1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * linear.c : Multiple Devices driver for Linux Copyright (C) 1994-96 Marc
4 * ZYNGIER <zyngier@ufr-info-p7.ibp.fr> or <maz@gloups.fdn.fr>
5 */
6
7 #include <linux/blkdev.h>
8 #include <linux/seq_file.h>
9 #include <linux/module.h>
10 #include <linux/slab.h>
11 #include <trace/events/block.h>
12 #include "md.h"
13
14 struct dev_info {
15 struct md_rdev *rdev;
16 sector_t end_sector;
17 };
18
19 struct linear_conf {
20 struct rcu_head rcu;
21 sector_t array_sectors;
22 /* a copy of mddev->raid_disks */
23 int raid_disks;
24 struct dev_info disks[] __counted_by(raid_disks);
25 };
26
27 /*
28 * find which device holds a particular offset
29 */
which_dev(struct mddev * mddev,sector_t sector)30 static inline struct dev_info *which_dev(struct mddev *mddev, sector_t sector)
31 {
32 int lo, mid, hi;
33 struct linear_conf *conf;
34
35 lo = 0;
36 hi = mddev->raid_disks - 1;
37 conf = mddev->private;
38
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;
53 }
54
linear_size(struct mddev * mddev,sector_t sectors,int raid_disks)55 static sector_t linear_size(struct mddev *mddev, sector_t sectors, int raid_disks)
56 {
57 struct linear_conf *conf;
58 sector_t array_sectors;
59
60 conf = mddev->private;
61 WARN_ONCE(sectors || raid_disks,
62 "%s does not support generic reshape\n", __func__);
63 array_sectors = conf->array_sectors;
64
65 return array_sectors;
66 }
67
linear_set_limits(struct mddev * mddev)68 static int linear_set_limits(struct mddev *mddev)
69 {
70 struct queue_limits lim;
71 int err;
72
73 md_init_stacking_limits(&lim);
74 lim.features |= BLK_FEAT_NOWAIT;
75 lim.max_hw_sectors = mddev->chunk_sectors;
76 lim.logical_block_size = mddev->logical_block_size;
77 lim.max_write_zeroes_sectors = mddev->chunk_sectors;
78 lim.max_hw_wzeroes_unmap_sectors = mddev->chunk_sectors;
79 lim.io_min = mddev->chunk_sectors << 9;
80 lim.features |= BLK_FEAT_ATOMIC_WRITES;
81 err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
82 if (err)
83 return err;
84
85 return queue_limits_set(mddev->gendisk->queue, &lim);
86 }
87
linear_conf(struct mddev * mddev,int raid_disks)88 static struct linear_conf *linear_conf(struct mddev *mddev, int raid_disks)
89 {
90 struct linear_conf *conf;
91 struct md_rdev *rdev;
92 int ret = -EINVAL;
93 int cnt;
94 int i;
95
96 conf = kzalloc_flex(*conf, disks, raid_disks);
97 if (!conf)
98 return ERR_PTR(-ENOMEM);
99
100 /*
101 * conf->raid_disks is copy of mddev->raid_disks. The reason to
102 * keep a copy of mddev->raid_disks in struct linear_conf is,
103 * mddev->raid_disks may not be consistent with pointers number of
104 * conf->disks[] when it is updated in linear_add() and used to
105 * iterate old conf->disks[] earray in linear_congested().
106 * Here conf->raid_disks is always consitent with number of
107 * pointers in conf->disks[] array, and mddev->private is updated
108 * with rcu_assign_pointer() in linear_addr(), such race can be
109 * avoided.
110 */
111 conf->raid_disks = raid_disks;
112
113 cnt = 0;
114 conf->array_sectors = 0;
115
116 rdev_for_each(rdev, mddev) {
117 int j = rdev->raid_disk;
118 struct dev_info *disk = conf->disks + j;
119 sector_t sectors;
120
121 if (j < 0 || j >= raid_disks || disk->rdev) {
122 pr_warn("md/linear:%s: disk numbering problem. Aborting!\n",
123 mdname(mddev));
124 goto out;
125 }
126
127 disk->rdev = rdev;
128 if (mddev->chunk_sectors) {
129 sectors = rdev->sectors;
130 sector_div(sectors, mddev->chunk_sectors);
131 rdev->sectors = sectors * mddev->chunk_sectors;
132 }
133
134 conf->array_sectors += rdev->sectors;
135 cnt++;
136 }
137 if (cnt != raid_disks) {
138 pr_warn("md/linear:%s: not enough drives present. Aborting!\n",
139 mdname(mddev));
140 goto out;
141 }
142
143 /*
144 * Here we calculate the device offsets.
145 */
146 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
147
148 for (i = 1; i < raid_disks; i++)
149 conf->disks[i].end_sector =
150 conf->disks[i-1].end_sector +
151 conf->disks[i].rdev->sectors;
152
153 if (!mddev_is_dm(mddev)) {
154 ret = linear_set_limits(mddev);
155 if (ret)
156 goto out;
157 }
158
159 return conf;
160
161 out:
162 kfree(conf);
163 return ERR_PTR(ret);
164 }
165
linear_run(struct mddev * mddev)166 static int linear_run(struct mddev *mddev)
167 {
168 struct linear_conf *conf;
169 int ret;
170
171 if (md_check_no_bitmap(mddev))
172 return -EINVAL;
173
174 conf = linear_conf(mddev, mddev->raid_disks);
175 if (IS_ERR(conf))
176 return PTR_ERR(conf);
177
178 mddev->private = conf;
179 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
180
181 ret = md_integrity_register(mddev);
182 if (ret) {
183 kfree(conf);
184 mddev->private = NULL;
185 }
186 return ret;
187 }
188
linear_add(struct mddev * mddev,struct md_rdev * rdev)189 static int linear_add(struct mddev *mddev, struct md_rdev *rdev)
190 {
191 /* Adding a drive to a linear array allows the array to grow.
192 * It is permitted if the new drive has a matching superblock
193 * already on it, with raid_disk equal to raid_disks.
194 * It is achieved by creating a new linear_private_data structure
195 * and swapping it in in-place of the current one.
196 * The current one is never freed until the array is stopped.
197 * This avoids races.
198 */
199 struct linear_conf *newconf, *oldconf;
200
201 if (rdev->saved_raid_disk != mddev->raid_disks)
202 return -EINVAL;
203
204 rdev->raid_disk = rdev->saved_raid_disk;
205 rdev->saved_raid_disk = -1;
206
207 newconf = linear_conf(mddev, mddev->raid_disks + 1);
208 if (IS_ERR(newconf))
209 return PTR_ERR(newconf);
210
211 /* newconf->raid_disks already keeps a copy of * the increased
212 * value of mddev->raid_disks, WARN_ONCE() is just used to make
213 * sure of this. It is possible that oldconf is still referenced
214 * in linear_congested(), therefore kfree_rcu() is used to free
215 * oldconf until no one uses it anymore.
216 */
217 oldconf = rcu_dereference_protected(mddev->private,
218 lockdep_is_held(&mddev->reconfig_mutex));
219 mddev->raid_disks++;
220 WARN_ONCE(mddev->raid_disks != newconf->raid_disks,
221 "copied raid_disks doesn't match mddev->raid_disks");
222 rcu_assign_pointer(mddev->private, newconf);
223 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
224 set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
225 kfree_rcu(oldconf, rcu);
226 return 0;
227 }
228
linear_free(struct mddev * mddev,void * priv)229 static void linear_free(struct mddev *mddev, void *priv)
230 {
231 struct linear_conf *conf = priv;
232
233 kfree(conf);
234 }
235
linear_make_request(struct mddev * mddev,struct bio * bio)236 static bool linear_make_request(struct mddev *mddev, struct bio *bio)
237 {
238 struct dev_info *tmp_dev;
239 sector_t start_sector, end_sector, data_offset;
240 sector_t bio_sector = bio->bi_iter.bi_sector;
241
242 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
243 && md_flush_request(mddev, bio))
244 return true;
245
246 tmp_dev = which_dev(mddev, bio_sector);
247 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
248 end_sector = tmp_dev->end_sector;
249 data_offset = tmp_dev->rdev->data_offset;
250
251 if (unlikely(bio_sector >= end_sector ||
252 bio_sector < start_sector))
253 goto out_of_bounds;
254
255 if (unlikely(is_rdev_broken(tmp_dev->rdev))) {
256 md_error(mddev, tmp_dev->rdev);
257 bio_io_error(bio);
258 return true;
259 }
260
261 if (unlikely(bio_end_sector(bio) > end_sector)) {
262 /* This bio crosses a device boundary, so we have to split it */
263 bio = bio_submit_split_bioset(bio, end_sector - bio_sector,
264 &mddev->bio_set);
265 if (!bio)
266 return true;
267 }
268
269 md_account_bio(mddev, &bio);
270 bio_set_dev(bio, tmp_dev->rdev->bdev);
271 bio->bi_iter.bi_sector = bio->bi_iter.bi_sector -
272 start_sector + data_offset;
273
274 if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
275 !bdev_max_discard_sectors(bio->bi_bdev))) {
276 /* Just ignore it */
277 bio_endio(bio);
278 } else {
279 if (mddev->gendisk)
280 trace_block_bio_remap(bio, disk_devt(mddev->gendisk),
281 bio_sector);
282 mddev_check_write_zeroes(mddev, bio);
283 submit_bio_noacct(bio);
284 }
285 return true;
286
287 out_of_bounds:
288 pr_err("md/linear:%s: make_request: Sector %llu out of bounds on dev %pg: %llu sectors, offset %llu\n",
289 mdname(mddev),
290 (unsigned long long)bio->bi_iter.bi_sector,
291 tmp_dev->rdev->bdev,
292 (unsigned long long)tmp_dev->rdev->sectors,
293 (unsigned long long)start_sector);
294 bio_io_error(bio);
295 return true;
296 }
297
linear_status(struct seq_file * seq,struct mddev * mddev)298 static void linear_status(struct seq_file *seq, struct mddev *mddev)
299 {
300 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
301 }
302
linear_error(struct mddev * mddev,struct md_rdev * rdev)303 static void linear_error(struct mddev *mddev, struct md_rdev *rdev)
304 {
305 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
306 char *md_name = mdname(mddev);
307
308 pr_crit("md/linear%s: Disk failure on %pg detected, failing array.\n",
309 md_name, rdev->bdev);
310 }
311 }
312
linear_quiesce(struct mddev * mddev,int state)313 static void linear_quiesce(struct mddev *mddev, int state)
314 {
315 }
316
317 static struct md_personality linear_personality = {
318 .head = {
319 .type = MD_PERSONALITY,
320 .id = ID_LINEAR,
321 .name = "linear",
322 .owner = THIS_MODULE,
323 },
324
325 .make_request = linear_make_request,
326 .run = linear_run,
327 .free = linear_free,
328 .status = linear_status,
329 .hot_add_disk = linear_add,
330 .size = linear_size,
331 .quiesce = linear_quiesce,
332 .error_handler = linear_error,
333 };
334
linear_init(void)335 static int __init linear_init(void)
336 {
337 return register_md_submodule(&linear_personality.head);
338 }
339
linear_exit(void)340 static void linear_exit(void)
341 {
342 unregister_md_submodule(&linear_personality.head);
343 }
344
345 module_init(linear_init);
346 module_exit(linear_exit);
347 MODULE_LICENSE("GPL");
348 MODULE_DESCRIPTION("Linear device concatenation personality for MD (deprecated)");
349 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
350 MODULE_ALIAS("md-linear");
351 MODULE_ALIAS("md-level--1");
352