blob: 34783a3c8b3c1af780a092b8d11018f702541481 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 raid0.c : Multiple Devices driver for Linux
NeilBrownf72ffdd2014-09-30 14:23:59 +10003 Copyright (C) 1994-96 Marc ZYNGIER
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 <zyngier@ufr-info-p7.ibp.fr> or
5 <maz@gloups.fdn.fr>
NeilBrownf72ffdd2014-09-30 14:23:59 +10006 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
Linus Torvalds1da177e2005-04-16 15:20:36 -07007
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.
NeilBrownf72ffdd2014-09-30 14:23:59 +100014
Linus Torvalds1da177e2005-04-16 15:20:36 -070015 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
NeilBrownf72ffdd2014-09-30 14:23:59 +100017 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
Linus Torvalds1da177e2005-04-16 15:20:36 -070018*/
19
NeilBrownbff61972009-03-31 14:33:13 +110020#include <linux/blkdev.h>
NeilBrownbff61972009-03-31 14:33:13 +110021#include <linux/seq_file.h>
Paul Gortmaker056075c2011-07-03 13:58:33 -040022#include <linux/module.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090023#include <linux/slab.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110024#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110025#include "raid0.h"
Trela, Maciej9af204c2010-03-08 16:02:44 +110026#include "raid5.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
NeilBrown5c675f82014-12-15 12:56:56 +110028static int raid0_congested(struct mddev *mddev, int bits)
NeilBrown26be34d2006-10-03 01:15:53 -070029{
NeilBrowne373ab12011-10-11 16:48:59 +110030 struct r0conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +110031 struct md_rdev **devlist = conf->devlist;
NeilBrown84707f32010-03-16 17:23:35 +110032 int raid_disks = conf->strip_zone[0].nb_dev;
NeilBrown26be34d2006-10-03 01:15:53 -070033 int i, ret = 0;
34
NeilBrown84707f32010-03-16 17:23:35 +110035 for (i = 0; i < raid_disks && !ret ; i++) {
Jens Axboe165125e2007-07-24 09:28:11 +020036 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
NeilBrown26be34d2006-10-03 01:15:53 -070037
38 ret |= bdi_congested(&q->backing_dev_info, bits);
39 }
40 return ret;
41}
42
raz ben yehuda46994192009-06-16 17:00:54 +100043/*
44 * inform the user of the raid configuration
45*/
NeilBrownfd01b882011-10-11 16:47:53 +110046static void dump_zones(struct mddev *mddev)
raz ben yehuda46994192009-06-16 17:00:54 +100047{
NeilBrown50de8df2011-10-07 14:23:22 +110048 int j, k;
raz ben yehuda46994192009-06-16 17:00:54 +100049 sector_t zone_size = 0;
50 sector_t zone_start = 0;
51 char b[BDEVNAME_SIZE];
NeilBrowne373ab12011-10-11 16:48:59 +110052 struct r0conf *conf = mddev->private;
NeilBrown84707f32010-03-16 17:23:35 +110053 int raid_disks = conf->strip_zone[0].nb_dev;
NeilBrown50de8df2011-10-07 14:23:22 +110054 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 yehuda46994192009-06-16 17:00:54 +100057 for (j = 0; j < conf->nr_strip_zones; j++) {
NeilBrown50de8df2011-10-07 14:23:22 +110058 printk(KERN_INFO "md: zone%d=[", j);
raz ben yehuda46994192009-06-16 17:00:54 +100059 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
NeilBrown50de8df2011-10-07 14:23:22 +110060 printk(KERN_CONT "%s%s", k?"/":"",
NeilBrown84707f32010-03-16 17:23:35 +110061 bdevname(conf->devlist[j*raid_disks
raz ben yehuda46994192009-06-16 17:00:54 +100062 + k]->bdev, b));
NeilBrownb5a20962010-05-03 15:06:27 +100063 printk(KERN_CONT "]\n");
raz ben yehuda46994192009-06-16 17:00:54 +100064
65 zone_size = conf->strip_zone[j].zone_end - zone_start;
NeilBrown50de8df2011-10-07 14:23:22 +110066 printk(KERN_INFO " zone-offset=%10lluKB, "
67 "device-offset=%10lluKB, size=%10lluKB\n",
raz ben yehuda46994192009-06-16 17:00:54 +100068 (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 }
raz ben yehuda46994192009-06-16 17:00:54 +100073}
74
NeilBrowne373ab12011-10-11 16:48:59 +110075static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076{
NeilBrowna9f326e2009-09-23 18:06:41 +100077 int i, c, err;
NeilBrown49f357a22009-06-16 16:50:35 +100078 sector_t curr_zone_end, sectors;
NeilBrown3cb03002011-10-11 16:45:26 +110079 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 struct strip_zone *zone;
81 int cnt;
82 char b[BDEVNAME_SIZE];
NeilBrown50de8df2011-10-07 14:23:22 +110083 char b2[BDEVNAME_SIZE];
NeilBrowne373ab12011-10-11 16:48:59 +110084 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
NeilBrown199dc6e2015-08-03 13:11:47 +100085 unsigned short blksize = 512;
Andre Nolled7b0032009-06-16 16:47:36 +100086
Dan Carpenter7dedd152016-04-14 12:31:49 +030087 *private_conf = ERR_PTR(-ENOMEM);
Andre Nolled7b0032009-06-16 16:47:36 +100088 if (!conf)
89 return -ENOMEM;
NeilBrowndafb20f2012-03-19 12:46:39 +110090 rdev_for_each(rdev1, mddev) {
NeilBrown50de8df2011-10-07 14:23:22 +110091 pr_debug("md/raid0:%s: looking at %s\n",
92 mdname(mddev),
93 bdevname(rdev1->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 c = 0;
NeilBrown13f26822009-06-18 08:48:55 +100095
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
NeilBrown199dc6e2015-08-03 13:11:47 +1000101 blksize = max(blksize, queue_logical_block_size(
102 rdev1->bdev->bd_disk->queue));
103
NeilBrowndafb20f2012-03-19 12:46:39 +1100104 rdev_for_each(rdev2, mddev) {
NeilBrown50de8df2011-10-07 14:23:22 +1100105 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 Torvalds1da177e2005-04-16 15:20:36 -0700112 if (rdev2 == rdev1) {
NeilBrown50de8df2011-10-07 14:23:22 +1100113 pr_debug("md/raid0:%s: END\n",
114 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700115 break;
116 }
Andre Nolldd8ac332009-03-31 14:33:13 +1100117 if (rdev2->sectors == rdev1->sectors) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 /*
119 * Not unique, don't count it as a new
120 * group
121 */
NeilBrown50de8df2011-10-07 14:23:22 +1100122 pr_debug("md/raid0:%s: EQUAL\n",
123 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 c = 1;
125 break;
126 }
NeilBrown50de8df2011-10-07 14:23:22 +1100127 pr_debug("md/raid0:%s: NOT EQUAL\n",
128 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 }
130 if (!c) {
NeilBrown50de8df2011-10-07 14:23:22 +1100131 pr_debug("md/raid0:%s: ==> UNIQUE\n",
132 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 conf->nr_strip_zones++;
NeilBrown50de8df2011-10-07 14:23:22 +1100134 pr_debug("md/raid0:%s: %d zones\n",
135 mdname(mddev), conf->nr_strip_zones);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 }
137 }
NeilBrown50de8df2011-10-07 14:23:22 +1100138 pr_debug("md/raid0:%s: FINAL %d zones\n",
139 mdname(mddev), conf->nr_strip_zones);
NeilBrown199dc6e2015-08-03 13:11:47 +1000140 /*
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 Nolled7b0032009-06-16 16:47:36 +1000152 err = -ENOMEM;
NeilBrown9ffae0c2006-01-06 00:20:32 -0800153 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 conf->nr_strip_zones, GFP_KERNEL);
155 if (!conf->strip_zone)
Andre Nolled7b0032009-06-16 16:47:36 +1000156 goto abort;
NeilBrown3cb03002011-10-11 16:45:26 +1100157 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 conf->nr_strip_zones*mddev->raid_disks,
159 GFP_KERNEL);
160 if (!conf->devlist)
Andre Nolled7b0032009-06-16 16:47:36 +1000161 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 /* 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;
NeilBrownb4145792009-06-16 16:50:52 +1000169 dev = conf->devlist;
Andre Nolled7b0032009-06-16 16:47:36 +1000170 err = -EINVAL;
NeilBrowndafb20f2012-03-19 12:46:39 +1100171 rdev_for_each(rdev1, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 int j = rdev1->raid_disk;
173
NeilBrowne93f68a2010-06-15 09:36:03 +0100174 if (mddev->level == 10) {
Trela, Maciej9af204c2010-03-08 16:02:44 +1100175 /* taking over a raid10-n2 array */
176 j /= 2;
NeilBrowne93f68a2010-06-15 09:36:03 +0100177 rdev1->new_raid_disk = j;
178 }
Trela, Maciej9af204c2010-03-08 16:02:44 +1100179
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100180 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
NeilBrownf96c9f32013-02-21 15:50:07 +1100188 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) {
NeilBrownb5a20962010-05-03 15:06:27 +1000195 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
196 "aborting!\n", mdname(mddev), j);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 goto abort;
198 }
NeilBrownb4145792009-06-16 16:50:52 +1000199 if (dev[j]) {
NeilBrownb5a20962010-05-03 15:06:27 +1000200 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
201 "aborting!\n", mdname(mddev), j);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 goto abort;
203 }
NeilBrownb4145792009-06-16 16:50:52 +1000204 dev[j] = rdev1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205
Andre Nolldd8ac332009-03-31 14:33:13 +1100206 if (!smallest || (rdev1->sectors < smallest->sectors))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 smallest = rdev1;
208 cnt++;
209 }
210 if (cnt != mddev->raid_disks) {
NeilBrownb5a20962010-05-03 15:06:27 +1000211 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
212 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 goto abort;
214 }
215 zone->nb_dev = cnt;
NeilBrown49f357a22009-06-16 16:50:35 +1000216 zone->zone_end = smallest->sectors * cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
NeilBrown49f357a22009-06-16 16:50:35 +1000218 curr_zone_end = zone->zone_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219
220 /* now do the other zones */
221 for (i = 1; i < conf->nr_strip_zones; i++)
222 {
NeilBrowna9f326e2009-09-23 18:06:41 +1000223 int j;
224
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 zone = conf->strip_zone + i;
NeilBrownb4145792009-06-16 16:50:52 +1000226 dev = conf->devlist + i * mddev->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
NeilBrown50de8df2011-10-07 14:23:22 +1100228 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
NeilBrownd27a43ab2009-06-16 16:46:46 +1000229 zone->dev_start = smallest->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 smallest = NULL;
231 c = 0;
232
233 for (j=0; j<cnt; j++) {
NeilBrownb4145792009-06-16 16:50:52 +1000234 rdev = conf->devlist[j];
NeilBrownd27a43ab2009-06-16 16:46:46 +1000235 if (rdev->sectors <= zone->dev_start) {
NeilBrown50de8df2011-10-07 14:23:22 +1100236 pr_debug("md/raid0:%s: checking %s ... nope\n",
237 mdname(mddev),
238 bdevname(rdev->bdev, b));
Andre Nolldd8ac332009-03-31 14:33:13 +1100239 continue;
240 }
NeilBrown50de8df2011-10-07 14:23:22 +1100241 pr_debug("md/raid0:%s: checking %s ..."
242 " contained as device %d\n",
243 mdname(mddev),
244 bdevname(rdev->bdev, b), c);
NeilBrownb4145792009-06-16 16:50:52 +1000245 dev[c] = rdev;
Andre Nolldd8ac332009-03-31 14:33:13 +1100246 c++;
247 if (!smallest || rdev->sectors < smallest->sectors) {
248 smallest = rdev;
NeilBrown50de8df2011-10-07 14:23:22 +1100249 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
250 mdname(mddev),
251 (unsigned long long)rdev->sectors);
Andre Nolldd8ac332009-03-31 14:33:13 +1100252 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 }
254
255 zone->nb_dev = c;
NeilBrown49f357a22009-06-16 16:50:35 +1000256 sectors = (smallest->sectors - zone->dev_start) * c;
NeilBrown50de8df2011-10-07 14:23:22 +1100257 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
258 mdname(mddev),
259 zone->nb_dev, (unsigned long long)sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
NeilBrown49f357a22009-06-16 16:50:35 +1000261 curr_zone_end += sectors;
NeilBrownd27a43ab2009-06-16 16:46:46 +1000262 zone->zone_end = curr_zone_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
NeilBrown50de8df2011-10-07 14:23:22 +1100264 pr_debug("md/raid0:%s: current zone start: %llu\n",
265 mdname(mddev),
266 (unsigned long long)smallest->sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
NeilBrown50de8df2011-10-07 14:23:22 +1100269 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100270 *private_conf = conf;
271
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 return 0;
Andre Noll5568a602009-06-16 16:47:21 +1000273abort:
Andre Nolled7b0032009-06-16 16:47:36 +1000274 kfree(conf->strip_zone);
275 kfree(conf->devlist);
276 kfree(conf);
NeilBrown58ebb342013-02-21 15:36:38 +1100277 *private_conf = ERR_PTR(err);
Andre Nolled7b0032009-06-16 16:47:36 +1000278 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279}
280
NeilBrownba13da42012-03-19 12:46:39 +1100281/* Find the zone which holds a particular offset
282 * Update *sectorp to be an offset in that zone
283 */
284static struct strip_zone *find_zone(struct r0conf *conf,
285 sector_t *sectorp)
286{
287 int i;
288 struct strip_zone *z = conf->strip_zone;
289 sector_t sector = *sectorp;
290
291 for (i = 0; i < conf->nr_strip_zones; i++)
292 if (sector < z[i].zone_end) {
293 if (i)
294 *sectorp = sector - z[i-1].zone_end;
295 return z + i;
296 }
297 BUG();
298}
299
300/*
301 * remaps the bio to the target device. we separate two flows.
NeilBrown47d68972015-04-10 13:19:04 +1000302 * power 2 flow and a general flow for the sake of performance
NeilBrownba13da42012-03-19 12:46:39 +1100303*/
304static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
305 sector_t sector, sector_t *sector_offset)
306{
307 unsigned int sect_in_chunk;
308 sector_t chunk;
309 struct r0conf *conf = mddev->private;
310 int raid_disks = conf->strip_zone[0].nb_dev;
311 unsigned int chunk_sects = mddev->chunk_sectors;
312
313 if (is_power_of_2(chunk_sects)) {
314 int chunksect_bits = ffz(~chunk_sects);
315 /* find the sector offset inside the chunk */
316 sect_in_chunk = sector & (chunk_sects - 1);
317 sector >>= chunksect_bits;
318 /* chunk in zone */
319 chunk = *sector_offset;
320 /* quotient is the chunk in real device*/
321 sector_div(chunk, zone->nb_dev << chunksect_bits);
322 } else{
323 sect_in_chunk = sector_div(sector, chunk_sects);
324 chunk = *sector_offset;
325 sector_div(chunk, chunk_sects * zone->nb_dev);
326 }
327 /*
328 * position the bio over the real device
329 * real sector = chunk in device + starting of zone
330 * + the position in the chunk
331 */
332 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
333 return conf->devlist[(zone - conf->strip_zone)*raid_disks
334 + sector_div(sector, zone->nb_dev)];
335}
336
NeilBrownfd01b882011-10-11 16:47:53 +1100337static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -0700338{
339 sector_t array_sectors = 0;
NeilBrown3cb03002011-10-11 16:45:26 +1100340 struct md_rdev *rdev;
Dan Williams80c3a6c2009-03-17 18:10:40 -0700341
342 WARN_ONCE(sectors || raid_disks,
343 "%s does not support generic reshape\n", __func__);
344
NeilBrowndafb20f2012-03-19 12:46:39 +1100345 rdev_for_each(rdev, mddev)
NeilBrowna6468532013-02-21 14:33:17 +1100346 array_sectors += (rdev->sectors &
347 ~(sector_t)(mddev->chunk_sectors-1));
Dan Williams80c3a6c2009-03-17 18:10:40 -0700348
349 return array_sectors;
350}
351
NeilBrownafa0f552014-12-15 12:56:58 +1100352static void raid0_free(struct mddev *mddev, void *priv);
majianpeng0366ef82012-04-02 09:48:37 +1000353
NeilBrownfd01b882011-10-11 16:47:53 +1100354static int raid0_run(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355{
NeilBrowne373ab12011-10-11 16:48:59 +1100356 struct r0conf *conf;
Andre Noll5568a602009-06-16 16:47:21 +1000357 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358
Andre Noll9d8f0362009-06-18 08:45:01 +1000359 if (mddev->chunk_sectors == 0) {
NeilBrownb5a20962010-05-03 15:06:27 +1000360 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
361 mdname(mddev));
NeilBrown2604b702006-01-06 00:20:36 -0800362 return -EINVAL;
363 }
Andre Noll0894cc32009-06-18 08:49:23 +1000364 if (md_check_no_bitmap(mddev))
365 return -EINVAL;
Heinz Mauelshagen753f2852015-02-13 19:48:01 +0100366
Trela, Maciej9af204c2010-03-08 16:02:44 +1100367 /* if private is not null, we are here after takeover */
368 if (mddev->private == NULL) {
369 ret = create_strip_zones(mddev, &conf);
370 if (ret < 0)
371 return ret;
372 mddev->private = conf;
373 }
374 conf = mddev->private;
NeilBrown199dc6e2015-08-03 13:11:47 +1000375 if (mddev->queue) {
376 struct md_rdev *rdev;
377 bool discard_supported = false;
378
NeilBrown199dc6e2015-08-03 13:11:47 +1000379 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
380 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
381 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
382
383 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
384 blk_queue_io_opt(mddev->queue,
385 (mddev->chunk_sectors << 9) * mddev->raid_disks);
386
NeilBrown66eefe52015-09-24 15:47:47 +1000387 rdev_for_each(rdev, mddev) {
388 disk_stack_limits(mddev->gendisk, rdev->bdev,
389 rdev->data_offset << 9);
390 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
391 discard_supported = true;
392 }
NeilBrown199dc6e2015-08-03 13:11:47 +1000393 if (!discard_supported)
394 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
395 else
396 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
397 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398
399 /* calculate array device size */
Dan Williams1f403622009-03-31 14:59:03 +1100400 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401
NeilBrownb5a20962010-05-03 15:06:27 +1000402 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
403 mdname(mddev),
404 (unsigned long long)mddev->array_sectors);
Heinz Mauelshagen753f2852015-02-13 19:48:01 +0100405
406 if (mddev->queue) {
407 /* calculate the max read-ahead size.
408 * For read-ahead of large files to be effective, we need to
409 * readahead at least twice a whole stripe. i.e. number of devices
410 * multiplied by chunk size times 2.
411 * If an individual device has an ra_pages greater than the
412 * chunk size, then we will not drive that device as hard as it
413 * wants. We consider this a configuration error: a larger
414 * chunksize should be used in that case.
415 */
Andre Noll9d8f0362009-06-18 08:45:01 +1000416 int stripe = mddev->raid_disks *
417 (mddev->chunk_sectors << 9) / PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
419 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
420 }
421
raz ben yehuda46994192009-06-16 17:00:54 +1000422 dump_zones(mddev);
majianpeng0366ef82012-04-02 09:48:37 +1000423
424 ret = md_integrity_register(mddev);
majianpeng0366ef82012-04-02 09:48:37 +1000425
426 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427}
428
NeilBrownafa0f552014-12-15 12:56:58 +1100429static void raid0_free(struct mddev *mddev, void *priv)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430{
NeilBrownafa0f552014-12-15 12:56:58 +1100431 struct r0conf *conf = priv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700433 kfree(conf->strip_zone);
Andre Nollfb5ab4b2009-06-16 16:48:19 +1000434 kfree(conf->devlist);
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700435 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436}
437
raz ben yehudafbb704e2009-06-16 17:02:05 +1000438/*
439 * Is io distribute over 1 or more chunks ?
440*/
NeilBrownfd01b882011-10-11 16:47:53 +1100441static inline int is_io_in_chunk_boundary(struct mddev *mddev,
raz ben yehudafbb704e2009-06-16 17:02:05 +1000442 unsigned int chunk_sects, struct bio *bio)
443{
NeilBrownd6e412e2009-06-18 08:47:00 +1000444 if (likely(is_power_of_2(chunk_sects))) {
Kent Overstreet4f024f32013-10-11 15:44:27 -0700445 return chunk_sects >=
446 ((bio->bi_iter.bi_sector & (chunk_sects-1))
Kent Overstreetaa8b57a2013-02-05 15:19:29 -0800447 + bio_sectors(bio));
raz ben yehudafbb704e2009-06-16 17:02:05 +1000448 } else{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700449 sector_t sector = bio->bi_iter.bi_sector;
raz ben yehudafbb704e2009-06-16 17:02:05 +1000450 return chunk_sects >= (sector_div(sector, chunk_sects)
Kent Overstreetaa8b57a2013-02-05 15:19:29 -0800451 + bio_sectors(bio));
raz ben yehudafbb704e2009-06-16 17:02:05 +1000452 }
453}
454
Linus Torvaldsb4fdcb02011-11-04 17:06:58 -0700455static void raid0_make_request(struct mddev *mddev, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 struct strip_zone *zone;
NeilBrown3cb03002011-10-11 16:45:26 +1100458 struct md_rdev *tmp_dev;
Kent Overstreet20d01892013-11-23 18:21:01 -0800459 struct bio *split;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460
Tejun Heoe9c74692010-09-03 11:56:18 +0200461 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
462 md_flush_request(mddev, bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200463 return;
NeilBrowne5dcdd82005-09-09 16:23:41 -0700464 }
465
Kent Overstreet20d01892013-11-23 18:21:01 -0800466 do {
Kent Overstreet4f024f32013-10-11 15:44:27 -0700467 sector_t sector = bio->bi_iter.bi_sector;
Kent Overstreet20d01892013-11-23 18:21:01 -0800468 unsigned chunk_sects = mddev->chunk_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469
Kent Overstreet20d01892013-11-23 18:21:01 -0800470 unsigned sectors = chunk_sects -
471 (likely(is_power_of_2(chunk_sects))
472 ? (sector & (chunk_sects-1))
473 : sector_div(sector, chunk_sects));
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200474
Eric Worka8115772015-05-18 23:26:23 -0700475 /* Restore due to sector_div */
476 sector = bio->bi_iter.bi_sector;
477
Kent Overstreet20d01892013-11-23 18:21:01 -0800478 if (sectors < bio_sectors(bio)) {
479 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
480 bio_chain(split, bio);
481 } else {
482 split = bio;
483 }
Shaohua Lic83057a2012-10-11 13:25:44 +1100484
Kent Overstreet20d01892013-11-23 18:21:01 -0800485 zone = find_zone(mddev->private, &sector);
486 tmp_dev = map_sector(mddev, zone, sector, &sector);
487 split->bi_bdev = tmp_dev->bdev;
488 split->bi_iter.bi_sector = sector + zone->dev_start +
489 tmp_dev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490
Kent Overstreet20d01892013-11-23 18:21:01 -0800491 if (unlikely((split->bi_rw & REQ_DISCARD) &&
492 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
493 /* Just ignore it */
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200494 bio_endio(split);
Kent Overstreet20d01892013-11-23 18:21:01 -0800495 } else
496 generic_make_request(split);
497 } while (split != bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498}
NeilBrown8299d7f2007-10-16 23:30:53 -0700499
NeilBrownfd01b882011-10-11 16:47:53 +1100500static void raid0_status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501{
Andre Noll9d8f0362009-06-18 08:45:01 +1000502 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 return;
504}
505
NeilBrownfd01b882011-10-11 16:47:53 +1100506static void *raid0_takeover_raid45(struct mddev *mddev)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100507{
NeilBrown3cb03002011-10-11 16:45:26 +1100508 struct md_rdev *rdev;
NeilBrowne373ab12011-10-11 16:48:59 +1100509 struct r0conf *priv_conf;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100510
511 if (mddev->degraded != 1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000512 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
513 mdname(mddev),
Trela, Maciej9af204c2010-03-08 16:02:44 +1100514 mddev->degraded);
515 return ERR_PTR(-EINVAL);
516 }
517
NeilBrowndafb20f2012-03-19 12:46:39 +1100518 rdev_for_each(rdev, mddev) {
Trela, Maciej9af204c2010-03-08 16:02:44 +1100519 /* check slot number for a disk */
520 if (rdev->raid_disk == mddev->raid_disks-1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000521 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
522 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100523 return ERR_PTR(-EINVAL);
524 }
NeilBrowneea136d2013-06-26 11:55:20 +1000525 rdev->sectors = mddev->dev_sectors;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100526 }
527
528 /* Set new parameters */
529 mddev->new_level = 0;
Maciej Trela001048a2010-06-16 11:55:14 +0100530 mddev->new_layout = 0;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100531 mddev->new_chunk_sectors = mddev->chunk_sectors;
532 mddev->raid_disks--;
533 mddev->delta_disks = -1;
534 /* make sure it will be not marked as dirty */
535 mddev->recovery_cp = MaxSector;
536
537 create_strip_zones(mddev, &priv_conf);
538 return priv_conf;
539}
540
NeilBrownfd01b882011-10-11 16:47:53 +1100541static void *raid0_takeover_raid10(struct mddev *mddev)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100542{
NeilBrowne373ab12011-10-11 16:48:59 +1100543 struct r0conf *priv_conf;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100544
545 /* Check layout:
546 * - far_copies must be 1
547 * - near_copies must be 2
548 * - disks number must be even
549 * - all mirrors must be already degraded
550 */
551 if (mddev->layout != ((1 << 8) + 2)) {
Masanari Iidae3d132d2015-10-16 21:14:29 +0900552 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takeover layout: 0x%x\n",
NeilBrownb5a20962010-05-03 15:06:27 +1000553 mdname(mddev),
Trela, Maciej9af204c2010-03-08 16:02:44 +1100554 mddev->layout);
555 return ERR_PTR(-EINVAL);
556 }
557 if (mddev->raid_disks & 1) {
Masanari Iidae3d132d2015-10-16 21:14:29 +0900558 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takeover Raid10 with odd disk number.\n",
NeilBrownb5a20962010-05-03 15:06:27 +1000559 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100560 return ERR_PTR(-EINVAL);
561 }
562 if (mddev->degraded != (mddev->raid_disks>>1)) {
NeilBrownb5a20962010-05-03 15:06:27 +1000563 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
564 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100565 return ERR_PTR(-EINVAL);
566 }
567
568 /* Set new parameters */
569 mddev->new_level = 0;
Maciej Trela001048a2010-06-16 11:55:14 +0100570 mddev->new_layout = 0;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100571 mddev->new_chunk_sectors = mddev->chunk_sectors;
572 mddev->delta_disks = - mddev->raid_disks / 2;
573 mddev->raid_disks += mddev->delta_disks;
574 mddev->degraded = 0;
575 /* make sure it will be not marked as dirty */
576 mddev->recovery_cp = MaxSector;
577
578 create_strip_zones(mddev, &priv_conf);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100579 return priv_conf;
580}
581
NeilBrownfd01b882011-10-11 16:47:53 +1100582static void *raid0_takeover_raid1(struct mddev *mddev)
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100583{
NeilBrowne373ab12011-10-11 16:48:59 +1100584 struct r0conf *priv_conf;
Jes Sorensen24b961f2012-04-01 23:48:38 +1000585 int chunksect;
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100586
587 /* Check layout:
588 * - (N - 1) mirror drives must be already faulty
589 */
590 if ((mddev->raid_disks - 1) != mddev->degraded) {
591 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
592 mdname(mddev));
593 return ERR_PTR(-EINVAL);
594 }
595
Jes Sorensen24b961f2012-04-01 23:48:38 +1000596 /*
597 * a raid1 doesn't have the notion of chunk size, so
598 * figure out the largest suitable size we can use.
599 */
600 chunksect = 64 * 2; /* 64K by default */
601
602 /* The array must be an exact multiple of chunksize */
603 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
604 chunksect >>= 1;
605
606 if ((chunksect << 9) < PAGE_SIZE)
607 /* array size does not allow a suitable chunk size */
608 return ERR_PTR(-EINVAL);
609
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100610 /* Set new parameters */
611 mddev->new_level = 0;
612 mddev->new_layout = 0;
Jes Sorensen24b961f2012-04-01 23:48:38 +1000613 mddev->new_chunk_sectors = chunksect;
614 mddev->chunk_sectors = chunksect;
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100615 mddev->delta_disks = 1 - mddev->raid_disks;
Krzysztof Wojcikf7bee802011-02-14 10:01:41 +1100616 mddev->raid_disks = 1;
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100617 /* make sure it will be not marked as dirty */
618 mddev->recovery_cp = MaxSector;
619
620 create_strip_zones(mddev, &priv_conf);
621 return priv_conf;
622}
623
NeilBrownfd01b882011-10-11 16:47:53 +1100624static void *raid0_takeover(struct mddev *mddev)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100625{
626 /* raid0 can take over:
Maciej Trela049d6c12010-06-16 11:56:12 +0100627 * raid4 - if all data disks are active.
Trela, Maciej9af204c2010-03-08 16:02:44 +1100628 * raid5 - providing it is Raid4 layout and one disk is faulty
629 * raid10 - assuming we have all necessary active disks
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100630 * raid1 - with (N -1) mirror drives faulty
Trela, Maciej9af204c2010-03-08 16:02:44 +1100631 */
NeilBrowna8461a62014-08-06 16:34:27 +1000632
633 if (mddev->bitmap) {
634 printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n",
635 mdname(mddev));
636 return ERR_PTR(-EBUSY);
637 }
Maciej Trela049d6c12010-06-16 11:56:12 +0100638 if (mddev->level == 4)
639 return raid0_takeover_raid45(mddev);
640
Trela, Maciej9af204c2010-03-08 16:02:44 +1100641 if (mddev->level == 5) {
642 if (mddev->layout == ALGORITHM_PARITY_N)
Maciej Trela049d6c12010-06-16 11:56:12 +0100643 return raid0_takeover_raid45(mddev);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100644
NeilBrownb5a20962010-05-03 15:06:27 +1000645 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
646 mdname(mddev), ALGORITHM_PARITY_N);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100647 }
648
649 if (mddev->level == 10)
650 return raid0_takeover_raid10(mddev);
651
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100652 if (mddev->level == 1)
653 return raid0_takeover_raid1(mddev);
654
655 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
656 mddev->level);
657
Trela, Maciej9af204c2010-03-08 16:02:44 +1100658 return ERR_PTR(-EINVAL);
659}
660
NeilBrownfd01b882011-10-11 16:47:53 +1100661static void raid0_quiesce(struct mddev *mddev, int state)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100662{
663}
664
NeilBrown84fc4b52011-10-11 16:49:58 +1100665static struct md_personality raid0_personality=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666{
667 .name = "raid0",
NeilBrown2604b702006-01-06 00:20:36 -0800668 .level = 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 .owner = THIS_MODULE,
670 .make_request = raid0_make_request,
671 .run = raid0_run,
NeilBrownafa0f552014-12-15 12:56:58 +1100672 .free = raid0_free,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 .status = raid0_status,
Dan Williams80c3a6c2009-03-17 18:10:40 -0700674 .size = raid0_size,
Trela, Maciej9af204c2010-03-08 16:02:44 +1100675 .takeover = raid0_takeover,
676 .quiesce = raid0_quiesce,
NeilBrown5c675f82014-12-15 12:56:56 +1100677 .congested = raid0_congested,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678};
679
680static int __init raid0_init (void)
681{
NeilBrown2604b702006-01-06 00:20:36 -0800682 return register_md_personality (&raid0_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683}
684
685static void raid0_exit (void)
686{
NeilBrown2604b702006-01-06 00:20:36 -0800687 unregister_md_personality (&raid0_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688}
689
690module_init(raid0_init);
691module_exit(raid0_exit);
692MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +1100693MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694MODULE_ALIAS("md-personality-2"); /* RAID0 */
NeilBrownd9d166c2006-01-06 00:20:51 -0800695MODULE_ALIAS("md-raid0");
NeilBrown2604b702006-01-06 00:20:36 -0800696MODULE_ALIAS("md-level-0");