownf53e29f2012-02-13 14:24:05 +1100735 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +1100736 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrown0d129222006-10-03 01:15:54 -0700737 if (rdev && !test_bit(Faulty, &rdev->flags)) {
Jens Axboe165125e2007-07-24 09:28:11 +0200738 struct request_queue *q = bdev_get_queue(rdev->bdev);
NeilBrown0d129222006-10-03 01:15:54 -0700739
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500740 BUG_ON(!q);
741
NeilBrown0d129222006-10-03 01:15:54 -0700742 /* Note the '|| 1' - when read_balance prefers
743 * non-congested targets, it can be removed
744 */
Alexander Beregalov91a9e992009-04-07 01:35:56 +0400745 if ((bits & (1<<BDI_async_congested)) || 1)
NeilBrown0d129222006-10-03 01:15:54 -0700746 ret |= bdi_congested(&q->backing_dev_info, bits);
747 else
748 ret &= bdi_congested(&q->backing_dev_info, bits);
749 }
750 }
751 rcu_read_unlock();
752 return ret;
753}
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500754EXPORT_SYMBOL_GPL(md_raid1_congested);
NeilBrown0d129222006-10-03 01:15:54 -0700755
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500756static int raid1_congested(void *data, int bits)
757{
NeilBrownfd01b882011-10-11 16:47:53 +1100758 struct mddev *mddev = data;
Jonathan Brassow1ed72422011-06-07 17:50:35 -0500759
760 return mddev_congested(mddev, bits) ||
761 md_raid1_congested(mddev, bits);
762}
NeilBrown0d129222006-10-03 01:15:54 -0700763
NeilBrowne8096362011-10-11 16:49:05 +1100764static void flush_pending_writes(struct r1conf *conf)
NeilBrowna35e63e2008-03-04 14:29:29 -0800765{
766 /* Any writes that have been queued but are awaiting
767 * bitmap updates get flushed here.
NeilBrowna35e63e2008-03-04 14:29:29 -0800768 */
NeilBrowna35e63e2008-03-04 14:29:29 -0800769 spin_lock_irq(&conf->device_lock);
770
771 if (conf->pending_bio_list.head) {
772 struct bio *bio;
773 bio = bio_list_get(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +1100774 conf->pending_count = 0;
NeilBrowna35e63e2008-03-04 14:29:29 -0800775 spin_unlock_irq(&conf->device_lock);
776 /* flush any pending bitmap writes to
777 * disk before proceeding w/ I/O */
778 bitmap_unplug(conf->mddev->bitmap);
NeilBrown34db0cd2011-10-11 16:50:01 +1100779 wake_up(&conf->wait_barrier);
NeilBrowna35e63e2008-03-04 14:29:29 -0800780
781 while (bio) { /* submit pending writes */
782 struct bio *next = bio->bi_next;
783 bio->bi_next = NULL;
Shaohua Li2ff8cc22012-10-11 13:28:54 +1100784 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
785 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
786 /* Just ignore it */
787 bio_endio(bio, 0);
788 else
789 generic_make_request(bio);
NeilBrowna35e63e2008-03-04 14:29:29 -0800790 bio = next;
791 }
NeilBrowna35e63e2008-03-04 14:29:29 -0800792 } else
793 spin_unlock_irq(&conf->device_lock);
Jens Axboe7eaceac2011-03-10 08:52:07 +0100794}
795
NeilBrown17999be2006-01-06 00:20:12 -0800796/* Barriers....
797 * Sometimes we need to suspend IO while we do something else,
798 * either some resync/recovery, or reconfigure the array.
799 * To do this we raise a 'barrier'.
800 * The 'barrier' is a counter that can be raised multiple times
801 * to count how many activities are happening which preclude
802 * normal IO.
803 * We can only raise the barrier if there is no pending IO.
804 * i.e. if nr_pending == 0.
805 * We choose only to raise the barrier if no-one is waiting for the
806 * barrier to go down. This means that as soon as an IO request
807 * is ready, no other operations which require a barrier will start
808 * until the IO request has had a chance.
809 *
810 * So: regular IO calls 'wait_barrier'. When that returns there
811 * is no backgroup IO happening, It must arrange to call
812 * allow_barrier when it has finished its IO.
813 * backgroup IO calls must call raise_barrier. Once that returns
814 * there is no normal IO happeing. It must arrange to call
815 * lower_barrier when the particular background IO completes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 */
817#define RESYNC_DEPTH 32
818
NeilBrowne8096362011-10-11 16:49:05 +1100819static void raise_barrier(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820{
821 spin_lock_irq(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800822
823 /* Wait until no block IO is waiting */
824 wait_event_lock_irq(conf->wait_barrier, !conf->nr_waiting,
Lukas Czernereed8c022012-11-30 11:42:40 +0100825 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800826
827 /* block any new IO from starting */
828 conf->barrier++;
829
NeilBrown046abee2010-10-26 15:46:20 +1100830 /* Now wait for all pending IO to complete */
NeilBrown17999be2006-01-06 00:20:12 -0800831 wait_event_lock_irq(conf->wait_barrier,
832 !conf->nr_pending && conf->barrier < RESYNC_DEPTH,
Lukas Czernereed8c022012-11-30 11:42:40 +0100833 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800834
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835 spin_unlock_irq(&conf->resync_lock);
836}
837
NeilBrowne8096362011-10-11 16:49:05 +1100838static void lower_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800839{
840 unsigned long flags;
NeilBrown709ae482009-12-14 12:49:51 +1100841 BUG_ON(conf->barrier <= 0);
NeilBrown17999be2006-01-06 00:20:12 -0800842 spin_lock_irqsave(&conf->resync_lock, flags);
843 conf->barrier--;
844 spin_unlock_irqrestore(&conf->resync_lock, flags);
845 wake_up(&conf->wait_barrier);
846}
847
NeilBrowne8096362011-10-11 16:49:05 +1100848static void wait_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800849{
850 spin_lock_irq(&conf->resync_lock);
851 if (conf->barrier) {
852 conf->nr_waiting++;
NeilBrownd6b42dc2012-03-19 12:46:38 +1100853 /* Wait for the barrier to drop.
854 * However if there are already pending
855 * requests (preventing the barrier from
856 * rising completely), and the
857 * pre-process bio queue isn't empty,
858 * then don't wait, as we need to empty
859 * that queue to get the nr_pending
860 * count down.
861 */
862 wait_event_lock_irq(conf->wait_barrier,
863 !conf->barrier ||
864 (conf->nr_pending &&
865 current->bio_list &&
866 !bio_list_empty(current->bio_list)),
Lukas Czernereed8c022012-11-30 11:42:40 +0100867 conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -0800868 conf->nr_waiting--;
869 }
870 conf->nr_pending++;
871 spin_unlock_irq(&conf->resync_lock);
872}
873
NeilBrowne8096362011-10-11 16:49:05 +1100874static void allow_barrier(struct r1conf *conf)
NeilBrown17999be2006-01-06 00:20:12 -0800875{
876 unsigned long flags;
877 spin_lock_irqsave(&conf->resync_lock, flags);
878 conf->nr_pending--;
879 spin_unlock_irqrestore(&conf->resync_lock, flags);
880 wake_up(&conf->wait_barrier);
881}
882
NeilBrowne2d59922013-06-12 11:01:22 +1000883static void freeze_array(struct r1conf *conf, int extra)
NeilBrownddaf22a2006-01-06 00:20:19 -0800884{
885 /* stop syncio and normal IO and wait for everything to
886 * go quite.
887 * We increment barrier and nr_waiting, and then
NeilBrowne2d59922013-06-12 11:01:22 +1000888 * wait until nr_pending match nr_queued+extra
NeilBrown1c830532008-03-04 14:29:35 -0800889 * This is called in the context of one normal IO request
890 * that has failed. Thus any sync request that might be pending
891 * will be blocked by nr_pending, and we need to wait for
892 * pending IO requests to complete or be queued for re-try.
NeilBrowne2d59922013-06-12 11:01:22 +1000893 * Thus the number queued (nr_queued) plus this request (extra)
NeilBrown1c830532008-03-04 14:29:35 -0800894 * must match the number of pending IOs (nr_pending) before
895 * we continue.
NeilBrownddaf22a2006-01-06 00:20:19 -0800896 */
897 spin_lock_irq(&conf->resync_lock);
898 conf->barrier++;
899 conf->nr_waiting++;
Lukas Czernereed8c022012-11-30 11:42:40 +0100900 wait_event_lock_irq_cmd(conf->wait_barrier,
NeilBrowne2d59922013-06-12 11:01:22 +1000901 conf->nr_pending == conf->nr_queued+extra,
Lukas Czernereed8c022012-11-30 11:42:40 +0100902 conf->resync_lock,
903 flush_pending_writes(conf));
NeilBrownddaf22a2006-01-06 00:20:19 -0800904 spin_unlock_irq(&conf->resync_lock);
905}
NeilBrowne8096362011-10-11 16:49:05 +1100906static void unfreeze_array(struct r1conf *conf)
NeilBrownddaf22a2006-01-06 00:20:19 -0800907{
908 /* reverse the effect of the freeze */
909 spin_lock_irq(&conf->resync_lock);
910 conf->barrier--;
911 conf->nr_waiting--;
912 wake_up(&conf->wait_barrier);
913 spin_unlock_irq(&conf->resync_lock);
914}
915
NeilBrown17999be2006-01-06 00:20:12 -0800916
NeilBrown4e780642010-10-19 12:54:01 +1100917/* duplicate the data pages for behind I/O
NeilBrown4e780642010-10-19 12:54:01 +1100918 */
NeilBrown9f2c9d12011-10-11 16:48:43 +1100919static void alloc_behind_pages(struct bio *bio, struct r1bio *r1_bio)
NeilBrown4b6d2872005-09-09 16:23:47 -0700920{
921 int i;
922 struct bio_vec *bvec;
NeilBrown2ca68f52011-07-28 11:32:10 +1000923 struct bio_vec *bvecs = kzalloc(bio->bi_vcnt * sizeof(struct bio_vec),
NeilBrown4b6d2872005-09-09 16:23:47 -0700924 GFP_NOIO);
NeilBrown2ca68f52011-07-28 11:32:10 +1000925 if (unlikely(!bvecs))
NeilBrownaf6d7b72011-05-11 14:51:19 +1000926 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700927
Kent Overstreetcb34e052012-09-05 15:22:02 -0700928 bio_for_each_segment_all(bvec, bio, i) {
NeilBrown2ca68f52011-07-28 11:32:10 +1000929 bvecs[i] = *bvec;
930 bvecs[i].bv_page = alloc_page(GFP_NOIO);
931 if (unlikely(!bvecs[i].bv_page))
NeilBrown4b6d2872005-09-09 16:23:47 -0700932 goto do_sync_io;
NeilBrown2ca68f52011-07-28 11:32:10 +1000933 memcpy(kmap(bvecs[i].bv_page) + bvec->bv_offset,
934 kmap(bvec->bv_page) + bvec->bv_offset, bvec->bv_len);
935 kunmap(bvecs[i].bv_page);
NeilBrown4b6d2872005-09-09 16:23:47 -0700936 kunmap(bvec->bv_page);
937 }
NeilBrown2ca68f52011-07-28 11:32:10 +1000938 r1_bio->behind_bvecs = bvecs;
NeilBrownaf6d7b72011-05-11 14:51:19 +1000939 r1_bio->behind_page_count = bio->bi_vcnt;
940 set_bit(R1BIO_BehindIO, &r1_bio->state);
941 return;
NeilBrown4b6d2872005-09-09 16:23:47 -0700942
943do_sync_io:
NeilBrownaf6d7b72011-05-11 14:51:19 +1000944 for (i = 0; i < bio->bi_vcnt; i++)
NeilBrown2ca68f52011-07-28 11:32:10 +1000945 if (bvecs[i].bv_page)
946 put_page(bvecs[i].bv_page);
947 kfree(bvecs);
NeilBrown36a4e1f2011-10-07 14:23:17 +1100948 pr_debug("%dB behind alloc failed, doing sync I/O\n", bio->bi_size);
NeilBrown4b6d2872005-09-09 16:23:47 -0700949}
950
NeilBrownf54a9d02012-08-02 08:33:20 +1000951struct raid1_plug_cb {
952 struct blk_plug_cb cb;
953 struct bio_list pending;
954 int pending_cnt;
955};
956
957static void raid1_unplug(struct blk_plug_cb *cb, bool from_schedule)
958{
959 struct raid1_plug_cb *plug = container_of(cb, struct raid1_plug_cb,
960 cb);
961 struct mddev *mddev = plug->cb.data;
962 struct r1conf *conf = mddev->private;
963 struct bio *bio;
964
NeilBrown874807a2012-11-27 12:14:40 +1100965 if (from_schedule || current->bio_list) {
NeilBrownf54a9d02012-08-02 08:33:20 +1000966 spin_lock_irq(&conf->device_lock);
967 bio_list_merge(&conf->pending_bio_list, &plug->pending);
968 conf->pending_count += plug->pending_cnt;
969 spin_unlock_irq(&conf->device_lock);
NeilBrownee0b0242013-02-25 12:38:29 +1100970 wake_up(&conf->wait_barrier);
NeilBrownf54a9d02012-08-02 08:33:20 +1000971 md_wakeup_thread(mddev->thread);
972 kfree(plug);
973 return;
974 }
975
976 /* we aren't scheduling, so we can do the write-out directly. */
977 bio = bio_list_get(&plug->pending);
978 bitmap_unplug(mddev->bitmap);
979 wake_up(&conf->wait_barrier);
980
981 while (bio) { /* submit pending writes */
982 struct bio *next = bio->bi_next;
983 bio->bi_next = NULL;
Shaohua Li32f9f572013-04-28 18:26:38 +0800984 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
985 !blk_queue_discard(bdev_get_queue(bio->bi_bdev))))
986 /* Just ignore it */
987 bio_endio(bio, 0);
988 else
989 generic_make_request(bio);
NeilBrownf54a9d02012-08-02 08:33:20 +1000990 bio = next;
991 }
992 kfree(plug);
993}
994
Linus Torvaldsb4fdcb02011-11-04 17:06:58 -0700995static void make_request(struct mddev *mddev, struct bio * bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996{
NeilBrowne8096362011-10-11 16:49:05 +1100997 struct r1conf *conf = mddev->private;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +1000998 struct raid1_info *mirror;
NeilBrown9f2c9d12011-10-11 16:48:43 +1100999 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 struct bio *read_bio;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001001 int i, disks;
NeilBrown84255d12008-05-23 13:04:32 -07001002 struct bitmap *bitmap;
NeilBrown191ea9b2005-06-21 17:17:23 -07001003 unsigned long flags;
Jens Axboea3623572005-11-01 09:26:16 +01001004 const int rw = bio_data_dir(bio);
NeilBrown2c7d46e2010-08-18 16:16:05 +10001005 const unsigned long do_sync = (bio->bi_rw & REQ_SYNC);
Tejun Heoe9c74692010-09-03 11:56:18 +02001006 const unsigned long do_flush_fua = (bio->bi_rw & (REQ_FLUSH | REQ_FUA));
Shaohua Li2ff8cc22012-10-11 13:28:54 +11001007 const unsigned long do_discard = (bio->bi_rw
1008 & (REQ_DISCARD | REQ_SECURE));
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11001009 const unsigned long do_same = (bio->bi_rw & REQ_WRITE_SAME);
NeilBrown3cb03002011-10-11 16:45:26 +11001010 struct md_rdev *blocked_rdev;
NeilBrownf54a9d02012-08-02 08:33:20 +10001011 struct blk_plug_cb *cb;
1012 struct raid1_plug_cb *plug = NULL;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001013 int first_clone;
1014 int sectors_handled;
1015 int max_sectors;
NeilBrown191ea9b2005-06-21 17:17:23 -07001016
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 /*
1018 * Register the new request and wait if the reconstruction
1019 * thread has put up a bar for new requests.
1020 * Continue immediately if no resync is active currently.
1021 */
NeilBrown62de6082006-05-01 12:15:47 -07001022
NeilBrown3d310eb2005-06-21 17:17:26 -07001023 md_write_start(mddev, bio); /* wait on superblock update early */
1024
NeilBrown6eef4b22009-12-14 12:49:51 +11001025 if (bio_data_dir(bio) == WRITE &&
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001026 bio_end_sector(bio) > mddev->suspend_lo &&
NeilBrown6eef4b22009-12-14 12:49:51 +11001027 bio->bi_sector < mddev->suspend_hi) {
1028 /* As the suspend_* range is controlled by
1029 * userspace, we want an interruptible
1030 * wait.
1031 */
1032 DEFINE_WAIT(w);
1033 for (;;) {
1034 flush_signals(current);
1035 prepare_to_wait(&conf->wait_barrier,
1036 &w, TASK_INTERRUPTIBLE);
Kent Overstreetf73a1c72012-09-25 15:05:12 -07001037 if (bio_end_sector(bio) <= mddev->suspend_lo ||
NeilBrown6eef4b22009-12-14 12:49:51 +11001038 bio->bi_sector >= mddev->suspend_hi)
1039 break;
1040 schedule();
1041 }
1042 finish_wait(&conf->wait_barrier, &w);
1043 }
NeilBrown62de6082006-05-01 12:15:47 -07001044
NeilBrown17999be2006-01-06 00:20:12 -08001045 wait_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
NeilBrown84255d12008-05-23 13:04:32 -07001047 bitmap = mddev->bitmap;
1048
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 /*
1050 * make_request() can abort the operation when READA is being
1051 * used and no empty request is available.
1052 *
1053 */
1054 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1055
1056 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001057 r1_bio->sectors = bio_sectors(bio);
NeilBrown191ea9b2005-06-21 17:17:23 -07001058 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 r1_bio->mddev = mddev;
1060 r1_bio->sector = bio->bi_sector;
1061
NeilBrownd2eb35a2011-07-28 11:31:48 +10001062 /* We might need to issue multiple reads to different
1063 * devices if there are bad blocks around, so we keep
1064 * track of the number of reads in bio->bi_phys_segments.
1065 * If this is 0, there is only one r1_bio and no locking
1066 * will be needed when requests complete. If it is
1067 * non-zero, then it is the number of not-completed requests.
1068 */
1069 bio->bi_phys_segments = 0;
1070 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
1071
Jens Axboea3623572005-11-01 09:26:16 +01001072 if (rw == READ) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 /*
1074 * read balancing logic:
1075 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10001076 int rdisk;
1077
1078read_again:
1079 rdisk = read_balance(conf, r1_bio, &max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 if (rdisk < 0) {
1082 /* couldn't find anywhere to read from */
1083 raid_end_bio_io(r1_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001084 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
1086 mirror = conf->mirrors + rdisk;
1087
NeilBrowne5551902010-03-31 11:21:44 +11001088 if (test_bit(WriteMostly, &mirror->rdev->flags) &&
1089 bitmap) {
1090 /* Reading from a write-mostly device must
1091 * take care not to over-take any writes
1092 * that are 'behind'
1093 */
1094 wait_event(bitmap->behind_wait,
1095 atomic_read(&bitmap->behind_writes) == 0);
1096 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 r1_bio->read_disk = rdisk;
1098
NeilBrowna167f662010-10-26 18:31:13 +11001099 read_bio = bio_clone_mddev(bio, GFP_NOIO, mddev);
NeilBrownd2eb35a2011-07-28 11:31:48 +10001100 md_trim_bio(read_bio, r1_bio->sector - bio->bi_sector,
1101 max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
1103 r1_bio->bios[rdisk] = read_bio;
1104
1105 read_bio->bi_sector = r1_bio->sector + mirror->rdev->data_offset;
1106 read_bio->bi_bdev = mirror->rdev->bdev;
1107 read_bio->bi_end_io = raid1_end_read_request;
Christoph Hellwig7b6d91d2010-08-07 18:20:39 +02001108 read_bio->bi_rw = READ | do_sync;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 read_bio->bi_private = r1_bio;
1110
NeilBrownd2eb35a2011-07-28 11:31:48 +10001111 if (max_sectors < r1_bio->sectors) {
1112 /* could not read all from this device, so we will
1113 * need another r1_bio.
1114 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10001115
1116 sectors_handled = (r1_bio->sector + max_sectors
1117 - bio->bi_sector);
1118 r1_bio->sectors = max_sectors;
1119 spin_lock_irq(&conf->device_lock);
1120 if (bio->bi_phys_segments == 0)
1121 bio->bi_phys_segments = 2;
1122 else
1123 bio->bi_phys_segments++;
1124 spin_unlock_irq(&conf->device_lock);
1125 /* Cannot call generic_make_request directly
1126 * as that will be queued in __make_request
1127 * and subsequent mempool_alloc might block waiting
1128 * for it. So hand bio over to raid1d.
1129 */
1130 reschedule_retry(r1_bio);
1131
1132 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1133
1134 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001135 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrownd2eb35a2011-07-28 11:31:48 +10001136 r1_bio->state = 0;
1137 r1_bio->mddev = mddev;
1138 r1_bio->sector = bio->bi_sector + sectors_handled;
1139 goto read_again;
1140 } else
1141 generic_make_request(read_bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +02001142 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 }
1144
1145 /*
1146 * WRITE:
1147 */
NeilBrown34db0cd2011-10-11 16:50:01 +11001148 if (conf->pending_count >= max_queued_requests) {
1149 md_wakeup_thread(mddev->thread);
1150 wait_event(conf->wait_barrier,
1151 conf->pending_count < max_queued_requests);
1152 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001153 /* first select target devices under rcu_lock and
Linus Torvalds1da177e2005-04-16 15:20:36 -07001154 * inc refcount on their rdev. Record them by setting
1155 * bios[x] to bio
NeilBrown1f68f0c2011-07-28 11:31:48 +10001156 * If there are known/acknowledged bad blocks on any device on
1157 * which we have seen a write error, we want to avoid writing those
1158 * blocks.
1159 * This potentially requires several writes to write around
1160 * the bad blocks. Each set of writes gets it's own r1bio
1161 * with a set of bios attached.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 */
NeilBrownc3b328a2011-04-18 18:25:43 +10001163
NeilBrown8f19ccb2011-12-23 10:17:56 +11001164 disks = conf->raid_disks * 2;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001165 retry_write:
1166 blocked_rdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167 rcu_read_lock();
NeilBrown1f68f0c2011-07-28 11:31:48 +10001168 max_sectors = r1_bio->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 for (i = 0; i < disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001170 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Dan Williams6bfe0b42008-04-30 00:52:32 -07001171 if (rdev && unlikely(test_bit(Blocked, &rdev->flags))) {
1172 atomic_inc(&rdev->nr_pending);
1173 blocked_rdev = rdev;
1174 break;
1175 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001176 r1_bio->bios[i] = NULL;
NeilBrown6b740b82012-03-19 12:46:39 +11001177 if (!rdev || test_bit(Faulty, &rdev->flags)
1178 || test_bit(Unmerged, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11001179 if (i < conf->raid_disks)
1180 set_bit(R1BIO_Degraded, &r1_bio->state);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001181 continue;
1182 }
1183
1184 atomic_inc(&rdev->nr_pending);
1185 if (test_bit(WriteErrorSeen, &rdev->flags)) {
1186 sector_t first_bad;
1187 int bad_sectors;
1188 int is_bad;
1189
1190 is_bad = is_badblock(rdev, r1_bio->sector,
1191 max_sectors,
1192 &first_bad, &bad_sectors);
1193 if (is_bad < 0) {
1194 /* mustn't write here until the bad block is
1195 * acknowledged*/
1196 set_bit(BlockedBadBlocks, &rdev->flags);
1197 blocked_rdev = rdev;
1198 break;
NeilBrown964147d2010-05-18 15:27:13 +10001199 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001200 if (is_bad && first_bad <= r1_bio->sector) {
1201 /* Cannot write here at all */
1202 bad_sectors -= (r1_bio->sector - first_bad);
1203 if (bad_sectors < max_sectors)
1204 /* mustn't write more than bad_sectors
1205 * to other devices yet
1206 */
1207 max_sectors = bad_sectors;
1208 rdev_dec_pending(rdev, mddev);
1209 /* We don't set R1BIO_Degraded as that
1210 * only applies if the disk is
1211 * missing, so it might be re-added,
1212 * and we want to know to recover this
1213 * chunk.
1214 * In this case the device is here,
1215 * and the fact that this chunk is not
1216 * in-sync is recorded in the bad
1217 * block log
1218 */
1219 continue;
1220 }
1221 if (is_bad) {
1222 int good_sectors = first_bad - r1_bio->sector;
1223 if (good_sectors < max_sectors)
1224 max_sectors = good_sectors;
1225 }
1226 }
1227 r1_bio->bios[i] = bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 }
1229 rcu_read_unlock();
1230
Dan Williams6bfe0b42008-04-30 00:52:32 -07001231 if (unlikely(blocked_rdev)) {
1232 /* Wait for this device to become unblocked */
1233 int j;
1234
1235 for (j = 0; j < i; j++)
1236 if (r1_bio->bios[j])
1237 rdev_dec_pending(conf->mirrors[j].rdev, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001238 r1_bio->state = 0;
Dan Williams6bfe0b42008-04-30 00:52:32 -07001239 allow_barrier(conf);
1240 md_wait_for_blocked_rdev(blocked_rdev, mddev);
1241 wait_barrier(conf);
1242 goto retry_write;
1243 }
1244
NeilBrown1f68f0c2011-07-28 11:31:48 +10001245 if (max_sectors < r1_bio->sectors) {
1246 /* We are splitting this write into multiple parts, so
1247 * we need to prepare for allocating another r1_bio.
1248 */
1249 r1_bio->sectors = max_sectors;
1250 spin_lock_irq(&conf->device_lock);
1251 if (bio->bi_phys_segments == 0)
1252 bio->bi_phys_segments = 2;
1253 else
1254 bio->bi_phys_segments++;
1255 spin_unlock_irq(&conf->device_lock);
NeilBrown191ea9b2005-06-21 17:17:23 -07001256 }
NeilBrown1f68f0c2011-07-28 11:31:48 +10001257 sectors_handled = r1_bio->sector + max_sectors - bio->bi_sector;
NeilBrown4b6d2872005-09-09 16:23:47 -07001258
NeilBrown4e780642010-10-19 12:54:01 +11001259 atomic_set(&r1_bio->remaining, 1);
NeilBrown4b6d2872005-09-09 16:23:47 -07001260 atomic_set(&r1_bio->behind_remaining, 0);
NeilBrown191ea9b2005-06-21 17:17:23 -07001261
NeilBrown1f68f0c2011-07-28 11:31:48 +10001262 first_clone = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 for (i = 0; i < disks; i++) {
1264 struct bio *mbio;
1265 if (!r1_bio->bios[i])
1266 continue;
1267
NeilBrowna167f662010-10-26 18:31:13 +11001268 mbio = bio_clone_mddev(bio, GFP_NOIO, mddev);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001269 md_trim_bio(mbio, r1_bio->sector - bio->bi_sector, max_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
NeilBrown1f68f0c2011-07-28 11:31:48 +10001271 if (first_clone) {
1272 /* do behind I/O ?
1273 * Not if there are too many, or cannot
1274 * allocate memory, or a reader on WriteMostly
1275 * is waiting for behind writes to flush */
1276 if (bitmap &&
1277 (atomic_read(&bitmap->behind_writes)
1278 < mddev->bitmap_info.max_write_behind) &&
1279 !waitqueue_active(&bitmap->behind_wait))
1280 alloc_behind_pages(mbio, r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281
NeilBrown1f68f0c2011-07-28 11:31:48 +10001282 bitmap_startwrite(bitmap, r1_bio->sector,
1283 r1_bio->sectors,
1284 test_bit(R1BIO_BehindIO,
1285 &r1_bio->state));
1286 first_clone = 0;
1287 }
NeilBrown2ca68f52011-07-28 11:32:10 +10001288 if (r1_bio->behind_bvecs) {
NeilBrown4b6d2872005-09-09 16:23:47 -07001289 struct bio_vec *bvec;
1290 int j;
1291
Kent Overstreetcb34e052012-09-05 15:22:02 -07001292 /*
1293 * We trimmed the bio, so _all is legit
NeilBrown4b6d2872005-09-09 16:23:47 -07001294 */
Kent Overstreetd74c6d52013-02-06 12:23:11 -08001295 bio_for_each_segment_all(bvec, mbio, j)
NeilBrown2ca68f52011-07-28 11:32:10 +10001296 bvec->bv_page = r1_bio->behind_bvecs[j].bv_page;
NeilBrown4b6d2872005-09-09 16:23:47 -07001297 if (test_bit(WriteMostly, &conf->mirrors[i].rdev->flags))
1298 atomic_inc(&r1_bio->behind_remaining);
1299 }
1300
NeilBrown1f68f0c2011-07-28 11:31:48 +10001301 r1_bio->bios[i] = mbio;
1302
1303 mbio->bi_sector = (r1_bio->sector +
1304 conf->mirrors[i].rdev->data_offset);
1305 mbio->bi_bdev = conf->mirrors[i].rdev->bdev;
1306 mbio->bi_end_io = raid1_end_write_request;
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11001307 mbio->bi_rw =
1308 WRITE | do_flush_fua | do_sync | do_discard | do_same;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001309 mbio->bi_private = r1_bio;
1310
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 atomic_inc(&r1_bio->remaining);
NeilBrownf54a9d02012-08-02 08:33:20 +10001312
1313 cb = blk_check_plugged(raid1_unplug, mddev, sizeof(*plug));
1314 if (cb)
1315 plug = container_of(cb, struct raid1_plug_cb, cb);
1316 else
1317 plug = NULL;
NeilBrown4e780642010-10-19 12:54:01 +11001318 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrownf54a9d02012-08-02 08:33:20 +10001319 if (plug) {
1320 bio_list_add(&plug->pending, mbio);
1321 plug->pending_cnt++;
1322 } else {
1323 bio_list_add(&conf->pending_bio_list, mbio);
1324 conf->pending_count++;
1325 }
NeilBrown4e780642010-10-19 12:54:01 +11001326 spin_unlock_irqrestore(&conf->device_lock, flags);
NeilBrownf54a9d02012-08-02 08:33:20 +10001327 if (!plug)
NeilBrownb357f042012-07-03 17:45:31 +10001328 md_wakeup_thread(mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329 }
NeilBrown079fa162011-09-10 17:21:23 +10001330 /* Mustn't call r1_bio_write_done before this next test,
1331 * as it could result in the bio being freed.
1332 */
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001333 if (sectors_handled < bio_sectors(bio)) {
NeilBrown079fa162011-09-10 17:21:23 +10001334 r1_bio_write_done(r1_bio);
NeilBrown1f68f0c2011-07-28 11:31:48 +10001335 /* We need another r1_bio. It has already been counted
1336 * in bio->bi_phys_segments
1337 */
1338 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
1339 r1_bio->master_bio = bio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001340 r1_bio->sectors = bio_sectors(bio) - sectors_handled;
NeilBrown1f68f0c2011-07-28 11:31:48 +10001341 r1_bio->state = 0;
1342 r1_bio->mddev = mddev;
1343 r1_bio->sector = bio->bi_sector + sectors_handled;
1344 goto retry_write;
1345 }
1346
NeilBrown079fa162011-09-10 17:21:23 +10001347 r1_bio_write_done(r1_bio);
1348
1349 /* In case raid1d snuck in to freeze_array */
1350 wake_up(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351}
1352
NeilBrownfd01b882011-10-11 16:47:53 +11001353static void status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354{
NeilBrowne8096362011-10-11 16:49:05 +11001355 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 int i;
1357
1358 seq_printf(seq, " [%d/%d] [", conf->raid_disks,
NeilBrown11ce99e2006-10-03 01:15:52 -07001359 conf->raid_disks - mddev->degraded);
NeilBrownddac7c72006-08-31 21:27:36 -07001360 rcu_read_lock();
1361 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001362 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 seq_printf(seq, "%s",
NeilBrownddac7c72006-08-31 21:27:36 -07001364 rdev && test_bit(In_sync, &rdev->flags) ? "U" : "_");
1365 }
1366 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 seq_printf(seq, "]");
1368}
1369
1370
NeilBrownfd01b882011-10-11 16:47:53 +11001371static void error(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372{
1373 char b[BDEVNAME_SIZE];
NeilBrowne8096362011-10-11 16:49:05 +11001374 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375
1376 /*
1377 * If it is not operational, then we have already marked it as dead
1378 * else if it is the last working disks, ignore the error, let the
1379 * next level up know.
1380 * else mark the drive as failed
1381 */
NeilBrownb2d444d2005-11-08 21:39:31 -08001382 if (test_bit(In_sync, &rdev->flags)
NeilBrown4044ba52009-01-09 08:31:11 +11001383 && (conf->raid_disks - mddev->degraded) == 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 /*
1385 * Don't fail the drive, act as though we were just a
NeilBrown4044ba52009-01-09 08:31:11 +11001386 * normal single drive.
1387 * However don't try a recovery from this drive as
1388 * it is very likely to fail.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 */
NeilBrown53890422011-07-27 11:00:36 +10001390 conf->recovery_disabled = mddev->recovery_disabled;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 return;
NeilBrown4044ba52009-01-09 08:31:11 +11001392 }
NeilBrownde393cd2011-07-28 11:31:48 +10001393 set_bit(Blocked, &rdev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001394 if (test_and_clear_bit(In_sync, &rdev->flags)) {
1395 unsigned long flags;
1396 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 mddev->degraded++;
NeilBrowndd00a992007-05-10 03:15:50 -07001398 set_bit(Faulty, &rdev->flags);
NeilBrownc04be0a2006-10-03 01:15:53 -07001399 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 /*
1401 * if recovery is running, make sure it aborts.
1402 */
NeilBrowndfc70642008-05-23 13:04:39 -07001403 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
NeilBrowndd00a992007-05-10 03:15:50 -07001404 } else
1405 set_bit(Faulty, &rdev->flags);
NeilBrown850b2b42006-10-03 01:15:46 -07001406 set_bit(MD_CHANGE_DEVS, &mddev->flags);
Joe Perches067032b2011-01-14 09:14:33 +11001407 printk(KERN_ALERT
1408 "md/raid1:%s: Disk failure on %s, disabling device.\n"
1409 "md/raid1:%s: Operation continuing on %d devices.\n",
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001410 mdname(mddev), bdevname(rdev->bdev, b),
1411 mdname(mddev), conf->raid_disks - mddev->degraded);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412}
1413
NeilBrowne8096362011-10-11 16:49:05 +11001414static void print_conf(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415{
1416 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001418 printk(KERN_DEBUG "RAID1 conf printout:\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 if (!conf) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001420 printk(KERN_DEBUG "(!conf)\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 return;
1422 }
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001423 printk(KERN_DEBUG " --- wd:%d rd:%d\n", conf->raid_disks - conf->mddev->degraded,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 conf->raid_disks);
1425
NeilBrownddac7c72006-08-31 21:27:36 -07001426 rcu_read_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427 for (i = 0; i < conf->raid_disks; i++) {
1428 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11001429 struct md_rdev *rdev = rcu_dereference(conf->mirrors[i].rdev);
NeilBrownddac7c72006-08-31 21:27:36 -07001430 if (rdev)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10001431 printk(KERN_DEBUG " disk %d, wo:%d, o:%d, dev:%s\n",
NeilBrownddac7c72006-08-31 21:27:36 -07001432 i, !test_bit(In_sync, &rdev->flags),
1433 !test_bit(Faulty, &rdev->flags),
1434 bdevname(rdev->bdev,b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 }
NeilBrownddac7c72006-08-31 21:27:36 -07001436 rcu_read_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437}
1438
NeilBrowne8096362011-10-11 16:49:05 +11001439static void close_sync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440{
NeilBrown17999be2006-01-06 00:20:12 -08001441 wait_barrier(conf);
1442 allow_barrier(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443
1444 mempool_destroy(conf->r1buf_pool);
1445 conf->r1buf_pool = NULL;
1446}
1447
NeilBrownfd01b882011-10-11 16:47:53 +11001448static int raid1_spare_active(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449{
1450 int i;
NeilBrowne8096362011-10-11 16:49:05 +11001451 struct r1conf *conf = mddev->private;
NeilBrown6b965622010-08-18 11:56:59 +10001452 int count = 0;
1453 unsigned long flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454
1455 /*
1456 * Find all failed disks within the RAID1 configuration
NeilBrownddac7c72006-08-31 21:27:36 -07001457 * and mark them readable.
1458 * Called under mddev lock, so rcu protection not needed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 */
1460 for (i = 0; i < conf->raid_disks; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11001461 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown8c7a2c22011-12-23 10:17:57 +11001462 struct md_rdev *repl = conf->mirrors[conf->raid_disks + i].rdev;
1463 if (repl
1464 && repl->recovery_offset == MaxSector
1465 && !test_bit(Faulty, &repl->flags)
1466 && !test_and_set_bit(In_sync, &repl->flags)) {
1467 /* replacement has just become active */
1468 if (!rdev ||
1469 !test_and_clear_bit(In_sync, &rdev->flags))
1470 count++;
1471 if (rdev) {
1472 /* Replaced device not technically
1473 * faulty, but we need to be sure
1474 * it gets removed and never re-added
1475 */
1476 set_bit(Faulty, &rdev->flags);
1477 sysfs_notify_dirent_safe(
1478 rdev->sysfs_state);
1479 }
1480 }
NeilBrownddac7c72006-08-31 21:27:36 -07001481 if (rdev
Lukasz Dorau61e49472013-10-24 12:55:17 +11001482 && rdev->recovery_offset == MaxSector
NeilBrownddac7c72006-08-31 21:27:36 -07001483 && !test_bit(Faulty, &rdev->flags)
NeilBrownc04be0a2006-10-03 01:15:53 -07001484 && !test_and_set_bit(In_sync, &rdev->flags)) {
NeilBrown6b965622010-08-18 11:56:59 +10001485 count++;
Jonathan Brassow654e8b52011-07-27 11:00:36 +10001486 sysfs_notify_dirent_safe(rdev->sysfs_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487 }
1488 }
NeilBrown6b965622010-08-18 11:56:59 +10001489 spin_lock_irqsave(&conf->device_lock, flags);
1490 mddev->degraded -= count;
1491 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492
1493 print_conf(conf);
NeilBrown6b965622010-08-18 11:56:59 +10001494 return count;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495}
1496
1497
NeilBrownfd01b882011-10-11 16:47:53 +11001498static int raid1_add_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499{
NeilBrowne8096362011-10-11 16:49:05 +11001500 struct r1conf *conf = mddev->private;
Neil Brown199050e2008-06-28 08:31:33 +10001501 int err = -EEXIST;
NeilBrown41158c72005-06-21 17:17:25 -07001502 int mirror = 0;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001503 struct raid1_info *p;
Neil Brown6c2fce22008-06-28 08:31:31 +10001504 int first = 0;
NeilBrown30194632011-12-23 10:17:56 +11001505 int last = conf->raid_disks - 1;
NeilBrown6b740b82012-03-19 12:46:39 +11001506 struct request_queue *q = bdev_get_queue(rdev->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507
NeilBrown53890422011-07-27 11:00:36 +10001508 if (mddev->recovery_disabled == conf->recovery_disabled)
1509 return -EBUSY;
1510
Neil Brown6c2fce22008-06-28 08:31:31 +10001511 if (rdev->raid_disk >= 0)
1512 first = last = rdev->raid_disk;
1513
NeilBrown6b740b82012-03-19 12:46:39 +11001514 if (q->merge_bvec_fn) {
1515 set_bit(Unmerged, &rdev->flags);
1516 mddev->merge_check_needed = 1;
1517 }
1518
NeilBrown7ef449d2011-12-23 10:17:57 +11001519 for (mirror = first; mirror <= last; mirror++) {
1520 p = conf->mirrors+mirror;
1521 if (!p->rdev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522
Jonathan Brassow9092c022013-05-02 14:19:24 -05001523 if (mddev->gendisk)
1524 disk_stack_limits(mddev->gendisk, rdev->bdev,
1525 rdev->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526
1527 p->head_position = 0;
1528 rdev->raid_disk = mirror;
Neil Brown199050e2008-06-28 08:31:33 +10001529 err = 0;
NeilBrown6aea114a2005-11-28 13:44:13 -08001530 /* As all devices are equivalent, we don't need a full recovery
1531 * if this was recently any drive of the array
1532 */
1533 if (rdev->saved_raid_disk < 0)
NeilBrown41158c72005-06-21 17:17:25 -07001534 conf->fullsync = 1;
Suzanne Woodd6065f72005-11-08 21:39:27 -08001535 rcu_assign_pointer(p->rdev, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 break;
1537 }
NeilBrown7ef449d2011-12-23 10:17:57 +11001538 if (test_bit(WantReplacement, &p->rdev->flags) &&
1539 p[conf->raid_disks].rdev == NULL) {
1540 /* Add this device as a replacement */
1541 clear_bit(In_sync, &rdev->flags);
1542 set_bit(Replacement, &rdev->flags);
1543 rdev->raid_disk = mirror;
1544 err = 0;
1545 conf->fullsync = 1;
1546 rcu_assign_pointer(p[conf->raid_disks].rdev, rdev);
1547 break;
1548 }
1549 }
NeilBrown6b740b82012-03-19 12:46:39 +11001550 if (err == 0 && test_bit(Unmerged, &rdev->flags)) {
1551 /* Some requests might not have seen this new
1552 * merge_bvec_fn. We must wait for them to complete
1553 * before merging the device fully.
1554 * First we make sure any code which has tested
1555 * our function has submitted the request, then
1556 * we wait for all outstanding requests to complete.
1557 */
1558 synchronize_sched();
NeilBrowne2d59922013-06-12 11:01:22 +10001559 freeze_array(conf, 0);
1560 unfreeze_array(conf);
NeilBrown6b740b82012-03-19 12:46:39 +11001561 clear_bit(Unmerged, &rdev->flags);
1562 }
Andre Nollac5e7112009-08-03 10:59:47 +10001563 md_integrity_add_rdev(rdev, mddev);
Jonathan Brassow9092c022013-05-02 14:19:24 -05001564 if (mddev->queue && blk_queue_discard(bdev_get_queue(rdev->bdev)))
Shaohua Li2ff8cc22012-10-11 13:28:54 +11001565 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 print_conf(conf);
Neil Brown199050e2008-06-28 08:31:33 +10001567 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568}
1569
NeilBrownb8321b62011-12-23 10:17:51 +11001570static int raid1_remove_disk(struct mddev *mddev, struct md_rdev *rdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571{
NeilBrowne8096362011-10-11 16:49:05 +11001572 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 int err = 0;
NeilBrownb8321b62011-12-23 10:17:51 +11001574 int number = rdev->raid_disk;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10001575 struct raid1_info *p = conf->mirrors + number;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576
NeilBrownb014f142011-12-23 10:17:56 +11001577 if (rdev != p->rdev)
1578 p = conf->mirrors + conf->raid_disks + number;
1579
Linus Torvalds1da177e2005-04-16 15:20:36 -07001580 print_conf(conf);
NeilBrownb8321b62011-12-23 10:17:51 +11001581 if (rdev == p->rdev) {
NeilBrownb2d444d2005-11-08 21:39:31 -08001582 if (test_bit(In_sync, &rdev->flags) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583 atomic_read(&rdev->nr_pending)) {
1584 err = -EBUSY;
1585 goto abort;
1586 }
NeilBrown046abee2010-10-26 15:46:20 +11001587 /* Only remove non-faulty devices if recovery
NeilBrowndfc70642008-05-23 13:04:39 -07001588 * is not possible.
1589 */
1590 if (!test_bit(Faulty, &rdev->flags) &&
NeilBrown53890422011-07-27 11:00:36 +10001591 mddev->recovery_disabled != conf->recovery_disabled &&
NeilBrowndfc70642008-05-23 13:04:39 -07001592 mddev->degraded < conf->raid_disks) {
1593 err = -EBUSY;
1594 goto abort;
1595 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 p->rdev = NULL;
Paul E. McKenneyfbd568a3e2005-05-01 08:59:04 -07001597 synchronize_rcu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 if (atomic_read(&rdev->nr_pending)) {
1599 /* lost the race, try later */
1600 err = -EBUSY;
1601 p->rdev = rdev;
Andre Nollac5e7112009-08-03 10:59:47 +10001602 goto abort;
NeilBrown8c7a2c22011-12-23 10:17:57 +11001603 } else if (conf->mirrors[conf->raid_disks + number].rdev) {
1604 /* We just removed a device that is being replaced.
1605 * Move down the replacement. We drain all IO before
1606 * doing this to avoid confusion.
1607 */
1608 struct md_rdev *repl =
1609 conf->mirrors[conf->raid_disks + number].rdev;
NeilBrowne2d59922013-06-12 11:01:22 +10001610 freeze_array(conf, 0);
NeilBrown8c7a2c22011-12-23 10:17:57 +11001611 clear_bit(Replacement, &repl->flags);
1612 p->rdev = repl;
1613 conf->mirrors[conf->raid_disks + number].rdev = NULL;
NeilBrowne2d59922013-06-12 11:01:22 +10001614 unfreeze_array(conf);
NeilBrownb014f142011-12-23 10:17:56 +11001615 clear_bit(WantReplacement, &rdev->flags);
NeilBrown8c7a2c22011-12-23 10:17:57 +11001616 } else
1617 clear_bit(WantReplacement, &rdev->flags);
Martin K. Petersena91a2782011-03-17 11:11:05 +01001618 err = md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 }
1620abort:
1621
1622 print_conf(conf);
1623 return err;
1624}
1625
1626
NeilBrown6712ecf2007-09-27 12:47:43 +02001627static void end_sync_read(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628{
NeilBrown9f2c9d12011-10-11 16:48:43 +11001629 struct r1bio *r1_bio = bio->bi_private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
NeilBrown0fc280f2011-10-07 14:22:55 +11001631 update_head_pos(r1_bio->read_disk, r1_bio);
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001632
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633 /*
1634 * we have read a block, now it needs to be re-written,
1635 * or re-read if the read failed.
1636 * We don't do much here, just schedule handling by raid1d
1637 */
NeilBrown69382e82006-01-06 00:20:22 -08001638 if (test_bit(BIO_UPTODATE, &bio->bi_flags))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrownd11c1712006-01-06 00:20:26 -08001640
1641 if (atomic_dec_and_test(&r1_bio->remaining))
1642 reschedule_retry(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001643}
1644
NeilBrown6712ecf2007-09-27 12:47:43 +02001645static void end_sync_write(struct bio *bio, int error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646{
1647 int uptodate = test_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrown9f2c9d12011-10-11 16:48:43 +11001648 struct r1bio *r1_bio = bio->bi_private;
NeilBrownfd01b882011-10-11 16:47:53 +11001649 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001650 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651 int mirror=0;
NeilBrown4367af52011-07-28 11:31:49 +10001652 sector_t first_bad;
1653 int bad_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654
Namhyung Kimba3ae3b2011-10-07 14:22:53 +11001655 mirror = find_bio_disk(r1_bio, bio);
1656
NeilBrown6b1117d2006-03-31 02:31:57 -08001657 if (!uptodate) {
NeilBrown57dab0b2010-10-19 10:03:39 +11001658 sector_t sync_blocks = 0;
NeilBrown6b1117d2006-03-31 02:31:57 -08001659 sector_t s = r1_bio->sector;
1660 long sectors_to_go = r1_bio->sectors;
1661 /* make sure these bits doesn't get cleared. */
1662 do {
NeilBrown5e3db642006-07-10 04:44:18 -07001663 bitmap_end_sync(mddev->bitmap, s,
NeilBrown6b1117d2006-03-31 02:31:57 -08001664 &sync_blocks, 1);
1665 s += sync_blocks;
1666 sectors_to_go -= sync_blocks;
1667 } while (sectors_to_go > 0);
NeilBrownd8f05d22011-07-28 11:33:00 +10001668 set_bit(WriteErrorSeen,
1669 &conf->mirrors[mirror].rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +11001670 if (!test_and_set_bit(WantReplacement,
1671 &conf->mirrors[mirror].rdev->flags))
1672 set_bit(MD_RECOVERY_NEEDED, &
1673 mddev->recovery);
NeilBrownd8f05d22011-07-28 11:33:00 +10001674 set_bit(R1BIO_WriteError, &r1_bio->state);
NeilBrown4367af52011-07-28 11:31:49 +10001675 } else if (is_badblock(conf->mirrors[mirror].rdev,
1676 r1_bio->sector,
1677 r1_bio->sectors,
NeilBrown3a9f28a2011-07-28 11:33:42 +10001678 &first_bad, &bad_sectors) &&
1679 !is_badblock(conf->mirrors[r1_bio->read_disk].rdev,
1680 r1_bio->sector,
1681 r1_bio->sectors,
1682 &first_bad, &bad_sectors)
1683 )
NeilBrown4367af52011-07-28 11:31:49 +10001684 set_bit(R1BIO_MadeGood, &r1_bio->state);
NeilBrowne3b97032005-08-04 12:53:34 -07001685
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown4367af52011-07-28 11:31:49 +10001687 int s = r1_bio->sectors;
NeilBrownd8f05d22011-07-28 11:33:00 +10001688 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1689 test_bit(R1BIO_WriteError, &r1_bio->state))
NeilBrown4367af52011-07-28 11:31:49 +10001690 reschedule_retry(r1_bio);
1691 else {
1692 put_buf(r1_bio);
1693 md_done_sync(mddev, s, uptodate);
1694 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696}
1697
NeilBrown3cb03002011-10-11 16:45:26 +11001698static int r1_sync_page_io(struct md_rdev *rdev, sector_t sector,
NeilBrownd8f05d22011-07-28 11:33:00 +10001699 int sectors, struct page *page, int rw)
1700{
1701 if (sync_page_io(rdev, sector, sectors << 9, page, rw, false))
1702 /* success */
1703 return 1;
NeilBrown19d67162011-12-23 10:17:57 +11001704 if (rw == WRITE) {
NeilBrownd8f05d22011-07-28 11:33:00 +10001705 set_bit(WriteErrorSeen, &rdev->flags);
NeilBrown19d67162011-12-23 10:17:57 +11001706 if (!test_and_set_bit(WantReplacement,
1707 &rdev->flags))
1708 set_bit(MD_RECOVERY_NEEDED, &
1709 rdev->mddev->recovery);
1710 }
NeilBrownd8f05d22011-07-28 11:33:00 +10001711 /* need to record an error - either for the block or the device */
1712 if (!rdev_set_badblocks(rdev, sector, sectors, 0))
1713 md_error(rdev->mddev, rdev);
1714 return 0;
1715}
1716
NeilBrown9f2c9d12011-10-11 16:48:43 +11001717static int fix_sync_read_error(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001718{
1719 /* Try some synchronous reads of other devices to get
1720 * good data, much like with normal read errors. Only
1721 * read into the pages we already have so we don't
1722 * need to re-issue the read request.
1723 * We don't need to freeze the array, because being in an
1724 * active sync request, there is no normal IO, and
1725 * no overlapping syncs.
NeilBrown06f60382011-07-28 11:31:48 +10001726 * We don't need to check is_badblock() again as we
1727 * made sure that anything with a bad block in range
1728 * will have bi_end_io clear.
NeilBrowna68e5872011-05-11 14:40:44 +10001729 */
NeilBrownfd01b882011-10-11 16:47:53 +11001730 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001731 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001732 struct bio *bio = r1_bio->bios[r1_bio->read_disk];
1733 sector_t sect = r1_bio->sector;
1734 int sectors = r1_bio->sectors;
1735 int idx = 0;
1736
1737 while(sectors) {
1738 int s = sectors;
1739 int d = r1_bio->read_disk;
1740 int success = 0;
NeilBrown3cb03002011-10-11 16:45:26 +11001741 struct md_rdev *rdev;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001742 int start;
NeilBrowna68e5872011-05-11 14:40:44 +10001743
1744 if (s > (PAGE_SIZE>>9))
1745 s = PAGE_SIZE >> 9;
1746 do {
1747 if (r1_bio->bios[d]->bi_end_io == end_sync_read) {
1748 /* No rcu protection needed here devices
1749 * can only be removed when no resync is
1750 * active, and resync is currently active
1751 */
1752 rdev = conf->mirrors[d].rdev;
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001753 if (sync_page_io(rdev, sect, s<<9,
NeilBrowna68e5872011-05-11 14:40:44 +10001754 bio->bi_io_vec[idx].bv_page,
1755 READ, false)) {
1756 success = 1;
1757 break;
1758 }
1759 }
1760 d++;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001761 if (d == conf->raid_disks * 2)
NeilBrowna68e5872011-05-11 14:40:44 +10001762 d = 0;
1763 } while (!success && d != r1_bio->read_disk);
1764
NeilBrown78d7f5f2011-05-11 14:48:56 +10001765 if (!success) {
NeilBrowna68e5872011-05-11 14:40:44 +10001766 char b[BDEVNAME_SIZE];
NeilBrown3a9f28a2011-07-28 11:33:42 +10001767 int abort = 0;
1768 /* Cannot read from anywhere, this block is lost.
1769 * Record a bad block on each device. If that doesn't
1770 * work just disable and interrupt the recovery.
1771 * Don't fail devices as that won't really help.
1772 */
NeilBrowna68e5872011-05-11 14:40:44 +10001773 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O read error"
1774 " for block %llu\n",
1775 mdname(mddev),
1776 bdevname(bio->bi_bdev, b),
1777 (unsigned long long)r1_bio->sector);
NeilBrown8f19ccb2011-12-23 10:17:56 +11001778 for (d = 0; d < conf->raid_disks * 2; d++) {
NeilBrown3a9f28a2011-07-28 11:33:42 +10001779 rdev = conf->mirrors[d].rdev;
1780 if (!rdev || test_bit(Faulty, &rdev->flags))
1781 continue;
1782 if (!rdev_set_badblocks(rdev, sect, s, 0))
1783 abort = 1;
1784 }
1785 if (abort) {
NeilBrownd890fa22011-10-26 11:54:39 +11001786 conf->recovery_disabled =
1787 mddev->recovery_disabled;
NeilBrown3a9f28a2011-07-28 11:33:42 +10001788 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
1789 md_done_sync(mddev, r1_bio->sectors, 0);
1790 put_buf(r1_bio);
1791 return 0;
1792 }
1793 /* Try next page */
1794 sectors -= s;
1795 sect += s;
1796 idx++;
1797 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001798 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001799
1800 start = d;
1801 /* write it back and re-read */
1802 while (d != r1_bio->read_disk) {
1803 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001804 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001805 d--;
1806 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1807 continue;
1808 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001809 if (r1_sync_page_io(rdev, sect, s,
1810 bio->bi_io_vec[idx].bv_page,
1811 WRITE) == 0) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001812 r1_bio->bios[d]->bi_end_io = NULL;
1813 rdev_dec_pending(rdev, mddev);
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001814 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001815 }
1816 d = start;
1817 while (d != r1_bio->read_disk) {
1818 if (d == 0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11001819 d = conf->raid_disks * 2;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001820 d--;
1821 if (r1_bio->bios[d]->bi_end_io != end_sync_read)
1822 continue;
1823 rdev = conf->mirrors[d].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10001824 if (r1_sync_page_io(rdev, sect, s,
1825 bio->bi_io_vec[idx].bv_page,
1826 READ) != 0)
Namhyung Kim9d3d8012011-07-27 11:00:36 +10001827 atomic_add(s, &rdev->corrected_errors);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001828 }
NeilBrowna68e5872011-05-11 14:40:44 +10001829 sectors -= s;
1830 sect += s;
1831 idx ++;
1832 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001833 set_bit(R1BIO_Uptodate, &r1_bio->state);
NeilBrown7ca78d52011-05-11 14:50:37 +10001834 set_bit(BIO_UPTODATE, &bio->bi_flags);
NeilBrowna68e5872011-05-11 14:40:44 +10001835 return 1;
1836}
1837
NeilBrown9f2c9d12011-10-11 16:48:43 +11001838static int process_checks(struct r1bio *r1_bio)
NeilBrowna68e5872011-05-11 14:40:44 +10001839{
1840 /* We have read all readable devices. If we haven't
1841 * got the block, then there is no hope left.
1842 * If we have, then we want to do a comparison
1843 * and skip the write if everything is the same.
1844 * If any blocks failed to read, then we need to
1845 * attempt an over-write
1846 */
NeilBrownfd01b882011-10-11 16:47:53 +11001847 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11001848 struct r1conf *conf = mddev->private;
NeilBrowna68e5872011-05-11 14:40:44 +10001849 int primary;
1850 int i;
majianpengf4380a92012-04-12 16:04:47 +10001851 int vcnt;
NeilBrowna68e5872011-05-11 14:40:44 +10001852
NeilBrown30bc9b52013-07-17 15:19:29 +10001853 /* Fix variable parts of all bios */
1854 vcnt = (r1_bio->sectors + PAGE_SIZE / 512 - 1) >> (PAGE_SHIFT - 9);
1855 for (i = 0; i < conf->raid_disks * 2; i++) {
1856 int j;
1857 int size;
1858 struct bio *b = r1_bio->bios[i];
1859 if (b->bi_end_io != end_sync_read)
1860 continue;
1861 /* fixup the bio for reuse */
1862 bio_reset(b);
1863 b->bi_vcnt = vcnt;
1864 b->bi_size = r1_bio->sectors << 9;
1865 b->bi_sector = r1_bio->sector +
1866 conf->mirrors[i].rdev->data_offset;
1867 b->bi_bdev = conf->mirrors[i].rdev->bdev;
1868 b->bi_end_io = end_sync_read;
1869 b->bi_private = r1_bio;
1870
1871 size = b->bi_size;
1872 for (j = 0; j < vcnt ; j++) {
1873 struct bio_vec *bi;
1874 bi = &b->bi_io_vec[j];
1875 bi->bv_offset = 0;
1876 if (size > PAGE_SIZE)
1877 bi->bv_len = PAGE_SIZE;
1878 else
1879 bi->bv_len = size;
1880 size -= PAGE_SIZE;
1881 }
1882 }
NeilBrown8f19ccb2011-12-23 10:17:56 +11001883 for (primary = 0; primary < conf->raid_disks * 2; primary++)
NeilBrowna68e5872011-05-11 14:40:44 +10001884 if (r1_bio->bios[primary]->bi_end_io == end_sync_read &&
1885 test_bit(BIO_UPTODATE, &r1_bio->bios[primary]->bi_flags)) {
1886 r1_bio->bios[primary]->bi_end_io = NULL;
1887 rdev_dec_pending(conf->mirrors[primary].rdev, mddev);
1888 break;
1889 }
1890 r1_bio->read_disk = primary;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001891 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown78d7f5f2011-05-11 14:48:56 +10001892 int j;
NeilBrown78d7f5f2011-05-11 14:48:56 +10001893 struct bio *pbio = r1_bio->bios[primary];
1894 struct bio *sbio = r1_bio->bios[i];
NeilBrowna68e5872011-05-11 14:40:44 +10001895
Kent Overstreet2aabaa62012-09-11 11:26:12 -07001896 if (sbio->bi_end_io != end_sync_read)
NeilBrown78d7f5f2011-05-11 14:48:56 +10001897 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001898
NeilBrown78d7f5f2011-05-11 14:48:56 +10001899 if (test_bit(BIO_UPTODATE, &sbio->bi_flags)) {
1900 for (j = vcnt; j-- ; ) {
1901 struct page *p, *s;
1902 p = pbio->bi_io_vec[j].bv_page;
1903 s = sbio->bi_io_vec[j].bv_page;
1904 if (memcmp(page_address(p),
1905 page_address(s),
NeilBrown5020ad72012-04-02 01:39:05 +10001906 sbio->bi_io_vec[j].bv_len))
NeilBrown78d7f5f2011-05-11 14:48:56 +10001907 break;
NeilBrowna68e5872011-05-11 14:40:44 +10001908 }
NeilBrown78d7f5f2011-05-11 14:48:56 +10001909 } else
1910 j = 0;
1911 if (j >= 0)
Jianpeng Ma7f7583d2012-10-11 14:17:59 +11001912 atomic64_add(r1_bio->sectors, &mddev->resync_mismatches);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001913 if (j < 0 || (test_bit(MD_RECOVERY_CHECK, &mddev->recovery)
1914 && test_bit(BIO_UPTODATE, &sbio->bi_flags))) {
1915 /* No need to write to this device. */
1916 sbio->bi_end_io = NULL;
1917 rdev_dec_pending(conf->mirrors[i].rdev, mddev);
1918 continue;
NeilBrowna68e5872011-05-11 14:40:44 +10001919 }
Kent Overstreetd3b45c22012-09-10 13:49:33 -07001920
1921 bio_copy_data(sbio, pbio);
NeilBrown78d7f5f2011-05-11 14:48:56 +10001922 }
NeilBrowna68e5872011-05-11 14:40:44 +10001923 return 0;
1924}
1925
NeilBrown9f2c9d12011-10-11 16:48:43 +11001926static void sync_request_write(struct mddev *mddev, struct r1bio *r1_bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927{
NeilBrowne8096362011-10-11 16:49:05 +11001928 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 int i;
NeilBrown8f19ccb2011-12-23 10:17:56 +11001930 int disks = conf->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 struct bio *bio, *wbio;
1932
1933 bio = r1_bio->bios[r1_bio->read_disk];
1934
NeilBrowna68e5872011-05-11 14:40:44 +10001935 if (!test_bit(R1BIO_Uptodate, &r1_bio->state))
1936 /* ouch - failed to read all of that. */
1937 if (!fix_sync_read_error(r1_bio))
1938 return;
NeilBrown7ca78d52011-05-11 14:50:37 +10001939
1940 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
1941 if (process_checks(r1_bio) < 0)
1942 return;
NeilBrownd11c1712006-01-06 00:20:26 -08001943 /*
1944 * schedule writes
1945 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 atomic_set(&r1_bio->remaining, 1);
1947 for (i = 0; i < disks ; i++) {
1948 wbio = r1_bio->bios[i];
NeilBrown3e198f72006-01-06 00:20:21 -08001949 if (wbio->bi_end_io == NULL ||
1950 (wbio->bi_end_io == end_sync_read &&
1951 (i == r1_bio->read_disk ||
1952 !test_bit(MD_RECOVERY_SYNC, &mddev->recovery))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953 continue;
1954
NeilBrown3e198f72006-01-06 00:20:21 -08001955 wbio->bi_rw = WRITE;
1956 wbio->bi_end_io = end_sync_write;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 atomic_inc(&r1_bio->remaining);
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08001958 md_sync_acct(conf->mirrors[i].rdev->bdev, bio_sectors(wbio));
NeilBrown191ea9b2005-06-21 17:17:23 -07001959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 generic_make_request(wbio);
1961 }
1962
1963 if (atomic_dec_and_test(&r1_bio->remaining)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07001964 /* if we're here, all write(s) have completed, so clean up */
NeilBrown58e94ae2012-07-19 15:59:18 +10001965 int s = r1_bio->sectors;
1966 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
1967 test_bit(R1BIO_WriteError, &r1_bio->state))
1968 reschedule_retry(r1_bio);
1969 else {
1970 put_buf(r1_bio);
1971 md_done_sync(mddev, s, 1);
1972 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 }
1974}
1975
1976/*
1977 * This is a kernel thread which:
1978 *
1979 * 1. Retries failed read operations on working mirrors.
1980 * 2. Updates the raid superblock when problems encounter.
NeilBrownd2eb35a2011-07-28 11:31:48 +10001981 * 3. Performs writes following reads for array synchronising.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982 */
1983
NeilBrowne8096362011-10-11 16:49:05 +11001984static void fix_read_error(struct r1conf *conf, int read_disk,
NeilBrown867868f2006-10-03 01:15:51 -07001985 sector_t sect, int sectors)
1986{
NeilBrownfd01b882011-10-11 16:47:53 +11001987 struct mddev *mddev = conf->mddev;
NeilBrown867868f2006-10-03 01:15:51 -07001988 while(sectors) {
1989 int s = sectors;
1990 int d = read_disk;
1991 int success = 0;
1992 int start;
NeilBrown3cb03002011-10-11 16:45:26 +11001993 struct md_rdev *rdev;
NeilBrown867868f2006-10-03 01:15:51 -07001994
1995 if (s > (PAGE_SIZE>>9))
1996 s = PAGE_SIZE >> 9;
1997
1998 do {
1999 /* Note: no rcu protection needed here
2000 * as this is synchronous in the raid1d thread
2001 * which is the thread that might remove
2002 * a device. If raid1d ever becomes multi-threaded....
2003 */
NeilBrownd2eb35a2011-07-28 11:31:48 +10002004 sector_t first_bad;
2005 int bad_sectors;
2006
NeilBrown867868f2006-10-03 01:15:51 -07002007 rdev = conf->mirrors[d].rdev;
2008 if (rdev &&
majianpengda8840a2012-05-22 13:55:03 +10002009 (test_bit(In_sync, &rdev->flags) ||
2010 (!test_bit(Faulty, &rdev->flags) &&
2011 rdev->recovery_offset >= sect + s)) &&
NeilBrownd2eb35a2011-07-28 11:31:48 +10002012 is_badblock(rdev, sect, s,
2013 &first_bad, &bad_sectors) == 0 &&
Jonathan Brassowccebd4c2011-01-14 09:14:33 +11002014 sync_page_io(rdev, sect, s<<9,
2015 conf->tmppage, READ, false))
NeilBrown867868f2006-10-03 01:15:51 -07002016 success = 1;
2017 else {
2018 d++;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002019 if (d == conf->raid_disks * 2)
NeilBrown867868f2006-10-03 01:15:51 -07002020 d = 0;
2021 }
2022 } while (!success && d != read_disk);
2023
2024 if (!success) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002025 /* Cannot read from anywhere - mark it bad */
NeilBrown3cb03002011-10-11 16:45:26 +11002026 struct md_rdev *rdev = conf->mirrors[read_disk].rdev;
NeilBrownd8f05d22011-07-28 11:33:00 +10002027 if (!rdev_set_badblocks(rdev, sect, s, 0))
2028 md_error(mddev, rdev);
NeilBrown867868f2006-10-03 01:15:51 -07002029 break;
2030 }
2031 /* write it back and re-read */
2032 start = d;
2033 while (d != read_disk) {
2034 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002035 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07002036 d--;
2037 rdev = conf->mirrors[d].rdev;
2038 if (rdev &&
NeilBrownd8f05d22011-07-28 11:33:00 +10002039 test_bit(In_sync, &rdev->flags))
2040 r1_sync_page_io(rdev, sect, s,
2041 conf->tmppage, WRITE);
NeilBrown867868f2006-10-03 01:15:51 -07002042 }
2043 d = start;
2044 while (d != read_disk) {
2045 char b[BDEVNAME_SIZE];
2046 if (d==0)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002047 d = conf->raid_disks * 2;
NeilBrown867868f2006-10-03 01:15:51 -07002048 d--;
2049 rdev = conf->mirrors[d].rdev;
2050 if (rdev &&
2051 test_bit(In_sync, &rdev->flags)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002052 if (r1_sync_page_io(rdev, sect, s,
2053 conf->tmppage, READ)) {
NeilBrown867868f2006-10-03 01:15:51 -07002054 atomic_add(s, &rdev->corrected_errors);
2055 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002056 "md/raid1:%s: read error corrected "
NeilBrown867868f2006-10-03 01:15:51 -07002057 "(%d sectors at %llu on %s)\n",
2058 mdname(mddev), s,
Randy Dunlap969b7552006-10-28 10:38:32 -07002059 (unsigned long long)(sect +
2060 rdev->data_offset),
NeilBrown867868f2006-10-03 01:15:51 -07002061 bdevname(rdev->bdev, b));
2062 }
2063 }
2064 }
2065 sectors -= s;
2066 sect += s;
2067 }
2068}
2069
NeilBrown9f2c9d12011-10-11 16:48:43 +11002070static int narrow_write_error(struct r1bio *r1_bio, int i)
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002071{
NeilBrownfd01b882011-10-11 16:47:53 +11002072 struct mddev *mddev = r1_bio->mddev;
NeilBrowne8096362011-10-11 16:49:05 +11002073 struct r1conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +11002074 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002075
2076 /* bio has the data to be written to device 'i' where
2077 * we just recently had a write error.
2078 * We repeatedly clone the bio and trim down to one block,
2079 * then try the write. Where the write fails we record
2080 * a bad block.
2081 * It is conceivable that the bio doesn't exactly align with
2082 * blocks. We must handle this somehow.
2083 *
2084 * We currently own a reference on the rdev.
2085 */
2086
2087 int block_sectors;
2088 sector_t sector;
2089 int sectors;
2090 int sect_to_write = r1_bio->sectors;
2091 int ok = 1;
2092
2093 if (rdev->badblocks.shift < 0)
2094 return 0;
2095
2096 block_sectors = 1 << rdev->badblocks.shift;
2097 sector = r1_bio->sector;
2098 sectors = ((sector + block_sectors)
2099 & ~(sector_t)(block_sectors - 1))
2100 - sector;
2101
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002102 while (sect_to_write) {
2103 struct bio *wbio;
2104 if (sectors > sect_to_write)
2105 sectors = sect_to_write;
2106 /* Write at 'sector' for 'sectors'*/
2107
Kent Overstreetb7838632012-09-10 15:17:11 -07002108 if (test_bit(R1BIO_BehindIO, &r1_bio->state)) {
2109 unsigned vcnt = r1_bio->behind_page_count;
2110 struct bio_vec *vec = r1_bio->behind_bvecs;
2111
2112 while (!vec->bv_page) {
2113 vec++;
2114 vcnt--;
2115 }
2116
2117 wbio = bio_alloc_mddev(GFP_NOIO, vcnt, mddev);
2118 memcpy(wbio->bi_io_vec, vec, vcnt * sizeof(struct bio_vec));
2119
2120 wbio->bi_vcnt = vcnt;
2121 } else {
2122 wbio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2123 }
2124
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002125 wbio->bi_rw = WRITE;
Kent Overstreetb7838632012-09-10 15:17:11 -07002126 wbio->bi_sector = r1_bio->sector;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002127 wbio->bi_size = r1_bio->sectors << 9;
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002128
2129 md_trim_bio(wbio, sector - r1_bio->sector, sectors);
2130 wbio->bi_sector += rdev->data_offset;
2131 wbio->bi_bdev = rdev->bdev;
2132 if (submit_bio_wait(WRITE, wbio) == 0)
2133 /* failure! */
2134 ok = rdev_set_badblocks(rdev, sector,
2135 sectors, 0)
2136 && ok;
2137
2138 bio_put(wbio);
2139 sect_to_write -= sectors;
2140 sector += sectors;
2141 sectors = block_sectors;
2142 }
2143 return ok;
2144}
2145
NeilBrowne8096362011-10-11 16:49:05 +11002146static void handle_sync_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002147{
2148 int m;
2149 int s = r1_bio->sectors;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002150 for (m = 0; m < conf->raid_disks * 2 ; m++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002151 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002152 struct bio *bio = r1_bio->bios[m];
2153 if (bio->bi_end_io == NULL)
2154 continue;
2155 if (test_bit(BIO_UPTODATE, &bio->bi_flags) &&
2156 test_bit(R1BIO_MadeGood, &r1_bio->state)) {
NeilBrownc6563a82012-05-21 09:27:00 +10002157 rdev_clear_badblocks(rdev, r1_bio->sector, s, 0);
NeilBrown62096bc2011-07-28 11:38:13 +10002158 }
2159 if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
2160 test_bit(R1BIO_WriteError, &r1_bio->state)) {
2161 if (!rdev_set_badblocks(rdev, r1_bio->sector, s, 0))
2162 md_error(conf->mddev, rdev);
2163 }
2164 }
2165 put_buf(r1_bio);
2166 md_done_sync(conf->mddev, s, 1);
2167}
2168
NeilBrowne8096362011-10-11 16:49:05 +11002169static void handle_write_finished(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002170{
2171 int m;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002172 for (m = 0; m < conf->raid_disks * 2 ; m++)
NeilBrown62096bc2011-07-28 11:38:13 +10002173 if (r1_bio->bios[m] == IO_MADE_GOOD) {
NeilBrown3cb03002011-10-11 16:45:26 +11002174 struct md_rdev *rdev = conf->mirrors[m].rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002175 rdev_clear_badblocks(rdev,
2176 r1_bio->sector,
NeilBrownc6563a82012-05-21 09:27:00 +10002177 r1_bio->sectors, 0);
NeilBrown62096bc2011-07-28 11:38:13 +10002178 rdev_dec_pending(rdev, conf->mddev);
2179 } else if (r1_bio->bios[m] != NULL) {
2180 /* This drive got a write error. We need to
2181 * narrow down and record precise write
2182 * errors.
2183 */
2184 if (!narrow_write_error(r1_bio, m)) {
2185 md_error(conf->mddev,
2186 conf->mirrors[m].rdev);
2187 /* an I/O failed, we can't clear the bitmap */
2188 set_bit(R1BIO_Degraded, &r1_bio->state);
2189 }
2190 rdev_dec_pending(conf->mirrors[m].rdev,
2191 conf->mddev);
2192 }
2193 if (test_bit(R1BIO_WriteError, &r1_bio->state))
2194 close_write(r1_bio);
2195 raid_end_bio_io(r1_bio);
2196}
2197
NeilBrowne8096362011-10-11 16:49:05 +11002198static void handle_read_error(struct r1conf *conf, struct r1bio *r1_bio)
NeilBrown62096bc2011-07-28 11:38:13 +10002199{
2200 int disk;
2201 int max_sectors;
NeilBrownfd01b882011-10-11 16:47:53 +11002202 struct mddev *mddev = conf->mddev;
NeilBrown62096bc2011-07-28 11:38:13 +10002203 struct bio *bio;
2204 char b[BDEVNAME_SIZE];
NeilBrown3cb03002011-10-11 16:45:26 +11002205 struct md_rdev *rdev;
NeilBrown62096bc2011-07-28 11:38:13 +10002206
2207 clear_bit(R1BIO_ReadError, &r1_bio->state);
2208 /* we got a read error. Maybe the drive is bad. Maybe just
2209 * the block and we can fix it.
2210 * We freeze all other IO, and try reading the block from
2211 * other devices. When we find one, we re-write
2212 * and check it that fixes the read error.
2213 * This is all done synchronously while the array is
2214 * frozen
2215 */
2216 if (mddev->ro == 0) {
NeilBrowne2d59922013-06-12 11:01:22 +10002217 freeze_array(conf, 1);
NeilBrown62096bc2011-07-28 11:38:13 +10002218 fix_read_error(conf, r1_bio->read_disk,
2219 r1_bio->sector, r1_bio->sectors);
2220 unfreeze_array(conf);
2221 } else
2222 md_error(mddev, conf->mirrors[r1_bio->read_disk].rdev);
NeilBrown7ad4d4a2012-10-11 13:44:30 +11002223 rdev_dec_pending(conf->mirrors[r1_bio->read_disk].rdev, conf->mddev);
NeilBrown62096bc2011-07-28 11:38:13 +10002224
2225 bio = r1_bio->bios[r1_bio->read_disk];
2226 bdevname(bio->bi_bdev, b);
2227read_more:
2228 disk = read_balance(conf, r1_bio, &max_sectors);
2229 if (disk == -1) {
2230 printk(KERN_ALERT "md/raid1:%s: %s: unrecoverable I/O"
2231 " read error for block %llu\n",
2232 mdname(mddev), b, (unsigned long long)r1_bio->sector);
2233 raid_end_bio_io(r1_bio);
2234 } else {
2235 const unsigned long do_sync
2236 = r1_bio->master_bio->bi_rw & REQ_SYNC;
2237 if (bio) {
2238 r1_bio->bios[r1_bio->read_disk] =
2239 mddev->ro ? IO_BLOCKED : NULL;
2240 bio_put(bio);
2241 }
2242 r1_bio->read_disk = disk;
2243 bio = bio_clone_mddev(r1_bio->master_bio, GFP_NOIO, mddev);
2244 md_trim_bio(bio, r1_bio->sector - bio->bi_sector, max_sectors);
2245 r1_bio->bios[r1_bio->read_disk] = bio;
2246 rdev = conf->mirrors[disk].rdev;
2247 printk_ratelimited(KERN_ERR
2248 "md/raid1:%s: redirecting sector %llu"
2249 " to other mirror: %s\n",
2250 mdname(mddev),
2251 (unsigned long long)r1_bio->sector,
2252 bdevname(rdev->bdev, b));
2253 bio->bi_sector = r1_bio->sector + rdev->data_offset;
2254 bio->bi_bdev = rdev->bdev;
2255 bio->bi_end_io = raid1_end_read_request;
2256 bio->bi_rw = READ | do_sync;
2257 bio->bi_private = r1_bio;
2258 if (max_sectors < r1_bio->sectors) {
2259 /* Drat - have to split this up more */
2260 struct bio *mbio = r1_bio->master_bio;
2261 int sectors_handled = (r1_bio->sector + max_sectors
2262 - mbio->bi_sector);
2263 r1_bio->sectors = max_sectors;
2264 spin_lock_irq(&conf->device_lock);
2265 if (mbio->bi_phys_segments == 0)
2266 mbio->bi_phys_segments = 2;
2267 else
2268 mbio->bi_phys_segments++;
2269 spin_unlock_irq(&conf->device_lock);
2270 generic_make_request(bio);
2271 bio = NULL;
2272
2273 r1_bio = mempool_alloc(conf->r1bio_pool, GFP_NOIO);
2274
2275 r1_bio->master_bio = mbio;
Kent Overstreetaa8b57a2013-02-05 15:19:29 -08002276 r1_bio->sectors = bio_sectors(mbio) - sectors_handled;
NeilBrown62096bc2011-07-28 11:38:13 +10002277 r1_bio->state = 0;
2278 set_bit(R1BIO_ReadError, &r1_bio->state);
2279 r1_bio->mddev = mddev;
2280 r1_bio->sector = mbio->bi_sector + sectors_handled;
2281
2282 goto read_more;
2283 } else
2284 generic_make_request(bio);
2285 }
2286}
2287
Shaohua Li4ed87312012-10-11 13:34:00 +11002288static void raid1d(struct md_thread *thread)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289{
Shaohua Li4ed87312012-10-11 13:34:00 +11002290 struct mddev *mddev = thread->mddev;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002291 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 unsigned long flags;
NeilBrowne8096362011-10-11 16:49:05 +11002293 struct r1conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 struct list_head *head = &conf->retry_list;
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002295 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
2297 md_check_recovery(mddev);
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002298
2299 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 for (;;) {
NeilBrowna35e63e2008-03-04 14:29:29 -08002301
NeilBrown0021b7b2012-07-31 09:08:14 +02002302 flush_pending_writes(conf);
NeilBrowna35e63e2008-03-04 14:29:29 -08002303
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304 spin_lock_irqsave(&conf->device_lock, flags);
NeilBrowna35e63e2008-03-04 14:29:29 -08002305 if (list_empty(head)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002306 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 break;
NeilBrowna35e63e2008-03-04 14:29:29 -08002308 }
NeilBrown9f2c9d12011-10-11 16:48:43 +11002309 r1_bio = list_entry(head->prev, struct r1bio, retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 list_del(head->prev);
NeilBrownddaf22a2006-01-06 00:20:19 -08002311 conf->nr_queued--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 spin_unlock_irqrestore(&conf->device_lock, flags);
2313
2314 mddev = r1_bio->mddev;
NeilBrown070ec552009-06-16 16:54:21 +10002315 conf = mddev->private;
NeilBrown4367af52011-07-28 11:31:49 +10002316 if (test_bit(R1BIO_IsSync, &r1_bio->state)) {
NeilBrownd8f05d22011-07-28 11:33:00 +10002317 if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002318 test_bit(R1BIO_WriteError, &r1_bio->state))
2319 handle_sync_write_finished(conf, r1_bio);
2320 else
NeilBrown4367af52011-07-28 11:31:49 +10002321 sync_request_write(mddev, r1_bio);
NeilBrowncd5ff9a12011-07-28 11:32:41 +10002322 } else if (test_bit(R1BIO_MadeGood, &r1_bio->state) ||
NeilBrown62096bc2011-07-28 11:38:13 +10002323 test_bit(R1BIO_WriteError, &r1_bio->state))
2324 handle_write_finished(conf, r1_bio);
2325 else if (test_bit(R1BIO_ReadError, &r1_bio->state))
2326 handle_read_error(conf, r1_bio);
2327 else
NeilBrownd2eb35a2011-07-28 11:31:48 +10002328 /* just a partial read to be scheduled from separate
2329 * context
2330 */
2331 generic_make_request(r1_bio->bios[r1_bio->read_disk]);
NeilBrown62096bc2011-07-28 11:38:13 +10002332
NeilBrown1d9d5242009-10-16 15:55:32 +11002333 cond_resched();
NeilBrownde393cd2011-07-28 11:31:48 +10002334 if (mddev->flags & ~(1<<MD_CHANGE_PENDING))
2335 md_check_recovery(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 }
NeilBrowne1dfa0a2011-04-18 18:25:41 +10002337 blk_finish_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338}
2339
2340
NeilBrowne8096362011-10-11 16:49:05 +11002341static int init_resync(struct r1conf *conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342{
2343 int buffs;
2344
2345 buffs = RESYNC_WINDOW / RESYNC_BLOCK_SIZE;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002346 BUG_ON(conf->r1buf_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 conf->r1buf_pool = mempool_create(buffs, r1buf_pool_alloc, r1buf_pool_free,
2348 conf->poolinfo);
2349 if (!conf->r1buf_pool)
2350 return -ENOMEM;
2351 conf->next_resync = 0;
2352 return 0;
2353}
2354
2355/*
2356 * perform a "sync" on one "block"
2357 *
2358 * We need to make sure that no normal I/O request - particularly write
2359 * requests - conflict with active sync requests.
2360 *
2361 * This is achieved by tracking pending requests and a 'barrier' concept
2362 * that can be installed to exclude normal IO requests.
2363 */
2364
NeilBrownfd01b882011-10-11 16:47:53 +11002365static sector_t sync_request(struct mddev *mddev, sector_t sector_nr, int *skipped, int go_faster)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366{
NeilBrowne8096362011-10-11 16:49:05 +11002367 struct r1conf *conf = mddev->private;
NeilBrown9f2c9d12011-10-11 16:48:43 +11002368 struct r1bio *r1_bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 struct bio *bio;
2370 sector_t max_sector, nr_sectors;
NeilBrown3e198f72006-01-06 00:20:21 -08002371 int disk = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372 int i;
NeilBrown3e198f72006-01-06 00:20:21 -08002373 int wonly = -1;
2374 int write_targets = 0, read_targets = 0;
NeilBrown57dab0b2010-10-19 10:03:39 +11002375 sector_t sync_blocks;
NeilBrowne3b97032005-08-04 12:53:34 -07002376 int still_degraded = 0;
NeilBrown06f60382011-07-28 11:31:48 +10002377 int good_sectors = RESYNC_SECTORS;
2378 int min_bad = 0; /* number of sectors that are bad in all devices */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379
2380 if (!conf->r1buf_pool)
2381 if (init_resync(conf))
NeilBrown57afd892005-06-21 17:17:13 -07002382 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383
Andre Noll58c0fed2009-03-31 14:33:13 +11002384 max_sector = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 if (sector_nr >= max_sector) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002386 /* If we aborted, we need to abort the
2387 * sync on the 'current' bitmap chunk (there will
2388 * only be one in raid1 resync.
2389 * We can find the current addess in mddev->curr_resync
2390 */
NeilBrown6a806c52005-07-15 03:56:35 -07002391 if (mddev->curr_resync < max_sector) /* aborted */
2392 bitmap_end_sync(mddev->bitmap, mddev->curr_resync,
NeilBrown191ea9b2005-06-21 17:17:23 -07002393 &sync_blocks, 1);
NeilBrown6a806c52005-07-15 03:56:35 -07002394 else /* completed sync */
NeilBrown191ea9b2005-06-21 17:17:23 -07002395 conf->fullsync = 0;
NeilBrown6a806c52005-07-15 03:56:35 -07002396
2397 bitmap_close_sync(mddev->bitmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398 close_sync(conf);
2399 return 0;
2400 }
2401
NeilBrown07d84d102006-06-26 00:27:56 -07002402 if (mddev->bitmap == NULL &&
2403 mddev->recovery_cp == MaxSector &&
NeilBrown6394cca2006-08-27 01:23:50 -07002404 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery) &&
NeilBrown07d84d102006-06-26 00:27:56 -07002405 conf->fullsync == 0) {
2406 *skipped = 1;
2407 return max_sector - sector_nr;
2408 }
NeilBrown6394cca2006-08-27 01:23:50 -07002409 /* before building a request, check if we can skip these blocks..
2410 * This call the bitmap_start_sync doesn't actually record anything
2411 */
NeilBrowne3b97032005-08-04 12:53:34 -07002412 if (!bitmap_start_sync(mddev->bitmap, sector_nr, &sync_blocks, 1) &&
NeilBrowne5de485f2005-11-08 21:39:38 -08002413 !conf->fullsync && !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002414 /* We can skip this block, and probably several more */
2415 *skipped = 1;
2416 return sync_blocks;
2417 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418 /*
NeilBrown17999be2006-01-06 00:20:12 -08002419 * If there is non-resync activity waiting for a turn,
2420 * and resync is going fast enough,
2421 * then let it though before starting on this new sync request.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 */
NeilBrown17999be2006-01-06 00:20:12 -08002423 if (!go_faster && conf->nr_waiting)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 msleep_interruptible(1000);
NeilBrown17999be2006-01-06 00:20:12 -08002425
NeilBrownb47490c2008-02-06 01:39:50 -08002426 bitmap_cond_end_sync(mddev->bitmap, sector_nr);
NeilBrown1c4588e2010-10-26 17:41:22 +11002427 r1_bio = mempool_alloc(conf->r1buf_pool, GFP_NOIO);
NeilBrown17999be2006-01-06 00:20:12 -08002428 raise_barrier(conf);
2429
2430 conf->next_resync = sector_nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431
NeilBrown3e198f72006-01-06 00:20:21 -08002432 rcu_read_lock();
2433 /*
2434 * If we get a correctably read error during resync or recovery,
2435 * we might want to read from a different device. So we
2436 * flag all drives that could conceivably be read from for READ,
2437 * and any others (which will be non-In_sync devices) for WRITE.
2438 * If a read fails, we try reading from something else for which READ
2439 * is OK.
2440 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441
Linus Torvalds1da177e2005-04-16 15:20:36 -07002442 r1_bio->mddev = mddev;
2443 r1_bio->sector = sector_nr;
NeilBrown191ea9b2005-06-21 17:17:23 -07002444 r1_bio->state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 set_bit(R1BIO_IsSync, &r1_bio->state);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446
NeilBrown8f19ccb2011-12-23 10:17:56 +11002447 for (i = 0; i < conf->raid_disks * 2; i++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002448 struct md_rdev *rdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 bio = r1_bio->bios[i];
Kent Overstreet2aabaa62012-09-11 11:26:12 -07002450 bio_reset(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451
NeilBrown3e198f72006-01-06 00:20:21 -08002452 rdev = rcu_dereference(conf->mirrors[i].rdev);
2453 if (rdev == NULL ||
NeilBrown06f60382011-07-28 11:31:48 +10002454 test_bit(Faulty, &rdev->flags)) {
NeilBrown8f19ccb2011-12-23 10:17:56 +11002455 if (i < conf->raid_disks)
2456 still_degraded = 1;
NeilBrown3e198f72006-01-06 00:20:21 -08002457 } else if (!test_bit(In_sync, &rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002458 bio->bi_rw = WRITE;
2459 bio->bi_end_io = end_sync_write;
2460 write_targets ++;
NeilBrown3e198f72006-01-06 00:20:21 -08002461 } else {
2462 /* may need to read from here */
NeilBrown06f60382011-07-28 11:31:48 +10002463 sector_t first_bad = MaxSector;
2464 int bad_sectors;
2465
2466 if (is_badblock(rdev, sector_nr, good_sectors,
2467 &first_bad, &bad_sectors)) {
2468 if (first_bad > sector_nr)
2469 good_sectors = first_bad - sector_nr;
2470 else {
2471 bad_sectors -= (sector_nr - first_bad);
2472 if (min_bad == 0 ||
2473 min_bad > bad_sectors)
2474 min_bad = bad_sectors;
2475 }
NeilBrown3e198f72006-01-06 00:20:21 -08002476 }
NeilBrown06f60382011-07-28 11:31:48 +10002477 if (sector_nr < first_bad) {
2478 if (test_bit(WriteMostly, &rdev->flags)) {
2479 if (wonly < 0)
2480 wonly = i;
2481 } else {
2482 if (disk < 0)
2483 disk = i;
2484 }
2485 bio->bi_rw = READ;
2486 bio->bi_end_io = end_sync_read;
2487 read_targets++;
Alexander Lyakasd57368a2012-07-17 13:17:55 +03002488 } else if (!test_bit(WriteErrorSeen, &rdev->flags) &&
2489 test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
2490 !test_bit(MD_RECOVERY_CHECK, &mddev->recovery)) {
2491 /*
2492 * The device is suitable for reading (InSync),
2493 * but has bad block(s) here. Let's try to correct them,
2494 * if we are doing resync or repair. Otherwise, leave
2495 * this device alone for this sync request.
2496 */
2497 bio->bi_rw = WRITE;
2498 bio->bi_end_io = end_sync_write;
2499 write_targets++;
NeilBrown06f60382011-07-28 11:31:48 +10002500 }
NeilBrown3e198f72006-01-06 00:20:21 -08002501 }
NeilBrown06f60382011-07-28 11:31:48 +10002502 if (bio->bi_end_io) {
2503 atomic_inc(&rdev->nr_pending);
2504 bio->bi_sector = sector_nr + rdev->data_offset;
2505 bio->bi_bdev = rdev->bdev;
2506 bio->bi_private = r1_bio;
2507 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 }
NeilBrown3e198f72006-01-06 00:20:21 -08002509 rcu_read_unlock();
2510 if (disk < 0)
2511 disk = wonly;
2512 r1_bio->read_disk = disk;
NeilBrown191ea9b2005-06-21 17:17:23 -07002513
NeilBrown06f60382011-07-28 11:31:48 +10002514 if (read_targets == 0 && min_bad > 0) {
2515 /* These sectors are bad on all InSync devices, so we
2516 * need to mark them bad on all write targets
2517 */
2518 int ok = 1;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002519 for (i = 0 ; i < conf->raid_disks * 2 ; i++)
NeilBrown06f60382011-07-28 11:31:48 +10002520 if (r1_bio->bios[i]->bi_end_io == end_sync_write) {
majianpenga42f9d82012-04-02 01:04:19 +10002521 struct md_rdev *rdev = conf->mirrors[i].rdev;
NeilBrown06f60382011-07-28 11:31:48 +10002522 ok = rdev_set_badblocks(rdev, sector_nr,
2523 min_bad, 0
2524 ) && ok;
2525 }
2526 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2527 *skipped = 1;
2528 put_buf(r1_bio);
2529
2530 if (!ok) {
2531 /* Cannot record the badblocks, so need to
2532 * abort the resync.
2533 * If there are multiple read targets, could just
2534 * fail the really bad ones ???
2535 */
2536 conf->recovery_disabled = mddev->recovery_disabled;
2537 set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2538 return 0;
2539 } else
2540 return min_bad;
2541
2542 }
2543 if (min_bad > 0 && min_bad < good_sectors) {
2544 /* only resync enough to reach the next bad->good
2545 * transition */
2546 good_sectors = min_bad;
2547 }
2548
NeilBrown3e198f72006-01-06 00:20:21 -08002549 if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && read_targets > 0)
2550 /* extra read targets are also write targets */
2551 write_targets += read_targets-1;
2552
2553 if (write_targets == 0 || read_targets == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 /* There is nowhere to write, so all non-sync
2555 * drives must be failed - so we are finished
2556 */
NeilBrownb7219cc2012-07-31 10:05:34 +10002557 sector_t rv;
2558 if (min_bad > 0)
2559 max_sector = sector_nr + min_bad;
2560 rv = max_sector - sector_nr;
NeilBrown57afd892005-06-21 17:17:13 -07002561 *skipped = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 put_buf(r1_bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 return rv;
2564 }
2565
NeilBrownc6207272008-02-06 01:39:52 -08002566 if (max_sector > mddev->resync_max)
2567 max_sector = mddev->resync_max; /* Don't do IO beyond here */
NeilBrown06f60382011-07-28 11:31:48 +10002568 if (max_sector > sector_nr + good_sectors)
2569 max_sector = sector_nr + good_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 nr_sectors = 0;
NeilBrown289e99e2005-06-21 17:17:24 -07002571 sync_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002572 do {
2573 struct page *page;
2574 int len = PAGE_SIZE;
2575 if (sector_nr + (len>>9) > max_sector)
2576 len = (max_sector - sector_nr) << 9;
2577 if (len == 0)
2578 break;
NeilBrown6a806c52005-07-15 03:56:35 -07002579 if (sync_blocks == 0) {
2580 if (!bitmap_start_sync(mddev->bitmap, sector_nr,
NeilBrowne5de485f2005-11-08 21:39:38 -08002581 &sync_blocks, still_degraded) &&
2582 !conf->fullsync &&
2583 !test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
NeilBrown6a806c52005-07-15 03:56:35 -07002584 break;
Eric Sesterhenn9e77c482006-04-01 01:08:49 +02002585 BUG_ON(sync_blocks < (PAGE_SIZE>>9));
NeilBrown7571ae82010-10-07 11:54:46 +11002586 if ((len >> 9) > sync_blocks)
NeilBrown6a806c52005-07-15 03:56:35 -07002587 len = sync_blocks<<9;
NeilBrownab7a30c2005-06-21 17:17:23 -07002588 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002589
NeilBrown8f19ccb2011-12-23 10:17:56 +11002590 for (i = 0 ; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 bio = r1_bio->bios[i];
2592 if (bio->bi_end_io) {
NeilBrownd11c1712006-01-06 00:20:26 -08002593 page = bio->bi_io_vec[bio->bi_vcnt].bv_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 if (bio_add_page(bio, page, len, 0) == 0) {
2595 /* stop here */
NeilBrownd11c1712006-01-06 00:20:26 -08002596 bio->bi_io_vec[bio->bi_vcnt].bv_page = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 while (i > 0) {
2598 i--;
2599 bio = r1_bio->bios[i];
NeilBrown6a806c52005-07-15 03:56:35 -07002600 if (bio->bi_end_io==NULL)
2601 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602 /* remove last page from this bio */
2603 bio->bi_vcnt--;
2604 bio->bi_size -= len;
2605 bio->bi_flags &= ~(1<< BIO_SEG_VALID);
2606 }
2607 goto bio_full;
2608 }
2609 }
2610 }
2611 nr_sectors += len>>9;
2612 sector_nr += len>>9;
NeilBrown191ea9b2005-06-21 17:17:23 -07002613 sync_blocks -= (len>>9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 } while (r1_bio->bios[disk]->bi_vcnt < RESYNC_PAGES);
2615 bio_full:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 r1_bio->sectors = nr_sectors;
2617
NeilBrownd11c1712006-01-06 00:20:26 -08002618 /* For a user-requested sync, we read all readable devices and do a
2619 * compare
2620 */
2621 if (test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery)) {
2622 atomic_set(&r1_bio->remaining, read_targets);
NeilBrown2d4f4f32012-07-09 11:34:13 +10002623 for (i = 0; i < conf->raid_disks * 2 && read_targets; i++) {
NeilBrownd11c1712006-01-06 00:20:26 -08002624 bio = r1_bio->bios[i];
2625 if (bio->bi_end_io == end_sync_read) {
NeilBrown2d4f4f32012-07-09 11:34:13 +10002626 read_targets--;
NeilBrownddac7c72006-08-31 21:27:36 -07002627 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002628 generic_make_request(bio);
2629 }
2630 }
2631 } else {
2632 atomic_set(&r1_bio->remaining, 1);
2633 bio = r1_bio->bios[r1_bio->read_disk];
NeilBrownddac7c72006-08-31 21:27:36 -07002634 md_sync_acct(bio->bi_bdev, nr_sectors);
NeilBrownd11c1712006-01-06 00:20:26 -08002635 generic_make_request(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636
NeilBrownd11c1712006-01-06 00:20:26 -08002637 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 return nr_sectors;
2639}
2640
NeilBrownfd01b882011-10-11 16:47:53 +11002641static sector_t raid1_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -07002642{
2643 if (sectors)
2644 return sectors;
2645
2646 return mddev->dev_sectors;
2647}
2648
NeilBrowne8096362011-10-11 16:49:05 +11002649static struct r1conf *setup_conf(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650{
NeilBrowne8096362011-10-11 16:49:05 +11002651 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002652 int i;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10002653 struct raid1_info *disk;
NeilBrown3cb03002011-10-11 16:45:26 +11002654 struct md_rdev *rdev;
NeilBrown709ae482009-12-14 12:49:51 +11002655 int err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656
NeilBrowne8096362011-10-11 16:49:05 +11002657 conf = kzalloc(sizeof(struct r1conf), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 if (!conf)
NeilBrown709ae482009-12-14 12:49:51 +11002659 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10002661 conf->mirrors = kzalloc(sizeof(struct raid1_info)
NeilBrown8f19ccb2011-12-23 10:17:56 +11002662 * mddev->raid_disks * 2,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 GFP_KERNEL);
2664 if (!conf->mirrors)
NeilBrown709ae482009-12-14 12:49:51 +11002665 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666
NeilBrownddaf22a2006-01-06 00:20:19 -08002667 conf->tmppage = alloc_page(GFP_KERNEL);
2668 if (!conf->tmppage)
NeilBrown709ae482009-12-14 12:49:51 +11002669 goto abort;
NeilBrownddaf22a2006-01-06 00:20:19 -08002670
NeilBrown709ae482009-12-14 12:49:51 +11002671 conf->poolinfo = kzalloc(sizeof(*conf->poolinfo), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672 if (!conf->poolinfo)
NeilBrown709ae482009-12-14 12:49:51 +11002673 goto abort;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002674 conf->poolinfo->raid_disks = mddev->raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 conf->r1bio_pool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2676 r1bio_pool_free,
2677 conf->poolinfo);
2678 if (!conf->r1bio_pool)
NeilBrown709ae482009-12-14 12:49:51 +11002679 goto abort;
2680
NeilBrowned9bfdf2009-10-16 15:55:44 +11002681 conf->poolinfo->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682
NeilBrownc19d5792011-12-23 10:17:57 +11002683 err = -EINVAL;
Neil Browne7e72bf2008-05-14 16:05:54 -07002684 spin_lock_init(&conf->device_lock);
NeilBrowndafb20f2012-03-19 12:46:39 +11002685 rdev_for_each(rdev, mddev) {
NeilBrownaba336b2012-05-31 15:39:11 +10002686 struct request_queue *q;
NeilBrown709ae482009-12-14 12:49:51 +11002687 int disk_idx = rdev->raid_disk;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 if (disk_idx >= mddev->raid_disks
2689 || disk_idx < 0)
2690 continue;
NeilBrownc19d5792011-12-23 10:17:57 +11002691 if (test_bit(Replacement, &rdev->flags))
NeilBrown02b898f2012-10-31 11:42:03 +11002692 disk = conf->mirrors + mddev->raid_disks + disk_idx;
NeilBrownc19d5792011-12-23 10:17:57 +11002693 else
2694 disk = conf->mirrors + disk_idx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695
NeilBrownc19d5792011-12-23 10:17:57 +11002696 if (disk->rdev)
2697 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698 disk->rdev = rdev;
NeilBrownaba336b2012-05-31 15:39:11 +10002699 q = bdev_get_queue(rdev->bdev);
2700 if (q->merge_bvec_fn)
2701 mddev->merge_check_needed = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702
2703 disk->head_position = 0;
Shaohua Li12cee5a2012-07-31 10:03:53 +10002704 disk->seq_start = MaxSector;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 }
2706 conf->raid_disks = mddev->raid_disks;
2707 conf->mddev = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 INIT_LIST_HEAD(&conf->retry_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709
2710 spin_lock_init(&conf->resync_lock);
NeilBrown17999be2006-01-06 00:20:12 -08002711 init_waitqueue_head(&conf->wait_barrier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002712
NeilBrown191ea9b2005-06-21 17:17:23 -07002713 bio_list_init(&conf->pending_bio_list);
NeilBrown34db0cd2011-10-11 16:50:01 +11002714 conf->pending_count = 0;
NeilBrownd890fa22011-10-26 11:54:39 +11002715 conf->recovery_disabled = mddev->recovery_disabled - 1;
NeilBrown191ea9b2005-06-21 17:17:23 -07002716
NeilBrownc19d5792011-12-23 10:17:57 +11002717 err = -EIO;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002718 for (i = 0; i < conf->raid_disks * 2; i++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719
2720 disk = conf->mirrors + i;
2721
NeilBrownc19d5792011-12-23 10:17:57 +11002722 if (i < conf->raid_disks &&
2723 disk[conf->raid_disks].rdev) {
2724 /* This slot has a replacement. */
2725 if (!disk->rdev) {
2726 /* No original, just make the replacement
2727 * a recovering spare
2728 */
2729 disk->rdev =
2730 disk[conf->raid_disks].rdev;
2731 disk[conf->raid_disks].rdev = NULL;
2732 } else if (!test_bit(In_sync, &disk->rdev->flags))
2733 /* Original is not in_sync - bad */
2734 goto abort;
2735 }
2736
NeilBrown5fd6c1d2006-06-26 00:27:40 -07002737 if (!disk->rdev ||
2738 !test_bit(In_sync, &disk->rdev->flags)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739 disk->head_position = 0;
Jonathan Brassow4f0a5e02012-05-22 13:55:31 +10002740 if (disk->rdev &&
2741 (disk->rdev->saved_raid_disk < 0))
NeilBrown918f0232007-08-22 14:01:52 -07002742 conf->fullsync = 1;
Shaohua Libe4d3282012-07-31 10:03:53 +10002743 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 }
NeilBrown709ae482009-12-14 12:49:51 +11002745
NeilBrown709ae482009-12-14 12:49:51 +11002746 err = -ENOMEM;
NeilBrown02326052012-07-03 15:56:52 +10002747 conf->thread = md_register_thread(raid1d, mddev, "raid1");
NeilBrown709ae482009-12-14 12:49:51 +11002748 if (!conf->thread) {
NeilBrown191ea9b2005-06-21 17:17:23 -07002749 printk(KERN_ERR
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002750 "md/raid1:%s: couldn't allocate thread\n",
NeilBrown191ea9b2005-06-21 17:17:23 -07002751 mdname(mddev));
NeilBrown709ae482009-12-14 12:49:51 +11002752 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 }
NeilBrown191ea9b2005-06-21 17:17:23 -07002754
NeilBrown709ae482009-12-14 12:49:51 +11002755 return conf;
2756
2757 abort:
2758 if (conf) {
2759 if (conf->r1bio_pool)
2760 mempool_destroy(conf->r1bio_pool);
2761 kfree(conf->mirrors);
2762 safe_put_page(conf->tmppage);
2763 kfree(conf->poolinfo);
2764 kfree(conf);
2765 }
2766 return ERR_PTR(err);
2767}
2768
majianpeng5220ea12012-04-02 09:48:38 +10002769static int stop(struct mddev *mddev);
NeilBrownfd01b882011-10-11 16:47:53 +11002770static int run(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11002771{
NeilBrowne8096362011-10-11 16:49:05 +11002772 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11002773 int i;
NeilBrown3cb03002011-10-11 16:45:26 +11002774 struct md_rdev *rdev;
majianpeng5220ea12012-04-02 09:48:38 +10002775 int ret;
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002776 bool discard_supported = false;
NeilBrown709ae482009-12-14 12:49:51 +11002777
2778 if (mddev->level != 1) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002779 printk(KERN_ERR "md/raid1:%s: raid level not set to mirroring (%d)\n",
NeilBrown709ae482009-12-14 12:49:51 +11002780 mdname(mddev), mddev->level);
2781 return -EIO;
2782 }
2783 if (mddev->reshape_position != MaxSector) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002784 printk(KERN_ERR "md/raid1:%s: reshape_position set but not supported\n",
NeilBrown709ae482009-12-14 12:49:51 +11002785 mdname(mddev));
2786 return -EIO;
2787 }
2788 /*
2789 * copy the already verified devices into our private RAID1
2790 * bookkeeping area. [whatever we allocate in run(),
2791 * should be freed in stop()]
2792 */
2793 if (mddev->private == NULL)
2794 conf = setup_conf(mddev);
2795 else
2796 conf = mddev->private;
2797
2798 if (IS_ERR(conf))
2799 return PTR_ERR(conf);
2800
Joe Lawrencec8dc9c62013-02-21 13:28:09 +11002801 if (mddev->queue)
H. Peter Anvin5026d7a2013-06-12 07:37:43 -07002802 blk_queue_max_write_same_sectors(mddev->queue, 0);
2803
NeilBrowndafb20f2012-03-19 12:46:39 +11002804 rdev_for_each(rdev, mddev) {
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002805 if (!mddev->gendisk)
2806 continue;
NeilBrown709ae482009-12-14 12:49:51 +11002807 disk_stack_limits(mddev->gendisk, rdev->bdev,
2808 rdev->data_offset << 9);
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002809 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
2810 discard_supported = true;
NeilBrown709ae482009-12-14 12:49:51 +11002811 }
2812
2813 mddev->degraded = 0;
2814 for (i=0; i < conf->raid_disks; i++)
2815 if (conf->mirrors[i].rdev == NULL ||
2816 !test_bit(In_sync, &conf->mirrors[i].rdev->flags) ||
2817 test_bit(Faulty, &conf->mirrors[i].rdev->flags))
2818 mddev->degraded++;
2819
2820 if (conf->raid_disks - mddev->degraded == 1)
2821 mddev->recovery_cp = MaxSector;
2822
Andre Noll8c6ac862009-06-18 08:48:06 +10002823 if (mddev->recovery_cp != MaxSector)
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002824 printk(KERN_NOTICE "md/raid1:%s: not clean"
Andre Noll8c6ac862009-06-18 08:48:06 +10002825 " -- starting background reconstruction\n",
2826 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827 printk(KERN_INFO
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002828 "md/raid1:%s: active with %d out of %d mirrors\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829 mdname(mddev), mddev->raid_disks - mddev->degraded,
2830 mddev->raid_disks);
NeilBrown709ae482009-12-14 12:49:51 +11002831
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 /*
2833 * Ok, everything is just fine now
2834 */
NeilBrown709ae482009-12-14 12:49:51 +11002835 mddev->thread = conf->thread;
2836 conf->thread = NULL;
2837 mddev->private = conf;
2838
Dan Williams1f403622009-03-31 14:59:03 +11002839 md_set_array_sectors(mddev, raid1_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002841 if (mddev->queue) {
2842 mddev->queue->backing_dev_info.congested_fn = raid1_congested;
2843 mddev->queue->backing_dev_info.congested_data = mddev;
NeilBrown6b740b82012-03-19 12:46:39 +11002844 blk_queue_merge_bvec(mddev->queue, raid1_mergeable_bvec);
Shaohua Li2ff8cc22012-10-11 13:28:54 +11002845
2846 if (discard_supported)
2847 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD,
2848 mddev->queue);
2849 else
2850 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD,
2851 mddev->queue);
Jonathan Brassow1ed72422011-06-07 17:50:35 -05002852 }
majianpeng5220ea12012-04-02 09:48:38 +10002853
2854 ret = md_integrity_register(mddev);
2855 if (ret)
2856 stop(mddev);
2857 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858}
2859
NeilBrownfd01b882011-10-11 16:47:53 +11002860static int stop(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861{
NeilBrowne8096362011-10-11 16:49:05 +11002862 struct r1conf *conf = mddev->private;
NeilBrown4b6d2872005-09-09 16:23:47 -07002863 struct bitmap *bitmap = mddev->bitmap;
NeilBrown4b6d2872005-09-09 16:23:47 -07002864
2865 /* wait for behind writes to complete */
NeilBrowne5551902010-03-31 11:21:44 +11002866 if (bitmap && atomic_read(&bitmap->behind_writes) > 0) {
NeilBrown9dd1e2f2010-05-03 14:30:35 +10002867 printk(KERN_INFO "md/raid1:%s: behind writes in progress - waiting to stop.\n",
2868 mdname(mddev));
NeilBrown4b6d2872005-09-09 16:23:47 -07002869 /* need to kick something here to make sure I/O goes? */
NeilBrowne5551902010-03-31 11:21:44 +11002870 wait_event(bitmap->behind_wait,
2871 atomic_read(&bitmap->behind_writes) == 0);
NeilBrown4b6d2872005-09-09 16:23:47 -07002872 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873
NeilBrown409c57f2009-03-31 14:39:39 +11002874 raise_barrier(conf);
2875 lower_barrier(conf);
2876
NeilBrown01f96c02011-09-21 15:30:20 +10002877 md_unregister_thread(&mddev->thread);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 if (conf->r1bio_pool)
2879 mempool_destroy(conf->r1bio_pool);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002880 kfree(conf->mirrors);
Hirokazu Takahashi0fea7ed2013-04-24 11:42:44 +10002881 safe_put_page(conf->tmppage);
Jesper Juhl990a8ba2005-06-21 17:17:30 -07002882 kfree(conf->poolinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 kfree(conf);
2884 mddev->private = NULL;
2885 return 0;
2886}
2887
NeilBrownfd01b882011-10-11 16:47:53 +11002888static int raid1_resize(struct mddev *mddev, sector_t sectors)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889{
2890 /* no resync is happening, and there is enough space
2891 * on all devices, so we can resize.
2892 * We need to make sure resync covers any new space.
2893 * If the array is shrinking we should possibly wait until
2894 * any io in the removed space completes, but it hardly seems
2895 * worth it.
2896 */
NeilBrowna4a61252012-05-22 13:55:27 +10002897 sector_t newsize = raid1_size(mddev, sectors, 0);
2898 if (mddev->external_size &&
2899 mddev->array_sectors > newsize)
Dan Williamsb522adc2009-03-31 15:00:31 +11002900 return -EINVAL;
NeilBrowna4a61252012-05-22 13:55:27 +10002901 if (mddev->bitmap) {
2902 int ret = bitmap_resize(mddev->bitmap, newsize, 0, 0);
2903 if (ret)
2904 return ret;
2905 }
2906 md_set_array_sectors(mddev, newsize);
Andre Nollf233ea52008-07-21 17:05:22 +10002907 set_capacity(mddev->gendisk, mddev->array_sectors);
NeilBrown449aad32009-08-03 10:59:58 +10002908 revalidate_disk(mddev->gendisk);
Dan Williamsb522adc2009-03-31 15:00:31 +11002909 if (sectors > mddev->dev_sectors &&
NeilBrownb0986362011-05-11 15:52:21 +10002910 mddev->recovery_cp > mddev->dev_sectors) {
Andre Noll58c0fed2009-03-31 14:33:13 +11002911 mddev->recovery_cp = mddev->dev_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002912 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2913 }
Dan Williamsb522adc2009-03-31 15:00:31 +11002914 mddev->dev_sectors = sectors;
NeilBrown4b5c7ae2005-07-27 11:43:28 -07002915 mddev->resync_max_sectors = sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916 return 0;
2917}
2918
NeilBrownfd01b882011-10-11 16:47:53 +11002919static int raid1_reshape(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920{
2921 /* We need to:
2922 * 1/ resize the r1bio_pool
2923 * 2/ resize conf->mirrors
2924 *
2925 * We allocate a new r1bio_pool if we can.
2926 * Then raise a device barrier and wait until all IO stops.
2927 * Then resize conf->mirrors and swap in the new r1bio pool.
NeilBrown6ea9c072005-06-21 17:17:09 -07002928 *
2929 * At the same time, we "pack" the devices so that all the missing
2930 * devices have the higher raid_disk numbers.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 */
2932 mempool_t *newpool, *oldpool;
2933 struct pool_info *newpoolinfo;
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10002934 struct raid1_info *newmirrors;
NeilBrowne8096362011-10-11 16:49:05 +11002935 struct r1conf *conf = mddev->private;
NeilBrown63c70c42006-03-27 01:18:13 -08002936 int cnt, raid_disks;
NeilBrownc04be0a2006-10-03 01:15:53 -07002937 unsigned long flags;
Dan Williamsb5470dc2008-06-27 21:44:04 -07002938 int d, d2, err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939
NeilBrown63c70c42006-03-27 01:18:13 -08002940 /* Cannot change chunk_size, layout, or level */
Andre Noll664e7c42009-06-18 08:45:27 +10002941 if (mddev->chunk_sectors != mddev->new_chunk_sectors ||
NeilBrown63c70c42006-03-27 01:18:13 -08002942 mddev->layout != mddev->new_layout ||
2943 mddev->level != mddev->new_level) {
Andre Noll664e7c42009-06-18 08:45:27 +10002944 mddev->new_chunk_sectors = mddev->chunk_sectors;
NeilBrown63c70c42006-03-27 01:18:13 -08002945 mddev->new_layout = mddev->layout;
2946 mddev->new_level = mddev->level;
2947 return -EINVAL;
2948 }
2949
Dan Williamsb5470dc2008-06-27 21:44:04 -07002950 err = md_allow_write(mddev);
2951 if (err)
2952 return err;
NeilBrown2a2275d2007-01-26 00:57:11 -08002953
NeilBrown63c70c42006-03-27 01:18:13 -08002954 raid_disks = mddev->raid_disks + mddev->delta_disks;
2955
NeilBrown6ea9c072005-06-21 17:17:09 -07002956 if (raid_disks < conf->raid_disks) {
2957 cnt=0;
2958 for (d= 0; d < conf->raid_disks; d++)
2959 if (conf->mirrors[d].rdev)
2960 cnt++;
2961 if (cnt > raid_disks)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 return -EBUSY;
NeilBrown6ea9c072005-06-21 17:17:09 -07002963 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964
2965 newpoolinfo = kmalloc(sizeof(*newpoolinfo), GFP_KERNEL);
2966 if (!newpoolinfo)
2967 return -ENOMEM;
2968 newpoolinfo->mddev = mddev;
NeilBrown8f19ccb2011-12-23 10:17:56 +11002969 newpoolinfo->raid_disks = raid_disks * 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970
2971 newpool = mempool_create(NR_RAID1_BIOS, r1bio_pool_alloc,
2972 r1bio_pool_free, newpoolinfo);
2973 if (!newpool) {
2974 kfree(newpoolinfo);
2975 return -ENOMEM;
2976 }
Jonathan Brassow0eaf8222012-07-31 10:03:52 +10002977 newmirrors = kzalloc(sizeof(struct raid1_info) * raid_disks * 2,
NeilBrown8f19ccb2011-12-23 10:17:56 +11002978 GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 if (!newmirrors) {
2980 kfree(newpoolinfo);
2981 mempool_destroy(newpool);
2982 return -ENOMEM;
2983 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984
NeilBrowne2d59922013-06-12 11:01:22 +10002985 freeze_array(conf, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986
2987 /* ok, everything is stopped */
2988 oldpool = conf->r1bio_pool;
2989 conf->r1bio_pool = newpool;
NeilBrown6ea9c072005-06-21 17:17:09 -07002990
NeilBrowna88aa782007-08-22 14:01:53 -07002991 for (d = d2 = 0; d < conf->raid_disks; d++) {
NeilBrown3cb03002011-10-11 16:45:26 +11002992 struct md_rdev *rdev = conf->mirrors[d].rdev;
NeilBrowna88aa782007-08-22 14:01:53 -07002993 if (rdev && rdev->raid_disk != d2) {
Namhyung Kim36fad852011-07-27 11:00:36 +10002994 sysfs_unlink_rdev(mddev, rdev);
NeilBrowna88aa782007-08-22 14:01:53 -07002995 rdev->raid_disk = d2;
Namhyung Kim36fad852011-07-27 11:00:36 +10002996 sysfs_unlink_rdev(mddev, rdev);
2997 if (sysfs_link_rdev(mddev, rdev))
NeilBrowna88aa782007-08-22 14:01:53 -07002998 printk(KERN_WARNING
Namhyung Kim36fad852011-07-27 11:00:36 +10002999 "md/raid1:%s: cannot register rd%d\n",
3000 mdname(mddev), rdev->raid_disk);
NeilBrown6ea9c072005-06-21 17:17:09 -07003001 }
NeilBrowna88aa782007-08-22 14:01:53 -07003002 if (rdev)
3003 newmirrors[d2++].rdev = rdev;
3004 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 kfree(conf->mirrors);
3006 conf->mirrors = newmirrors;
3007 kfree(conf->poolinfo);
3008 conf->poolinfo = newpoolinfo;
3009
NeilBrownc04be0a2006-10-03 01:15:53 -07003010 spin_lock_irqsave(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011 mddev->degraded += (raid_disks - conf->raid_disks);
NeilBrownc04be0a2006-10-03 01:15:53 -07003012 spin_unlock_irqrestore(&conf->device_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013 conf->raid_disks = mddev->raid_disks = raid_disks;
NeilBrown63c70c42006-03-27 01:18:13 -08003014 mddev->delta_disks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015
NeilBrowne2d59922013-06-12 11:01:22 +10003016 unfreeze_array(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017
3018 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3019 md_wakeup_thread(mddev->thread);
3020
3021 mempool_destroy(oldpool);
3022 return 0;
3023}
3024
NeilBrownfd01b882011-10-11 16:47:53 +11003025static void raid1_quiesce(struct mddev *mddev, int state)
NeilBrown36fa3062005-09-09 16:23:45 -07003026{
NeilBrowne8096362011-10-11 16:49:05 +11003027 struct r1conf *conf = mddev->private;
NeilBrown36fa3062005-09-09 16:23:45 -07003028
3029 switch(state) {
NeilBrown6eef4b22009-12-14 12:49:51 +11003030 case 2: /* wake for suspend */
3031 wake_up(&conf->wait_barrier);
3032 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07003033 case 1:
NeilBrown17999be2006-01-06 00:20:12 -08003034 raise_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07003035 break;
NeilBrown9e6603d2005-09-09 16:23:48 -07003036 case 0:
NeilBrown17999be2006-01-06 00:20:12 -08003037 lower_barrier(conf);
NeilBrown36fa3062005-09-09 16:23:45 -07003038 break;
3039 }
NeilBrown36fa3062005-09-09 16:23:45 -07003040}
3041
NeilBrownfd01b882011-10-11 16:47:53 +11003042static void *raid1_takeover(struct mddev *mddev)
NeilBrown709ae482009-12-14 12:49:51 +11003043{
3044 /* raid1 can take over:
3045 * raid5 with 2 devices, any layout or chunk size
3046 */
3047 if (mddev->level == 5 && mddev->raid_disks == 2) {
NeilBrowne8096362011-10-11 16:49:05 +11003048 struct r1conf *conf;
NeilBrown709ae482009-12-14 12:49:51 +11003049 mddev->new_level = 1;
3050 mddev->new_layout = 0;
3051 mddev->new_chunk_sectors = 0;
3052 conf = setup_conf(mddev);
3053 if (!IS_ERR(conf))
3054 conf->barrier = 1;
3055 return conf;
3056 }
3057 return ERR_PTR(-EINVAL);
3058}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059
NeilBrown84fc4b52011-10-11 16:49:58 +11003060static struct md_personality raid1_personality =
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061{
3062 .name = "raid1",
NeilBrown2604b702006-01-06 00:20:36 -08003063 .level = 1,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064 .owner = THIS_MODULE,
3065 .make_request = make_request,
3066 .run = run,
3067 .stop = stop,
3068 .status = status,
3069 .error_handler = error,
3070 .hot_add_disk = raid1_add_disk,
3071 .hot_remove_disk= raid1_remove_disk,
3072 .spare_active = raid1_spare_active,
3073 .sync_request = sync_request,
3074 .resize = raid1_resize,
Dan Williams80c3a6c2009-03-17 18:10:40 -07003075 .size = raid1_size,
NeilBrown63c70c42006-03-27 01:18:13 -08003076 .check_reshape = raid1_reshape,
NeilBrown36fa3062005-09-09 16:23:45 -07003077 .quiesce = raid1_quiesce,
NeilBrown709ae482009-12-14 12:49:51 +11003078 .takeover = raid1_takeover,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079};
3080
3081static int __init raid_init(void)
3082{
NeilBrown2604b702006-01-06 00:20:36 -08003083 return register_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084}
3085
3086static void raid_exit(void)
3087{
NeilBrown2604b702006-01-06 00:20:36 -08003088 unregister_md_personality(&raid1_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089}
3090
3091module_init(raid_init);
3092module_exit(raid_exit);
3093MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +11003094MODULE_DESCRIPTION("RAID1 (mirroring) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095MODULE_ALIAS("md-personality-3"); /* RAID1 */
NeilBrownd9d166c2006-01-06 00:20:51 -08003096MODULE_ALIAS("md-raid1");
NeilBrown2604b702006-01-06 00:20:36 -08003097MODULE_ALIAS("md-level-1");
NeilBrown34db0cd2011-10-11 16:50:01 +11003098
3099module_param(max_queued_requests, int, S_IRUGO|S_IWUSR);