Commit | Line | Data |
---|---|---|
af1a8899 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | linear.c : Multiple Devices driver for Linux | |
4 | Copyright (C) 1994-96 Marc ZYNGIER | |
5 | <zyngier@ufr-info-p7.ibp.fr> or | |
6 | <maz@gloups.fdn.fr> | |
7 | ||
8 | Linear mode management functions. | |
9 | ||
1da177e4 LT |
10 | */ |
11 | ||
bff61975 N |
12 | #include <linux/blkdev.h> |
13 | #include <linux/raid/md_u.h> | |
bff61975 | 14 | #include <linux/seq_file.h> |
056075c7 | 15 | #include <linux/module.h> |
5a0e3ad6 | 16 | #include <linux/slab.h> |
109e3765 | 17 | #include <trace/events/block.h> |
43b2e5d8 | 18 | #include "md.h" |
935fe098 | 19 | #include "md-linear.h" |
1da177e4 | 20 | |
1da177e4 | 21 | /* |
f72ffdd6 | 22 | * find which device holds a particular offset |
1da177e4 | 23 | */ |
a7120771 | 24 | static inline struct dev_info *which_dev(struct mddev *mddev, sector_t sector) |
1da177e4 | 25 | { |
aece3d1f | 26 | int lo, mid, hi; |
e849b938 | 27 | struct linear_conf *conf; |
1da177e4 | 28 | |
aece3d1f SS |
29 | lo = 0; |
30 | hi = mddev->raid_disks - 1; | |
3be260cc | 31 | conf = mddev->private; |
1da177e4 | 32 | |
aece3d1f SS |
33 | /* |
34 | * Binary Search | |
35 | */ | |
36 | ||
37 | while (hi > lo) { | |
38 | ||
39 | mid = (hi + lo) / 2; | |
40 | if (sector < conf->disks[mid].end_sector) | |
41 | hi = mid; | |
42 | else | |
43 | lo = mid + 1; | |
44 | } | |
45 | ||
46 | return conf->disks + lo; | |
1da177e4 LT |
47 | } |
48 | ||
fd01b88c | 49 | static sector_t linear_size(struct mddev *mddev, sector_t sectors, int raid_disks) |
80c3a6ce | 50 | { |
e849b938 | 51 | struct linear_conf *conf; |
af11c397 | 52 | sector_t array_sectors; |
80c3a6ce | 53 | |
3be260cc | 54 | conf = mddev->private; |
80c3a6ce DW |
55 | WARN_ONCE(sectors || raid_disks, |
56 | "%s does not support generic reshape\n", __func__); | |
af11c397 | 57 | array_sectors = conf->array_sectors; |
80c3a6ce | 58 | |
af11c397 | 59 | return array_sectors; |
80c3a6ce DW |
60 | } |
61 | ||
e849b938 | 62 | static struct linear_conf *linear_conf(struct mddev *mddev, int raid_disks) |
1da177e4 | 63 | { |
e849b938 | 64 | struct linear_conf *conf; |
3cb03002 | 65 | struct md_rdev *rdev; |
45d4582f | 66 | int i, cnt; |
1da177e4 | 67 | |
f1e5b623 | 68 | conf = kzalloc(struct_size(conf, disks, raid_disks), GFP_KERNEL); |
1da177e4 | 69 | if (!conf) |
7c7546cc N |
70 | return NULL; |
71 | ||
1da177e4 | 72 | cnt = 0; |
d6e22150 | 73 | conf->array_sectors = 0; |
1da177e4 | 74 | |
dafb20fa | 75 | rdev_for_each(rdev, mddev) { |
1da177e4 | 76 | int j = rdev->raid_disk; |
a7120771 | 77 | struct dev_info *disk = conf->disks + j; |
13f2682b | 78 | sector_t sectors; |
1da177e4 | 79 | |
13864515 | 80 | if (j < 0 || j >= raid_disks || disk->rdev) { |
a2e202af N |
81 | pr_warn("md/linear:%s: disk numbering problem. Aborting!\n", |
82 | mdname(mddev)); | |
1da177e4 LT |
83 | goto out; |
84 | } | |
85 | ||
86 | disk->rdev = rdev; | |
13f2682b N |
87 | if (mddev->chunk_sectors) { |
88 | sectors = rdev->sectors; | |
89 | sector_div(sectors, mddev->chunk_sectors); | |
90 | rdev->sectors = sectors * mddev->chunk_sectors; | |
91 | } | |
1da177e4 | 92 | |
8f6c2e4b MP |
93 | disk_stack_limits(mddev->gendisk, rdev->bdev, |
94 | rdev->data_offset << 9); | |
1da177e4 | 95 | |
dd8ac336 | 96 | conf->array_sectors += rdev->sectors; |
1da177e4 LT |
97 | cnt++; |
98 | } | |
7c7546cc | 99 | if (cnt != raid_disks) { |
a2e202af N |
100 | pr_warn("md/linear:%s: not enough drives present. Aborting!\n", |
101 | mdname(mddev)); | |
1da177e4 LT |
102 | goto out; |
103 | } | |
104 | ||
105 | /* | |
45d4582f | 106 | * Here we calculate the device offsets. |
1da177e4 | 107 | */ |
4db7cdc8 SS |
108 | conf->disks[0].end_sector = conf->disks[0].rdev->sectors; |
109 | ||
a778b73f | 110 | for (i = 1; i < raid_disks; i++) |
4db7cdc8 SS |
111 | conf->disks[i].end_sector = |
112 | conf->disks[i-1].end_sector + | |
113 | conf->disks[i].rdev->sectors; | |
15945fee | 114 | |
03a9e24e | 115 | /* |
116 | * conf->raid_disks is copy of mddev->raid_disks. The reason to | |
117 | * keep a copy of mddev->raid_disks in struct linear_conf is, | |
118 | * mddev->raid_disks may not be consistent with pointers number of | |
119 | * conf->disks[] when it is updated in linear_add() and used to | |
120 | * iterate old conf->disks[] earray in linear_congested(). | |
121 | * Here conf->raid_disks is always consitent with number of | |
122 | * pointers in conf->disks[] array, and mddev->private is updated | |
123 | * with rcu_assign_pointer() in linear_addr(), such race can be | |
124 | * avoided. | |
125 | */ | |
126 | conf->raid_disks = raid_disks; | |
127 | ||
7c7546cc N |
128 | return conf; |
129 | ||
130 | out: | |
131 | kfree(conf); | |
132 | return NULL; | |
133 | } | |
134 | ||
fd01b88c | 135 | static int linear_run (struct mddev *mddev) |
7c7546cc | 136 | { |
e849b938 | 137 | struct linear_conf *conf; |
98d5561b | 138 | int ret; |
7c7546cc | 139 | |
0894cc30 AN |
140 | if (md_check_no_bitmap(mddev)) |
141 | return -EINVAL; | |
7c7546cc N |
142 | conf = linear_conf(mddev, mddev->raid_disks); |
143 | ||
144 | if (!conf) | |
145 | return 1; | |
146 | mddev->private = conf; | |
1f403624 | 147 | md_set_array_sectors(mddev, linear_size(mddev, 0, 0)); |
7c7546cc | 148 | |
98d5561b | 149 | ret = md_integrity_register(mddev); |
150 | if (ret) { | |
151 | kfree(conf); | |
152 | mddev->private = NULL; | |
153 | } | |
154 | return ret; | |
7c7546cc | 155 | } |
1da177e4 | 156 | |
fd01b88c | 157 | static int linear_add(struct mddev *mddev, struct md_rdev *rdev) |
7c7546cc N |
158 | { |
159 | /* Adding a drive to a linear array allows the array to grow. | |
160 | * It is permitted if the new drive has a matching superblock | |
161 | * already on it, with raid_disk equal to raid_disks. | |
162 | * It is achieved by creating a new linear_private_data structure | |
163 | * and swapping it in in-place of the current one. | |
164 | * The current one is never freed until the array is stopped. | |
165 | * This avoids races. | |
166 | */ | |
e849b938 | 167 | struct linear_conf *newconf, *oldconf; |
7c7546cc | 168 | |
a778b73f | 169 | if (rdev->saved_raid_disk != mddev->raid_disks) |
7c7546cc N |
170 | return -EINVAL; |
171 | ||
a778b73f | 172 | rdev->raid_disk = rdev->saved_raid_disk; |
09cd9270 | 173 | rdev->saved_raid_disk = -1; |
a778b73f | 174 | |
7c7546cc N |
175 | newconf = linear_conf(mddev,mddev->raid_disks+1); |
176 | ||
177 | if (!newconf) | |
178 | return -ENOMEM; | |
179 | ||
03a9e24e | 180 | /* newconf->raid_disks already keeps a copy of * the increased |
181 | * value of mddev->raid_disks, WARN_ONCE() is just used to make | |
182 | * sure of this. It is possible that oldconf is still referenced | |
183 | * in linear_congested(), therefore kfree_rcu() is used to free | |
184 | * oldconf until no one uses it anymore. | |
185 | */ | |
3be260cc | 186 | mddev_suspend(mddev); |
d939cdfd SL |
187 | oldconf = rcu_dereference_protected(mddev->private, |
188 | lockdep_is_held(&mddev->reconfig_mutex)); | |
7c7546cc | 189 | mddev->raid_disks++; |
03a9e24e | 190 | WARN_ONCE(mddev->raid_disks != newconf->raid_disks, |
191 | "copied raid_disks doesn't match mddev->raid_disks"); | |
192 | rcu_assign_pointer(mddev->private, newconf); | |
1f403624 | 193 | md_set_array_sectors(mddev, linear_size(mddev, 0, 0)); |
2c247c51 | 194 | set_capacity_and_notify(mddev->gendisk, mddev->array_sectors); |
3be260cc | 195 | mddev_resume(mddev); |
03a9e24e | 196 | kfree_rcu(oldconf, rcu); |
7c7546cc | 197 | return 0; |
1da177e4 LT |
198 | } |
199 | ||
afa0f557 | 200 | static void linear_free(struct mddev *mddev, void *priv) |
1da177e4 | 201 | { |
afa0f557 | 202 | struct linear_conf *conf = priv; |
af11c397 | 203 | |
495d3573 | 204 | kfree(conf); |
1da177e4 LT |
205 | } |
206 | ||
cc27b0c7 | 207 | static bool linear_make_request(struct mddev *mddev, struct bio *bio) |
1da177e4 | 208 | { |
a7120771 | 209 | struct dev_info *tmp_dev; |
20d0189b | 210 | sector_t start_sector, end_sector, data_offset; |
868f604b | 211 | sector_t bio_sector = bio->bi_iter.bi_sector; |
1da177e4 | 212 | |
775d7831 DJ |
213 | if (unlikely(bio->bi_opf & REQ_PREFLUSH) |
214 | && md_flush_request(mddev, bio)) | |
cc27b0c7 | 215 | return true; |
e5dcdd80 | 216 | |
868f604b N |
217 | tmp_dev = which_dev(mddev, bio_sector); |
218 | start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors; | |
219 | end_sector = tmp_dev->end_sector; | |
220 | data_offset = tmp_dev->rdev->data_offset; | |
221 | ||
222 | if (unlikely(bio_sector >= end_sector || | |
223 | bio_sector < start_sector)) | |
224 | goto out_of_bounds; | |
225 | ||
62f7b198 GP |
226 | if (unlikely(is_mddev_broken(tmp_dev->rdev, "linear"))) { |
227 | bio_io_error(bio); | |
228 | return true; | |
229 | } | |
230 | ||
868f604b N |
231 | if (unlikely(bio_end_sector(bio) > end_sector)) { |
232 | /* This bio crosses a device boundary, so we have to split it */ | |
233 | struct bio *split = bio_split(bio, end_sector - bio_sector, | |
afeee514 | 234 | GFP_NOIO, &mddev->bio_set); |
868f604b | 235 | bio_chain(split, bio); |
ed00aabd | 236 | submit_bio_noacct(bio); |
868f604b N |
237 | bio = split; |
238 | } | |
f1cad2b6 | 239 | |
74d46992 | 240 | bio_set_dev(bio, tmp_dev->rdev->bdev); |
868f604b N |
241 | bio->bi_iter.bi_sector = bio->bi_iter.bi_sector - |
242 | start_sector + data_offset; | |
243 | ||
244 | if (unlikely((bio_op(bio) == REQ_OP_DISCARD) && | |
70200574 | 245 | !bdev_max_discard_sectors(bio->bi_bdev))) { |
868f604b N |
246 | /* Just ignore it */ |
247 | bio_endio(bio); | |
248 | } else { | |
249 | if (mddev->gendisk) | |
1c02fca6 | 250 | trace_block_bio_remap(bio, disk_devt(mddev->gendisk), |
868f604b | 251 | bio_sector); |
e265eb3a | 252 | mddev_check_write_zeroes(mddev, bio); |
ed00aabd | 253 | submit_bio_noacct(bio); |
868f604b | 254 | } |
cc27b0c7 | 255 | return true; |
20d0189b KO |
256 | |
257 | out_of_bounds: | |
913cce5a | 258 | pr_err("md/linear:%s: make_request: Sector %llu out of bounds on dev %pg: %llu sectors, offset %llu\n", |
20d0189b KO |
259 | mdname(mddev), |
260 | (unsigned long long)bio->bi_iter.bi_sector, | |
913cce5a | 261 | tmp_dev->rdev->bdev, |
20d0189b KO |
262 | (unsigned long long)tmp_dev->rdev->sectors, |
263 | (unsigned long long)start_sector); | |
264 | bio_io_error(bio); | |
cc27b0c7 | 265 | return true; |
1da177e4 LT |
266 | } |
267 | ||
fd01b88c | 268 | static void linear_status (struct seq_file *seq, struct mddev *mddev) |
1da177e4 | 269 | { |
9d8f0363 | 270 | seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2); |
1da177e4 LT |
271 | } |
272 | ||
3be260cc N |
273 | static void linear_quiesce(struct mddev *mddev, int state) |
274 | { | |
275 | } | |
276 | ||
84fc4b56 | 277 | static struct md_personality linear_personality = |
1da177e4 LT |
278 | { |
279 | .name = "linear", | |
2604b703 | 280 | .level = LEVEL_LINEAR, |
1da177e4 LT |
281 | .owner = THIS_MODULE, |
282 | .make_request = linear_make_request, | |
283 | .run = linear_run, | |
afa0f557 | 284 | .free = linear_free, |
1da177e4 | 285 | .status = linear_status, |
7c7546cc | 286 | .hot_add_disk = linear_add, |
80c3a6ce | 287 | .size = linear_size, |
3be260cc | 288 | .quiesce = linear_quiesce, |
1da177e4 LT |
289 | }; |
290 | ||
291 | static int __init linear_init (void) | |
292 | { | |
2604b703 | 293 | return register_md_personality (&linear_personality); |
1da177e4 LT |
294 | } |
295 | ||
296 | static void linear_exit (void) | |
297 | { | |
2604b703 | 298 | unregister_md_personality (&linear_personality); |
1da177e4 LT |
299 | } |
300 | ||
1da177e4 LT |
301 | module_init(linear_init); |
302 | module_exit(linear_exit); | |
303 | MODULE_LICENSE("GPL"); | |
608f52e3 | 304 | MODULE_DESCRIPTION("Linear device concatenation personality for MD (deprecated)"); |
d9d166c2 N |
305 | MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/ |
306 | MODULE_ALIAS("md-linear"); | |
2604b703 | 307 | MODULE_ALIAS("md-level--1"); |