blob: 24b359717a7e8917a4955e36651e6eb4009dc393 [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>
Paul Gortmaker056075c2011-07-03 13:58:33 -040023#include <linux/module.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090024#include <linux/slab.h>
NeilBrown43b2e5d2009-03-31 14:33:13 +110025#include "md.h"
Christoph Hellwigef740c32009-03-31 14:27:03 +110026#include "raid0.h"
Trela, Maciej9af204c2010-03-08 16:02:44 +110027#include "raid5.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
NeilBrown26be34d2006-10-03 01:15:53 -070029static int raid0_congested(void *data, int bits)
30{
NeilBrownfd01b882011-10-11 16:47:53 +110031 struct mddev *mddev = data;
NeilBrowne373ab12011-10-11 16:48:59 +110032 struct r0conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +110033 struct md_rdev **devlist = conf->devlist;
NeilBrown84707f32010-03-16 17:23:35 +110034 int raid_disks = conf->strip_zone[0].nb_dev;
NeilBrown26be34d2006-10-03 01:15:53 -070035 int i, ret = 0;
36
NeilBrown3fa841d2009-09-23 18:10:29 +100037 if (mddev_congested(mddev, bits))
38 return 1;
39
NeilBrown84707f32010-03-16 17:23:35 +110040 for (i = 0; i < raid_disks && !ret ; i++) {
Jens Axboe165125e2007-07-24 09:28:11 +020041 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
NeilBrown26be34d2006-10-03 01:15:53 -070042
43 ret |= bdi_congested(&q->backing_dev_info, bits);
44 }
45 return ret;
46}
47
raz ben yehuda46994192009-06-16 17:00:54 +100048/*
49 * inform the user of the raid configuration
50*/
NeilBrownfd01b882011-10-11 16:47:53 +110051static void dump_zones(struct mddev *mddev)
raz ben yehuda46994192009-06-16 17:00:54 +100052{
NeilBrown50de8df2011-10-07 14:23:22 +110053 int j, k;
raz ben yehuda46994192009-06-16 17:00:54 +100054 sector_t zone_size = 0;
55 sector_t zone_start = 0;
56 char b[BDEVNAME_SIZE];
NeilBrowne373ab12011-10-11 16:48:59 +110057 struct r0conf *conf = mddev->private;
NeilBrown84707f32010-03-16 17:23:35 +110058 int raid_disks = conf->strip_zone[0].nb_dev;
NeilBrown50de8df2011-10-07 14:23:22 +110059 printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n",
60 mdname(mddev),
61 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
raz ben yehuda46994192009-06-16 17:00:54 +100062 for (j = 0; j < conf->nr_strip_zones; j++) {
NeilBrown50de8df2011-10-07 14:23:22 +110063 printk(KERN_INFO "md: zone%d=[", j);
raz ben yehuda46994192009-06-16 17:00:54 +100064 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
NeilBrown50de8df2011-10-07 14:23:22 +110065 printk(KERN_CONT "%s%s", k?"/":"",
NeilBrown84707f32010-03-16 17:23:35 +110066 bdevname(conf->devlist[j*raid_disks
raz ben yehuda46994192009-06-16 17:00:54 +100067 + k]->bdev, b));
NeilBrownb5a20962010-05-03 15:06:27 +100068 printk(KERN_CONT "]\n");
raz ben yehuda46994192009-06-16 17:00:54 +100069
70 zone_size = conf->strip_zone[j].zone_end - zone_start;
NeilBrown50de8df2011-10-07 14:23:22 +110071 printk(KERN_INFO " zone-offset=%10lluKB, "
72 "device-offset=%10lluKB, size=%10lluKB\n",
raz ben yehuda46994192009-06-16 17:00:54 +100073 (unsigned long long)zone_start>>1,
74 (unsigned long long)conf->strip_zone[j].dev_start>>1,
75 (unsigned long long)zone_size>>1);
76 zone_start = conf->strip_zone[j].zone_end;
77 }
NeilBrown50de8df2011-10-07 14:23:22 +110078 printk(KERN_INFO "\n");
raz ben yehuda46994192009-06-16 17:00:54 +100079}
80
NeilBrowne373ab12011-10-11 16:48:59 +110081static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -070082{
NeilBrowna9f326e2009-09-23 18:06:41 +100083 int i, c, err;
NeilBrown49f357a22009-06-16 16:50:35 +100084 sector_t curr_zone_end, sectors;
NeilBrown3cb03002011-10-11 16:45:26 +110085 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 struct strip_zone *zone;
87 int cnt;
88 char b[BDEVNAME_SIZE];
NeilBrown50de8df2011-10-07 14:23:22 +110089 char b2[BDEVNAME_SIZE];
NeilBrowne373ab12011-10-11 16:48:59 +110090 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
Shaohua Lic83057a2012-10-11 13:25:44 +110091 bool discard_supported = false;
Andre Nolled7b0032009-06-16 16:47:36 +100092
93 if (!conf)
94 return -ENOMEM;
NeilBrowndafb20f2012-03-19 12:46:39 +110095 rdev_for_each(rdev1, mddev) {
NeilBrown50de8df2011-10-07 14:23:22 +110096 pr_debug("md/raid0:%s: looking at %s\n",
97 mdname(mddev),
98 bdevname(rdev1->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 c = 0;
NeilBrown13f26822009-06-18 08:48:55 +1000100
101 /* round size to chunk_size */
102 sectors = rdev1->sectors;
103 sector_div(sectors, mddev->chunk_sectors);
104 rdev1->sectors = sectors * mddev->chunk_sectors;
105
NeilBrowndafb20f2012-03-19 12:46:39 +1100106 rdev_for_each(rdev2, mddev) {
NeilBrown50de8df2011-10-07 14:23:22 +1100107 pr_debug("md/raid0:%s: comparing %s(%llu)"
108 " with %s(%llu)\n",
109 mdname(mddev),
110 bdevname(rdev1->bdev,b),
111 (unsigned long long)rdev1->sectors,
112 bdevname(rdev2->bdev,b2),
113 (unsigned long long)rdev2->sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114 if (rdev2 == rdev1) {
NeilBrown50de8df2011-10-07 14:23:22 +1100115 pr_debug("md/raid0:%s: END\n",
116 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 break;
118 }
Andre Nolldd8ac332009-03-31 14:33:13 +1100119 if (rdev2->sectors == rdev1->sectors) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 /*
121 * Not unique, don't count it as a new
122 * group
123 */
NeilBrown50de8df2011-10-07 14:23:22 +1100124 pr_debug("md/raid0:%s: EQUAL\n",
125 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 c = 1;
127 break;
128 }
NeilBrown50de8df2011-10-07 14:23:22 +1100129 pr_debug("md/raid0:%s: NOT EQUAL\n",
130 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 }
132 if (!c) {
NeilBrown50de8df2011-10-07 14:23:22 +1100133 pr_debug("md/raid0:%s: ==> UNIQUE\n",
134 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 conf->nr_strip_zones++;
NeilBrown50de8df2011-10-07 14:23:22 +1100136 pr_debug("md/raid0:%s: %d zones\n",
137 mdname(mddev), conf->nr_strip_zones);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 }
139 }
NeilBrown50de8df2011-10-07 14:23:22 +1100140 pr_debug("md/raid0:%s: FINAL %d zones\n",
141 mdname(mddev), conf->nr_strip_zones);
Andre Nolled7b0032009-06-16 16:47:36 +1000142 err = -ENOMEM;
NeilBrown9ffae0c2006-01-06 00:20:32 -0800143 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 conf->nr_strip_zones, GFP_KERNEL);
145 if (!conf->strip_zone)
Andre Nolled7b0032009-06-16 16:47:36 +1000146 goto abort;
NeilBrown3cb03002011-10-11 16:45:26 +1100147 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 conf->nr_strip_zones*mddev->raid_disks,
149 GFP_KERNEL);
150 if (!conf->devlist)
Andre Nolled7b0032009-06-16 16:47:36 +1000151 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 /* The first zone must contain all devices, so here we check that
154 * there is a proper alignment of slots to devices and find them all
155 */
156 zone = &conf->strip_zone[0];
157 cnt = 0;
158 smallest = NULL;
NeilBrownb4145792009-06-16 16:50:52 +1000159 dev = conf->devlist;
Andre Nolled7b0032009-06-16 16:47:36 +1000160 err = -EINVAL;
NeilBrowndafb20f2012-03-19 12:46:39 +1100161 rdev_for_each(rdev1, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 int j = rdev1->raid_disk;
163
NeilBrowne93f68a2010-06-15 09:36:03 +0100164 if (mddev->level == 10) {
Trela, Maciej9af204c2010-03-08 16:02:44 +1100165 /* taking over a raid10-n2 array */
166 j /= 2;
NeilBrowne93f68a2010-06-15 09:36:03 +0100167 rdev1->new_raid_disk = j;
168 }
Trela, Maciej9af204c2010-03-08 16:02:44 +1100169
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100170 if (mddev->level == 1) {
171 /* taiking over a raid1 array-
172 * we have only one active disk
173 */
174 j = 0;
175 rdev1->new_raid_disk = j;
176 }
177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 if (j < 0 || j >= mddev->raid_disks) {
NeilBrownb5a20962010-05-03 15:06:27 +1000179 printk(KERN_ERR "md/raid0:%s: bad disk number %d - "
180 "aborting!\n", mdname(mddev), j);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 goto abort;
182 }
NeilBrownb4145792009-06-16 16:50:52 +1000183 if (dev[j]) {
NeilBrownb5a20962010-05-03 15:06:27 +1000184 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
185 "aborting!\n", mdname(mddev), j);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 goto abort;
187 }
NeilBrownb4145792009-06-16 16:50:52 +1000188 dev[j] = rdev1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +1000190 disk_stack_limits(mddev->gendisk, rdev1->bdev,
191 rdev1->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192
NeilBrownba13da42012-03-19 12:46:39 +1100193 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
194 conf->has_merge_bvec = 1;
195
Andre Nolldd8ac332009-03-31 14:33:13 +1100196 if (!smallest || (rdev1->sectors < smallest->sectors))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 smallest = rdev1;
198 cnt++;
Shaohua Lic83057a2012-10-11 13:25:44 +1100199
200 if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
201 discard_supported = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 }
203 if (cnt != mddev->raid_disks) {
NeilBrownb5a20962010-05-03 15:06:27 +1000204 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
205 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 goto abort;
207 }
208 zone->nb_dev = cnt;
NeilBrown49f357a22009-06-16 16:50:35 +1000209 zone->zone_end = smallest->sectors * cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
NeilBrown49f357a22009-06-16 16:50:35 +1000211 curr_zone_end = zone->zone_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
213 /* now do the other zones */
214 for (i = 1; i < conf->nr_strip_zones; i++)
215 {
NeilBrowna9f326e2009-09-23 18:06:41 +1000216 int j;
217
Linus Torvalds1da177e2005-04-16 15:20:36 -0700218 zone = conf->strip_zone + i;
NeilBrownb4145792009-06-16 16:50:52 +1000219 dev = conf->devlist + i * mddev->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
NeilBrown50de8df2011-10-07 14:23:22 +1100221 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
NeilBrownd27a43ab2009-06-16 16:46:46 +1000222 zone->dev_start = smallest->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 smallest = NULL;
224 c = 0;
225
226 for (j=0; j<cnt; j++) {
NeilBrownb4145792009-06-16 16:50:52 +1000227 rdev = conf->devlist[j];
NeilBrownd27a43ab2009-06-16 16:46:46 +1000228 if (rdev->sectors <= zone->dev_start) {
NeilBrown50de8df2011-10-07 14:23:22 +1100229 pr_debug("md/raid0:%s: checking %s ... nope\n",
230 mdname(mddev),
231 bdevname(rdev->bdev, b));
Andre Nolldd8ac332009-03-31 14:33:13 +1100232 continue;
233 }
NeilBrown50de8df2011-10-07 14:23:22 +1100234 pr_debug("md/raid0:%s: checking %s ..."
235 " contained as device %d\n",
236 mdname(mddev),
237 bdevname(rdev->bdev, b), c);
NeilBrownb4145792009-06-16 16:50:52 +1000238 dev[c] = rdev;
Andre Nolldd8ac332009-03-31 14:33:13 +1100239 c++;
240 if (!smallest || rdev->sectors < smallest->sectors) {
241 smallest = rdev;
NeilBrown50de8df2011-10-07 14:23:22 +1100242 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
243 mdname(mddev),
244 (unsigned long long)rdev->sectors);
Andre Nolldd8ac332009-03-31 14:33:13 +1100245 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 }
247
248 zone->nb_dev = c;
NeilBrown49f357a22009-06-16 16:50:35 +1000249 sectors = (smallest->sectors - zone->dev_start) * c;
NeilBrown50de8df2011-10-07 14:23:22 +1100250 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
251 mdname(mddev),
252 zone->nb_dev, (unsigned long long)sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
NeilBrown49f357a22009-06-16 16:50:35 +1000254 curr_zone_end += sectors;
NeilBrownd27a43ab2009-06-16 16:46:46 +1000255 zone->zone_end = curr_zone_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
NeilBrown50de8df2011-10-07 14:23:22 +1100257 pr_debug("md/raid0:%s: current zone start: %llu\n",
258 mdname(mddev),
259 (unsigned long long)smallest->sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 }
NeilBrown26be34d2006-10-03 01:15:53 -0700261 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
262 mddev->queue->backing_dev_info.congested_data = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263
raz ben yehuda92e59b62009-06-16 17:00:57 +1000264 /*
265 * now since we have the hard sector sizes, we can make sure
266 * chunk size is a multiple of that sector size
267 */
Andre Noll9d8f0362009-06-18 08:45:01 +1000268 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
NeilBrownb5a20962010-05-03 15:06:27 +1000269 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
raz ben yehuda92e59b62009-06-16 17:00:57 +1000270 mdname(mddev),
Andre Noll9d8f0362009-06-18 08:45:01 +1000271 mddev->chunk_sectors << 9);
raz ben yehuda92e59b62009-06-16 17:00:57 +1000272 goto abort;
273 }
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +1000274
275 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
276 blk_queue_io_opt(mddev->queue,
277 (mddev->chunk_sectors << 9) * mddev->raid_disks);
278
Shaohua Lic83057a2012-10-11 13:25:44 +1100279 if (!discard_supported)
280 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
281 else
282 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
283
NeilBrown50de8df2011-10-07 14:23:22 +1100284 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100285 *private_conf = conf;
286
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return 0;
Andre Noll5568a602009-06-16 16:47:21 +1000288abort:
Andre Nolled7b0032009-06-16 16:47:36 +1000289 kfree(conf->strip_zone);
290 kfree(conf->devlist);
291 kfree(conf);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100292 *private_conf = NULL;
Andre Nolled7b0032009-06-16 16:47:36 +1000293 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294}
295
NeilBrownba13da42012-03-19 12:46:39 +1100296/* Find the zone which holds a particular offset
297 * Update *sectorp to be an offset in that zone
298 */
299static struct strip_zone *find_zone(struct r0conf *conf,
300 sector_t *sectorp)
301{
302 int i;
303 struct strip_zone *z = conf->strip_zone;
304 sector_t sector = *sectorp;
305
306 for (i = 0; i < conf->nr_strip_zones; i++)
307 if (sector < z[i].zone_end) {
308 if (i)
309 *sectorp = sector - z[i-1].zone_end;
310 return z + i;
311 }
312 BUG();
313}
314
315/*
316 * remaps the bio to the target device. we separate two flows.
317 * power 2 flow and a general flow for the sake of perfromance
318*/
319static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
320 sector_t sector, sector_t *sector_offset)
321{
322 unsigned int sect_in_chunk;
323 sector_t chunk;
324 struct r0conf *conf = mddev->private;
325 int raid_disks = conf->strip_zone[0].nb_dev;
326 unsigned int chunk_sects = mddev->chunk_sectors;
327
328 if (is_power_of_2(chunk_sects)) {
329 int chunksect_bits = ffz(~chunk_sects);
330 /* find the sector offset inside the chunk */
331 sect_in_chunk = sector & (chunk_sects - 1);
332 sector >>= chunksect_bits;
333 /* chunk in zone */
334 chunk = *sector_offset;
335 /* quotient is the chunk in real device*/
336 sector_div(chunk, zone->nb_dev << chunksect_bits);
337 } else{
338 sect_in_chunk = sector_div(sector, chunk_sects);
339 chunk = *sector_offset;
340 sector_div(chunk, chunk_sects * zone->nb_dev);
341 }
342 /*
343 * position the bio over the real device
344 * real sector = chunk in device + starting of zone
345 * + the position in the chunk
346 */
347 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
348 return conf->devlist[(zone - conf->strip_zone)*raid_disks
349 + sector_div(sector, zone->nb_dev)];
350}
351
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352/**
NeilBrownba13da42012-03-19 12:46:39 +1100353 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 * @q: request queue
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200355 * @bvm: properties of new bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 * @biovec: the request that could be merged to it.
357 *
358 * Return amount of bytes we can accept at this offset
359 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200360static int raid0_mergeable_bvec(struct request_queue *q,
361 struct bvec_merge_data *bvm,
362 struct bio_vec *biovec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
NeilBrownfd01b882011-10-11 16:47:53 +1100364 struct mddev *mddev = q->queuedata;
NeilBrownba13da42012-03-19 12:46:39 +1100365 struct r0conf *conf = mddev->private;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200366 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
NeilBrownba13da42012-03-19 12:46:39 +1100367 sector_t sector_offset = sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 int max;
Andre Noll9d8f0362009-06-18 08:45:01 +1000369 unsigned int chunk_sectors = mddev->chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200370 unsigned int bio_sectors = bvm->bi_size >> 9;
NeilBrownba13da42012-03-19 12:46:39 +1100371 struct strip_zone *zone;
372 struct md_rdev *rdev;
373 struct request_queue *subq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374
NeilBrownd6e412e2009-06-18 08:47:00 +1000375 if (is_power_of_2(chunk_sectors))
raz ben yehudafbb704e2009-06-16 17:02:05 +1000376 max = (chunk_sectors - ((sector & (chunk_sectors-1))
377 + bio_sectors)) << 9;
378 else
379 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
380 + bio_sectors)) << 9;
NeilBrownba13da42012-03-19 12:46:39 +1100381 if (max < 0)
382 max = 0; /* bio_add cannot handle a negative return */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 if (max <= biovec->bv_len && bio_sectors == 0)
384 return biovec->bv_len;
NeilBrownba13da42012-03-19 12:46:39 +1100385 if (max < biovec->bv_len)
386 /* too small already, no need to check further */
387 return max;
388 if (!conf->has_merge_bvec)
389 return max;
390
391 /* May need to check subordinate device */
392 sector = sector_offset;
393 zone = find_zone(mddev->private, &sector_offset);
394 rdev = map_sector(mddev, zone, sector, &sector_offset);
395 subq = bdev_get_queue(rdev->bdev);
396 if (subq->merge_bvec_fn) {
397 bvm->bi_bdev = rdev->bdev;
398 bvm->bi_sector = sector_offset + zone->dev_start +
399 rdev->data_offset;
400 return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
401 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 return max;
403}
404
NeilBrownfd01b882011-10-11 16:47:53 +1100405static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -0700406{
407 sector_t array_sectors = 0;
NeilBrown3cb03002011-10-11 16:45:26 +1100408 struct md_rdev *rdev;
Dan Williams80c3a6c2009-03-17 18:10:40 -0700409
410 WARN_ONCE(sectors || raid_disks,
411 "%s does not support generic reshape\n", __func__);
412
NeilBrowndafb20f2012-03-19 12:46:39 +1100413 rdev_for_each(rdev, mddev)
Dan Williams80c3a6c2009-03-17 18:10:40 -0700414 array_sectors += rdev->sectors;
415
416 return array_sectors;
417}
418
majianpeng0366ef82012-04-02 09:48:37 +1000419static int raid0_stop(struct mddev *mddev);
420
NeilBrownfd01b882011-10-11 16:47:53 +1100421static int raid0_run(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
NeilBrowne373ab12011-10-11 16:48:59 +1100423 struct r0conf *conf;
Andre Noll5568a602009-06-16 16:47:21 +1000424 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
Andre Noll9d8f0362009-06-18 08:45:01 +1000426 if (mddev->chunk_sectors == 0) {
NeilBrownb5a20962010-05-03 15:06:27 +1000427 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
428 mdname(mddev));
NeilBrown2604b702006-01-06 00:20:36 -0800429 return -EINVAL;
430 }
Andre Noll0894cc32009-06-18 08:49:23 +1000431 if (md_check_no_bitmap(mddev))
432 return -EINVAL;
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500433 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400434 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
Shaohua Lic83057a2012-10-11 13:25:44 +1100435 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436
Trela, Maciej9af204c2010-03-08 16:02:44 +1100437 /* if private is not null, we are here after takeover */
438 if (mddev->private == NULL) {
439 ret = create_strip_zones(mddev, &conf);
440 if (ret < 0)
441 return ret;
442 mddev->private = conf;
443 }
444 conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
446 /* calculate array device size */
Dan Williams1f403622009-03-31 14:59:03 +1100447 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
NeilBrownb5a20962010-05-03 15:06:27 +1000449 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
450 mdname(mddev),
451 (unsigned long long)mddev->array_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 /* calculate the max read-ahead size.
453 * For read-ahead of large files to be effective, we need to
454 * readahead at least twice a whole stripe. i.e. number of devices
455 * multiplied by chunk size times 2.
456 * If an individual device has an ra_pages greater than the
457 * chunk size, then we will not drive that device as hard as it
458 * wants. We consider this a configuration error: a larger
459 * chunksize should be used in that case.
460 */
461 {
Andre Noll9d8f0362009-06-18 08:45:01 +1000462 int stripe = mddev->raid_disks *
463 (mddev->chunk_sectors << 9) / PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
465 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
466 }
467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
raz ben yehuda46994192009-06-16 17:00:54 +1000469 dump_zones(mddev);
majianpeng0366ef82012-04-02 09:48:37 +1000470
471 ret = md_integrity_register(mddev);
472 if (ret)
473 raid0_stop(mddev);
474
475 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476}
477
NeilBrownfd01b882011-10-11 16:47:53 +1100478static int raid0_stop(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479{
NeilBrowne373ab12011-10-11 16:48:59 +1100480 struct r0conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
482 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700483 kfree(conf->strip_zone);
Andre Nollfb5ab4b2009-06-16 16:48:19 +1000484 kfree(conf->devlist);
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700485 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 mddev->private = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 return 0;
488}
489
raz ben yehudafbb704e2009-06-16 17:02:05 +1000490/*
491 * Is io distribute over 1 or more chunks ?
492*/
NeilBrownfd01b882011-10-11 16:47:53 +1100493static inline int is_io_in_chunk_boundary(struct mddev *mddev,
raz ben yehudafbb704e2009-06-16 17:02:05 +1000494 unsigned int chunk_sects, struct bio *bio)
495{
NeilBrownd6e412e2009-06-18 08:47:00 +1000496 if (likely(is_power_of_2(chunk_sects))) {
raz ben yehudafbb704e2009-06-16 17:02:05 +1000497 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
498 + (bio->bi_size >> 9));
499 } else{
500 sector_t sector = bio->bi_sector;
501 return chunk_sects >= (sector_div(sector, chunk_sects)
502 + (bio->bi_size >> 9));
503 }
504}
505
Linus Torvaldsb4fdcb02011-11-04 17:06:58 -0700506static void raid0_make_request(struct mddev *mddev, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507{
raz ben yehudafbb704e2009-06-16 17:02:05 +1000508 unsigned int chunk_sects;
509 sector_t sector_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700510 struct strip_zone *zone;
NeilBrown3cb03002011-10-11 16:45:26 +1100511 struct md_rdev *tmp_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512
Tejun Heoe9c74692010-09-03 11:56:18 +0200513 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
514 md_flush_request(mddev, bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200515 return;
NeilBrowne5dcdd82005-09-09 16:23:41 -0700516 }
517
Andre Noll9d8f0362009-06-18 08:45:01 +1000518 chunk_sects = mddev->chunk_sectors;
raz ben yehudafbb704e2009-06-16 17:02:05 +1000519 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
520 sector_t sector = bio->bi_sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 struct bio_pair *bp;
522 /* Sanity check -- queue functions should prevent this happening */
Shaohua Lic83057a2012-10-11 13:25:44 +1100523 if ((bio->bi_vcnt != 1 && bio->bi_vcnt != 0) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 bio->bi_idx != 0)
525 goto bad_map;
526 /* This is a one page bio that upper layers
527 * refuse to split for us, so we need to split it.
528 */
NeilBrownd6e412e2009-06-18 08:47:00 +1000529 if (likely(is_power_of_2(chunk_sects)))
raz ben yehudafbb704e2009-06-16 17:02:05 +1000530 bp = bio_split(bio, chunk_sects - (sector &
531 (chunk_sects-1)));
532 else
533 bp = bio_split(bio, chunk_sects -
534 sector_div(sector, chunk_sects));
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200535 raid0_make_request(mddev, &bp->bio1);
536 raid0_make_request(mddev, &bp->bio2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 bio_pair_release(bp);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200538 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540
raz ben yehudafbb704e2009-06-16 17:02:05 +1000541 sector_offset = bio->bi_sector;
NeilBrownba13da42012-03-19 12:46:39 +1100542 zone = find_zone(mddev->private, &sector_offset);
raz ben yehudafbb704e2009-06-16 17:02:05 +1000543 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
544 &sector_offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 bio->bi_bdev = tmp_dev->bdev;
raz ben yehudafbb704e2009-06-16 17:02:05 +1000546 bio->bi_sector = sector_offset + zone->dev_start +
547 tmp_dev->data_offset;
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200548
Shaohua Lic83057a2012-10-11 13:25:44 +1100549 if (unlikely((bio->bi_rw & REQ_DISCARD) &&
550 !blk_queue_discard(bdev_get_queue(bio->bi_bdev)))) {
551 /* Just ignore it */
552 bio_endio(bio, 0);
553 return;
554 }
555
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200556 generic_make_request(bio);
557 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
559bad_map:
NeilBrownb5a20962010-05-03 15:06:27 +1000560 printk("md/raid0:%s: make_request bug: can't convert block across chunks"
561 " or bigger than %dk %llu %d\n",
562 mdname(mddev), chunk_sects / 2,
563 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
NeilBrown6712ecf2007-09-27 12:47:43 +0200565 bio_io_error(bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200566 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567}
NeilBrown8299d7f2007-10-16 23:30:53 -0700568
NeilBrownfd01b882011-10-11 16:47:53 +1100569static void raid0_status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570{
Andre Noll9d8f0362009-06-18 08:45:01 +1000571 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 return;
573}
574
NeilBrownfd01b882011-10-11 16:47:53 +1100575static void *raid0_takeover_raid45(struct mddev *mddev)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100576{
NeilBrown3cb03002011-10-11 16:45:26 +1100577 struct md_rdev *rdev;
NeilBrowne373ab12011-10-11 16:48:59 +1100578 struct r0conf *priv_conf;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100579
580 if (mddev->degraded != 1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000581 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
582 mdname(mddev),
Trela, Maciej9af204c2010-03-08 16:02:44 +1100583 mddev->degraded);
584 return ERR_PTR(-EINVAL);
585 }
586
NeilBrowndafb20f2012-03-19 12:46:39 +1100587 rdev_for_each(rdev, mddev) {
Trela, Maciej9af204c2010-03-08 16:02:44 +1100588 /* check slot number for a disk */
589 if (rdev->raid_disk == mddev->raid_disks-1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000590 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
591 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100592 return ERR_PTR(-EINVAL);
593 }
594 }
595
596 /* Set new parameters */
597 mddev->new_level = 0;
Maciej Trela001048a2010-06-16 11:55:14 +0100598 mddev->new_layout = 0;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100599 mddev->new_chunk_sectors = mddev->chunk_sectors;
600 mddev->raid_disks--;
601 mddev->delta_disks = -1;
602 /* make sure it will be not marked as dirty */
603 mddev->recovery_cp = MaxSector;
604
605 create_strip_zones(mddev, &priv_conf);
606 return priv_conf;
607}
608
NeilBrownfd01b882011-10-11 16:47:53 +1100609static void *raid0_takeover_raid10(struct mddev *mddev)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100610{
NeilBrowne373ab12011-10-11 16:48:59 +1100611 struct r0conf *priv_conf;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100612
613 /* Check layout:
614 * - far_copies must be 1
615 * - near_copies must be 2
616 * - disks number must be even
617 * - all mirrors must be already degraded
618 */
619 if (mddev->layout != ((1 << 8) + 2)) {
NeilBrownb5a20962010-05-03 15:06:27 +1000620 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
621 mdname(mddev),
Trela, Maciej9af204c2010-03-08 16:02:44 +1100622 mddev->layout);
623 return ERR_PTR(-EINVAL);
624 }
625 if (mddev->raid_disks & 1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000626 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
627 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100628 return ERR_PTR(-EINVAL);
629 }
630 if (mddev->degraded != (mddev->raid_disks>>1)) {
NeilBrownb5a20962010-05-03 15:06:27 +1000631 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
632 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100633 return ERR_PTR(-EINVAL);
634 }
635
636 /* Set new parameters */
637 mddev->new_level = 0;
Maciej Trela001048a2010-06-16 11:55:14 +0100638 mddev->new_layout = 0;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100639 mddev->new_chunk_sectors = mddev->chunk_sectors;
640 mddev->delta_disks = - mddev->raid_disks / 2;
641 mddev->raid_disks += mddev->delta_disks;
642 mddev->degraded = 0;
643 /* make sure it will be not marked as dirty */
644 mddev->recovery_cp = MaxSector;
645
646 create_strip_zones(mddev, &priv_conf);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100647 return priv_conf;
648}
649
NeilBrownfd01b882011-10-11 16:47:53 +1100650static void *raid0_takeover_raid1(struct mddev *mddev)
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100651{
NeilBrowne373ab12011-10-11 16:48:59 +1100652 struct r0conf *priv_conf;
Jes Sorensen24b961f2012-04-01 23:48:38 +1000653 int chunksect;
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100654
655 /* Check layout:
656 * - (N - 1) mirror drives must be already faulty
657 */
658 if ((mddev->raid_disks - 1) != mddev->degraded) {
659 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
660 mdname(mddev));
661 return ERR_PTR(-EINVAL);
662 }
663
Jes Sorensen24b961f2012-04-01 23:48:38 +1000664 /*
665 * a raid1 doesn't have the notion of chunk size, so
666 * figure out the largest suitable size we can use.
667 */
668 chunksect = 64 * 2; /* 64K by default */
669
670 /* The array must be an exact multiple of chunksize */
671 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
672 chunksect >>= 1;
673
674 if ((chunksect << 9) < PAGE_SIZE)
675 /* array size does not allow a suitable chunk size */
676 return ERR_PTR(-EINVAL);
677
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100678 /* Set new parameters */
679 mddev->new_level = 0;
680 mddev->new_layout = 0;
Jes Sorensen24b961f2012-04-01 23:48:38 +1000681 mddev->new_chunk_sectors = chunksect;
682 mddev->chunk_sectors = chunksect;
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100683 mddev->delta_disks = 1 - mddev->raid_disks;
Krzysztof Wojcikf7bee802011-02-14 10:01:41 +1100684 mddev->raid_disks = 1;
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100685 /* make sure it will be not marked as dirty */
686 mddev->recovery_cp = MaxSector;
687
688 create_strip_zones(mddev, &priv_conf);
689 return priv_conf;
690}
691
NeilBrownfd01b882011-10-11 16:47:53 +1100692static void *raid0_takeover(struct mddev *mddev)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100693{
694 /* raid0 can take over:
Maciej Trela049d6c12010-06-16 11:56:12 +0100695 * raid4 - if all data disks are active.
Trela, Maciej9af204c2010-03-08 16:02:44 +1100696 * raid5 - providing it is Raid4 layout and one disk is faulty
697 * raid10 - assuming we have all necessary active disks
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100698 * raid1 - with (N -1) mirror drives faulty
Trela, Maciej9af204c2010-03-08 16:02:44 +1100699 */
Maciej Trela049d6c12010-06-16 11:56:12 +0100700 if (mddev->level == 4)
701 return raid0_takeover_raid45(mddev);
702
Trela, Maciej9af204c2010-03-08 16:02:44 +1100703 if (mddev->level == 5) {
704 if (mddev->layout == ALGORITHM_PARITY_N)
Maciej Trela049d6c12010-06-16 11:56:12 +0100705 return raid0_takeover_raid45(mddev);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100706
NeilBrownb5a20962010-05-03 15:06:27 +1000707 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
708 mdname(mddev), ALGORITHM_PARITY_N);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100709 }
710
711 if (mddev->level == 10)
712 return raid0_takeover_raid10(mddev);
713
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100714 if (mddev->level == 1)
715 return raid0_takeover_raid1(mddev);
716
717 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
718 mddev->level);
719
Trela, Maciej9af204c2010-03-08 16:02:44 +1100720 return ERR_PTR(-EINVAL);
721}
722
NeilBrownfd01b882011-10-11 16:47:53 +1100723static void raid0_quiesce(struct mddev *mddev, int state)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100724{
725}
726
NeilBrown84fc4b52011-10-11 16:49:58 +1100727static struct md_personality raid0_personality=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728{
729 .name = "raid0",
NeilBrown2604b702006-01-06 00:20:36 -0800730 .level = 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731 .owner = THIS_MODULE,
732 .make_request = raid0_make_request,
733 .run = raid0_run,
734 .stop = raid0_stop,
735 .status = raid0_status,
Dan Williams80c3a6c2009-03-17 18:10:40 -0700736 .size = raid0_size,
Trela, Maciej9af204c2010-03-08 16:02:44 +1100737 .takeover = raid0_takeover,
738 .quiesce = raid0_quiesce,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739};
740
741static int __init raid0_init (void)
742{
NeilBrown2604b702006-01-06 00:20:36 -0800743 return register_md_personality (&raid0_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744}
745
746static void raid0_exit (void)
747{
NeilBrown2604b702006-01-06 00:20:36 -0800748 unregister_md_personality (&raid0_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749}
750
751module_init(raid0_init);
752module_exit(raid0_exit);
753MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +1100754MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755MODULE_ALIAS("md-personality-2"); /* RAID0 */
NeilBrownd9d166c2006-01-06 00:20:51 -0800756MODULE_ALIAS("md-raid0");
NeilBrown2604b702006-01-06 00:20:36 -0800757MODULE_ALIAS("md-level-0");