Loading...
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/raid/md_u.h>
9#include <linux/seq_file.h>
10#include <linux/module.h>
11#include <linux/slab.h>
12#include <trace/events/block.h>
13#include "md.h"
14
15struct dev_info {
16 struct md_rdev *rdev;
17 sector_t end_sector;
18};
19
20struct linear_conf {
21 struct rcu_head rcu;
22 sector_t array_sectors;
23 /* a copy of mddev->raid_disks */
24 int raid_disks;
25 struct dev_info disks[] __counted_by(raid_disks);
26};
27
28/*
29 * find which device holds a particular offset
30 */
31static inline struct dev_info *which_dev(struct mddev *mddev, sector_t sector)
32{
33 int lo, mid, hi;
34 struct linear_conf *conf;
35
36 lo = 0;
37 hi = mddev->raid_disks - 1;
38 conf = mddev->private;
39
40 /*
41 * Binary Search
42 */
43
44 while (hi > lo) {
45
46 mid = (hi + lo) / 2;
47 if (sector < conf->disks[mid].end_sector)
48 hi = mid;
49 else
50 lo = mid + 1;
51 }
52
53 return conf->disks + lo;
54}
55
56static sector_t linear_size(struct mddev *mddev, sector_t sectors, int raid_disks)
57{
58 struct linear_conf *conf;
59 sector_t array_sectors;
60
61 conf = mddev->private;
62 WARN_ONCE(sectors || raid_disks,
63 "%s does not support generic reshape\n", __func__);
64 array_sectors = conf->array_sectors;
65
66 return array_sectors;
67}
68
69static int linear_set_limits(struct mddev *mddev)
70{
71 struct queue_limits lim;
72 int err;
73
74 md_init_stacking_limits(&lim);
75 lim.max_hw_sectors = mddev->chunk_sectors;
76 lim.max_write_zeroes_sectors = mddev->chunk_sectors;
77 lim.io_min = mddev->chunk_sectors << 9;
78 err = mddev_stack_rdev_limits(mddev, &lim, MDDEV_STACK_INTEGRITY);
79 if (err)
80 return err;
81
82 return queue_limits_set(mddev->gendisk->queue, &lim);
83}
84
85static struct linear_conf *linear_conf(struct mddev *mddev, int raid_disks)
86{
87 struct linear_conf *conf;
88 struct md_rdev *rdev;
89 int ret = -EINVAL;
90 int cnt;
91 int i;
92
93 conf = kzalloc(struct_size(conf, disks, raid_disks), GFP_KERNEL);
94 if (!conf)
95 return ERR_PTR(-ENOMEM);
96
97 /*
98 * conf->raid_disks is copy of mddev->raid_disks. The reason to
99 * keep a copy of mddev->raid_disks in struct linear_conf is,
100 * mddev->raid_disks may not be consistent with pointers number of
101 * conf->disks[] when it is updated in linear_add() and used to
102 * iterate old conf->disks[] earray in linear_congested().
103 * Here conf->raid_disks is always consitent with number of
104 * pointers in conf->disks[] array, and mddev->private is updated
105 * with rcu_assign_pointer() in linear_addr(), such race can be
106 * avoided.
107 */
108 conf->raid_disks = raid_disks;
109
110 cnt = 0;
111 conf->array_sectors = 0;
112
113 rdev_for_each(rdev, mddev) {
114 int j = rdev->raid_disk;
115 struct dev_info *disk = conf->disks + j;
116 sector_t sectors;
117
118 if (j < 0 || j >= raid_disks || disk->rdev) {
119 pr_warn("md/linear:%s: disk numbering problem. Aborting!\n",
120 mdname(mddev));
121 goto out;
122 }
123
124 disk->rdev = rdev;
125 if (mddev->chunk_sectors) {
126 sectors = rdev->sectors;
127 sector_div(sectors, mddev->chunk_sectors);
128 rdev->sectors = sectors * mddev->chunk_sectors;
129 }
130
131 conf->array_sectors += rdev->sectors;
132 cnt++;
133 }
134 if (cnt != raid_disks) {
135 pr_warn("md/linear:%s: not enough drives present. Aborting!\n",
136 mdname(mddev));
137 goto out;
138 }
139
140 /*
141 * Here we calculate the device offsets.
142 */
143 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
144
145 for (i = 1; i < raid_disks; i++)
146 conf->disks[i].end_sector =
147 conf->disks[i-1].end_sector +
148 conf->disks[i].rdev->sectors;
149
150 if (!mddev_is_dm(mddev)) {
151 ret = linear_set_limits(mddev);
152 if (ret)
153 goto out;
154 }
155
156 return conf;
157
158out:
159 kfree(conf);
160 return ERR_PTR(ret);
161}
162
163static int linear_run(struct mddev *mddev)
164{
165 struct linear_conf *conf;
166 int ret;
167
168 if (md_check_no_bitmap(mddev))
169 return -EINVAL;
170
171 conf = linear_conf(mddev, mddev->raid_disks);
172 if (IS_ERR(conf))
173 return PTR_ERR(conf);
174
175 mddev->private = conf;
176 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
177
178 ret = md_integrity_register(mddev);
179 if (ret) {
180 kfree(conf);
181 mddev->private = NULL;
182 }
183 return ret;
184}
185
186static int linear_add(struct mddev *mddev, struct md_rdev *rdev)
187{
188 /* Adding a drive to a linear array allows the array to grow.
189 * It is permitted if the new drive has a matching superblock
190 * already on it, with raid_disk equal to raid_disks.
191 * It is achieved by creating a new linear_private_data structure
192 * and swapping it in in-place of the current one.
193 * The current one is never freed until the array is stopped.
194 * This avoids races.
195 */
196 struct linear_conf *newconf, *oldconf;
197
198 if (rdev->saved_raid_disk != mddev->raid_disks)
199 return -EINVAL;
200
201 rdev->raid_disk = rdev->saved_raid_disk;
202 rdev->saved_raid_disk = -1;
203
204 newconf = linear_conf(mddev, mddev->raid_disks + 1);
205 if (IS_ERR(newconf))
206 return PTR_ERR(newconf);
207
208 /* newconf->raid_disks already keeps a copy of * the increased
209 * value of mddev->raid_disks, WARN_ONCE() is just used to make
210 * sure of this. It is possible that oldconf is still referenced
211 * in linear_congested(), therefore kfree_rcu() is used to free
212 * oldconf until no one uses it anymore.
213 */
214 oldconf = rcu_dereference_protected(mddev->private,
215 lockdep_is_held(&mddev->reconfig_mutex));
216 mddev->raid_disks++;
217 WARN_ONCE(mddev->raid_disks != newconf->raid_disks,
218 "copied raid_disks doesn't match mddev->raid_disks");
219 rcu_assign_pointer(mddev->private, newconf);
220 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
221 set_capacity_and_notify(mddev->gendisk, mddev->array_sectors);
222 kfree_rcu(oldconf, rcu);
223 return 0;
224}
225
226static void linear_free(struct mddev *mddev, void *priv)
227{
228 struct linear_conf *conf = priv;
229
230 kfree(conf);
231}
232
233static bool linear_make_request(struct mddev *mddev, struct bio *bio)
234{
235 struct dev_info *tmp_dev;
236 sector_t start_sector, end_sector, data_offset;
237 sector_t bio_sector = bio->bi_iter.bi_sector;
238
239 if (unlikely(bio->bi_opf & REQ_PREFLUSH)
240 && md_flush_request(mddev, bio))
241 return true;
242
243 tmp_dev = which_dev(mddev, bio_sector);
244 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
245 end_sector = tmp_dev->end_sector;
246 data_offset = tmp_dev->rdev->data_offset;
247
248 if (unlikely(bio_sector >= end_sector ||
249 bio_sector < start_sector))
250 goto out_of_bounds;
251
252 if (unlikely(is_rdev_broken(tmp_dev->rdev))) {
253 md_error(mddev, tmp_dev->rdev);
254 bio_io_error(bio);
255 return true;
256 }
257
258 if (unlikely(bio_end_sector(bio) > end_sector)) {
259 /* This bio crosses a device boundary, so we have to split it */
260 struct bio *split = bio_split(bio, end_sector - bio_sector,
261 GFP_NOIO, &mddev->bio_set);
262
263 if (IS_ERR(split)) {
264 bio->bi_status = errno_to_blk_status(PTR_ERR(split));
265 bio_endio(bio);
266 return true;
267 }
268
269 bio_chain(split, bio);
270 submit_bio_noacct(bio);
271 bio = split;
272 }
273
274 md_account_bio(mddev, &bio);
275 bio_set_dev(bio, tmp_dev->rdev->bdev);
276 bio->bi_iter.bi_sector = bio->bi_iter.bi_sector -
277 start_sector + data_offset;
278
279 if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
280 !bdev_max_discard_sectors(bio->bi_bdev))) {
281 /* Just ignore it */
282 bio_endio(bio);
283 } else {
284 if (mddev->gendisk)
285 trace_block_bio_remap(bio, disk_devt(mddev->gendisk),
286 bio_sector);
287 mddev_check_write_zeroes(mddev, bio);
288 submit_bio_noacct(bio);
289 }
290 return true;
291
292out_of_bounds:
293 pr_err("md/linear:%s: make_request: Sector %llu out of bounds on dev %pg: %llu sectors, offset %llu\n",
294 mdname(mddev),
295 (unsigned long long)bio->bi_iter.bi_sector,
296 tmp_dev->rdev->bdev,
297 (unsigned long long)tmp_dev->rdev->sectors,
298 (unsigned long long)start_sector);
299 bio_io_error(bio);
300 return true;
301}
302
303static void linear_status(struct seq_file *seq, struct mddev *mddev)
304{
305 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
306}
307
308static void linear_error(struct mddev *mddev, struct md_rdev *rdev)
309{
310 if (!test_and_set_bit(MD_BROKEN, &mddev->flags)) {
311 char *md_name = mdname(mddev);
312
313 pr_crit("md/linear%s: Disk failure on %pg detected, failing array.\n",
314 md_name, rdev->bdev);
315 }
316}
317
318static void linear_quiesce(struct mddev *mddev, int state)
319{
320}
321
322static struct md_personality linear_personality = {
323 .name = "linear",
324 .level = LEVEL_LINEAR,
325 .owner = THIS_MODULE,
326 .make_request = linear_make_request,
327 .run = linear_run,
328 .free = linear_free,
329 .status = linear_status,
330 .hot_add_disk = linear_add,
331 .size = linear_size,
332 .quiesce = linear_quiesce,
333 .error_handler = linear_error,
334};
335
336static int __init linear_init(void)
337{
338 return register_md_personality(&linear_personality);
339}
340
341static void linear_exit(void)
342{
343 unregister_md_personality(&linear_personality);
344}
345
346module_init(linear_init);
347module_exit(linear_exit);
348MODULE_LICENSE("GPL");
349MODULE_DESCRIPTION("Linear device concatenation personality for MD (deprecated)");
350MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
351MODULE_ALIAS("md-linear");
352MODULE_ALIAS("md-level--1");
1// SPDX-License-Identifier: GPL-2.0-or-later
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
10*/
11
12#include <linux/blkdev.h>
13#include <linux/raid/md_u.h>
14#include <linux/seq_file.h>
15#include <linux/module.h>
16#include <linux/slab.h>
17#include <trace/events/block.h>
18#include "md.h"
19#include "md-linear.h"
20
21/*
22 * find which device holds a particular offset
23 */
24static inline struct dev_info *which_dev(struct mddev *mddev, sector_t sector)
25{
26 int lo, mid, hi;
27 struct linear_conf *conf;
28
29 lo = 0;
30 hi = mddev->raid_disks - 1;
31 conf = mddev->private;
32
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;
47}
48
49/*
50 * In linear_congested() conf->raid_disks is used as a copy of
51 * mddev->raid_disks to iterate conf->disks[], because conf->raid_disks
52 * and conf->disks[] are created in linear_conf(), they are always
53 * consitent with each other, but mddev->raid_disks does not.
54 */
55static int linear_congested(struct mddev *mddev, int bits)
56{
57 struct linear_conf *conf;
58 int i, ret = 0;
59
60 rcu_read_lock();
61 conf = rcu_dereference(mddev->private);
62
63 for (i = 0; i < conf->raid_disks && !ret ; i++) {
64 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
65 ret |= bdi_congested(q->backing_dev_info, bits);
66 }
67
68 rcu_read_unlock();
69 return ret;
70}
71
72static sector_t linear_size(struct mddev *mddev, sector_t sectors, int raid_disks)
73{
74 struct linear_conf *conf;
75 sector_t array_sectors;
76
77 conf = mddev->private;
78 WARN_ONCE(sectors || raid_disks,
79 "%s does not support generic reshape\n", __func__);
80 array_sectors = conf->array_sectors;
81
82 return array_sectors;
83}
84
85static struct linear_conf *linear_conf(struct mddev *mddev, int raid_disks)
86{
87 struct linear_conf *conf;
88 struct md_rdev *rdev;
89 int i, cnt;
90 bool discard_supported = false;
91
92 conf = kzalloc(struct_size(conf, disks, raid_disks), GFP_KERNEL);
93 if (!conf)
94 return NULL;
95
96 cnt = 0;
97 conf->array_sectors = 0;
98
99 rdev_for_each(rdev, mddev) {
100 int j = rdev->raid_disk;
101 struct dev_info *disk = conf->disks + j;
102 sector_t sectors;
103
104 if (j < 0 || j >= raid_disks || disk->rdev) {
105 pr_warn("md/linear:%s: disk numbering problem. Aborting!\n",
106 mdname(mddev));
107 goto out;
108 }
109
110 disk->rdev = rdev;
111 if (mddev->chunk_sectors) {
112 sectors = rdev->sectors;
113 sector_div(sectors, mddev->chunk_sectors);
114 rdev->sectors = sectors * mddev->chunk_sectors;
115 }
116
117 disk_stack_limits(mddev->gendisk, rdev->bdev,
118 rdev->data_offset << 9);
119
120 conf->array_sectors += rdev->sectors;
121 cnt++;
122
123 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
124 discard_supported = true;
125 }
126 if (cnt != raid_disks) {
127 pr_warn("md/linear:%s: not enough drives present. Aborting!\n",
128 mdname(mddev));
129 goto out;
130 }
131
132 if (!discard_supported)
133 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
134 else
135 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
136
137 /*
138 * Here we calculate the device offsets.
139 */
140 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
141
142 for (i = 1; i < raid_disks; i++)
143 conf->disks[i].end_sector =
144 conf->disks[i-1].end_sector +
145 conf->disks[i].rdev->sectors;
146
147 /*
148 * conf->raid_disks is copy of mddev->raid_disks. The reason to
149 * keep a copy of mddev->raid_disks in struct linear_conf is,
150 * mddev->raid_disks may not be consistent with pointers number of
151 * conf->disks[] when it is updated in linear_add() and used to
152 * iterate old conf->disks[] earray in linear_congested().
153 * Here conf->raid_disks is always consitent with number of
154 * pointers in conf->disks[] array, and mddev->private is updated
155 * with rcu_assign_pointer() in linear_addr(), such race can be
156 * avoided.
157 */
158 conf->raid_disks = raid_disks;
159
160 return conf;
161
162out:
163 kfree(conf);
164 return NULL;
165}
166
167static int linear_run (struct mddev *mddev)
168{
169 struct linear_conf *conf;
170 int ret;
171
172 if (md_check_no_bitmap(mddev))
173 return -EINVAL;
174 conf = linear_conf(mddev, mddev->raid_disks);
175
176 if (!conf)
177 return 1;
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
189static 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
209 if (!newconf)
210 return -ENOMEM;
211
212 /* newconf->raid_disks already keeps a copy of * the increased
213 * value of mddev->raid_disks, WARN_ONCE() is just used to make
214 * sure of this. It is possible that oldconf is still referenced
215 * in linear_congested(), therefore kfree_rcu() is used to free
216 * oldconf until no one uses it anymore.
217 */
218 mddev_suspend(mddev);
219 oldconf = rcu_dereference_protected(mddev->private,
220 lockdep_is_held(&mddev->reconfig_mutex));
221 mddev->raid_disks++;
222 WARN_ONCE(mddev->raid_disks != newconf->raid_disks,
223 "copied raid_disks doesn't match mddev->raid_disks");
224 rcu_assign_pointer(mddev->private, newconf);
225 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
226 set_capacity(mddev->gendisk, mddev->array_sectors);
227 mddev_resume(mddev);
228 revalidate_disk(mddev->gendisk);
229 kfree_rcu(oldconf, rcu);
230 return 0;
231}
232
233static void linear_free(struct mddev *mddev, void *priv)
234{
235 struct linear_conf *conf = priv;
236
237 kfree(conf);
238}
239
240static bool linear_make_request(struct mddev *mddev, struct bio *bio)
241{
242 char b[BDEVNAME_SIZE];
243 struct dev_info *tmp_dev;
244 sector_t start_sector, end_sector, data_offset;
245 sector_t bio_sector = bio->bi_iter.bi_sector;
246
247 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
248 md_flush_request(mddev, bio);
249 return true;
250 }
251
252 tmp_dev = which_dev(mddev, bio_sector);
253 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
254 end_sector = tmp_dev->end_sector;
255 data_offset = tmp_dev->rdev->data_offset;
256
257 if (unlikely(bio_sector >= end_sector ||
258 bio_sector < start_sector))
259 goto out_of_bounds;
260
261 if (unlikely(is_mddev_broken(tmp_dev->rdev, "linear"))) {
262 bio_io_error(bio);
263 return true;
264 }
265
266 if (unlikely(bio_end_sector(bio) > end_sector)) {
267 /* This bio crosses a device boundary, so we have to split it */
268 struct bio *split = bio_split(bio, end_sector - bio_sector,
269 GFP_NOIO, &mddev->bio_set);
270 bio_chain(split, bio);
271 generic_make_request(bio);
272 bio = split;
273 }
274
275 bio_set_dev(bio, tmp_dev->rdev->bdev);
276 bio->bi_iter.bi_sector = bio->bi_iter.bi_sector -
277 start_sector + data_offset;
278
279 if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
280 !blk_queue_discard(bio->bi_disk->queue))) {
281 /* Just ignore it */
282 bio_endio(bio);
283 } else {
284 if (mddev->gendisk)
285 trace_block_bio_remap(bio->bi_disk->queue,
286 bio, disk_devt(mddev->gendisk),
287 bio_sector);
288 mddev_check_writesame(mddev, bio);
289 mddev_check_write_zeroes(mddev, bio);
290 generic_make_request(bio);
291 }
292 return true;
293
294out_of_bounds:
295 pr_err("md/linear:%s: make_request: Sector %llu out of bounds on dev %s: %llu sectors, offset %llu\n",
296 mdname(mddev),
297 (unsigned long long)bio->bi_iter.bi_sector,
298 bdevname(tmp_dev->rdev->bdev, b),
299 (unsigned long long)tmp_dev->rdev->sectors,
300 (unsigned long long)start_sector);
301 bio_io_error(bio);
302 return true;
303}
304
305static void linear_status (struct seq_file *seq, struct mddev *mddev)
306{
307 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
308}
309
310static void linear_quiesce(struct mddev *mddev, int state)
311{
312}
313
314static struct md_personality linear_personality =
315{
316 .name = "linear",
317 .level = LEVEL_LINEAR,
318 .owner = THIS_MODULE,
319 .make_request = linear_make_request,
320 .run = linear_run,
321 .free = linear_free,
322 .status = linear_status,
323 .hot_add_disk = linear_add,
324 .size = linear_size,
325 .quiesce = linear_quiesce,
326 .congested = linear_congested,
327};
328
329static int __init linear_init (void)
330{
331 return register_md_personality (&linear_personality);
332}
333
334static void linear_exit (void)
335{
336 unregister_md_personality (&linear_personality);
337}
338
339module_init(linear_init);
340module_exit(linear_exit);
341MODULE_LICENSE("GPL");
342MODULE_DESCRIPTION("Linear device concatenation personality for MD");
343MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
344MODULE_ALIAS("md-linear");
345MODULE_ALIAS("md-level--1");