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