blob: ba6b85de96d209635e640efb19a6569bfd9e8a38 [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
NeilBrown26be34d2006-10-03 01:15:53 -070028static int raid0_congested(void *data, int bits)
29{
NeilBrownfd01b882011-10-11 16:47:53 +110030 struct mddev *mddev = data;
NeilBrowne373ab12011-10-11 16:48:59 +110031 struct r0conf *conf = mddev->private;
NeilBrown3cb03002011-10-11 16:45:26 +110032 struct md_rdev **devlist = conf->devlist;
NeilBrown84707f32010-03-16 17:23:35 +110033 int raid_disks = conf->strip_zone[0].nb_dev;
NeilBrown26be34d2006-10-03 01:15:53 -070034 int i, ret = 0;
35
NeilBrown3fa841d2009-09-23 18:10:29 +100036 if (mddev_congested(mddev, bits))
37 return 1;
38
NeilBrown84707f32010-03-16 17:23:35 +110039 for (i = 0; i < raid_disks && !ret ; i++) {
Jens Axboe165125e2007-07-24 09:28:11 +020040 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
NeilBrown26be34d2006-10-03 01:15:53 -070041
42 ret |= bdi_congested(&q->backing_dev_info, bits);
43 }
44 return ret;
45}
46
raz ben yehuda46994192009-06-16 17:00:54 +100047/*
48 * inform the user of the raid configuration
49*/
NeilBrownfd01b882011-10-11 16:47:53 +110050static void dump_zones(struct mddev *mddev)
raz ben yehuda46994192009-06-16 17:00:54 +100051{
NeilBrown50de8df2011-10-07 14:23:22 +110052 int j, k;
raz ben yehuda46994192009-06-16 17:00:54 +100053 sector_t zone_size = 0;
54 sector_t zone_start = 0;
55 char b[BDEVNAME_SIZE];
NeilBrowne373ab12011-10-11 16:48:59 +110056 struct r0conf *conf = mddev->private;
NeilBrown84707f32010-03-16 17:23:35 +110057 int raid_disks = conf->strip_zone[0].nb_dev;
NeilBrown50de8df2011-10-07 14:23:22 +110058 printk(KERN_INFO "md: RAID0 configuration for %s - %d zone%s\n",
59 mdname(mddev),
60 conf->nr_strip_zones, conf->nr_strip_zones==1?"":"s");
raz ben yehuda46994192009-06-16 17:00:54 +100061 for (j = 0; j < conf->nr_strip_zones; j++) {
NeilBrown50de8df2011-10-07 14:23:22 +110062 printk(KERN_INFO "md: zone%d=[", j);
raz ben yehuda46994192009-06-16 17:00:54 +100063 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
NeilBrown50de8df2011-10-07 14:23:22 +110064 printk(KERN_CONT "%s%s", k?"/":"",
NeilBrown84707f32010-03-16 17:23:35 +110065 bdevname(conf->devlist[j*raid_disks
raz ben yehuda46994192009-06-16 17:00:54 +100066 + k]->bdev, b));
NeilBrownb5a20962010-05-03 15:06:27 +100067 printk(KERN_CONT "]\n");
raz ben yehuda46994192009-06-16 17:00:54 +100068
69 zone_size = conf->strip_zone[j].zone_end - zone_start;
NeilBrown50de8df2011-10-07 14:23:22 +110070 printk(KERN_INFO " zone-offset=%10lluKB, "
71 "device-offset=%10lluKB, size=%10lluKB\n",
raz ben yehuda46994192009-06-16 17:00:54 +100072 (unsigned long long)zone_start>>1,
73 (unsigned long long)conf->strip_zone[j].dev_start>>1,
74 (unsigned long long)zone_size>>1);
75 zone_start = conf->strip_zone[j].zone_end;
76 }
NeilBrown50de8df2011-10-07 14:23:22 +110077 printk(KERN_INFO "\n");
raz ben yehuda46994192009-06-16 17:00:54 +100078}
79
NeilBrowne373ab12011-10-11 16:48:59 +110080static int create_strip_zones(struct mddev *mddev, struct r0conf **private_conf)
Linus Torvalds1da177e2005-04-16 15:20:36 -070081{
NeilBrowna9f326e2009-09-23 18:06:41 +100082 int i, c, err;
NeilBrown49f357a22009-06-16 16:50:35 +100083 sector_t curr_zone_end, sectors;
NeilBrown3cb03002011-10-11 16:45:26 +110084 struct md_rdev *smallest, *rdev1, *rdev2, *rdev, **dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 struct strip_zone *zone;
86 int cnt;
87 char b[BDEVNAME_SIZE];
NeilBrown50de8df2011-10-07 14:23:22 +110088 char b2[BDEVNAME_SIZE];
NeilBrowne373ab12011-10-11 16:48:59 +110089 struct r0conf *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
Shaohua Lic83057a2012-10-11 13:25:44 +110090 bool discard_supported = false;
Andre Nolled7b0032009-06-16 16:47:36 +100091
92 if (!conf)
93 return -ENOMEM;
NeilBrowndafb20f2012-03-19 12:46:39 +110094 rdev_for_each(rdev1, mddev) {
NeilBrown50de8df2011-10-07 14:23:22 +110095 pr_debug("md/raid0:%s: looking at %s\n",
96 mdname(mddev),
97 bdevname(rdev1->bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 c = 0;
NeilBrown13f26822009-06-18 08:48:55 +100099
100 /* round size to chunk_size */
101 sectors = rdev1->sectors;
102 sector_div(sectors, mddev->chunk_sectors);
103 rdev1->sectors = sectors * mddev->chunk_sectors;
104
NeilBrowndafb20f2012-03-19 12:46:39 +1100105 rdev_for_each(rdev2, mddev) {
NeilBrown50de8df2011-10-07 14:23:22 +1100106 pr_debug("md/raid0:%s: comparing %s(%llu)"
107 " with %s(%llu)\n",
108 mdname(mddev),
109 bdevname(rdev1->bdev,b),
110 (unsigned long long)rdev1->sectors,
111 bdevname(rdev2->bdev,b2),
112 (unsigned long long)rdev2->sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 if (rdev2 == rdev1) {
NeilBrown50de8df2011-10-07 14:23:22 +1100114 pr_debug("md/raid0:%s: END\n",
115 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 break;
117 }
Andre Nolldd8ac332009-03-31 14:33:13 +1100118 if (rdev2->sectors == rdev1->sectors) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 /*
120 * Not unique, don't count it as a new
121 * group
122 */
NeilBrown50de8df2011-10-07 14:23:22 +1100123 pr_debug("md/raid0:%s: EQUAL\n",
124 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 c = 1;
126 break;
127 }
NeilBrown50de8df2011-10-07 14:23:22 +1100128 pr_debug("md/raid0:%s: NOT EQUAL\n",
129 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130 }
131 if (!c) {
NeilBrown50de8df2011-10-07 14:23:22 +1100132 pr_debug("md/raid0:%s: ==> UNIQUE\n",
133 mdname(mddev));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 conf->nr_strip_zones++;
NeilBrown50de8df2011-10-07 14:23:22 +1100135 pr_debug("md/raid0:%s: %d zones\n",
136 mdname(mddev), conf->nr_strip_zones);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 }
138 }
NeilBrown50de8df2011-10-07 14:23:22 +1100139 pr_debug("md/raid0:%s: FINAL %d zones\n",
140 mdname(mddev), conf->nr_strip_zones);
Andre Nolled7b0032009-06-16 16:47:36 +1000141 err = -ENOMEM;
NeilBrown9ffae0c2006-01-06 00:20:32 -0800142 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 conf->nr_strip_zones, GFP_KERNEL);
144 if (!conf->strip_zone)
Andre Nolled7b0032009-06-16 16:47:36 +1000145 goto abort;
NeilBrown3cb03002011-10-11 16:45:26 +1100146 conf->devlist = kzalloc(sizeof(struct md_rdev*)*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 conf->nr_strip_zones*mddev->raid_disks,
148 GFP_KERNEL);
149 if (!conf->devlist)
Andre Nolled7b0032009-06-16 16:47:36 +1000150 goto abort;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 /* The first zone must contain all devices, so here we check that
153 * there is a proper alignment of slots to devices and find them all
154 */
155 zone = &conf->strip_zone[0];
156 cnt = 0;
157 smallest = NULL;
NeilBrownb4145792009-06-16 16:50:52 +1000158 dev = conf->devlist;
Andre Nolled7b0032009-06-16 16:47:36 +1000159 err = -EINVAL;
NeilBrowndafb20f2012-03-19 12:46:39 +1100160 rdev_for_each(rdev1, mddev) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 int j = rdev1->raid_disk;
162
NeilBrowne93f68a2010-06-15 09:36:03 +0100163 if (mddev->level == 10) {
Trela, Maciej9af204c2010-03-08 16:02:44 +1100164 /* taking over a raid10-n2 array */
165 j /= 2;
NeilBrowne93f68a2010-06-15 09:36:03 +0100166 rdev1->new_raid_disk = j;
167 }
Trela, Maciej9af204c2010-03-08 16:02:44 +1100168
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100169 if (mddev->level == 1) {
170 /* taiking over a raid1 array-
171 * we have only one active disk
172 */
173 j = 0;
174 rdev1->new_raid_disk = j;
175 }
176
NeilBrownf96c9f32013-02-21 15:50:07 +1100177 if (j < 0) {
178 printk(KERN_ERR
179 "md/raid0:%s: remove inactive devices before converting to RAID0\n",
180 mdname(mddev));
181 goto abort;
182 }
183 if (j >= mddev->raid_disks) {
NeilBrownb5a20962010-05-03 15:06:27 +1000184 printk(KERN_ERR "md/raid0:%s: bad disk number %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 if (dev[j]) {
NeilBrownb5a20962010-05-03 15:06:27 +1000189 printk(KERN_ERR "md/raid0:%s: multiple devices for %d - "
190 "aborting!\n", mdname(mddev), j);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 goto abort;
192 }
NeilBrownb4145792009-06-16 16:50:52 +1000193 dev[j] = rdev1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +1000195 disk_stack_limits(mddev->gendisk, rdev1->bdev,
196 rdev1->data_offset << 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
NeilBrownba13da42012-03-19 12:46:39 +1100198 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn)
199 conf->has_merge_bvec = 1;
200
Andre Nolldd8ac332009-03-31 14:33:13 +1100201 if (!smallest || (rdev1->sectors < smallest->sectors))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 smallest = rdev1;
203 cnt++;
Shaohua Lic83057a2012-10-11 13:25:44 +1100204
205 if (blk_queue_discard(bdev_get_queue(rdev1->bdev)))
206 discard_supported = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 }
208 if (cnt != mddev->raid_disks) {
NeilBrownb5a20962010-05-03 15:06:27 +1000209 printk(KERN_ERR "md/raid0:%s: too few disks (%d of %d) - "
210 "aborting!\n", mdname(mddev), cnt, mddev->raid_disks);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 goto abort;
212 }
213 zone->nb_dev = cnt;
NeilBrown49f357a22009-06-16 16:50:35 +1000214 zone->zone_end = smallest->sectors * cnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
NeilBrown49f357a22009-06-16 16:50:35 +1000216 curr_zone_end = zone->zone_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
218 /* now do the other zones */
219 for (i = 1; i < conf->nr_strip_zones; i++)
220 {
NeilBrowna9f326e2009-09-23 18:06:41 +1000221 int j;
222
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 zone = conf->strip_zone + i;
NeilBrownb4145792009-06-16 16:50:52 +1000224 dev = conf->devlist + i * mddev->raid_disks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
NeilBrown50de8df2011-10-07 14:23:22 +1100226 pr_debug("md/raid0:%s: zone %d\n", mdname(mddev), i);
NeilBrownd27a43ab2009-06-16 16:46:46 +1000227 zone->dev_start = smallest->sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 smallest = NULL;
229 c = 0;
230
231 for (j=0; j<cnt; j++) {
NeilBrownb4145792009-06-16 16:50:52 +1000232 rdev = conf->devlist[j];
NeilBrownd27a43ab2009-06-16 16:46:46 +1000233 if (rdev->sectors <= zone->dev_start) {
NeilBrown50de8df2011-10-07 14:23:22 +1100234 pr_debug("md/raid0:%s: checking %s ... nope\n",
235 mdname(mddev),
236 bdevname(rdev->bdev, b));
Andre Nolldd8ac332009-03-31 14:33:13 +1100237 continue;
238 }
NeilBrown50de8df2011-10-07 14:23:22 +1100239 pr_debug("md/raid0:%s: checking %s ..."
240 " contained as device %d\n",
241 mdname(mddev),
242 bdevname(rdev->bdev, b), c);
NeilBrownb4145792009-06-16 16:50:52 +1000243 dev[c] = rdev;
Andre Nolldd8ac332009-03-31 14:33:13 +1100244 c++;
245 if (!smallest || rdev->sectors < smallest->sectors) {
246 smallest = rdev;
NeilBrown50de8df2011-10-07 14:23:22 +1100247 pr_debug("md/raid0:%s: (%llu) is smallest!.\n",
248 mdname(mddev),
249 (unsigned long long)rdev->sectors);
Andre Nolldd8ac332009-03-31 14:33:13 +1100250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251 }
252
253 zone->nb_dev = c;
NeilBrown49f357a22009-06-16 16:50:35 +1000254 sectors = (smallest->sectors - zone->dev_start) * c;
NeilBrown50de8df2011-10-07 14:23:22 +1100255 pr_debug("md/raid0:%s: zone->nb_dev: %d, sectors: %llu\n",
256 mdname(mddev),
257 zone->nb_dev, (unsigned long long)sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
NeilBrown49f357a22009-06-16 16:50:35 +1000259 curr_zone_end += sectors;
NeilBrownd27a43ab2009-06-16 16:46:46 +1000260 zone->zone_end = curr_zone_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
NeilBrown50de8df2011-10-07 14:23:22 +1100262 pr_debug("md/raid0:%s: current zone start: %llu\n",
263 mdname(mddev),
264 (unsigned long long)smallest->sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 }
NeilBrown26be34d2006-10-03 01:15:53 -0700266 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
267 mddev->queue->backing_dev_info.congested_data = mddev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268
raz ben yehuda92e59b62009-06-16 17:00:57 +1000269 /*
270 * now since we have the hard sector sizes, we can make sure
271 * chunk size is a multiple of that sector size
272 */
Andre Noll9d8f0362009-06-18 08:45:01 +1000273 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
NeilBrownb5a20962010-05-03 15:06:27 +1000274 printk(KERN_ERR "md/raid0:%s: chunk_size of %d not valid\n",
raz ben yehuda92e59b62009-06-16 17:00:57 +1000275 mdname(mddev),
Andre Noll9d8f0362009-06-18 08:45:01 +1000276 mddev->chunk_sectors << 9);
raz ben yehuda92e59b62009-06-16 17:00:57 +1000277 goto abort;
278 }
Martin K. Petersen8f6c2e42009-07-01 11:13:45 +1000279
280 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
281 blk_queue_io_opt(mddev->queue,
282 (mddev->chunk_sectors << 9) * mddev->raid_disks);
283
Shaohua Lic83057a2012-10-11 13:25:44 +1100284 if (!discard_supported)
285 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
286 else
287 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, mddev->queue);
288
NeilBrown50de8df2011-10-07 14:23:22 +1100289 pr_debug("md/raid0:%s: done.\n", mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100290 *private_conf = conf;
291
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 return 0;
Andre Noll5568a602009-06-16 16:47:21 +1000293abort:
Andre Nolled7b0032009-06-16 16:47:36 +1000294 kfree(conf->strip_zone);
295 kfree(conf->devlist);
296 kfree(conf);
NeilBrown58ebb342013-02-21 15:36:38 +1100297 *private_conf = ERR_PTR(err);
Andre Nolled7b0032009-06-16 16:47:36 +1000298 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299}
300
NeilBrownba13da42012-03-19 12:46:39 +1100301/* Find the zone which holds a particular offset
302 * Update *sectorp to be an offset in that zone
303 */
304static struct strip_zone *find_zone(struct r0conf *conf,
305 sector_t *sectorp)
306{
307 int i;
308 struct strip_zone *z = conf->strip_zone;
309 sector_t sector = *sectorp;
310
311 for (i = 0; i < conf->nr_strip_zones; i++)
312 if (sector < z[i].zone_end) {
313 if (i)
314 *sectorp = sector - z[i-1].zone_end;
315 return z + i;
316 }
317 BUG();
318}
319
320/*
321 * remaps the bio to the target device. we separate two flows.
322 * power 2 flow and a general flow for the sake of perfromance
323*/
324static struct md_rdev *map_sector(struct mddev *mddev, struct strip_zone *zone,
325 sector_t sector, sector_t *sector_offset)
326{
327 unsigned int sect_in_chunk;
328 sector_t chunk;
329 struct r0conf *conf = mddev->private;
330 int raid_disks = conf->strip_zone[0].nb_dev;
331 unsigned int chunk_sects = mddev->chunk_sectors;
332
333 if (is_power_of_2(chunk_sects)) {
334 int chunksect_bits = ffz(~chunk_sects);
335 /* find the sector offset inside the chunk */
336 sect_in_chunk = sector & (chunk_sects - 1);
337 sector >>= chunksect_bits;
338 /* chunk in zone */
339 chunk = *sector_offset;
340 /* quotient is the chunk in real device*/
341 sector_div(chunk, zone->nb_dev << chunksect_bits);
342 } else{
343 sect_in_chunk = sector_div(sector, chunk_sects);
344 chunk = *sector_offset;
345 sector_div(chunk, chunk_sects * zone->nb_dev);
346 }
347 /*
348 * position the bio over the real device
349 * real sector = chunk in device + starting of zone
350 * + the position in the chunk
351 */
352 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
353 return conf->devlist[(zone - conf->strip_zone)*raid_disks
354 + sector_div(sector, zone->nb_dev)];
355}
356
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357/**
NeilBrownba13da42012-03-19 12:46:39 +1100358 * raid0_mergeable_bvec -- tell bio layer if two requests can be merged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 * @q: request queue
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200360 * @bvm: properties of new bio
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 * @biovec: the request that could be merged to it.
362 *
363 * Return amount of bytes we can accept at this offset
364 */
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200365static int raid0_mergeable_bvec(struct request_queue *q,
366 struct bvec_merge_data *bvm,
367 struct bio_vec *biovec)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368{
NeilBrownfd01b882011-10-11 16:47:53 +1100369 struct mddev *mddev = q->queuedata;
NeilBrownba13da42012-03-19 12:46:39 +1100370 struct r0conf *conf = mddev->private;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200371 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
NeilBrownba13da42012-03-19 12:46:39 +1100372 sector_t sector_offset = sector;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 int max;
Andre Noll9d8f0362009-06-18 08:45:01 +1000374 unsigned int chunk_sectors = mddev->chunk_sectors;
Alasdair G Kergoncc371e62008-07-03 09:53:43 +0200375 unsigned int bio_sectors = bvm->bi_size >> 9;
NeilBrownba13da42012-03-19 12:46:39 +1100376 struct strip_zone *zone;
377 struct md_rdev *rdev;
378 struct request_queue *subq;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
NeilBrownd6e412e2009-06-18 08:47:00 +1000380 if (is_power_of_2(chunk_sectors))
raz ben yehudafbb704e2009-06-16 17:02:05 +1000381 max = (chunk_sectors - ((sector & (chunk_sectors-1))
382 + bio_sectors)) << 9;
383 else
384 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
385 + bio_sectors)) << 9;
NeilBrownba13da42012-03-19 12:46:39 +1100386 if (max < 0)
387 max = 0; /* bio_add cannot handle a negative return */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 if (max <= biovec->bv_len && bio_sectors == 0)
389 return biovec->bv_len;
NeilBrownba13da42012-03-19 12:46:39 +1100390 if (max < biovec->bv_len)
391 /* too small already, no need to check further */
392 return max;
393 if (!conf->has_merge_bvec)
394 return max;
395
396 /* May need to check subordinate device */
397 sector = sector_offset;
398 zone = find_zone(mddev->private, &sector_offset);
399 rdev = map_sector(mddev, zone, sector, &sector_offset);
400 subq = bdev_get_queue(rdev->bdev);
401 if (subq->merge_bvec_fn) {
402 bvm->bi_bdev = rdev->bdev;
403 bvm->bi_sector = sector_offset + zone->dev_start +
404 rdev->data_offset;
405 return min(max, subq->merge_bvec_fn(subq, bvm, biovec));
406 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 return max;
408}
409
NeilBrownfd01b882011-10-11 16:47:53 +1100410static sector_t raid0_size(struct mddev *mddev, sector_t sectors, int raid_disks)
Dan Williams80c3a6c2009-03-17 18:10:40 -0700411{
412 sector_t array_sectors = 0;
NeilBrown3cb03002011-10-11 16:45:26 +1100413 struct md_rdev *rdev;
Dan Williams80c3a6c2009-03-17 18:10:40 -0700414
415 WARN_ONCE(sectors || raid_disks,
416 "%s does not support generic reshape\n", __func__);
417
NeilBrowndafb20f2012-03-19 12:46:39 +1100418 rdev_for_each(rdev, mddev)
NeilBrowna6468532013-02-21 14:33:17 +1100419 array_sectors += (rdev->sectors &
420 ~(sector_t)(mddev->chunk_sectors-1));
Dan Williams80c3a6c2009-03-17 18:10:40 -0700421
422 return array_sectors;
423}
424
majianpeng0366ef82012-04-02 09:48:37 +1000425static int raid0_stop(struct mddev *mddev);
426
NeilBrownfd01b882011-10-11 16:47:53 +1100427static int raid0_run(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428{
NeilBrowne373ab12011-10-11 16:48:59 +1100429 struct r0conf *conf;
Andre Noll5568a602009-06-16 16:47:21 +1000430 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431
Andre Noll9d8f0362009-06-18 08:45:01 +1000432 if (mddev->chunk_sectors == 0) {
NeilBrownb5a20962010-05-03 15:06:27 +1000433 printk(KERN_ERR "md/raid0:%s: chunk size must be set.\n",
434 mdname(mddev));
NeilBrown2604b702006-01-06 00:20:36 -0800435 return -EINVAL;
436 }
Andre Noll0894cc32009-06-18 08:49:23 +1000437 if (md_check_no_bitmap(mddev))
438 return -EINVAL;
Martin K. Petersen086fa5f2010-02-26 00:20:38 -0500439 blk_queue_max_hw_sectors(mddev->queue, mddev->chunk_sectors);
Martin K. Petersen4363ac72012-09-18 12:19:27 -0400440 blk_queue_max_write_same_sectors(mddev->queue, mddev->chunk_sectors);
Shaohua Lic83057a2012-10-11 13:25:44 +1100441 blk_queue_max_discard_sectors(mddev->queue, mddev->chunk_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
Trela, Maciej9af204c2010-03-08 16:02:44 +1100443 /* if private is not null, we are here after takeover */
444 if (mddev->private == NULL) {
445 ret = create_strip_zones(mddev, &conf);
446 if (ret < 0)
447 return ret;
448 mddev->private = conf;
449 }
450 conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
452 /* calculate array device size */
Dan Williams1f403622009-03-31 14:59:03 +1100453 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454
NeilBrownb5a20962010-05-03 15:06:27 +1000455 printk(KERN_INFO "md/raid0:%s: md_size is %llu sectors.\n",
456 mdname(mddev),
457 (unsigned long long)mddev->array_sectors);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 /* calculate the max read-ahead size.
459 * For read-ahead of large files to be effective, we need to
460 * readahead at least twice a whole stripe. i.e. number of devices
461 * multiplied by chunk size times 2.
462 * If an individual device has an ra_pages greater than the
463 * chunk size, then we will not drive that device as hard as it
464 * wants. We consider this a configuration error: a larger
465 * chunksize should be used in that case.
466 */
467 {
Andre Noll9d8f0362009-06-18 08:45:01 +1000468 int stripe = mddev->raid_disks *
469 (mddev->chunk_sectors << 9) / PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
471 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
472 }
473
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
raz ben yehuda46994192009-06-16 17:00:54 +1000475 dump_zones(mddev);
majianpeng0366ef82012-04-02 09:48:37 +1000476
477 ret = md_integrity_register(mddev);
478 if (ret)
479 raid0_stop(mddev);
480
481 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
NeilBrownfd01b882011-10-11 16:47:53 +1100484static int raid0_stop(struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485{
NeilBrowne373ab12011-10-11 16:48:59 +1100486 struct r0conf *conf = mddev->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
488 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700489 kfree(conf->strip_zone);
Andre Nollfb5ab4b2009-06-16 16:48:19 +1000490 kfree(conf->devlist);
Jesper Juhl990a8ba2005-06-21 17:17:30 -0700491 kfree(conf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492 mddev->private = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700493 return 0;
494}
495
raz ben yehudafbb704e2009-06-16 17:02:05 +1000496/*
497 * Is io distribute over 1 or more chunks ?
498*/
NeilBrownfd01b882011-10-11 16:47:53 +1100499static inline int is_io_in_chunk_boundary(struct mddev *mddev,
raz ben yehudafbb704e2009-06-16 17:02:05 +1000500 unsigned int chunk_sects, struct bio *bio)
501{
NeilBrownd6e412e2009-06-18 08:47:00 +1000502 if (likely(is_power_of_2(chunk_sects))) {
Kent Overstreet4f024f32013-10-11 15:44:27 -0700503 return chunk_sects >=
504 ((bio->bi_iter.bi_sector & (chunk_sects-1))
Kent Overstreetaa8b57a2013-02-05 15:19:29 -0800505 + bio_sectors(bio));
raz ben yehudafbb704e2009-06-16 17:02:05 +1000506 } else{
Kent Overstreet4f024f32013-10-11 15:44:27 -0700507 sector_t sector = bio->bi_iter.bi_sector;
raz ben yehudafbb704e2009-06-16 17:02:05 +1000508 return chunk_sects >= (sector_div(sector, chunk_sects)
Kent Overstreetaa8b57a2013-02-05 15:19:29 -0800509 + bio_sectors(bio));
raz ben yehudafbb704e2009-06-16 17:02:05 +1000510 }
511}
512
Linus Torvaldsb4fdcb02011-11-04 17:06:58 -0700513static void raid0_make_request(struct mddev *mddev, struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 struct strip_zone *zone;
NeilBrown3cb03002011-10-11 16:45:26 +1100516 struct md_rdev *tmp_dev;
Kent Overstreet20d01892013-11-23 18:21:01 -0800517 struct bio *split;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518
Tejun Heoe9c74692010-09-03 11:56:18 +0200519 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
520 md_flush_request(mddev, bio);
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200521 return;
NeilBrowne5dcdd82005-09-09 16:23:41 -0700522 }
523
Kent Overstreet20d01892013-11-23 18:21:01 -0800524 do {
Kent Overstreet4f024f32013-10-11 15:44:27 -0700525 sector_t sector = bio->bi_iter.bi_sector;
Kent Overstreet20d01892013-11-23 18:21:01 -0800526 unsigned chunk_sects = mddev->chunk_sectors;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
Kent Overstreet20d01892013-11-23 18:21:01 -0800528 unsigned sectors = chunk_sects -
529 (likely(is_power_of_2(chunk_sects))
530 ? (sector & (chunk_sects-1))
531 : sector_div(sector, chunk_sects));
Christoph Hellwig5a7bbad2011-09-12 12:12:01 +0200532
Kent Overstreet20d01892013-11-23 18:21:01 -0800533 if (sectors < bio_sectors(bio)) {
534 split = bio_split(bio, sectors, GFP_NOIO, fs_bio_set);
535 bio_chain(split, bio);
536 } else {
537 split = bio;
538 }
Shaohua Lic83057a2012-10-11 13:25:44 +1100539
Kent Overstreet20d01892013-11-23 18:21:01 -0800540 zone = find_zone(mddev->private, &sector);
541 tmp_dev = map_sector(mddev, zone, sector, &sector);
542 split->bi_bdev = tmp_dev->bdev;
543 split->bi_iter.bi_sector = sector + zone->dev_start +
544 tmp_dev->data_offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545
Kent Overstreet20d01892013-11-23 18:21:01 -0800546 if (unlikely((split->bi_rw & REQ_DISCARD) &&
547 !blk_queue_discard(bdev_get_queue(split->bi_bdev)))) {
548 /* Just ignore it */
549 bio_endio(split, 0);
550 } else
551 generic_make_request(split);
552 } while (split != bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553}
NeilBrown8299d7f2007-10-16 23:30:53 -0700554
NeilBrownfd01b882011-10-11 16:47:53 +1100555static void raid0_status(struct seq_file *seq, struct mddev *mddev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556{
Andre Noll9d8f0362009-06-18 08:45:01 +1000557 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 return;
559}
560
NeilBrownfd01b882011-10-11 16:47:53 +1100561static void *raid0_takeover_raid45(struct mddev *mddev)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100562{
NeilBrown3cb03002011-10-11 16:45:26 +1100563 struct md_rdev *rdev;
NeilBrowne373ab12011-10-11 16:48:59 +1100564 struct r0conf *priv_conf;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100565
566 if (mddev->degraded != 1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000567 printk(KERN_ERR "md/raid0:%s: raid5 must be degraded! Degraded disks: %d\n",
568 mdname(mddev),
Trela, Maciej9af204c2010-03-08 16:02:44 +1100569 mddev->degraded);
570 return ERR_PTR(-EINVAL);
571 }
572
NeilBrowndafb20f2012-03-19 12:46:39 +1100573 rdev_for_each(rdev, mddev) {
Trela, Maciej9af204c2010-03-08 16:02:44 +1100574 /* check slot number for a disk */
575 if (rdev->raid_disk == mddev->raid_disks-1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000576 printk(KERN_ERR "md/raid0:%s: raid5 must have missing parity disk!\n",
577 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100578 return ERR_PTR(-EINVAL);
579 }
NeilBrowneea136d2013-06-26 11:55:20 +1000580 rdev->sectors = mddev->dev_sectors;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100581 }
582
583 /* Set new parameters */
584 mddev->new_level = 0;
Maciej Trela001048a2010-06-16 11:55:14 +0100585 mddev->new_layout = 0;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100586 mddev->new_chunk_sectors = mddev->chunk_sectors;
587 mddev->raid_disks--;
588 mddev->delta_disks = -1;
589 /* make sure it will be not marked as dirty */
590 mddev->recovery_cp = MaxSector;
591
592 create_strip_zones(mddev, &priv_conf);
593 return priv_conf;
594}
595
NeilBrownfd01b882011-10-11 16:47:53 +1100596static void *raid0_takeover_raid10(struct mddev *mddev)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100597{
NeilBrowne373ab12011-10-11 16:48:59 +1100598 struct r0conf *priv_conf;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100599
600 /* Check layout:
601 * - far_copies must be 1
602 * - near_copies must be 2
603 * - disks number must be even
604 * - all mirrors must be already degraded
605 */
606 if (mddev->layout != ((1 << 8) + 2)) {
NeilBrownb5a20962010-05-03 15:06:27 +1000607 printk(KERN_ERR "md/raid0:%s:: Raid0 cannot takover layout: 0x%x\n",
608 mdname(mddev),
Trela, Maciej9af204c2010-03-08 16:02:44 +1100609 mddev->layout);
610 return ERR_PTR(-EINVAL);
611 }
612 if (mddev->raid_disks & 1) {
NeilBrownb5a20962010-05-03 15:06:27 +1000613 printk(KERN_ERR "md/raid0:%s: Raid0 cannot takover Raid10 with odd disk number.\n",
614 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100615 return ERR_PTR(-EINVAL);
616 }
617 if (mddev->degraded != (mddev->raid_disks>>1)) {
NeilBrownb5a20962010-05-03 15:06:27 +1000618 printk(KERN_ERR "md/raid0:%s: All mirrors must be already degraded!\n",
619 mdname(mddev));
Trela, Maciej9af204c2010-03-08 16:02:44 +1100620 return ERR_PTR(-EINVAL);
621 }
622
623 /* Set new parameters */
624 mddev->new_level = 0;
Maciej Trela001048a2010-06-16 11:55:14 +0100625 mddev->new_layout = 0;
Trela, Maciej9af204c2010-03-08 16:02:44 +1100626 mddev->new_chunk_sectors = mddev->chunk_sectors;
627 mddev->delta_disks = - mddev->raid_disks / 2;
628 mddev->raid_disks += mddev->delta_disks;
629 mddev->degraded = 0;
630 /* make sure it will be not marked as dirty */
631 mddev->recovery_cp = MaxSector;
632
633 create_strip_zones(mddev, &priv_conf);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100634 return priv_conf;
635}
636
NeilBrownfd01b882011-10-11 16:47:53 +1100637static void *raid0_takeover_raid1(struct mddev *mddev)
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100638{
NeilBrowne373ab12011-10-11 16:48:59 +1100639 struct r0conf *priv_conf;
Jes Sorensen24b961f2012-04-01 23:48:38 +1000640 int chunksect;
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100641
642 /* Check layout:
643 * - (N - 1) mirror drives must be already faulty
644 */
645 if ((mddev->raid_disks - 1) != mddev->degraded) {
646 printk(KERN_ERR "md/raid0:%s: (N - 1) mirrors drives must be already faulty!\n",
647 mdname(mddev));
648 return ERR_PTR(-EINVAL);
649 }
650
Jes Sorensen24b961f2012-04-01 23:48:38 +1000651 /*
652 * a raid1 doesn't have the notion of chunk size, so
653 * figure out the largest suitable size we can use.
654 */
655 chunksect = 64 * 2; /* 64K by default */
656
657 /* The array must be an exact multiple of chunksize */
658 while (chunksect && (mddev->array_sectors & (chunksect - 1)))
659 chunksect >>= 1;
660
661 if ((chunksect << 9) < PAGE_SIZE)
662 /* array size does not allow a suitable chunk size */
663 return ERR_PTR(-EINVAL);
664
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100665 /* Set new parameters */
666 mddev->new_level = 0;
667 mddev->new_layout = 0;
Jes Sorensen24b961f2012-04-01 23:48:38 +1000668 mddev->new_chunk_sectors = chunksect;
669 mddev->chunk_sectors = chunksect;
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100670 mddev->delta_disks = 1 - mddev->raid_disks;
Krzysztof Wojcikf7bee802011-02-14 10:01:41 +1100671 mddev->raid_disks = 1;
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100672 /* make sure it will be not marked as dirty */
673 mddev->recovery_cp = MaxSector;
674
675 create_strip_zones(mddev, &priv_conf);
676 return priv_conf;
677}
678
NeilBrownfd01b882011-10-11 16:47:53 +1100679static void *raid0_takeover(struct mddev *mddev)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100680{
681 /* raid0 can take over:
Maciej Trela049d6c12010-06-16 11:56:12 +0100682 * raid4 - if all data disks are active.
Trela, Maciej9af204c2010-03-08 16:02:44 +1100683 * raid5 - providing it is Raid4 layout and one disk is faulty
684 * raid10 - assuming we have all necessary active disks
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100685 * raid1 - with (N -1) mirror drives faulty
Trela, Maciej9af204c2010-03-08 16:02:44 +1100686 */
NeilBrowna8461a62014-08-06 16:34:27 +1000687
688 if (mddev->bitmap) {
689 printk(KERN_ERR "md/raid0: %s: cannot takeover array with bitmap\n",
690 mdname(mddev));
691 return ERR_PTR(-EBUSY);
692 }
Maciej Trela049d6c12010-06-16 11:56:12 +0100693 if (mddev->level == 4)
694 return raid0_takeover_raid45(mddev);
695
Trela, Maciej9af204c2010-03-08 16:02:44 +1100696 if (mddev->level == 5) {
697 if (mddev->layout == ALGORITHM_PARITY_N)
Maciej Trela049d6c12010-06-16 11:56:12 +0100698 return raid0_takeover_raid45(mddev);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100699
NeilBrownb5a20962010-05-03 15:06:27 +1000700 printk(KERN_ERR "md/raid0:%s: Raid can only takeover Raid5 with layout: %d\n",
701 mdname(mddev), ALGORITHM_PARITY_N);
Trela, Maciej9af204c2010-03-08 16:02:44 +1100702 }
703
704 if (mddev->level == 10)
705 return raid0_takeover_raid10(mddev);
706
Krzysztof Wojcikfc3a08b2011-01-31 13:47:13 +1100707 if (mddev->level == 1)
708 return raid0_takeover_raid1(mddev);
709
710 printk(KERN_ERR "Takeover from raid%i to raid0 not supported\n",
711 mddev->level);
712
Trela, Maciej9af204c2010-03-08 16:02:44 +1100713 return ERR_PTR(-EINVAL);
714}
715
NeilBrownfd01b882011-10-11 16:47:53 +1100716static void raid0_quiesce(struct mddev *mddev, int state)
Trela, Maciej9af204c2010-03-08 16:02:44 +1100717{
718}
719
NeilBrown84fc4b52011-10-11 16:49:58 +1100720static struct md_personality raid0_personality=
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721{
722 .name = "raid0",
NeilBrown2604b702006-01-06 00:20:36 -0800723 .level = 0,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724 .owner = THIS_MODULE,
725 .make_request = raid0_make_request,
726 .run = raid0_run,
727 .stop = raid0_stop,
728 .status = raid0_status,
Dan Williams80c3a6c2009-03-17 18:10:40 -0700729 .size = raid0_size,
Trela, Maciej9af204c2010-03-08 16:02:44 +1100730 .takeover = raid0_takeover,
731 .quiesce = raid0_quiesce,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732};
733
734static int __init raid0_init (void)
735{
NeilBrown2604b702006-01-06 00:20:36 -0800736 return register_md_personality (&raid0_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
738
739static void raid0_exit (void)
740{
NeilBrown2604b702006-01-06 00:20:36 -0800741 unregister_md_personality (&raid0_personality);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742}
743
744module_init(raid0_init);
745module_exit(raid0_exit);
746MODULE_LICENSE("GPL");
NeilBrown0efb9e62009-12-14 12:49:58 +1100747MODULE_DESCRIPTION("RAID0 (striping) personality for MD");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748MODULE_ALIAS("md-personality-2"); /* RAID0 */
NeilBrownd9d166c2006-01-06 00:20:51 -0800749MODULE_ALIAS("md-raid0");
NeilBrown2604b702006-01-06 00:20:36 -0800750MODULE_ALIAS("md-level-0");