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/slab.h>
23#include "md.h"
24#include "linear.h"
25
26/*
27 * find which device holds a particular offset
28 */
29static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
30{
31 int lo, mid, hi;
32 linear_conf_t *conf;
33
34 lo = 0;
35 hi = mddev->raid_disks - 1;
36 conf = rcu_dereference(mddev->private);
37
38 /*
39 * Binary Search
40 */
41
42 while (hi > lo) {
43
44 mid = (hi + lo) / 2;
45 if (sector < conf->disks[mid].end_sector)
46 hi = mid;
47 else
48 lo = mid + 1;
49 }
50
51 return conf->disks + lo;
52}
53
54/**
55 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
56 * @q: request queue
57 * @bvm: properties of new bio
58 * @biovec: the request that could be merged to it.
59 *
60 * Return amount of bytes we can take at this offset
61 */
62static int linear_mergeable_bvec(struct request_queue *q,
63 struct bvec_merge_data *bvm,
64 struct bio_vec *biovec)
65{
66 mddev_t *mddev = q->queuedata;
67 dev_info_t *dev0;
68 unsigned long maxsectors, bio_sectors = bvm->bi_size >> 9;
69 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
70
71 rcu_read_lock();
72 dev0 = which_dev(mddev, sector);
73 maxsectors = dev0->end_sector - sector;
74 rcu_read_unlock();
75
76 if (maxsectors < bio_sectors)
77 maxsectors = 0;
78 else
79 maxsectors -= bio_sectors;
80
81 if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
82 return biovec->bv_len;
83 /* The bytes available at this offset could be really big,
84 * so we cap at 2^31 to avoid overflow */
85 if (maxsectors > (1 << (31-9)))
86 return 1<<31;
87 return maxsectors << 9;
88}
89
90static int linear_congested(void *data, int bits)
91{
92 mddev_t *mddev = data;
93 linear_conf_t *conf;
94 int i, ret = 0;
95
96 if (mddev_congested(mddev, bits))
97 return 1;
98
99 rcu_read_lock();
100 conf = rcu_dereference(mddev->private);
101
102 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
103 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
104 ret |= bdi_congested(&q->backing_dev_info, bits);
105 }
106
107 rcu_read_unlock();
108 return ret;
109}
110
111static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
112{
113 linear_conf_t *conf;
114 sector_t array_sectors;
115
116 rcu_read_lock();
117 conf = rcu_dereference(mddev->private);
118 WARN_ONCE(sectors || raid_disks,
119 "%s does not support generic reshape\n", __func__);
120 array_sectors = conf->array_sectors;
121 rcu_read_unlock();
122
123 return array_sectors;
124}
125
126static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
127{
128 linear_conf_t *conf;
129 mdk_rdev_t *rdev;
130 int i, cnt;
131
132 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
133 GFP_KERNEL);
134 if (!conf)
135 return NULL;
136
137 cnt = 0;
138 conf->array_sectors = 0;
139
140 list_for_each_entry(rdev, &mddev->disks, same_set) {
141 int j = rdev->raid_disk;
142 dev_info_t *disk = conf->disks + j;
143 sector_t sectors;
144
145 if (j < 0 || j >= raid_disks || disk->rdev) {
146 printk(KERN_ERR "md/linear:%s: disk numbering problem. Aborting!\n",
147 mdname(mddev));
148 goto out;
149 }
150
151 disk->rdev = rdev;
152 if (mddev->chunk_sectors) {
153 sectors = rdev->sectors;
154 sector_div(sectors, mddev->chunk_sectors);
155 rdev->sectors = sectors * mddev->chunk_sectors;
156 }
157
158 disk_stack_limits(mddev->gendisk, rdev->bdev,
159 rdev->data_offset << 9);
160 /* as we don't honour merge_bvec_fn, we must never risk
161 * violating it, so limit max_segments to 1 lying within
162 * a single page.
163 */
164 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
165 blk_queue_max_segments(mddev->queue, 1);
166 blk_queue_segment_boundary(mddev->queue,
167 PAGE_CACHE_SIZE - 1);
168 }
169
170 conf->array_sectors += rdev->sectors;
171 cnt++;
172
173 }
174 if (cnt != raid_disks) {
175 printk(KERN_ERR "md/linear:%s: not enough drives present. Aborting!\n",
176 mdname(mddev));
177 goto out;
178 }
179
180 /*
181 * Here we calculate the device offsets.
182 */
183 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
184
185 for (i = 1; i < raid_disks; i++)
186 conf->disks[i].end_sector =
187 conf->disks[i-1].end_sector +
188 conf->disks[i].rdev->sectors;
189
190 return conf;
191
192out:
193 kfree(conf);
194 return NULL;
195}
196
197static int linear_run (mddev_t *mddev)
198{
199 linear_conf_t *conf;
200
201 if (md_check_no_bitmap(mddev))
202 return -EINVAL;
203 conf = linear_conf(mddev, mddev->raid_disks);
204
205 if (!conf)
206 return 1;
207 mddev->private = conf;
208 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
209
210 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
211 mddev->queue->backing_dev_info.congested_fn = linear_congested;
212 mddev->queue->backing_dev_info.congested_data = mddev;
213 return md_integrity_register(mddev);
214}
215
216static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
217{
218 /* Adding a drive to a linear array allows the array to grow.
219 * It is permitted if the new drive has a matching superblock
220 * already on it, with raid_disk equal to raid_disks.
221 * It is achieved by creating a new linear_private_data structure
222 * and swapping it in in-place of the current one.
223 * The current one is never freed until the array is stopped.
224 * This avoids races.
225 */
226 linear_conf_t *newconf, *oldconf;
227
228 if (rdev->saved_raid_disk != mddev->raid_disks)
229 return -EINVAL;
230
231 rdev->raid_disk = rdev->saved_raid_disk;
232
233 newconf = linear_conf(mddev,mddev->raid_disks+1);
234
235 if (!newconf)
236 return -ENOMEM;
237
238 oldconf = rcu_dereference(mddev->private);
239 mddev->raid_disks++;
240 rcu_assign_pointer(mddev->private, newconf);
241 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
242 set_capacity(mddev->gendisk, mddev->array_sectors);
243 revalidate_disk(mddev->gendisk);
244 kfree_rcu(oldconf, rcu);
245 return 0;
246}
247
248static int linear_stop (mddev_t *mddev)
249{
250 linear_conf_t *conf = mddev->private;
251
252 /*
253 * We do not require rcu protection here since
254 * we hold reconfig_mutex for both linear_add and
255 * linear_stop, so they cannot race.
256 * We should make sure any old 'conf's are properly
257 * freed though.
258 */
259 rcu_barrier();
260 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
261 kfree(conf);
262 mddev->private = NULL;
263
264 return 0;
265}
266
267static int linear_make_request (mddev_t *mddev, struct bio *bio)
268{
269 dev_info_t *tmp_dev;
270 sector_t start_sector;
271
272 if (unlikely(bio->bi_rw & REQ_FLUSH)) {
273 md_flush_request(mddev, bio);
274 return 0;
275 }
276
277 rcu_read_lock();
278 tmp_dev = which_dev(mddev, bio->bi_sector);
279 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
280
281
282 if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
283 || (bio->bi_sector < start_sector))) {
284 char b[BDEVNAME_SIZE];
285
286 printk(KERN_ERR
287 "md/linear:%s: make_request: Sector %llu out of bounds on "
288 "dev %s: %llu sectors, offset %llu\n",
289 mdname(mddev),
290 (unsigned long long)bio->bi_sector,
291 bdevname(tmp_dev->rdev->bdev, b),
292 (unsigned long long)tmp_dev->rdev->sectors,
293 (unsigned long long)start_sector);
294 rcu_read_unlock();
295 bio_io_error(bio);
296 return 0;
297 }
298 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
299 tmp_dev->end_sector)) {
300 /* This bio crosses a device boundary, so we have to
301 * split it.
302 */
303 struct bio_pair *bp;
304 sector_t end_sector = tmp_dev->end_sector;
305
306 rcu_read_unlock();
307
308 bp = bio_split(bio, end_sector - bio->bi_sector);
309
310 if (linear_make_request(mddev, &bp->bio1))
311 generic_make_request(&bp->bio1);
312 if (linear_make_request(mddev, &bp->bio2))
313 generic_make_request(&bp->bio2);
314 bio_pair_release(bp);
315 return 0;
316 }
317
318 bio->bi_bdev = tmp_dev->rdev->bdev;
319 bio->bi_sector = bio->bi_sector - start_sector
320 + tmp_dev->rdev->data_offset;
321 rcu_read_unlock();
322
323 return 1;
324}
325
326static void linear_status (struct seq_file *seq, mddev_t *mddev)
327{
328
329 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
330}
331
332
333static struct mdk_personality linear_personality =
334{
335 .name = "linear",
336 .level = LEVEL_LINEAR,
337 .owner = THIS_MODULE,
338 .make_request = linear_make_request,
339 .run = linear_run,
340 .stop = linear_stop,
341 .status = linear_status,
342 .hot_add_disk = linear_add,
343 .size = linear_size,
344};
345
346static int __init linear_init (void)
347{
348 return register_md_personality (&linear_personality);
349}
350
351static void linear_exit (void)
352{
353 unregister_md_personality (&linear_personality);
354}
355
356
357module_init(linear_init);
358module_exit(linear_exit);
359MODULE_LICENSE("GPL");
360MODULE_DESCRIPTION("Linear device concatenation personality for MD");
361MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
362MODULE_ALIAS("md-linear");
363MODULE_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");