blob: 637a96855edb2b6c9c99e988fcdb78b696e703b2 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 raid0.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
4 <zyngier@ufr-info-p7.ibp.fr> or
5 <maz@gloups.fdn.fr>
6 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
7
8
9 RAID-0 management functions.
10
11 This program is free software; you can redistribute it and/or modify
12 it under the terms of the GNU General Public License as published by
13 the Free Software Foundation; either version 2, or (at your option)
14 any later version.
15
16 You should have received a copy of the GNU General Public License
17 (for example /usr/src/linux/COPYING); if not, write to the Free
18 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
19*/
20
NeilBrownbff61972009-03-31 14:33:13 +110021#include <linux/blkdev.h>
NeilBrownbff61972009-03-31 14:33:13 +110022#include <linux/seq_file.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
Jens Axboe165125e2007-07-24 09:28:11 +020028static void raid0_unplug(struct request_queue *q)
Linus Torvalds1da177e2005-04-16 15:20:36 -070029{
30 mddev_t *mddev = q->queuedata;
NeilBrown070ec552009-06-16 16:54:21 +100031 raid0_conf_t *conf = mddev->private;
NeilBrownb4145792009-06-16 16:50:52 +100032 mdk_rdev_t **devlist = conf->devlist;
NeilBrown84707f32010-03-16 17:23:35 +110033 int raid_disks = conf->strip_zone[0].nb_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 int i;
35
NeilBrown84707f32010-03-16 17:23:35 +110036 for (i=0; i < raid_disks; i++) {
Jens Axboe165125e2007-07-24 09:28:11 +020037 struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Alan D. Brunelle2ad8b1e2007-11-07 14:26:56 -050039 blk_unplug(r_queue);
Linus Torvalds1da177e2005-04-16 15:20:36 -070040 }
41}
42
NeilBrown26be34d2006-10-03 01:15:53 -070043static int raid0_congested(void *data, int bits)
44{
45 mddev_t *mddev = data;
NeilBrown070ec552009-06-16 16:54:21 +100046 raid0_conf_t *conf = mddev->private;
NeilBrownb4145792009-06-16 16:50:52 +100047 mdk_rdev_t **devlist = conf->devlist;
NeilBrown84707f32010-03-16 17:23:35 +110048 int raid_disks = conf->strip_zone[0].nb_dev;
NeilBrown26be34d2006-10-03 01:15:53 -070049 int i, ret = 0;
50
NeilBrown3fa841d2009-09-23 18:10:29 +100051 if (mddev_congested(mddev, bits))
52 return 1;
53
NeilBrown84707f32010-03-16 17:23:35 +110054 for (i = 0; i < raid_disks && !ret ; i++) {
Jens Axboe165125e2007-07-24 09:28:11 +020055 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
NeilBrown26be34d2006-10-03 01:15:53 -070056
57 ret |= bdi_congested(&q->backing_dev_info, bits);
58 }
59 return ret;
60}
61
raz ben yehuda46994192009-06-16 17:00:54 +100062/*
63 * inform the user of the raid configuration
64*/
65static void dump_zones(mddev_t *mddev)
66{
67 int j, k, h;
68 sector_t zone_size = 0;
69 sector_t zone_start = 0;
70 char b[BDEVNAME_SIZE];
71 raid0_conf_t *conf = mddev->private;
NeilBrown84707f32010-03-16 17:23:35 +110072 int raid_disks = conf->strip_zone[0].nb_dev;
raz ben yehuda46994192009-06-16 17:00:54 +100073 printk(KERN_INFO "******* %s configuration *********\n",
74 mdname(mddev));
75 h = 0;
76 for (j = 0; j < conf->nr_strip_zones; j++) {
77 printk(KERN_INFO "zone%d=[", j);
78 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
NeilBrownb5a20962010-05-03 15:06:27 +100079 printk(KERN_CONT "%s/",
NeilBrown84707f32010-03-16 17:23:35 +110080 bdevname(conf->devlist[j*raid_disks
raz ben yehuda46994192009-06-16 17:00:54 +100081 + k]->bdev, b));
NeilBrownb5a20962010-05-03 15:06:27 +100082 printk(KERN_CONT "]\n");
raz ben yehuda46994192009-06-16 17:00:54 +100083
84 zone_size = conf->strip_zone[j].zone_end - zone_start;
85 printk(KERN_INFO " zone offset=%llukb "
86 "device offset=%llukb size=%llukb\n",
87 (unsigned long long)zone_start>>1,
88 (unsigned long long)conf->strip_zone[j].dev_start>>1,
89 (unsigned long long)zone_size>>1);
90 zone_start = conf->strip_zone[j].zone_end;
91 }
92 printk(KERN_INFO "**********************************\n\n");
93}
94
Trela, Maciej9af204c2010-03-08 16:02:44 +110095static int create_strip_zones(mddev_t *mddev, raid0_conf_t **private_conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -070096{
NeilBrowna9f326e2009-09-23 18:06:41 +100097 int i, c, err;
NeilBrown49f357a22009-06-16 16:50:35 +100098 sector_t curr_zone_end, sectors;
NeilBrownb4145792009-06-16 16:50:52 +100099 mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev, **dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 struct strip_zone *zone;
101 int cnt;
102 char b[BDEVNAME_SIZE];
Andre Nolled7b0032009-06-16 16:47:36 +1000103 raid0_conf_t *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
104
105 if (!conf)
106 return -ENOMEM;
Cheng Renquan159ec1f2009-01-09 08:31:08 +1100107 list_for_each_entry(rdev1, &mddev->disks, same_set) {
NeilBrownb5a20962010-05-03 15:06:27 +1000108 printk(KERN_INFO "md/raid0:%s: looking at %s\n",
109 mdname(mddev),
110 bdevname(rdev1->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 c = 0;
NeilBrown13f26822009-06-18 08:48:55 +1000112
113 /* round size to chunk_size */
114 sectors = rdev1->sectors;
115 sector_div(sectors, mddev->chunk_sectors);
116 rdev1->sectors = sectors * mddev->chunk_sectors;
117
Cheng Renquan159ec1f2009-01-09 08:31:08 +1100118 list_for_each_entry(rdev2, &mddev->disks, same_set) {
NeilBrownb5a20962010-05-03 15:06:27 +1000119 printk(KERN_INFO "md/raid0:%s: comparing %s(%llu)",
120 mdname(mddev),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 bdevname(rdev1->bdev,b),
Andre Nolldd8ac332009-03-31 14:33:13 +1100122 (unsigned long long)rdev1->sectors);
NeilBrownb5a20962010-05-03 15:06:27 +1000123 printk(KERN_CONT " with %s(%llu)\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124 bdevname(rdev2->bdev,b),
Andre Nolldd8ac332009-03-31 14:33:13 +1100125 (unsigned long long)rdev2->sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 if (rdev2 == rdev1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000127 printk(KERN_INFO "md/raid0:%s: END\n",
128 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 break;
130 }
Andre Nolldd8ac332009-03-31 14:33:13 +1100131 if (rdev2->sectors == rdev1->sectors) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 /*
133 * Not unique, don't count it as a new
134 * group
135 */
NeilBrownb5a20962010-05-03 15:06:27 +1000136 printk(KERN_INFO "md/raid0:%s: EQUAL\n",
137 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 c = 1;
139 break;
140 }
NeilBrownb5a20962010-05-03 15:06:27 +1000141 printk(KERN_INFO "md/raid0:%s: NOT EQUAL\n",
142 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 }
144 if (!c) {
NeilBrownb5a20962010-05-03 15:06:27 +1000145 printk(KERN_INFO "md/raid0:%s: ==> UNIQUE\n",
146 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 conf->nr_strip_zones++;
NeilBrownb5a20962010-05-03 15:06:27 +1000148 printk(KERN_INFO "md/raid0:%s: %d zones\n",
149 mdname(mddev), conf->nr_strip_zones);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 }
151 }
NeilBrownb5a20962010-05-03 15:06:27 +1000152 printk(KERN_INFO "md/raid0:%s: FINAL %d zones\n",
153 mdname(mddev), conf->nr_strip_zones);
Andre Nolled7b0032009-06-16 16:47:36 +1000154 err = -ENOMEM;
NeilBrown9ffae0c2006-01-06 00:20:32 -0800155 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 conf->nr_strip_zones, GFP_KERNEL);
157 if (!conf->strip_zone)
Andre Nolled7b0032009-06-16 16:47:36 +1000158 goto abort;
NeilBrown9ffae0c2006-01-06 00:20:32 -0800159 conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 conf->nr_strip_zones*mddev->raid_disks,
161 GFP_KERNEL);
162 if (!conf->devlist)
Andre Nolled7b0032009-06-16 16:47:36 +1000163 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 /* The first zone must contain all devices, so here we check that
166 * there is a proper alignment of slots to devices and find them all
167 */
168 zone = &conf->strip_zone[0];
169 cnt = 0;
170 smallest = NULL;
NeilBrownb4145792009-06-16 16:50:52 +1000171 dev = conf->devlist;
Andre Nolled7b0032009-06-16 16:47:36 +1000172 err = -EINVAL;
Cheng Renquan159ec1f2009-01-09 08:31:08 +1100173 list_for_each_entry(rdev1, &mddev->disks, same_set) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 int j = rdev1->raid_disk;
175
NeilBrowne93f68a2010-06-15 09:36:03 +0100176 if (mddev->level == 10) {
Trela, Maciej9af204c2010-03-08 16:02:44 +1100177 /* taking over a raid10-n2 array */
178 j /= 2;
NeilBrowne93f68a2010-06-15 09:36:03 +0100179 rdev1->new_raid_disk = j;
180 }
Trela, Maciej9af204c2010-03-08 16:02:44 +1100181
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100182 if (mddev->level == 1) {
183 /* taiking over a raid1 array-
184 * we have only one active disk
185 */
186 j = 0;
187 rdev1->new_raid_disk = j;
188 }
189
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 if (j < 0 || j >= mddev->raid_disks) {
NeilBrownb5a20962010-05-03 15:06:27 +1000191 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
192 "aborting!\n", mdname(mddev), j);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 goto abort;
194 }
NeilBrownb4145792009-06-16 16:50:52 +1000195 if (dev[j]) {
NeilBrownb5a20962010-05-03 15:06:27 +1000196 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
197 "aborting!\n", mdname(mddev), j);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 goto abort;
199 }
NeilBrownb4145792009-06-16 16:50:52 +1000200 dev[j] = rdev1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +1000202 disk_stack_limits(mddev->gendisk, rdev1->bdev,
203 rdev1->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 /* as we don't honour merge_bvec_fn, we must never risk
NeilBrown627a2d32010-03-08 16:44:38 +1100205 * violating it, so limit ->max_segments to 1, lying within
206 * a single page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 */
208
NeilBrown627a2d32010-03-08 16:44:38 +1100209 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn) {
210 blk_queue_max_segments(mddev->queue, 1);
211 blk_queue_segment_boundary(mddev->queue,
212 PAGE_CACHE_SIZE - 1);
213 }
Andre Nolldd8ac332009-03-31 14:33:13 +1100214 if (!smallest || (rdev1->sectors < smallest->sectors))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 smallest = rdev1;
216 cnt++;
217 }
218 if (cnt != mddev->raid_disks) {
NeilBrownb5a20962010-05-03 15:06:27 +1000219 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
220 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 goto abort;
222 }
223 zone->nb_dev = cnt;
NeilBrown49f357a22009-06-16 16:50:35 +1000224 zone->zone_end = smallest->sectors * cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
NeilBrown49f357a22009-06-16 16:50:35 +1000226 curr_zone_end = zone->zone_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227
228 /* now do the other zones */
229 for (i = 1; i < conf->nr_strip_zones; i++)
230 {
NeilBrowna9f326e2009-09-23 18:06:41 +1000231 int j;
232
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 zone = conf->strip_zone + i;
NeilBrownb4145792009-06-16 16:50:52 +1000234 dev = conf->devlist + i * mddev->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235
NeilBrownb5a20962010-05-03 15:06:27 +1000236 printk(KERN_INFO "md/raid0:%s: zone %d\n",
237 mdname(mddev), i);
NeilBrownd27a43a2009-06-16 16:46:46 +1000238 zone->dev_start = smallest->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 smallest = NULL;
240 c = 0;
241
242 for (j=0; j<cnt; j++) {
NeilBrownb4145792009-06-16 16:50:52 +1000243 rdev = conf->devlist[j];
NeilBrownb5a20962010-05-03 15:06:27 +1000244 printk(KERN_INFO "md/raid0:%s: checking %s ...",
245 mdname(mddev),
246 bdevname(rdev->bdev, b));
NeilBrownd27a43a2009-06-16 16:46:46 +1000247 if (rdev->sectors <= zone->dev_start) {
NeilBrownb5a20962010-05-03 15:06:27 +1000248 printk(KERN_CONT " nope.\n");
Andre Nolldd8ac332009-03-31 14:33:13 +1100249 continue;
250 }
NeilBrownb5a20962010-05-03 15:06:27 +1000251 printk(KERN_CONT " contained as device %d\n", c);
NeilBrownb4145792009-06-16 16:50:52 +1000252 dev[c] = rdev;
Andre Nolldd8ac332009-03-31 14:33:13 +1100253 c++;
254 if (!smallest || rdev->sectors < smallest->sectors) {
255 smallest = rdev;
NeilBrownb5a20962010-05-03 15:06:27 +1000256 printk(KERN_INFO "md/raid0:%s: (%llu) is smallest!.\n",
257 mdname(mddev),
258 (unsigned long long)rdev->sectors);
Andre Nolldd8ac332009-03-31 14:33:13 +1100259 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 }
261
262 zone->nb_dev = c;
NeilBrown49f357a22009-06-16 16:50:35 +1000263 sectors = (smallest->sectors - zone->dev_start) * c;
NeilBrownb5a20962010-05-03 15:06:27 +1000264 printk(KERN_INFO "md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
265 mdname(mddev),
266 zone->nb_dev, (unsigned long long)sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
NeilBrown49f357a22009-06-16 16:50:35 +1000268 curr_zone_end += sectors;
NeilBrownd27a43a2009-06-16 16:46:46 +1000269 zone->zone_end = curr_zone_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270
NeilBrownb5a20962010-05-03 15:06:27 +1000271 printk(KERN_INFO "md/raid0:%s: current zone start: %llu\n",
272 mdname(mddev),
273 (unsigned long long)smallest->sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 mddev->queue->unplug_fn = raid0_unplug;
NeilBrown26be34d2006-10-03 01:15:53 -0700276 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
277 mddev->queue->backing_dev_info.congested_data = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278
raz ben yehuda92e59b62009-06-16 17:00:57 +1000279 /*
280 * now since we have the hard sector sizes, we can make sure
281 * chunk size is a multiple of that sector size
282 */
Andre Noll9d8f0362009-06-18 08:45:01 +1000283 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
NeilBrownb5a20962010-05-03 15:06:27 +1000284 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
raz ben yehuda92e59b62009-06-16 17:00:57 +1000285 mdname(mddev),
Andre Noll9d8f0362009-06-18 08:45:01 +1000286 mddev->chunk_sectors << 9);
raz ben yehuda92e59b62009-06-16 17:00:57 +1000287 goto abort;
288 }
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +1000289
290 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
291 blk_queue_io_opt(mddev->queue,
292 (mddev->chunk_sectors << 9) * mddev->raid_disks);
293
NeilBrownb5a20962010-05-03 15:06:27 +1000294 printk(KERN_INFO "md/raid0:%s: done.\n", mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100295 *private_conf = conf;
296
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297 return 0;
Andre Noll5568a602009-06-16 16:47:21 +1000298abort:
Andre Nolled7b0032009-06-16 16:47:36 +1000299 kfree(conf->strip_zone);
300 kfree(conf->devlist);
301 kfree(conf);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100302 *private_conf = NULL;
Andre Nolled7b0032009-06-16 16:47:36 +1000303 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700304}
305
306/**
307 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
308 * @q: request queue
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200309 * @bvm: properties of new bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 * @biovec: the request that could be merged to it.
311 *
312 * Return amount of bytes we can accept at this offset
313 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200314static int raid0_mergeable_bvec(struct request_queue *q,
315 struct bvec_merge_data *bvm,
316 struct bio_vec *biovec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317{
318 mddev_t *mddev = q->queuedata;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200319 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320 int max;
Andre Noll9d8f0362009-06-18 08:45:01 +1000321 unsigned int chunk_sectors = mddev->chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200322 unsigned int bio_sectors = bvm->bi_size >> 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323
NeilBrownd6e412e2009-06-18 08:47:00 +1000324 if (is_power_of_2(chunk_sectors))
raz ben yehudafbb704e2009-06-16 17:02:05 +1000325 max = (chunk_sectors - ((sector & (chunk_sectors-1))
326 + bio_sectors)) << 9;
327 else
328 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
329 + bio_sectors)) << 9;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 if (max < 0) max = 0; /* bio_add cannot handle a negative return */
331 if (max <= biovec->bv_len && bio_sectors == 0)
332 return biovec->bv_len;
333 else
334 return max;
335}
336
Dan Williams80c3a6c2009-03-17 18:10:40 -0700337static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
338{
339 sector_t array_sectors = 0;
340 mdk_rdev_t *rdev;
341
342 WARN_ONCE(sectors || raid_disks,
343 "%s does not support generic reshape\n", __func__);
344
345 list_for_each_entry(rdev, &mddev->disks, same_set)
346 array_sectors += rdev->sectors;
347
348 return array_sectors;
349}
350
Andre Noll8f79cfc2009-06-16 16:47:10 +1000351static int raid0_run(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352{
Trela, Maciej9af204c2010-03-08 16:02:44 +1100353 raid0_conf_t *conf;
Andre Noll5568a602009-06-16 16:47:21 +1000354 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355
Andre Noll9d8f0362009-06-18 08:45:01 +1000356 if (mddev->chunk_sectors == 0) {
NeilBrownb5a20962010-05-03 15:06:27 +1000357 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
358 mdname(mddev));
NeilBrown2604b702006-01-06 00:20:36 -0800359 return -EINVAL;
360 }
Andre Noll0894cc32009-06-18 08:49:23 +1000361 if (md_check_no_bitmap(mddev))
362 return -EINVAL;
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500363 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
Neil Browne7e72bf2008-05-14 16:05:54 -0700364 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365
Trela, Maciej9af204c2010-03-08 16:02:44 +1100366 /* if private is not null, we are here after takeover */
367 if (mddev->private == NULL) {
368 ret = create_strip_zones(mddev, &conf);
369 if (ret < 0)
370 return ret;
371 mddev->private = conf;
372 }
373 conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
375 /* calculate array device size */
Dan Williams1f403622009-03-31 14:59:03 +1100376 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
NeilBrownb5a20962010-05-03 15:06:27 +1000378 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
379 mdname(mddev),
380 (unsigned long long)mddev->array_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 /* calculate the max read-ahead size.
382 * For read-ahead of large files to be effective, we need to
383 * readahead at least twice a whole stripe. i.e. number of devices
384 * multiplied by chunk size times 2.
385 * If an individual device has an ra_pages greater than the
386 * chunk size, then we will not drive that device as hard as it
387 * wants. We consider this a configuration error: a larger
388 * chunksize should be used in that case.
389 */
390 {
Andre Noll9d8f0362009-06-18 08:45:01 +1000391 int stripe = mddev->raid_disks *
392 (mddev->chunk_sectors << 9) / PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700393 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
394 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
395 }
396
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
raz ben yehuda46994192009-06-16 17:00:54 +1000398 dump_zones(mddev);
Andre Nollac5e7112009-08-03 10:59:47 +1000399 md_integrity_register(mddev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401}
402
Andre Nollfb5ab4b2009-06-16 16:48:19 +1000403static int raid0_stop(mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404{
NeilBrown070ec552009-06-16 16:54:21 +1000405 raid0_conf_t *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
407 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700408 kfree(conf->strip_zone);
Andre Nollfb5ab4b2009-06-16 16:48:19 +1000409 kfree(conf->devlist);
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700410 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 mddev->private = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 return 0;
413}
414
NeilBrown49f357a22009-06-16 16:50:35 +1000415/* Find the zone which holds a particular offset
416 * Update *sectorp to be an offset in that zone
417 */
Andre Nolldc582662009-06-16 16:18:43 +1000418static struct strip_zone *find_zone(struct raid0_private_data *conf,
NeilBrown49f357a22009-06-16 16:50:35 +1000419 sector_t *sectorp)
Andre Nolldc582662009-06-16 16:18:43 +1000420{
421 int i;
422 struct strip_zone *z = conf->strip_zone;
NeilBrown49f357a22009-06-16 16:50:35 +1000423 sector_t sector = *sectorp;
Andre Nolldc582662009-06-16 16:18:43 +1000424
425 for (i = 0; i < conf->nr_strip_zones; i++)
NeilBrown49f357a22009-06-16 16:50:35 +1000426 if (sector < z[i].zone_end) {
427 if (i)
428 *sectorp = sector - z[i-1].zone_end;
Andre Nolldc582662009-06-16 16:18:43 +1000429 return z + i;
NeilBrown49f357a22009-06-16 16:50:35 +1000430 }
Andre Nolldc582662009-06-16 16:18:43 +1000431 BUG();
432}
433
raz ben yehudafbb704e2009-06-16 17:02:05 +1000434/*
435 * remaps the bio to the target device. we separate two flows.
436 * power 2 flow and a general flow for the sake of perfromance
437*/
438static mdk_rdev_t *map_sector(mddev_t *mddev, struct strip_zone *zone,
439 sector_t sector, sector_t *sector_offset)
440{
441 unsigned int sect_in_chunk;
442 sector_t chunk;
443 raid0_conf_t *conf = mddev->private;
NeilBrown84707f32010-03-16 17:23:35 +1100444 int raid_disks = conf->strip_zone[0].nb_dev;
Andre Noll9d8f0362009-06-18 08:45:01 +1000445 unsigned int chunk_sects = mddev->chunk_sectors;
raz ben yehudafbb704e2009-06-16 17:02:05 +1000446
NeilBrownd6e412e2009-06-18 08:47:00 +1000447 if (is_power_of_2(chunk_sects)) {
raz ben yehudafbb704e2009-06-16 17:02:05 +1000448 int chunksect_bits = ffz(~chunk_sects);
449 /* find the sector offset inside the chunk */
450 sect_in_chunk = sector & (chunk_sects - 1);
451 sector >>= chunksect_bits;
452 /* chunk in zone */
453 chunk = *sector_offset;
454 /* quotient is the chunk in real device*/
455 sector_div(chunk, zone->nb_dev << chunksect_bits);
456 } else{
457 sect_in_chunk = sector_div(sector, chunk_sects);
458 chunk = *sector_offset;
459 sector_div(chunk, chunk_sects * zone->nb_dev);
460 }
461 /*
462 * position the bio over the real device
463 * real sector = chunk in device + starting of zone
464 * + the position in the chunk
465 */
466 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
NeilBrown84707f32010-03-16 17:23:35 +1100467 return conf->devlist[(zone - conf->strip_zone)*raid_disks
raz ben yehudafbb704e2009-06-16 17:02:05 +1000468 + sector_div(sector, zone->nb_dev)];
469}
470
471/*
472 * Is io distribute over 1 or more chunks ?
473*/
474static inline int is_io_in_chunk_boundary(mddev_t *mddev,
475 unsigned int chunk_sects, struct bio *bio)
476{
NeilBrownd6e412e2009-06-18 08:47:00 +1000477 if (likely(is_power_of_2(chunk_sects))) {
raz ben yehudafbb704e2009-06-16 17:02:05 +1000478 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
479 + (bio->bi_size >> 9));
480 } else{
481 sector_t sector = bio->bi_sector;
482 return chunk_sects >= (sector_div(sector, chunk_sects)
483 + (bio->bi_size >> 9));
484 }
485}
486
NeilBrown21a52c62010-04-01 15:02:13 +1100487static int raid0_make_request(mddev_t *mddev, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488{
raz ben yehudafbb704e2009-06-16 17:02:05 +1000489 unsigned int chunk_sects;
490 sector_t sector_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 struct strip_zone *zone;
492 mdk_rdev_t *tmp_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493
Tejun Heoe9c74692010-09-03 11:56:18 +0200494 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
495 md_flush_request(mddev, bio);
NeilBrowne5dcdd82005-09-09 16:23:41 -0700496 return 0;
497 }
498
Andre Noll9d8f0362009-06-18 08:45:01 +1000499 chunk_sects = mddev->chunk_sectors;
raz ben yehudafbb704e2009-06-16 17:02:05 +1000500 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
501 sector_t sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 struct bio_pair *bp;
503 /* Sanity check -- queue functions should prevent this happening */
504 if (bio->bi_vcnt != 1 ||
505 bio->bi_idx != 0)
506 goto bad_map;
507 /* This is a one page bio that upper layers
508 * refuse to split for us, so we need to split it.
509 */
NeilBrownd6e412e2009-06-18 08:47:00 +1000510 if (likely(is_power_of_2(chunk_sects)))
raz ben yehudafbb704e2009-06-16 17:02:05 +1000511 bp = bio_split(bio, chunk_sects - (sector &
512 (chunk_sects-1)));
513 else
514 bp = bio_split(bio, chunk_sects -
515 sector_div(sector, chunk_sects));
NeilBrown21a52c62010-04-01 15:02:13 +1100516 if (raid0_make_request(mddev, &bp->bio1))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 generic_make_request(&bp->bio1);
NeilBrown21a52c62010-04-01 15:02:13 +1100518 if (raid0_make_request(mddev, &bp->bio2))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519 generic_make_request(&bp->bio2);
520
521 bio_pair_release(bp);
522 return 0;
523 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
raz ben yehudafbb704e2009-06-16 17:02:05 +1000525 sector_offset = bio->bi_sector;
526 zone = find_zone(mddev->private, &sector_offset);
527 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
528 &sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529 bio->bi_bdev = tmp_dev->bdev;
raz ben yehudafbb704e2009-06-16 17:02:05 +1000530 bio->bi_sector = sector_offset + zone->dev_start +
531 tmp_dev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 /*
533 * Let the main block layer submit the IO and resolve recursion:
534 */
535 return 1;
536
537bad_map:
NeilBrownb5a20962010-05-03 15:06:27 +1000538 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
539 " or bigger than %dk %llu %d\n",
540 mdname(mddev), chunk_sects / 2,
541 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542
NeilBrown6712ecf2007-09-27 12:47:43 +0200543 bio_io_error(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544 return 0;
545}
NeilBrown8299d7f2007-10-16 23:30:53 -0700546
raz ben yehuda1b961422009-06-16 16:57:40 +1000547static void raid0_status(struct seq_file *seq, mddev_t *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548{
549#undef MD_DEBUG
550#ifdef MD_DEBUG
551 int j, k, h;
552 char b[BDEVNAME_SIZE];
NeilBrown070ec552009-06-16 16:54:21 +1000553 raid0_conf_t *conf = mddev->private;
NeilBrown84707f32010-03-16 17:23:35 +1100554 int raid_disks = conf->strip_zone[0].nb_dev;
NeilBrown8299d7f2007-10-16 23:30:53 -0700555
raz ben yehuda1b961422009-06-16 16:57:40 +1000556 sector_t zone_size;
557 sector_t zone_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 h = 0;
raz ben yehuda1b961422009-06-16 16:57:40 +1000559
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 for (j = 0; j < conf->nr_strip_zones; j++) {
561 seq_printf(seq, " z%d", j);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 seq_printf(seq, "=[");
563 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
NeilBrown8299d7f2007-10-16 23:30:53 -0700564 seq_printf(seq, "%s/", bdevname(
NeilBrown84707f32010-03-16 17:23:35 +1100565 conf->devlist[j*raid_disks + k]
raz ben yehuda1b961422009-06-16 16:57:40 +1000566 ->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
raz ben yehuda1b961422009-06-16 16:57:40 +1000568 zone_size = conf->strip_zone[j].zone_end - zone_start;
569 seq_printf(seq, "] ze=%lld ds=%lld s=%lld\n",
570 (unsigned long long)zone_start>>1,
571 (unsigned long long)conf->strip_zone[j].dev_start>>1,
572 (unsigned long long)zone_size>>1);
573 zone_start = conf->strip_zone[j].zone_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 }
575#endif
Andre Noll9d8f0362009-06-18 08:45:01 +1000576 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 return;
578}
579
Maciej Trela049d6c12010-06-16 11:56:12 +0100580static void *raid0_takeover_raid45(mddev_t *mddev)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100581{
582 mdk_rdev_t *rdev;
583 raid0_conf_t *priv_conf;
584
585 if (mddev->degraded != 1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000586 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
587 mdname(mddev),
Trela, Maciej9af204c2010-03-08 16:02:44 +1100588 mddev->degraded);
589 return ERR_PTR(-EINVAL);
590 }
591
592 list_for_each_entry(rdev, &mddev->disks, same_set) {
593 /* check slot number for a disk */
594 if (rdev->raid_disk == mddev->raid_disks-1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000595 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
596 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100597 return ERR_PTR(-EINVAL);
598 }
599 }
600
601 /* Set new parameters */
602 mddev->new_level = 0;
Maciej Trela001048a2010-06-16 11:55:14 +0100603 mddev->new_layout = 0;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100604 mddev->new_chunk_sectors = mddev->chunk_sectors;
605 mddev->raid_disks--;
606 mddev->delta_disks = -1;
607 /* make sure it will be not marked as dirty */
608 mddev->recovery_cp = MaxSector;
609
610 create_strip_zones(mddev, &priv_conf);
611 return priv_conf;
612}
613
614static void *raid0_takeover_raid10(mddev_t *mddev)
615{
616 raid0_conf_t *priv_conf;
617
618 /* Check layout:
619 * - far_copies must be 1
620 * - near_copies must be 2
621 * - disks number must be even
622 * - all mirrors must be already degraded
623 */
624 if (mddev->layout != ((1 << 8) + 2)) {
NeilBrownb5a20962010-05-03 15:06:27 +1000625 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
626 mdname(mddev),
Trela, Maciej9af204c2010-03-08 16:02:44 +1100627 mddev->layout);
628 return ERR_PTR(-EINVAL);
629 }
630 if (mddev->raid_disks & 1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000631 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
632 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100633 return ERR_PTR(-EINVAL);
634 }
635 if (mddev->degraded != (mddev->raid_disks>>1)) {
NeilBrownb5a20962010-05-03 15:06:27 +1000636 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
637 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100638 return ERR_PTR(-EINVAL);
639 }
640
641 /* Set new parameters */
642 mddev->new_level = 0;
Maciej Trela001048a2010-06-16 11:55:14 +0100643 mddev->new_layout = 0;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100644 mddev->new_chunk_sectors = mddev->chunk_sectors;
645 mddev->delta_disks = - mddev->raid_disks / 2;
646 mddev->raid_disks += mddev->delta_disks;
647 mddev->degraded = 0;
648 /* make sure it will be not marked as dirty */
649 mddev->recovery_cp = MaxSector;
650
651 create_strip_zones(mddev, &priv_conf);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100652 return priv_conf;
653}
654
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100655static void *raid0_takeover_raid1(mddev_t *mddev)
656{
657 raid0_conf_t *priv_conf;
658
659 /* Check layout:
660 * - (N - 1) mirror drives must be already faulty
661 */
662 if ((mddev->raid_disks - 1) != mddev->degraded) {
663 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
664 mdname(mddev));
665 return ERR_PTR(-EINVAL);
666 }
667
668 /* Set new parameters */
669 mddev->new_level = 0;
670 mddev->new_layout = 0;
671 mddev->new_chunk_sectors = 128; /* by default set chunk size to 64k */
672 mddev->delta_disks = 1 - mddev->raid_disks;
673 /* make sure it will be not marked as dirty */
674 mddev->recovery_cp = MaxSector;
675
676 create_strip_zones(mddev, &priv_conf);
677 return priv_conf;
678}
679
Trela, Maciej9af204c2010-03-08 16:02:44 +1100680static void *raid0_takeover(mddev_t *mddev)
681{
682 /* raid0 can take over:
Maciej Trela049d6c12010-06-16 11:56:12 +0100683 * raid4 - if all data disks are active.
Trela, Maciej9af204c2010-03-08 16:02:44 +1100684 * raid5 - providing it is Raid4 layout and one disk is faulty
685 * raid10 - assuming we have all necessary active disks
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100686 * raid1 - with (N -1) mirror drives faulty
Trela, Maciej9af204c2010-03-08 16:02:44 +1100687 */
Maciej Trela049d6c12010-06-16 11:56:12 +0100688 if (mddev->level == 4)
689 return raid0_takeover_raid45(mddev);
690
Trela, Maciej9af204c2010-03-08 16:02:44 +1100691 if (mddev->level == 5) {
692 if (mddev->layout == ALGORITHM_PARITY_N)
Maciej Trela049d6c12010-06-16 11:56:12 +0100693 return raid0_takeover_raid45(mddev);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100694
NeilBrownb5a20962010-05-03 15:06:27 +1000695 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
696 mdname(mddev), ALGORITHM_PARITY_N);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100697 }
698
699 if (mddev->level == 10)
700 return raid0_takeover_raid10(mddev);
701
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100702 if (mddev->level == 1)
703 return raid0_takeover_raid1(mddev);
704
705 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
706 mddev->level);
707
Trela, Maciej9af204c2010-03-08 16:02:44 +1100708 return ERR_PTR(-EINVAL);
709}
710
711static void raid0_quiesce(mddev_t *mddev, int state)
712{
713}
714
NeilBrown2604b702006-01-06 00:20:36 -0800715static struct mdk_personality raid0_personality=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716{
717 .name = "raid0",
NeilBrown2604b702006-01-06 00:20:36 -0800718 .level = 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 .owner = THIS_MODULE,
720 .make_request = raid0_make_request,
721 .run = raid0_run,
722 .stop = raid0_stop,
723 .status = raid0_status,
Dan Williams80c3a6c2009-03-17 18:10:40 -0700724 .size = raid0_size,
Trela, Maciej9af204c2010-03-08 16:02:44 +1100725 .takeover = raid0_takeover,
726 .quiesce = raid0_quiesce,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727};
728
729static int __init raid0_init (void)
730{
NeilBrown2604b702006-01-06 00:20:36 -0800731 return register_md_personality (&raid0_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732}
733
734static void raid0_exit (void)
735{
NeilBrown2604b702006-01-06 00:20:36 -0800736 unregister_md_personality (&raid0_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
738
739module_init(raid0_init);
740module_exit(raid0_exit);
741MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +1100742MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743MODULE_ALIAS("md-personality-2"); /* RAID0 */
NeilBrownd9d166c2006-01-06 00:20:51 -0800744MODULE_ALIAS("md-raid0");
NeilBrown2604b702006-01-06 00:20:36 -0800745MODULE_ALIAS("md-level-0");