]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blame - drivers/md/linear.c
md: raid5 crash during degradation
[mirror_ubuntu-zesty-kernel.git] / drivers / md / linear.c
CommitLineData
1da177e4
LT
1/*
2 linear.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
7 Linear mode management functions.
8
9 This program is free software; you can redistribute it and/or modify
10 it under the terms of the GNU General Public License as published by
11 the Free Software Foundation; either version 2, or (at your option)
12 any later version.
13
14 You should have received a copy of the GNU General Public License
15 (for example /usr/src/linux/COPYING); if not, write to the Free
16 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
17*/
18
bff61975
N
19#include <linux/blkdev.h>
20#include <linux/raid/md_u.h>
bff61975 21#include <linux/seq_file.h>
056075c7 22#include <linux/module.h>
5a0e3ad6 23#include <linux/slab.h>
43b2e5d8 24#include "md.h"
ef740c37 25#include "linear.h"
1da177e4 26
1da177e4
LT
27/*
28 * find which device holds a particular offset
29 */
a7120771 30static inline struct dev_info *which_dev(struct mddev *mddev, sector_t sector)
1da177e4 31{
aece3d1f 32 int lo, mid, hi;
e849b938 33 struct linear_conf *conf;
1da177e4 34
aece3d1f
SS
35 lo = 0;
36 hi = mddev->raid_disks - 1;
af11c397 37 conf = rcu_dereference(mddev->private);
1da177e4 38
aece3d1f
SS
39 /*
40 * Binary Search
41 */
42
43 while (hi > lo) {
44
45 mid = (hi + lo) / 2;
46 if (sector < conf->disks[mid].end_sector)
47 hi = mid;
48 else
49 lo = mid + 1;
50 }
51
52 return conf->disks + lo;
1da177e4
LT
53}
54
55/**
15945fee 56 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
1da177e4 57 * @q: request queue
cc371e66 58 * @bvm: properties of new bio
1da177e4
LT
59 * @biovec: the request that could be merged to it.
60 *
61 * Return amount of bytes we can take at this offset
62 */
cc371e66
AK
63static int linear_mergeable_bvec(struct request_queue *q,
64 struct bvec_merge_data *bvm,
65 struct bio_vec *biovec)
1da177e4 66{
fd01b88c 67 struct mddev *mddev = q->queuedata;
a7120771 68 struct dev_info *dev0;
cc371e66
AK
69 unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
70 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
1da177e4 71
af11c397 72 rcu_read_lock();
1da177e4 73 dev0 = which_dev(mddev, sector);
4db7cdc8 74 maxsectors = dev0->end_sector - sector;
af11c397 75 rcu_read_unlock();
1da177e4
LT
76
77 if (maxsectors < bio_sectors)
78 maxsectors = 0;
79 else
80 maxsectors -= bio_sectors;
81
82 if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
83 return biovec->bv_len;
84 /* The bytes available at this offset could be really big,
85 * so we cap at 2^31 to avoid overflow */
86 if (maxsectors > (1 << (31-9)))
87 return 1<<31;
88 return maxsectors << 9;
89}
90
26be34dc
N
91static int linear_congested(void *data, int bits)
92{
fd01b88c 93 struct mddev *mddev = data;
e849b938 94 struct linear_conf *conf;
26be34dc
N
95 int i, ret = 0;
96
3fa841d7
N
97 if (mddev_congested(mddev, bits))
98 return 1;
99
af11c397
S
100 rcu_read_lock();
101 conf = rcu_dereference(mddev->private);
102
26be34dc 103 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
165125e1 104 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
26be34dc
N
105 ret |= bdi_congested(&q->backing_dev_info, bits);
106 }
af11c397
S
107
108 rcu_read_unlock();
26be34dc
N
109 return ret;
110}
111
fd01b88c 112static sector_t linear_size(struct mddev *mddev, sector_t sectors, int raid_disks)
80c3a6ce 113{
e849b938 114 struct linear_conf *conf;
af11c397 115 sector_t array_sectors;
80c3a6ce 116
af11c397
S
117 rcu_read_lock();
118 conf = rcu_dereference(mddev->private);
80c3a6ce
DW
119 WARN_ONCE(sectors || raid_disks,
120 "%s does not support generic reshape\n", __func__);
af11c397
S
121 array_sectors = conf->array_sectors;
122 rcu_read_unlock();
80c3a6ce 123
af11c397 124 return array_sectors;
80c3a6ce
DW
125}
126
e849b938 127static struct linear_conf *linear_conf(struct mddev *mddev, int raid_disks)
1da177e4 128{
e849b938 129 struct linear_conf *conf;
3cb03002 130 struct md_rdev *rdev;
45d4582f 131 int i, cnt;
1da177e4 132
a7120771 133 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(struct dev_info),
1da177e4
LT
134 GFP_KERNEL);
135 if (!conf)
7c7546cc
N
136 return NULL;
137
1da177e4 138 cnt = 0;
d6e22150 139 conf->array_sectors = 0;
1da177e4 140
159ec1fc 141 list_for_each_entry(rdev, &mddev->disks, same_set) {
1da177e4 142 int j = rdev->raid_disk;
a7120771 143 struct dev_info *disk = conf->disks + j;
13f2682b 144 sector_t sectors;
1da177e4 145
13864515 146 if (j < 0 || j >= raid_disks || disk->rdev) {
2dc40f80
N
147 printk(KERN_ERR "md/linear:%s: disk numbering problem. Aborting!\n",
148 mdname(mddev));
1da177e4
LT
149 goto out;
150 }
151
152 disk->rdev = rdev;
13f2682b
N
153 if (mddev->chunk_sectors) {
154 sectors = rdev->sectors;
155 sector_div(sectors, mddev->chunk_sectors);
156 rdev->sectors = sectors * mddev->chunk_sectors;
157 }
1da177e4 158
8f6c2e4b
MP
159 disk_stack_limits(mddev->gendisk, rdev->bdev,
160 rdev->data_offset << 9);
1da177e4 161 /* as we don't honour merge_bvec_fn, we must never risk
627a2d3c
N
162 * violating it, so limit max_segments to 1 lying within
163 * a single page.
1da177e4 164 */
627a2d3c
N
165 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
166 blk_queue_max_segments(mddev->queue, 1);
167 blk_queue_segment_boundary(mddev->queue,
168 PAGE_CACHE_SIZE - 1);
169 }
1da177e4 170
dd8ac336 171 conf->array_sectors += rdev->sectors;
1da177e4 172 cnt++;
4db7cdc8 173
1da177e4 174 }
7c7546cc 175 if (cnt != raid_disks) {
2dc40f80
N
176 printk(KERN_ERR "md/linear:%s: not enough drives present. Aborting!\n",
177 mdname(mddev));
1da177e4
LT
178 goto out;
179 }
180
181 /*
45d4582f 182 * Here we calculate the device offsets.
1da177e4 183 */
4db7cdc8
SS
184 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
185
a778b73f 186 for (i = 1; i < raid_disks; i++)
4db7cdc8
SS
187 conf->disks[i].end_sector =
188 conf->disks[i-1].end_sector +
189 conf->disks[i].rdev->sectors;
15945fee 190
7c7546cc
N
191 return conf;
192
193out:
194 kfree(conf);
195 return NULL;
196}
197
fd01b88c 198static int linear_run (struct mddev *mddev)
7c7546cc 199{
e849b938 200 struct linear_conf *conf;
7c7546cc 201
0894cc30
AN
202 if (md_check_no_bitmap(mddev))
203 return -EINVAL;
7c7546cc
N
204 conf = linear_conf(mddev, mddev->raid_disks);
205
206 if (!conf)
207 return 1;
208 mddev->private = conf;
1f403624 209 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
7c7546cc 210
1da177e4 211 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
26be34dc
N
212 mddev->queue->backing_dev_info.congested_fn = linear_congested;
213 mddev->queue->backing_dev_info.congested_data = mddev;
a91a2785 214 return md_integrity_register(mddev);
7c7546cc 215}
1da177e4 216
fd01b88c 217static int linear_add(struct mddev *mddev, struct md_rdev *rdev)
7c7546cc
N
218{
219 /* Adding a drive to a linear array allows the array to grow.
220 * It is permitted if the new drive has a matching superblock
221 * already on it, with raid_disk equal to raid_disks.
222 * It is achieved by creating a new linear_private_data structure
223 * and swapping it in in-place of the current one.
224 * The current one is never freed until the array is stopped.
225 * This avoids races.
226 */
e849b938 227 struct linear_conf *newconf, *oldconf;
7c7546cc 228
a778b73f 229 if (rdev->saved_raid_disk != mddev->raid_disks)
7c7546cc
N
230 return -EINVAL;
231
a778b73f
N
232 rdev->raid_disk = rdev->saved_raid_disk;
233
7c7546cc
N
234 newconf = linear_conf(mddev,mddev->raid_disks+1);
235
236 if (!newconf)
237 return -ENOMEM;
238
495d3573 239 oldconf = rcu_dereference(mddev->private);
7c7546cc 240 mddev->raid_disks++;
af11c397 241 rcu_assign_pointer(mddev->private, newconf);
1f403624 242 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
f233ea5c 243 set_capacity(mddev->gendisk, mddev->array_sectors);
449aad3e 244 revalidate_disk(mddev->gendisk);
b119cbab 245 kfree_rcu(oldconf, rcu);
7c7546cc 246 return 0;
1da177e4
LT
247}
248
fd01b88c 249static int linear_stop (struct mddev *mddev)
1da177e4 250{
e849b938 251 struct linear_conf *conf = mddev->private;
af11c397
S
252
253 /*
254 * We do not require rcu protection here since
255 * we hold reconfig_mutex for both linear_add and
256 * linear_stop, so they cannot race.
495d3573
N
257 * We should make sure any old 'conf's are properly
258 * freed though.
af11c397 259 */
495d3573 260 rcu_barrier();
1da177e4 261 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
495d3573 262 kfree(conf);
ef2f80ff 263 mddev->private = NULL;
1da177e4
LT
264
265 return 0;
266}
267
b4fdcb02 268static void linear_make_request(struct mddev *mddev, struct bio *bio)
1da177e4 269{
a7120771 270 struct dev_info *tmp_dev;
4db7cdc8 271 sector_t start_sector;
1da177e4 272
e9c7469b
TH
273 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
274 md_flush_request(mddev, bio);
5a7bbad2 275 return;
e5dcdd80
N
276 }
277
af11c397 278 rcu_read_lock();
1da177e4 279 tmp_dev = which_dev(mddev, bio->bi_sector);
4db7cdc8
SS
280 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
281
af11c397 282
4db7cdc8
SS
283 if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
284 || (bio->bi_sector < start_sector))) {
1da177e4
LT
285 char b[BDEVNAME_SIZE];
286
2dc40f80
N
287 printk(KERN_ERR
288 "md/linear:%s: make_request: Sector %llu out of bounds on "
289 "dev %s: %llu sectors, offset %llu\n",
290 mdname(mddev),
291 (unsigned long long)bio->bi_sector,
292 bdevname(tmp_dev->rdev->bdev, b),
293 (unsigned long long)tmp_dev->rdev->sectors,
294 (unsigned long long)start_sector);
af11c397 295 rcu_read_unlock();
6712ecf8 296 bio_io_error(bio);
5a7bbad2 297 return;
1da177e4
LT
298 }
299 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
4db7cdc8 300 tmp_dev->end_sector)) {
1da177e4
LT
301 /* This bio crosses a device boundary, so we have to
302 * split it.
303 */
304 struct bio_pair *bp;
af11c397
S
305 sector_t end_sector = tmp_dev->end_sector;
306
307 rcu_read_unlock();
6283815d 308
af11c397 309 bp = bio_split(bio, end_sector - bio->bi_sector);
6283815d 310
5a7bbad2
CH
311 linear_make_request(mddev, &bp->bio1);
312 linear_make_request(mddev, &bp->bio2);
1da177e4 313 bio_pair_release(bp);
5a7bbad2 314 return;
1da177e4
LT
315 }
316
317 bio->bi_bdev = tmp_dev->rdev->bdev;
4db7cdc8 318 bio->bi_sector = bio->bi_sector - start_sector
6283815d 319 + tmp_dev->rdev->data_offset;
af11c397 320 rcu_read_unlock();
5a7bbad2 321 generic_make_request(bio);
1da177e4
LT
322}
323
fd01b88c 324static void linear_status (struct seq_file *seq, struct mddev *mddev)
1da177e4
LT
325{
326
9d8f0363 327 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
1da177e4
LT
328}
329
330
84fc4b56 331static struct md_personality linear_personality =
1da177e4
LT
332{
333 .name = "linear",
2604b703 334 .level = LEVEL_LINEAR,
1da177e4
LT
335 .owner = THIS_MODULE,
336 .make_request = linear_make_request,
337 .run = linear_run,
338 .stop = linear_stop,
339 .status = linear_status,
7c7546cc 340 .hot_add_disk = linear_add,
80c3a6ce 341 .size = linear_size,
1da177e4
LT
342};
343
344static int __init linear_init (void)
345{
2604b703 346 return register_md_personality (&linear_personality);
1da177e4
LT
347}
348
349static void linear_exit (void)
350{
2604b703 351 unregister_md_personality (&linear_personality);
1da177e4
LT
352}
353
354
355module_init(linear_init);
356module_exit(linear_exit);
357MODULE_LICENSE("GPL");
0efb9e61 358MODULE_DESCRIPTION("Linear device concatenation personality for MD");
d9d166c2
N
359MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
360MODULE_ALIAS("md-linear");
2604b703 361MODULE_ALIAS("md-level--1");