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 "md-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(struct_size(conf, disks, raid_disks), GFP_KERNEL);
100 if (!conf)
101 return NULL;
102
103 cnt = 0;
104 conf->array_sectors = 0;
105
106 rdev_for_each(rdev, mddev) {
107 int j = rdev->raid_disk;
108 struct dev_info *disk = conf->disks + j;
109 sector_t sectors;
110
111 if (j < 0 || j >= raid_disks || disk->rdev) {
112 pr_warn("md/linear:%s: disk numbering problem. Aborting!\n",
113 mdname(mddev));
114 goto out;
115 }
116
117 disk->rdev = rdev;
118 if (mddev->chunk_sectors) {
119 sectors = rdev->sectors;
120 sector_div(sectors, mddev->chunk_sectors);
121 rdev->sectors = sectors * mddev->chunk_sectors;
122 }
123
124 disk_stack_limits(mddev->gendisk, rdev->bdev,
125 rdev->data_offset << 9);
126
127 conf->array_sectors += rdev->sectors;
128 cnt++;
129
130 if (blk_queue_discard(bdev_get_queue(rdev->bdev)))
131 discard_supported = true;
132 }
133 if (cnt != raid_disks) {
134 pr_warn("md/linear:%s: not enough drives present. Aborting!\n",
135 mdname(mddev));
136 goto out;
137 }
138
139 if (!discard_supported)
140 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, mddev->queue);
141 else
142 blk_queue_flag_set(QUEUE_FLAG_DISCARD, mddev->queue);
143
144 /*
145 * Here we calculate the device offsets.
146 */
147 conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
148
149 for (i = 1; i < raid_disks; i++)
150 conf->disks[i].end_sector =
151 conf->disks[i-1].end_sector +
152 conf->disks[i].rdev->sectors;
153
154 /*
155 * conf->raid_disks is copy of mddev->raid_disks. The reason to
156 * keep a copy of mddev->raid_disks in struct linear_conf is,
157 * mddev->raid_disks may not be consistent with pointers number of
158 * conf->disks[] when it is updated in linear_add() and used to
159 * iterate old conf->disks[] earray in linear_congested().
160 * Here conf->raid_disks is always consitent with number of
161 * pointers in conf->disks[] array, and mddev->private is updated
162 * with rcu_assign_pointer() in linear_addr(), such race can be
163 * avoided.
164 */
165 conf->raid_disks = raid_disks;
166
167 return conf;
168
169out:
170 kfree(conf);
171 return NULL;
172}
173
174static int linear_run (struct mddev *mddev)
175{
176 struct linear_conf *conf;
177 int ret;
178
179 if (md_check_no_bitmap(mddev))
180 return -EINVAL;
181 conf = linear_conf(mddev, mddev->raid_disks);
182
183 if (!conf)
184 return 1;
185 mddev->private = conf;
186 md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
187
188 ret = md_integrity_register(mddev);
189 if (ret) {
190 kfree(conf);
191 mddev->private = NULL;
192 }
193 return ret;
194}
195
196static int linear_add(struct mddev *mddev, struct md_rdev *rdev)
197{
198 /* Adding a drive to a linear array allows the array to grow.
199 * It is permitted if the new drive has a matching superblock
200 * already on it, with raid_disk equal to raid_disks.
201 * It is achieved by creating a new linear_private_data structure
202 * and swapping it in in-place of the current one.
203 * The current one is never freed until the array is stopped.
204 * This avoids races.
205 */
206 struct linear_conf *newconf, *oldconf;
207
208 if (rdev->saved_raid_disk != mddev->raid_disks)
209 return -EINVAL;
210
211 rdev->raid_disk = rdev->saved_raid_disk;
212 rdev->saved_raid_disk = -1;
213
214 newconf = linear_conf(mddev,mddev->raid_disks+1);
215
216 if (!newconf)
217 return -ENOMEM;
218
219 /* newconf->raid_disks already keeps a copy of * the increased
220 * value of mddev->raid_disks, WARN_ONCE() is just used to make
221 * sure of this. It is possible that oldconf is still referenced
222 * in linear_congested(), therefore kfree_rcu() is used to free
223 * oldconf until no one uses it anymore.
224 */
225 mddev_suspend(mddev);
226 oldconf = rcu_dereference_protected(mddev->private,
227 lockdep_is_held(&mddev->reconfig_mutex));
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 bool linear_make_request(struct mddev *mddev, struct bio *bio)
248{
249 char b[BDEVNAME_SIZE];
250 struct dev_info *tmp_dev;
251 sector_t start_sector, end_sector, data_offset;
252 sector_t bio_sector = bio->bi_iter.bi_sector;
253
254 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
255 md_flush_request(mddev, bio);
256 return true;
257 }
258
259 tmp_dev = which_dev(mddev, bio_sector);
260 start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
261 end_sector = tmp_dev->end_sector;
262 data_offset = tmp_dev->rdev->data_offset;
263
264 if (unlikely(bio_sector >= end_sector ||
265 bio_sector < start_sector))
266 goto out_of_bounds;
267
268 if (unlikely(bio_end_sector(bio) > end_sector)) {
269 /* This bio crosses a device boundary, so we have to split it */
270 struct bio *split = bio_split(bio, end_sector - bio_sector,
271 GFP_NOIO, &mddev->bio_set);
272 bio_chain(split, bio);
273 generic_make_request(bio);
274 bio = split;
275 }
276
277 bio_set_dev(bio, tmp_dev->rdev->bdev);
278 bio->bi_iter.bi_sector = bio->bi_iter.bi_sector -
279 start_sector + data_offset;
280
281 if (unlikely((bio_op(bio) == REQ_OP_DISCARD) &&
282 !blk_queue_discard(bio->bi_disk->queue))) {
283 /* Just ignore it */
284 bio_endio(bio);
285 } else {
286 if (mddev->gendisk)
287 trace_block_bio_remap(bio->bi_disk->queue,
288 bio, disk_devt(mddev->gendisk),
289 bio_sector);
290 mddev_check_writesame(mddev, bio);
291 mddev_check_write_zeroes(mddev, bio);
292 generic_make_request(bio);
293 }
294 return true;
295
296out_of_bounds:
297 pr_err("md/linear:%s: make_request: Sector %llu out of bounds on dev %s: %llu sectors, offset %llu\n",
298 mdname(mddev),
299 (unsigned long long)bio->bi_iter.bi_sector,
300 bdevname(tmp_dev->rdev->bdev, b),
301 (unsigned long long)tmp_dev->rdev->sectors,
302 (unsigned long long)start_sector);
303 bio_io_error(bio);
304 return true;
305}
306
307static void linear_status (struct seq_file *seq, struct mddev *mddev)
308{
309 seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
310}
311
312static void linear_quiesce(struct mddev *mddev, int state)
313{
314}
315
316static struct md_personality linear_personality =
317{
318 .name = "linear",
319 .level = LEVEL_LINEAR,
320 .owner = THIS_MODULE,
321 .make_request = linear_make_request,
322 .run = linear_run,
323 .free = linear_free,
324 .status = linear_status,
325 .hot_add_disk = linear_add,
326 .size = linear_size,
327 .quiesce = linear_quiesce,
328 .congested = linear_congested,
329};
330
331static int __init linear_init (void)
332{
333 return register_md_personality (&linear_personality);
334}
335
336static void linear_exit (void)
337{
338 unregister_md_personality (&linear_personality);
339}
340
341module_init(linear_init);
342module_exit(linear_exit);
343MODULE_LICENSE("GPL");
344MODULE_DESCRIPTION("Linear device concatenation personality for MD");
345MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
346MODULE_ALIAS("md-linear");
347MODULE_ALIAS("md-level--1");
348