Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | raid0.c : Multiple Devices driver for Linux |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 3 | Copyright (C) 1994-96 Marc ZYNGIER |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 4 | <zyngier@ufr-info-p7.ibp.fr> or |
| 5 | <maz@gloups.fdn.fr> |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 6 | Copyright (C) 1999, 2000 Ingo Molnar, Red Hat |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | |
| 8 | RAID-0 management functions. |
| 9 | |
| 10 | This program is free software; you can redistribute it and/or modify |
| 11 | it under the terms of the GNU General Public License as published by |
| 12 | the Free Software Foundation; either version 2, or (at your option) |
| 13 | any later version. |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 14 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 15 | You should have received a copy of the GNU General Public License |
| 16 | (for example /usr/src/linux/COPYING); if not, write to the Free |
NeilBrown | f72ffdd | 2014-09-30 14:23:59 +1000 | [diff] [blame] | 17 | Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | */ |
| 19 | |
NeilBrown | bff6197 | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 20 | #include <linux/blkdev.h> |
NeilBrown | bff6197 | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 21 | #include <linux/seq_file.h> |
Paul Gortmaker | 056075c | 2011-07-03 13:58:33 -0400 | [diff] [blame] | 22 | #include <linux/module.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 23 | #include <linux/slab.h> |
NeilBrown | 43b2e5d | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 24 | #include "md.h" |
Christoph Hellwig | ef740c3 | 2009-03-31 14:27:03 +1100 | [diff] [blame] | 25 | #include "raid0.h" |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 26 | #include "raid5.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | |
NeilBrown | 5c675f8 | 2014-12-15 12:56:56 +1100 | [diff] [blame] | 28 | static int raid0_congested(struct mddev *mddev, int bits) |
NeilBrown | 26be34d | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 29 | { |
NeilBrown | e373ab1 | 2011-10-11 16:48:59 +1100 | [diff] [blame] | 30 | struct r0conf *conf = mddev->private; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 31 | struct md_rdev **devlist = conf->devlist; |
NeilBrown | 84707f3 | 2010-03-16 17:23:35 +1100 | [diff] [blame] | 32 | int raid_disks = conf->strip_zone[0].nb_dev; |
NeilBrown | 26be34d | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 33 | int i, ret = 0; |
| 34 | |
NeilBrown | 84707f3 | 2010-03-16 17:23:35 +1100 | [diff] [blame] | 35 | for (i = 0; i < raid_disks && !ret ; i++) { |
Jens Axboe | 165125e | 2007-07-24 09:28:11 +0200 | [diff] [blame] | 36 | struct request_queue *q = bdev_get_queue(devlist[i]->bdev); |
NeilBrown | 26be34d | 2006-10-03 01:15:53 -0700 | [diff] [blame] | 37 | |
| 38 | ret |= bdi_congested(&q->backing_dev_info, bits); |
| 39 | } |
| 40 | return ret; |
| 41 | } |
| 42 | |
raz ben yehuda | 4699419 | 2009-06-16 17:00:54 +1000 | [diff] [blame] | 43 | /* |
| 44 | * inform the user of the raid configuration |
| 45 | */ |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 46 | static void dump_zones(struct mddev *mddev) |
raz ben yehuda | 4699419 | 2009-06-16 17:00:54 +1000 | [diff] [blame] | 47 | { |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 48 | int j, k; |
raz ben yehuda | 4699419 | 2009-06-16 17:00:54 +1000 | [diff] [blame] | 49 | sector_t zone_size = 0; |
| 50 | sector_t zone_start = 0; |
| 51 | char b[BDEVNAME_SIZE]; |
NeilBrown | e373ab1 | 2011-10-11 16:48:59 +1100 | [diff] [blame] | 52 | struct r0conf *conf = mddev->private; |
NeilBrown | 84707f3 | 2010-03-16 17:23:35 +1100 | [diff] [blame] | 53 | int raid_disks = conf->strip_zone[0].nb_dev; |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 54 | printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n", |
| 55 | mdname(mddev), |
| 56 | conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s"); |
raz ben yehuda | 4699419 | 2009-06-16 17:00:54 +1000 | [diff] [blame] | 57 | for (j = 0; j < conf->nr_strip_zones; j++) { |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 58 | printk(KERN_INFO "md: zone%d=[", j); |
raz ben yehuda | 4699419 | 2009-06-16 17:00:54 +1000 | [diff] [blame] | 59 | for (k = 0; k < conf->strip_zone[j].nb_dev; k++) |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 60 | printk(KERN_CONT "%s%s", k?"/":"", |
NeilBrown | 84707f3 | 2010-03-16 17:23:35 +1100 | [diff] [blame] | 61 | bdevname(conf->devlist[j*raid_disks |
raz ben yehuda | 4699419 | 2009-06-16 17:00:54 +1000 | [diff] [blame] | 62 | + k]->bdev, b)); |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 63 | printk(KERN_CONT "]\n"); |
raz ben yehuda | 4699419 | 2009-06-16 17:00:54 +1000 | [diff] [blame] | 64 | |
| 65 | zone_size = conf->strip_zone[j].zone_end - zone_start; |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 66 | printk(KERN_INFO " zone-offset=%10lluKB, " |
| 67 | "device-offset=%10lluKB, size=%10lluKB\n", |
raz ben yehuda | 4699419 | 2009-06-16 17:00:54 +1000 | [diff] [blame] | 68 | (unsigned long long)zone_start>>1, |
| 69 | (unsigned long long)conf->strip_zone[j].dev_start>>1, |
| 70 | (unsigned long long)zone_size>>1); |
| 71 | zone_start = conf->strip_zone[j].zone_end; |
| 72 | } |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 73 | printk(KERN_INFO "\n"); |
raz ben yehuda | 4699419 | 2009-06-16 17:00:54 +1000 | [diff] [blame] | 74 | } |
| 75 | |
NeilBrown | e373ab1 | 2011-10-11 16:48:59 +1100 | [diff] [blame] | 76 | static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 77 | { |
NeilBrown | a9f326e | 2009-09-23 18:06:41 +1000 | [diff] [blame] | 78 | int i, c, err; |
NeilBrown | 49f357a2 | 2009-06-16 16:50:35 +1000 | [diff] [blame] | 79 | sector_t curr_zone_end, sectors; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 80 | struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 81 | struct strip_zone *zone; |
| 82 | int cnt; |
| 83 | char b[BDEVNAME_SIZE]; |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 84 | char b2[BDEVNAME_SIZE]; |
NeilBrown | e373ab1 | 2011-10-11 16:48:59 +1100 | [diff] [blame] | 85 | struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL); |
NeilBrown | 199dc6e | 2015-08-03 13:11:47 +1000 | [diff] [blame^] | 86 | unsigned short blksize = 512; |
Andre Noll | ed7b003 | 2009-06-16 16:47:36 +1000 | [diff] [blame] | 87 | |
| 88 | if (!conf) |
| 89 | return -ENOMEM; |
NeilBrown | dafb20f | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 90 | rdev_for_each(rdev1, mddev) { |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 91 | pr_debug("md/raid0:%s: looking at %s\n", |
| 92 | mdname(mddev), |
| 93 | bdevname(rdev1->bdev, b)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | c = 0; |
NeilBrown | 13f2682 | 2009-06-18 08:48:55 +1000 | [diff] [blame] | 95 | |
| 96 | /* round size to chunk_size */ |
| 97 | sectors = rdev1->sectors; |
| 98 | sector_div(sectors, mddev->chunk_sectors); |
| 99 | rdev1->sectors = sectors * mddev->chunk_sectors; |
| 100 | |
NeilBrown | 199dc6e | 2015-08-03 13:11:47 +1000 | [diff] [blame^] | 101 | blksize = max(blksize, queue_logical_block_size( |
| 102 | rdev1->bdev->bd_disk->queue)); |
| 103 | |
NeilBrown | dafb20f | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 104 | rdev_for_each(rdev2, mddev) { |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 105 | pr_debug("md/raid0:%s: comparing %s(%llu)" |
| 106 | " with %s(%llu)\n", |
| 107 | mdname(mddev), |
| 108 | bdevname(rdev1->bdev,b), |
| 109 | (unsigned long long)rdev1->sectors, |
| 110 | bdevname(rdev2->bdev,b2), |
| 111 | (unsigned long long)rdev2->sectors); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | if (rdev2 == rdev1) { |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 113 | pr_debug("md/raid0:%s: END\n", |
| 114 | mdname(mddev)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | break; |
| 116 | } |
Andre Noll | dd8ac33 | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 117 | if (rdev2->sectors == rdev1->sectors) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 118 | /* |
| 119 | * Not unique, don't count it as a new |
| 120 | * group |
| 121 | */ |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 122 | pr_debug("md/raid0:%s: EQUAL\n", |
| 123 | mdname(mddev)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | c = 1; |
| 125 | break; |
| 126 | } |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 127 | pr_debug("md/raid0:%s: NOT EQUAL\n", |
| 128 | mdname(mddev)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 129 | } |
| 130 | if (!c) { |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 131 | pr_debug("md/raid0:%s: ==> UNIQUE\n", |
| 132 | mdname(mddev)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | conf->nr_strip_zones++; |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 134 | pr_debug("md/raid0:%s: %d zones\n", |
| 135 | mdname(mddev), conf->nr_strip_zones); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 136 | } |
| 137 | } |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 138 | pr_debug("md/raid0:%s: FINAL %d zones\n", |
| 139 | mdname(mddev), conf->nr_strip_zones); |
NeilBrown | 199dc6e | 2015-08-03 13:11:47 +1000 | [diff] [blame^] | 140 | /* |
| 141 | * now since we have the hard sector sizes, we can make sure |
| 142 | * chunk size is a multiple of that sector size |
| 143 | */ |
| 144 | if ((mddev->chunk_sectors << 9) % blksize) { |
| 145 | printk(KERN_ERR "md/raid0:%s: chunk_size of %d not multiple of block size %d\n", |
| 146 | mdname(mddev), |
| 147 | mddev->chunk_sectors << 9, blksize); |
| 148 | err = -EINVAL; |
| 149 | goto abort; |
| 150 | } |
| 151 | |
Andre Noll | ed7b003 | 2009-06-16 16:47:36 +1000 | [diff] [blame] | 152 | err = -ENOMEM; |
NeilBrown | 9ffae0c | 2006-01-06 00:20:32 -0800 | [diff] [blame] | 153 | conf->strip_zone = kzalloc(sizeof(struct strip_zone)* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 154 | conf->nr_strip_zones, GFP_KERNEL); |
| 155 | if (!conf->strip_zone) |
Andre Noll | ed7b003 | 2009-06-16 16:47:36 +1000 | [diff] [blame] | 156 | goto abort; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 157 | conf->devlist = kzalloc(sizeof(struct md_rdev*)* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 158 | conf->nr_strip_zones*mddev->raid_disks, |
| 159 | GFP_KERNEL); |
| 160 | if (!conf->devlist) |
Andre Noll | ed7b003 | 2009-06-16 16:47:36 +1000 | [diff] [blame] | 161 | goto abort; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 162 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 163 | /* The first zone must contain all devices, so here we check that |
| 164 | * there is a proper alignment of slots to devices and find them all |
| 165 | */ |
| 166 | zone = &conf->strip_zone[0]; |
| 167 | cnt = 0; |
| 168 | smallest = NULL; |
NeilBrown | b414579 | 2009-06-16 16:50:52 +1000 | [diff] [blame] | 169 | dev = conf->devlist; |
Andre Noll | ed7b003 | 2009-06-16 16:47:36 +1000 | [diff] [blame] | 170 | err = -EINVAL; |
NeilBrown | dafb20f | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 171 | rdev_for_each(rdev1, mddev) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | int j = rdev1->raid_disk; |
| 173 | |
NeilBrown | e93f68a | 2010-06-15 09:36:03 +0100 | [diff] [blame] | 174 | if (mddev->level == 10) { |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 175 | /* taking over a raid10-n2 array */ |
| 176 | j /= 2; |
NeilBrown | e93f68a | 2010-06-15 09:36:03 +0100 | [diff] [blame] | 177 | rdev1->new_raid_disk = j; |
| 178 | } |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 179 | |
Krzysztof Wojcik | fc3a08b | 2011-01-31 13:47:13 +1100 | [diff] [blame] | 180 | if (mddev->level == 1) { |
| 181 | /* taiking over a raid1 array- |
| 182 | * we have only one active disk |
| 183 | */ |
| 184 | j = 0; |
| 185 | rdev1->new_raid_disk = j; |
| 186 | } |
| 187 | |
NeilBrown | f96c9f3 | 2013-02-21 15:50:07 +1100 | [diff] [blame] | 188 | if (j < 0) { |
| 189 | printk(KERN_ERR |
| 190 | "md/raid0:%s: remove inactive devices before converting to RAID0\n", |
| 191 | mdname(mddev)); |
| 192 | goto abort; |
| 193 | } |
| 194 | if (j >= mddev->raid_disks) { |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 195 | printk(KERN_ERR "md/raid0:%s: bad disk number %d - " |
| 196 | "aborting!\n", mdname(mddev), j); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | goto abort; |
| 198 | } |
NeilBrown | b414579 | 2009-06-16 16:50:52 +1000 | [diff] [blame] | 199 | if (dev[j]) { |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 200 | printk(KERN_ERR "md/raid0:%s: multiple devices for %d - " |
| 201 | "aborting!\n", mdname(mddev), j); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | goto abort; |
| 203 | } |
NeilBrown | b414579 | 2009-06-16 16:50:52 +1000 | [diff] [blame] | 204 | dev[j] = rdev1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | |
NeilBrown | ba13da4 | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 206 | if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) |
| 207 | conf->has_merge_bvec = 1; |
| 208 | |
Andre Noll | dd8ac33 | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 209 | if (!smallest || (rdev1->sectors < smallest->sectors)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | smallest = rdev1; |
| 211 | cnt++; |
| 212 | } |
| 213 | if (cnt != mddev->raid_disks) { |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 214 | printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - " |
| 215 | "aborting!\n", mdname(mddev), cnt, mddev->raid_disks); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 216 | goto abort; |
| 217 | } |
| 218 | zone->nb_dev = cnt; |
NeilBrown | 49f357a2 | 2009-06-16 16:50:35 +1000 | [diff] [blame] | 219 | zone->zone_end = smallest->sectors * cnt; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 220 | |
NeilBrown | 49f357a2 | 2009-06-16 16:50:35 +1000 | [diff] [blame] | 221 | curr_zone_end = zone->zone_end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 222 | |
| 223 | /* now do the other zones */ |
| 224 | for (i = 1; i < conf->nr_strip_zones; i++) |
| 225 | { |
NeilBrown | a9f326e | 2009-09-23 18:06:41 +1000 | [diff] [blame] | 226 | int j; |
| 227 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 228 | zone = conf->strip_zone + i; |
NeilBrown | b414579 | 2009-06-16 16:50:52 +1000 | [diff] [blame] | 229 | dev = conf->devlist + i * mddev->raid_disks; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 230 | |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 231 | pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i); |
NeilBrown | d27a43ab | 2009-06-16 16:46:46 +1000 | [diff] [blame] | 232 | zone->dev_start = smallest->sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 233 | smallest = NULL; |
| 234 | c = 0; |
| 235 | |
| 236 | for (j=0; j<cnt; j++) { |
NeilBrown | b414579 | 2009-06-16 16:50:52 +1000 | [diff] [blame] | 237 | rdev = conf->devlist[j]; |
NeilBrown | d27a43ab | 2009-06-16 16:46:46 +1000 | [diff] [blame] | 238 | if (rdev->sectors <= zone->dev_start) { |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 239 | pr_debug("md/raid0:%s: checking %s ... nope\n", |
| 240 | mdname(mddev), |
| 241 | bdevname(rdev->bdev, b)); |
Andre Noll | dd8ac33 | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 242 | continue; |
| 243 | } |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 244 | pr_debug("md/raid0:%s: checking %s ..." |
| 245 | " contained as device %d\n", |
| 246 | mdname(mddev), |
| 247 | bdevname(rdev->bdev, b), c); |
NeilBrown | b414579 | 2009-06-16 16:50:52 +1000 | [diff] [blame] | 248 | dev[c] = rdev; |
Andre Noll | dd8ac33 | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 249 | c++; |
| 250 | if (!smallest || rdev->sectors < smallest->sectors) { |
| 251 | smallest = rdev; |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 252 | pr_debug("md/raid0:%s: (%llu) is smallest!.\n", |
| 253 | mdname(mddev), |
| 254 | (unsigned long long)rdev->sectors); |
Andre Noll | dd8ac33 | 2009-03-31 14:33:13 +1100 | [diff] [blame] | 255 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 256 | } |
| 257 | |
| 258 | zone->nb_dev = c; |
NeilBrown | 49f357a2 | 2009-06-16 16:50:35 +1000 | [diff] [blame] | 259 | sectors = (smallest->sectors - zone->dev_start) * c; |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 260 | pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n", |
| 261 | mdname(mddev), |
| 262 | zone->nb_dev, (unsigned long long)sectors); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 263 | |
NeilBrown | 49f357a2 | 2009-06-16 16:50:35 +1000 | [diff] [blame] | 264 | curr_zone_end += sectors; |
NeilBrown | d27a43ab | 2009-06-16 16:46:46 +1000 | [diff] [blame] | 265 | zone->zone_end = curr_zone_end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 267 | pr_debug("md/raid0:%s: current zone start: %llu\n", |
| 268 | mdname(mddev), |
| 269 | (unsigned long long)smallest->sectors); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 270 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | |
NeilBrown | 50de8df | 2011-10-07 14:23:22 +1100 | [diff] [blame] | 272 | pr_debug("md/raid0:%s: done.\n", mdname(mddev)); |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 273 | *private_conf = conf; |
| 274 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | return 0; |
Andre Noll | 5568a60 | 2009-06-16 16:47:21 +1000 | [diff] [blame] | 276 | abort: |
Andre Noll | ed7b003 | 2009-06-16 16:47:36 +1000 | [diff] [blame] | 277 | kfree(conf->strip_zone); |
| 278 | kfree(conf->devlist); |
| 279 | kfree(conf); |
NeilBrown | 58ebb34 | 2013-02-21 15:36:38 +1100 | [diff] [blame] | 280 | *private_conf = ERR_PTR(err); |
Andre Noll | ed7b003 | 2009-06-16 16:47:36 +1000 | [diff] [blame] | 281 | return err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 282 | } |
| 283 | |
NeilBrown | ba13da4 | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 284 | /* Find the zone which holds a particular offset |
| 285 | * Update *sectorp to be an offset in that zone |
| 286 | */ |
| 287 | static struct strip_zone *find_zone(struct r0conf *conf, |
| 288 | sector_t *sectorp) |
| 289 | { |
| 290 | int i; |
| 291 | struct strip_zone *z = conf->strip_zone; |
| 292 | sector_t sector = *sectorp; |
| 293 | |
| 294 | for (i = 0; i < conf->nr_strip_zones; i++) |
| 295 | if (sector < z[i].zone_end) { |
| 296 | if (i) |
| 297 | *sectorp = sector - z[i-1].zone_end; |
| 298 | return z + i; |
| 299 | } |
| 300 | BUG(); |
| 301 | } |
| 302 | |
| 303 | /* |
| 304 | * remaps the bio to the target device. we separate two flows. |
NeilBrown | 47d6897 | 2015-04-10 13:19:04 +1000 | [diff] [blame] | 305 | * power 2 flow and a general flow for the sake of performance |
NeilBrown | ba13da4 | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 306 | */ |
| 307 | static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone, |
| 308 | sector_t sector, sector_t *sector_offset) |
| 309 | { |
| 310 | unsigned int sect_in_chunk; |
| 311 | sector_t chunk; |
| 312 | struct r0conf *conf = mddev->private; |
| 313 | int raid_disks = conf->strip_zone[0].nb_dev; |
| 314 | unsigned int chunk_sects = mddev->chunk_sectors; |
| 315 | |
| 316 | if (is_power_of_2(chunk_sects)) { |
| 317 | int chunksect_bits = ffz(~chunk_sects); |
| 318 | /* find the sector offset inside the chunk */ |
| 319 | sect_in_chunk = sector & (chunk_sects - 1); |
| 320 | sector >>= chunksect_bits; |
| 321 | /* chunk in zone */ |
| 322 | chunk = *sector_offset; |
| 323 | /* quotient is the chunk in real device*/ |
| 324 | sector_div(chunk, zone->nb_dev << chunksect_bits); |
| 325 | } else{ |
| 326 | sect_in_chunk = sector_div(sector, chunk_sects); |
| 327 | chunk = *sector_offset; |
| 328 | sector_div(chunk, chunk_sects * zone->nb_dev); |
| 329 | } |
| 330 | /* |
| 331 | * position the bio over the real device |
| 332 | * real sector = chunk in device + starting of zone |
| 333 | * + the position in the chunk |
| 334 | */ |
| 335 | *sector_offset = (chunk * chunk_sects) + sect_in_chunk; |
| 336 | return conf->devlist[(zone - conf->strip_zone)*raid_disks |
| 337 | + sector_div(sector, zone->nb_dev)]; |
| 338 | } |
| 339 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 340 | /** |
NeilBrown | ba13da4 | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 341 | * raid0_mergeable_bvec -- tell bio layer if two requests can be merged |
NeilBrown | 64590f4 | 2014-12-15 12:56:57 +1100 | [diff] [blame] | 342 | * @mddev: the md device |
Alasdair G Kergon | cc371e6 | 2008-07-03 09:53:43 +0200 | [diff] [blame] | 343 | * @bvm: properties of new bio |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 344 | * @biovec: the request that could be merged to it. |
| 345 | * |
| 346 | * Return amount of bytes we can accept at this offset |
| 347 | */ |
NeilBrown | 64590f4 | 2014-12-15 12:56:57 +1100 | [diff] [blame] | 348 | static int raid0_mergeable_bvec(struct mddev *mddev, |
Alasdair G Kergon | cc371e6 | 2008-07-03 09:53:43 +0200 | [diff] [blame] | 349 | struct bvec_merge_data *bvm, |
| 350 | struct bio_vec *biovec) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 351 | { |
NeilBrown | ba13da4 | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 352 | struct r0conf *conf = mddev->private; |
Alasdair G Kergon | cc371e6 | 2008-07-03 09:53:43 +0200 | [diff] [blame] | 353 | sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev); |
NeilBrown | ba13da4 | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 354 | sector_t sector_offset = sector; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 355 | int max; |
Andre Noll | 9d8f036 | 2009-06-18 08:45:01 +1000 | [diff] [blame] | 356 | unsigned int chunk_sectors = mddev->chunk_sectors; |
Alasdair G Kergon | cc371e6 | 2008-07-03 09:53:43 +0200 | [diff] [blame] | 357 | unsigned int bio_sectors = bvm->bi_size >> 9; |
NeilBrown | ba13da4 | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 358 | struct strip_zone *zone; |
| 359 | struct md_rdev *rdev; |
| 360 | struct request_queue *subq; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 361 | |
NeilBrown | d6e412e | 2009-06-18 08:47:00 +1000 | [diff] [blame] | 362 | if (is_power_of_2(chunk_sectors)) |
raz ben yehuda | fbb704e | 2009-06-16 17:02:05 +1000 | [diff] [blame] | 363 | max = (chunk_sectors - ((sector & (chunk_sectors-1)) |
| 364 | + bio_sectors)) << 9; |
| 365 | else |
| 366 | max = (chunk_sectors - (sector_div(sector, chunk_sectors) |
| 367 | + bio_sectors)) << 9; |
NeilBrown | ba13da4 | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 368 | if (max < 0) |
| 369 | max = 0; /* bio_add cannot handle a negative return */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 370 | if (max <= biovec->bv_len && bio_sectors == 0) |
| 371 | return biovec->bv_len; |
NeilBrown | ba13da4 | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 372 | if (max < biovec->bv_len) |
| 373 | /* too small already, no need to check further */ |
| 374 | return max; |
| 375 | if (!conf->has_merge_bvec) |
| 376 | return max; |
| 377 | |
| 378 | /* May need to check subordinate device */ |
| 379 | sector = sector_offset; |
| 380 | zone = find_zone(mddev->private, §or_offset); |
| 381 | rdev = map_sector(mddev, zone, sector, §or_offset); |
| 382 | subq = bdev_get_queue(rdev->bdev); |
| 383 | if (subq->merge_bvec_fn) { |
| 384 | bvm->bi_bdev = rdev->bdev; |
| 385 | bvm->bi_sector = sector_offset + zone->dev_start + |
| 386 | rdev->data_offset; |
| 387 | return min(max, subq->merge_bvec_fn(subq, bvm, biovec)); |
| 388 | } else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 389 | return max; |
| 390 | } |
| 391 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 392 | static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
Dan Williams | 80c3a6c | 2009-03-17 18:10:40 -0700 | [diff] [blame] | 393 | { |
| 394 | sector_t array_sectors = 0; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 395 | struct md_rdev *rdev; |
Dan Williams | 80c3a6c | 2009-03-17 18:10:40 -0700 | [diff] [blame] | 396 | |
| 397 | WARN_ONCE(sectors || raid_disks, |
| 398 | "%s does not support generic reshape\n", __func__); |
| 399 | |
NeilBrown | dafb20f | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 400 | rdev_for_each(rdev, mddev) |
NeilBrown | a646853 | 2013-02-21 14:33:17 +1100 | [diff] [blame] | 401 | array_sectors += (rdev->sectors & |
| 402 | ~(sector_t)(mddev->chunk_sectors-1)); |
Dan Williams | 80c3a6c | 2009-03-17 18:10:40 -0700 | [diff] [blame] | 403 | |
| 404 | return array_sectors; |
| 405 | } |
| 406 | |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 407 | static void raid0_free(struct mddev *mddev, void *priv); |
majianpeng | 0366ef8 | 2012-04-02 09:48:37 +1000 | [diff] [blame] | 408 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 409 | static int raid0_run(struct mddev *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 410 | { |
NeilBrown | e373ab1 | 2011-10-11 16:48:59 +1100 | [diff] [blame] | 411 | struct r0conf *conf; |
Andre Noll | 5568a60 | 2009-06-16 16:47:21 +1000 | [diff] [blame] | 412 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 413 | |
Andre Noll | 9d8f036 | 2009-06-18 08:45:01 +1000 | [diff] [blame] | 414 | if (mddev->chunk_sectors == 0) { |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 415 | printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n", |
| 416 | mdname(mddev)); |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 417 | return -EINVAL; |
| 418 | } |
Andre Noll | 0894cc3 | 2009-06-18 08:49:23 +1000 | [diff] [blame] | 419 | if (md_check_no_bitmap(mddev)) |
| 420 | return -EINVAL; |
Heinz Mauelshagen | 753f285 | 2015-02-13 19:48:01 +0100 | [diff] [blame] | 421 | |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 422 | /* if private is not null, we are here after takeover */ |
| 423 | if (mddev->private == NULL) { |
| 424 | ret = create_strip_zones(mddev, &conf); |
| 425 | if (ret < 0) |
| 426 | return ret; |
| 427 | mddev->private = conf; |
| 428 | } |
| 429 | conf = mddev->private; |
NeilBrown | 199dc6e | 2015-08-03 13:11:47 +1000 | [diff] [blame^] | 430 | if (mddev->queue) { |
| 431 | struct md_rdev *rdev; |
| 432 | bool discard_supported = false; |
| 433 | |
| 434 | rdev_for_each(rdev, mddev) { |
| 435 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
| 436 | rdev->data_offset << 9); |
| 437 | if (blk_queue_discard(bdev_get_queue(rdev->bdev))) |
| 438 | discard_supported = true; |
| 439 | } |
| 440 | blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors); |
| 441 | blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors); |
| 442 | blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors); |
| 443 | |
| 444 | blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9); |
| 445 | blk_queue_io_opt(mddev->queue, |
| 446 | (mddev->chunk_sectors << 9) * mddev->raid_disks); |
| 447 | |
| 448 | if (!discard_supported) |
| 449 | queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
| 450 | else |
| 451 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue); |
| 452 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | |
| 454 | /* calculate array device size */ |
Dan Williams | 1f40362 | 2009-03-31 14:59:03 +1100 | [diff] [blame] | 455 | md_set_array_sectors(mddev, raid0_size(mddev, 0, 0)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 457 | printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n", |
| 458 | mdname(mddev), |
| 459 | (unsigned long long)mddev->array_sectors); |
Heinz Mauelshagen | 753f285 | 2015-02-13 19:48:01 +0100 | [diff] [blame] | 460 | |
| 461 | if (mddev->queue) { |
| 462 | /* calculate the max read-ahead size. |
| 463 | * For read-ahead of large files to be effective, we need to |
| 464 | * readahead at least twice a whole stripe. i.e. number of devices |
| 465 | * multiplied by chunk size times 2. |
| 466 | * If an individual device has an ra_pages greater than the |
| 467 | * chunk size, then we will not drive that device as hard as it |
| 468 | * wants. We consider this a configuration error: a larger |
| 469 | * chunksize should be used in that case. |
| 470 | */ |
Andre Noll | 9d8f036 | 2009-06-18 08:45:01 +1000 | [diff] [blame] | 471 | int stripe = mddev->raid_disks * |
| 472 | (mddev->chunk_sectors << 9) / PAGE_SIZE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 473 | if (mddev->queue->backing_dev_info.ra_pages < 2* stripe) |
| 474 | mddev->queue->backing_dev_info.ra_pages = 2* stripe; |
| 475 | } |
| 476 | |
raz ben yehuda | 4699419 | 2009-06-16 17:00:54 +1000 | [diff] [blame] | 477 | dump_zones(mddev); |
majianpeng | 0366ef8 | 2012-04-02 09:48:37 +1000 | [diff] [blame] | 478 | |
| 479 | ret = md_integrity_register(mddev); |
majianpeng | 0366ef8 | 2012-04-02 09:48:37 +1000 | [diff] [blame] | 480 | |
| 481 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 482 | } |
| 483 | |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 484 | static void raid0_free(struct mddev *mddev, void *priv) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 485 | { |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 486 | struct r0conf *conf = priv; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 487 | |
Jesper Juhl | 990a8ba | 2005-06-21 17:17:30 -0700 | [diff] [blame] | 488 | kfree(conf->strip_zone); |
Andre Noll | fb5ab4b5 | 2009-06-16 16:48:19 +1000 | [diff] [blame] | 489 | kfree(conf->devlist); |
Jesper Juhl | 990a8ba | 2005-06-21 17:17:30 -0700 | [diff] [blame] | 490 | kfree(conf); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 491 | } |
| 492 | |
raz ben yehuda | fbb704e | 2009-06-16 17:02:05 +1000 | [diff] [blame] | 493 | /* |
| 494 | * Is io distribute over 1 or more chunks ? |
| 495 | */ |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 496 | static inline int is_io_in_chunk_boundary(struct mddev *mddev, |
raz ben yehuda | fbb704e | 2009-06-16 17:02:05 +1000 | [diff] [blame] | 497 | unsigned int chunk_sects, struct bio *bio) |
| 498 | { |
NeilBrown | d6e412e | 2009-06-18 08:47:00 +1000 | [diff] [blame] | 499 | if (likely(is_power_of_2(chunk_sects))) { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 500 | return chunk_sects >= |
| 501 | ((bio->bi_iter.bi_sector & (chunk_sects-1)) |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 502 | + bio_sectors(bio)); |
raz ben yehuda | fbb704e | 2009-06-16 17:02:05 +1000 | [diff] [blame] | 503 | } else{ |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 504 | sector_t sector = bio->bi_iter.bi_sector; |
raz ben yehuda | fbb704e | 2009-06-16 17:02:05 +1000 | [diff] [blame] | 505 | return chunk_sects >= (sector_div(sector, chunk_sects) |
Kent Overstreet | aa8b57a | 2013-02-05 15:19:29 -0800 | [diff] [blame] | 506 | + bio_sectors(bio)); |
raz ben yehuda | fbb704e | 2009-06-16 17:02:05 +1000 | [diff] [blame] | 507 | } |
| 508 | } |
| 509 | |
Linus Torvalds | b4fdcb0 | 2011-11-04 17:06:58 -0700 | [diff] [blame] | 510 | static void raid0_make_request(struct mddev *mddev, struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 511 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 512 | struct strip_zone *zone; |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 513 | struct md_rdev *tmp_dev; |
Kent Overstreet | 20d0189 | 2013-11-23 18:21:01 -0800 | [diff] [blame] | 514 | struct bio *split; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 515 | |
Tejun Heo | e9c7469 | 2010-09-03 11:56:18 +0200 | [diff] [blame] | 516 | if (unlikely(bio->bi_rw & REQ_FLUSH)) { |
| 517 | md_flush_request(mddev, bio); |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 518 | return; |
NeilBrown | e5dcdd8 | 2005-09-09 16:23:41 -0700 | [diff] [blame] | 519 | } |
| 520 | |
Kent Overstreet | 20d0189 | 2013-11-23 18:21:01 -0800 | [diff] [blame] | 521 | do { |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 522 | sector_t sector = bio->bi_iter.bi_sector; |
Kent Overstreet | 20d0189 | 2013-11-23 18:21:01 -0800 | [diff] [blame] | 523 | unsigned chunk_sects = mddev->chunk_sectors; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 524 | |
Kent Overstreet | 20d0189 | 2013-11-23 18:21:01 -0800 | [diff] [blame] | 525 | unsigned sectors = chunk_sects - |
| 526 | (likely(is_power_of_2(chunk_sects)) |
| 527 | ? (sector & (chunk_sects-1)) |
| 528 | : sector_div(sector, chunk_sects)); |
Christoph Hellwig | 5a7bbad | 2011-09-12 12:12:01 +0200 | [diff] [blame] | 529 | |
Eric Work | a811577 | 2015-05-18 23:26:23 -0700 | [diff] [blame] | 530 | /* Restore due to sector_div */ |
| 531 | sector = bio->bi_iter.bi_sector; |
| 532 | |
Kent Overstreet | 20d0189 | 2013-11-23 18:21:01 -0800 | [diff] [blame] | 533 | if (sectors < bio_sectors(bio)) { |
| 534 | split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set); |
| 535 | bio_chain(split, bio); |
| 536 | } else { |
| 537 | split = bio; |
| 538 | } |
Shaohua Li | c83057a | 2012-10-11 13:25:44 +1100 | [diff] [blame] | 539 | |
Kent Overstreet | 20d0189 | 2013-11-23 18:21:01 -0800 | [diff] [blame] | 540 | zone = find_zone(mddev->private, §or); |
| 541 | tmp_dev = map_sector(mddev, zone, sector, §or); |
| 542 | split->bi_bdev = tmp_dev->bdev; |
| 543 | split->bi_iter.bi_sector = sector + zone->dev_start + |
| 544 | tmp_dev->data_offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 545 | |
Kent Overstreet | 20d0189 | 2013-11-23 18:21:01 -0800 | [diff] [blame] | 546 | if (unlikely((split->bi_rw & REQ_DISCARD) && |
| 547 | !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) { |
| 548 | /* Just ignore it */ |
| 549 | bio_endio(split, 0); |
| 550 | } else |
| 551 | generic_make_request(split); |
| 552 | } while (split != bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 553 | } |
NeilBrown | 8299d7f | 2007-10-16 23:30:53 -0700 | [diff] [blame] | 554 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 555 | static void raid0_status(struct seq_file *seq, struct mddev *mddev) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 556 | { |
Andre Noll | 9d8f036 | 2009-06-18 08:45:01 +1000 | [diff] [blame] | 557 | seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 558 | return; |
| 559 | } |
| 560 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 561 | static void *raid0_takeover_raid45(struct mddev *mddev) |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 562 | { |
NeilBrown | 3cb0300 | 2011-10-11 16:45:26 +1100 | [diff] [blame] | 563 | struct md_rdev *rdev; |
NeilBrown | e373ab1 | 2011-10-11 16:48:59 +1100 | [diff] [blame] | 564 | struct r0conf *priv_conf; |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 565 | |
| 566 | if (mddev->degraded != 1) { |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 567 | printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n", |
| 568 | mdname(mddev), |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 569 | mddev->degraded); |
| 570 | return ERR_PTR(-EINVAL); |
| 571 | } |
| 572 | |
NeilBrown | dafb20f | 2012-03-19 12:46:39 +1100 | [diff] [blame] | 573 | rdev_for_each(rdev, mddev) { |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 574 | /* check slot number for a disk */ |
| 575 | if (rdev->raid_disk == mddev->raid_disks-1) { |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 576 | printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n", |
| 577 | mdname(mddev)); |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 578 | return ERR_PTR(-EINVAL); |
| 579 | } |
NeilBrown | eea136d | 2013-06-26 11:55:20 +1000 | [diff] [blame] | 580 | rdev->sectors = mddev->dev_sectors; |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 581 | } |
| 582 | |
| 583 | /* Set new parameters */ |
| 584 | mddev->new_level = 0; |
Maciej Trela | 001048a | 2010-06-16 11:55:14 +0100 | [diff] [blame] | 585 | mddev->new_layout = 0; |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 586 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
| 587 | mddev->raid_disks--; |
| 588 | mddev->delta_disks = -1; |
| 589 | /* make sure it will be not marked as dirty */ |
| 590 | mddev->recovery_cp = MaxSector; |
| 591 | |
| 592 | create_strip_zones(mddev, &priv_conf); |
| 593 | return priv_conf; |
| 594 | } |
| 595 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 596 | static void *raid0_takeover_raid10(struct mddev *mddev) |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 597 | { |
NeilBrown | e373ab1 | 2011-10-11 16:48:59 +1100 | [diff] [blame] | 598 | struct r0conf *priv_conf; |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 599 | |
| 600 | /* Check layout: |
| 601 | * - far_copies must be 1 |
| 602 | * - near_copies must be 2 |
| 603 | * - disks number must be even |
| 604 | * - all mirrors must be already degraded |
| 605 | */ |
| 606 | if (mddev->layout != ((1 << 8) + 2)) { |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 607 | printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n", |
| 608 | mdname(mddev), |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 609 | mddev->layout); |
| 610 | return ERR_PTR(-EINVAL); |
| 611 | } |
| 612 | if (mddev->raid_disks & 1) { |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 613 | printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n", |
| 614 | mdname(mddev)); |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 615 | return ERR_PTR(-EINVAL); |
| 616 | } |
| 617 | if (mddev->degraded != (mddev->raid_disks>>1)) { |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 618 | printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n", |
| 619 | mdname(mddev)); |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 620 | return ERR_PTR(-EINVAL); |
| 621 | } |
| 622 | |
| 623 | /* Set new parameters */ |
| 624 | mddev->new_level = 0; |
Maciej Trela | 001048a | 2010-06-16 11:55:14 +0100 | [diff] [blame] | 625 | mddev->new_layout = 0; |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 626 | mddev->new_chunk_sectors = mddev->chunk_sectors; |
| 627 | mddev->delta_disks = - mddev->raid_disks / 2; |
| 628 | mddev->raid_disks += mddev->delta_disks; |
| 629 | mddev->degraded = 0; |
| 630 | /* make sure it will be not marked as dirty */ |
| 631 | mddev->recovery_cp = MaxSector; |
| 632 | |
| 633 | create_strip_zones(mddev, &priv_conf); |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 634 | return priv_conf; |
| 635 | } |
| 636 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 637 | static void *raid0_takeover_raid1(struct mddev *mddev) |
Krzysztof Wojcik | fc3a08b | 2011-01-31 13:47:13 +1100 | [diff] [blame] | 638 | { |
NeilBrown | e373ab1 | 2011-10-11 16:48:59 +1100 | [diff] [blame] | 639 | struct r0conf *priv_conf; |
Jes Sorensen | 24b961f | 2012-04-01 23:48:38 +1000 | [diff] [blame] | 640 | int chunksect; |
Krzysztof Wojcik | fc3a08b | 2011-01-31 13:47:13 +1100 | [diff] [blame] | 641 | |
| 642 | /* Check layout: |
| 643 | * - (N - 1) mirror drives must be already faulty |
| 644 | */ |
| 645 | if ((mddev->raid_disks - 1) != mddev->degraded) { |
| 646 | printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n", |
| 647 | mdname(mddev)); |
| 648 | return ERR_PTR(-EINVAL); |
| 649 | } |
| 650 | |
Jes Sorensen | 24b961f | 2012-04-01 23:48:38 +1000 | [diff] [blame] | 651 | /* |
| 652 | * a raid1 doesn't have the notion of chunk size, so |
| 653 | * figure out the largest suitable size we can use. |
| 654 | */ |
| 655 | chunksect = 64 * 2; /* 64K by default */ |
| 656 | |
| 657 | /* The array must be an exact multiple of chunksize */ |
| 658 | while (chunksect && (mddev->array_sectors & (chunksect - 1))) |
| 659 | chunksect >>= 1; |
| 660 | |
| 661 | if ((chunksect << 9) < PAGE_SIZE) |
| 662 | /* array size does not allow a suitable chunk size */ |
| 663 | return ERR_PTR(-EINVAL); |
| 664 | |
Krzysztof Wojcik | fc3a08b | 2011-01-31 13:47:13 +1100 | [diff] [blame] | 665 | /* Set new parameters */ |
| 666 | mddev->new_level = 0; |
| 667 | mddev->new_layout = 0; |
Jes Sorensen | 24b961f | 2012-04-01 23:48:38 +1000 | [diff] [blame] | 668 | mddev->new_chunk_sectors = chunksect; |
| 669 | mddev->chunk_sectors = chunksect; |
Krzysztof Wojcik | fc3a08b | 2011-01-31 13:47:13 +1100 | [diff] [blame] | 670 | mddev->delta_disks = 1 - mddev->raid_disks; |
Krzysztof Wojcik | f7bee80 | 2011-02-14 10:01:41 +1100 | [diff] [blame] | 671 | mddev->raid_disks = 1; |
Krzysztof Wojcik | fc3a08b | 2011-01-31 13:47:13 +1100 | [diff] [blame] | 672 | /* make sure it will be not marked as dirty */ |
| 673 | mddev->recovery_cp = MaxSector; |
| 674 | |
| 675 | create_strip_zones(mddev, &priv_conf); |
| 676 | return priv_conf; |
| 677 | } |
| 678 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 679 | static void *raid0_takeover(struct mddev *mddev) |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 680 | { |
| 681 | /* raid0 can take over: |
Maciej Trela | 049d6c1 | 2010-06-16 11:56:12 +0100 | [diff] [blame] | 682 | * raid4 - if all data disks are active. |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 683 | * raid5 - providing it is Raid4 layout and one disk is faulty |
| 684 | * raid10 - assuming we have all necessary active disks |
Krzysztof Wojcik | fc3a08b | 2011-01-31 13:47:13 +1100 | [diff] [blame] | 685 | * raid1 - with (N -1) mirror drives faulty |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 686 | */ |
NeilBrown | a8461a6 | 2014-08-06 16:34:27 +1000 | [diff] [blame] | 687 | |
| 688 | if (mddev->bitmap) { |
| 689 | printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n", |
| 690 | mdname(mddev)); |
| 691 | return ERR_PTR(-EBUSY); |
| 692 | } |
Maciej Trela | 049d6c1 | 2010-06-16 11:56:12 +0100 | [diff] [blame] | 693 | if (mddev->level == 4) |
| 694 | return raid0_takeover_raid45(mddev); |
| 695 | |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 696 | if (mddev->level == 5) { |
| 697 | if (mddev->layout == ALGORITHM_PARITY_N) |
Maciej Trela | 049d6c1 | 2010-06-16 11:56:12 +0100 | [diff] [blame] | 698 | return raid0_takeover_raid45(mddev); |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 699 | |
NeilBrown | b5a2096 | 2010-05-03 15:06:27 +1000 | [diff] [blame] | 700 | printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n", |
| 701 | mdname(mddev), ALGORITHM_PARITY_N); |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 702 | } |
| 703 | |
| 704 | if (mddev->level == 10) |
| 705 | return raid0_takeover_raid10(mddev); |
| 706 | |
Krzysztof Wojcik | fc3a08b | 2011-01-31 13:47:13 +1100 | [diff] [blame] | 707 | if (mddev->level == 1) |
| 708 | return raid0_takeover_raid1(mddev); |
| 709 | |
| 710 | printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n", |
| 711 | mddev->level); |
| 712 | |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 713 | return ERR_PTR(-EINVAL); |
| 714 | } |
| 715 | |
NeilBrown | fd01b88 | 2011-10-11 16:47:53 +1100 | [diff] [blame] | 716 | static void raid0_quiesce(struct mddev *mddev, int state) |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 717 | { |
| 718 | } |
| 719 | |
NeilBrown | 84fc4b5 | 2011-10-11 16:49:58 +1100 | [diff] [blame] | 720 | static struct md_personality raid0_personality= |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 721 | { |
| 722 | .name = "raid0", |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 723 | .level = 0, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 724 | .owner = THIS_MODULE, |
| 725 | .make_request = raid0_make_request, |
| 726 | .run = raid0_run, |
NeilBrown | afa0f55 | 2014-12-15 12:56:58 +1100 | [diff] [blame] | 727 | .free = raid0_free, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 728 | .status = raid0_status, |
Dan Williams | 80c3a6c | 2009-03-17 18:10:40 -0700 | [diff] [blame] | 729 | .size = raid0_size, |
Trela, Maciej | 9af204c | 2010-03-08 16:02:44 +1100 | [diff] [blame] | 730 | .takeover = raid0_takeover, |
| 731 | .quiesce = raid0_quiesce, |
NeilBrown | 5c675f8 | 2014-12-15 12:56:56 +1100 | [diff] [blame] | 732 | .congested = raid0_congested, |
NeilBrown | 64590f4 | 2014-12-15 12:56:57 +1100 | [diff] [blame] | 733 | .mergeable_bvec = raid0_mergeable_bvec, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 734 | }; |
| 735 | |
| 736 | static int __init raid0_init (void) |
| 737 | { |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 738 | return register_md_personality (&raid0_personality); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 739 | } |
| 740 | |
| 741 | static void raid0_exit (void) |
| 742 | { |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 743 | unregister_md_personality (&raid0_personality); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 744 | } |
| 745 | |
| 746 | module_init(raid0_init); |
| 747 | module_exit(raid0_exit); |
| 748 | MODULE_LICENSE("GPL"); |
NeilBrown | 0efb9e6 | 2009-12-14 12:49:58 +1100 | [diff] [blame] | 749 | MODULE_DESCRIPTION("RAID0 (striping) personality for MD"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 750 | MODULE_ALIAS("md-personality-2"); /* RAID0 */ |
NeilBrown | d9d166c | 2006-01-06 00:20:51 -0800 | [diff] [blame] | 751 | MODULE_ALIAS("md-raid0"); |
NeilBrown | 2604b70 | 2006-01-06 00:20:36 -0800 | [diff] [blame] | 752 | MODULE_ALIAS("md-level-0"); |