[PATCH] md: fix some small races in bitmap plugging in raid5
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / drivers / md / linear.c
CommitLineData
1da177e4
LT
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/module.h>
20
21#include <linux/raid/md.h>
22#include <linux/slab.h>
23#include <linux/raid/linear.h>
24
25#define MAJOR_NR MD_MAJOR
26#define MD_DRIVER
27#define MD_PERSONALITY
28
29/*
30 * find which device holds a particular offset
31 */
32static inline dev_info_t *which_dev(mddev_t *mddev, sector_t sector)
33{
34 dev_info_t *hash;
35 linear_conf_t *conf = mddev_to_conf(mddev);
36 sector_t block = sector >> 1;
37
38 /*
39 * sector_div(a,b) returns the remainer and sets a to a/b
40 */
15945fee
N
41 block >>= conf->preshift;
42 (void)sector_div(block, conf->hash_spacing);
1da177e4
LT
43 hash = conf->hash_table[block];
44
45 while ((sector>>1) >= (hash->size + hash->offset))
46 hash++;
47 return hash;
48}
49
50/**
15945fee 51 * linear_mergeable_bvec -- tell bio layer if two requests can be merged
1da177e4
LT
52 * @q: request queue
53 * @bio: the buffer head that's been built up so far
54 * @biovec: the request that could be merged to it.
55 *
56 * Return amount of bytes we can take at this offset
57 */
58static int linear_mergeable_bvec(request_queue_t *q, struct bio *bio, struct bio_vec *biovec)
59{
60 mddev_t *mddev = q->queuedata;
61 dev_info_t *dev0;
62 unsigned long maxsectors, bio_sectors = bio->bi_size >> 9;
63 sector_t sector = bio->bi_sector + get_start_sect(bio->bi_bdev);
64
65 dev0 = which_dev(mddev, sector);
66 maxsectors = (dev0->size << 1) - (sector - (dev0->offset<<1));
67
68 if (maxsectors < bio_sectors)
69 maxsectors = 0;
70 else
71 maxsectors -= bio_sectors;
72
73 if (maxsectors <= (PAGE_SIZE >> 9 ) && bio_sectors == 0)
74 return biovec->bv_len;
75 /* The bytes available at this offset could be really big,
76 * so we cap at 2^31 to avoid overflow */
77 if (maxsectors > (1 << (31-9)))
78 return 1<<31;
79 return maxsectors << 9;
80}
81
82static void linear_unplug(request_queue_t *q)
83{
84 mddev_t *mddev = q->queuedata;
85 linear_conf_t *conf = mddev_to_conf(mddev);
86 int i;
87
88 for (i=0; i < mddev->raid_disks; i++) {
89 request_queue_t *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
90 if (r_queue->unplug_fn)
91 r_queue->unplug_fn(r_queue);
92 }
93}
94
95static int linear_issue_flush(request_queue_t *q, struct gendisk *disk,
96 sector_t *error_sector)
97{
98 mddev_t *mddev = q->queuedata;
99 linear_conf_t *conf = mddev_to_conf(mddev);
100 int i, ret = 0;
101
102 for (i=0; i < mddev->raid_disks && ret == 0; i++) {
103 struct block_device *bdev = conf->disks[i].rdev->bdev;
104 request_queue_t *r_queue = bdev_get_queue(bdev);
105
106 if (!r_queue->issue_flush_fn)
107 ret = -EOPNOTSUPP;
108 else
109 ret = r_queue->issue_flush_fn(r_queue, bdev->bd_disk, error_sector);
110 }
111 return ret;
112}
113
7c7546cc 114static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
1da177e4
LT
115{
116 linear_conf_t *conf;
117 dev_info_t **table;
118 mdk_rdev_t *rdev;
119 int i, nb_zone, cnt;
15945fee 120 sector_t min_spacing;
1da177e4
LT
121 sector_t curr_offset;
122 struct list_head *tmp;
123
7c7546cc 124 conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
1da177e4
LT
125 GFP_KERNEL);
126 if (!conf)
7c7546cc
N
127 return NULL;
128
1da177e4
LT
129 mddev->private = conf;
130
1da177e4 131 cnt = 0;
7c7546cc 132 conf->array_size = 0;
1da177e4
LT
133
134 ITERATE_RDEV(mddev,rdev,tmp) {
135 int j = rdev->raid_disk;
136 dev_info_t *disk = conf->disks + j;
137
7c7546cc 138 if (j < 0 || j > raid_disks || disk->rdev) {
1da177e4
LT
139 printk("linear: disk numbering problem. Aborting!\n");
140 goto out;
141 }
142
143 disk->rdev = rdev;
144
145 blk_queue_stack_limits(mddev->queue,
146 rdev->bdev->bd_disk->queue);
147 /* as we don't honour merge_bvec_fn, we must never risk
148 * violating it, so limit ->max_sector to one PAGE, as
149 * a one page request is never in violation.
150 */
151 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
152 mddev->queue->max_sectors > (PAGE_SIZE>>9))
153 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
154
155 disk->size = rdev->size;
7c7546cc 156 conf->array_size += rdev->size;
1da177e4 157
1da177e4
LT
158 cnt++;
159 }
7c7546cc 160 if (cnt != raid_disks) {
1da177e4
LT
161 printk("linear: not enough drives present. Aborting!\n");
162 goto out;
163 }
164
15945fee
N
165 min_spacing = mddev->array_size;
166 sector_div(min_spacing, PAGE_SIZE/sizeof(struct dev_info *));
167
168 /* min_spacing is the minimum spacing that will fit the hash
169 * table in one PAGE. This may be much smaller than needed.
170 * We find the smallest non-terminal set of consecutive devices
171 * that is larger than min_spacing as use the size of that as
172 * the actual spacing
173 */
174 conf->hash_spacing = mddev->array_size;
175 for (i=0; i < cnt-1 ; i++) {
176 sector_t sz = 0;
177 int j;
178 for (j=i; i<cnt-1 && sz < min_spacing ; j++)
179 sz += conf->disks[j].size;
180 if (sz >= min_spacing && sz < conf->hash_spacing)
181 conf->hash_spacing = sz;
182 }
183
184 /* hash_spacing may be too large for sector_div to work with,
185 * so we might need to pre-shift
186 */
187 conf->preshift = 0;
188 if (sizeof(sector_t) > sizeof(u32)) {
189 sector_t space = conf->hash_spacing;
190 while (space > (sector_t)(~(u32)0)) {
191 space >>= 1;
192 conf->preshift++;
193 }
194 }
1da177e4
LT
195 /*
196 * This code was restructured to work around a gcc-2.95.3 internal
197 * compiler error. Alter it with care.
198 */
199 {
200 sector_t sz;
201 unsigned round;
202 unsigned long base;
203
7c7546cc 204 sz = conf->array_size >> conf->preshift;
15945fee
N
205 sz += 1; /* force round-up */
206 base = conf->hash_spacing >> conf->preshift;
1da177e4 207 round = sector_div(sz, base);
15945fee 208 nb_zone = sz + (round ? 1 : 0);
1da177e4 209 }
15945fee
N
210 BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
211
212 conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
1da177e4
LT
213 GFP_KERNEL);
214 if (!conf->hash_table)
215 goto out;
216
217 /*
218 * Here we generate the linear hash table
15945fee 219 * First calculate the device offsets.
1da177e4 220 */
15945fee
N
221 conf->disks[0].offset = 0;
222 for (i=1; i<mddev->raid_disks; i++)
223 conf->disks[i].offset =
224 conf->disks[i-1].offset +
225 conf->disks[i-1].size;
226
1da177e4 227 table = conf->hash_table;
1da177e4 228 curr_offset = 0;
15945fee
N
229 i = 0;
230 for (curr_offset = 0;
231 curr_offset < mddev->array_size;
232 curr_offset += conf->hash_spacing) {
1da177e4 233
15945fee
N
234 while (i < mddev->raid_disks-1 &&
235 curr_offset >= conf->disks[i+1].offset)
236 i++;
1da177e4 237
15945fee
N
238 *table ++ = conf->disks + i;
239 }
240
241 if (conf->preshift) {
242 conf->hash_spacing >>= conf->preshift;
243 /* round hash_spacing up so that when we divide by it,
244 * we err on the side of "too-low", which is safest.
1da177e4 245 */
15945fee 246 conf->hash_spacing++;
1da177e4 247 }
15945fee
N
248
249 BUG_ON(table - conf->hash_table > nb_zone);
1da177e4 250
7c7546cc
N
251 return conf;
252
253out:
254 kfree(conf);
255 return NULL;
256}
257
258static int linear_run (mddev_t *mddev)
259{
260 linear_conf_t *conf;
261
262 conf = linear_conf(mddev, mddev->raid_disks);
263
264 if (!conf)
265 return 1;
266 mddev->private = conf;
267 mddev->array_size = conf->array_size;
268
1da177e4
LT
269 blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
270 mddev->queue->unplug_fn = linear_unplug;
271 mddev->queue->issue_flush_fn = linear_issue_flush;
272 return 0;
7c7546cc 273}
1da177e4 274
7c7546cc
N
275static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
276{
277 /* Adding a drive to a linear array allows the array to grow.
278 * It is permitted if the new drive has a matching superblock
279 * already on it, with raid_disk equal to raid_disks.
280 * It is achieved by creating a new linear_private_data structure
281 * and swapping it in in-place of the current one.
282 * The current one is never freed until the array is stopped.
283 * This avoids races.
284 */
285 linear_conf_t *newconf;
286
287 if (rdev->raid_disk != mddev->raid_disks)
288 return -EINVAL;
289
290 newconf = linear_conf(mddev,mddev->raid_disks+1);
291
292 if (!newconf)
293 return -ENOMEM;
294
295 newconf->prev = mddev_to_conf(mddev);
296 mddev->private = newconf;
297 mddev->raid_disks++;
298 mddev->array_size = newconf->array_size;
299 set_capacity(mddev->gendisk, mddev->array_size << 1);
300 return 0;
1da177e4
LT
301}
302
303static int linear_stop (mddev_t *mddev)
304{
305 linear_conf_t *conf = mddev_to_conf(mddev);
306
307 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
7c7546cc
N
308 do {
309 linear_conf_t *t = conf->prev;
310 kfree(conf->hash_table);
311 kfree(conf);
312 conf = t;
313 } while (conf);
1da177e4
LT
314
315 return 0;
316}
317
318static int linear_make_request (request_queue_t *q, struct bio *bio)
319{
a362357b 320 const int rw = bio_data_dir(bio);
1da177e4
LT
321 mddev_t *mddev = q->queuedata;
322 dev_info_t *tmp_dev;
323 sector_t block;
324
e5dcdd80
N
325 if (unlikely(bio_barrier(bio))) {
326 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
327 return 0;
328 }
329
a362357b
JA
330 disk_stat_inc(mddev->gendisk, ios[rw]);
331 disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
1da177e4
LT
332
333 tmp_dev = which_dev(mddev, bio->bi_sector);
334 block = bio->bi_sector >> 1;
335
336 if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
337 || block < tmp_dev->offset)) {
338 char b[BDEVNAME_SIZE];
339
340 printk("linear_make_request: Block %llu out of bounds on "
341 "dev %s size %llu offset %llu\n",
342 (unsigned long long)block,
343 bdevname(tmp_dev->rdev->bdev, b),
344 (unsigned long long)tmp_dev->size,
345 (unsigned long long)tmp_dev->offset);
346 bio_io_error(bio, bio->bi_size);
347 return 0;
348 }
349 if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
350 (tmp_dev->offset + tmp_dev->size)<<1)) {
351 /* This bio crosses a device boundary, so we have to
352 * split it.
353 */
354 struct bio_pair *bp;
29ac8e05
N
355 bp = bio_split(bio, bio_split_pool,
356 ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector);
1da177e4
LT
357 if (linear_make_request(q, &bp->bio1))
358 generic_make_request(&bp->bio1);
359 if (linear_make_request(q, &bp->bio2))
360 generic_make_request(&bp->bio2);
361 bio_pair_release(bp);
362 return 0;
363 }
364
365 bio->bi_bdev = tmp_dev->rdev->bdev;
366 bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset;
367
368 return 1;
369}
370
371static void linear_status (struct seq_file *seq, mddev_t *mddev)
372{
373
374#undef MD_DEBUG
375#ifdef MD_DEBUG
376 int j;
377 linear_conf_t *conf = mddev_to_conf(mddev);
378 sector_t s = 0;
379
380 seq_printf(seq, " ");
15945fee 381 for (j = 0; j < mddev->raid_disks; j++)
1da177e4
LT
382 {
383 char b[BDEVNAME_SIZE];
384 s += conf->smallest_size;
385 seq_printf(seq, "[%s",
386 bdevname(conf->hash_table[j][0].rdev->bdev,b));
387
388 while (s > conf->hash_table[j][0].offset +
389 conf->hash_table[j][0].size)
390 seq_printf(seq, "/%s] ",
391 bdevname(conf->hash_table[j][1].rdev->bdev,b));
392 else
393 seq_printf(seq, "] ");
394 }
395 seq_printf(seq, "\n");
396#endif
397 seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
398}
399
400
2604b703 401static struct mdk_personality linear_personality =
1da177e4
LT
402{
403 .name = "linear",
2604b703 404 .level = LEVEL_LINEAR,
1da177e4
LT
405 .owner = THIS_MODULE,
406 .make_request = linear_make_request,
407 .run = linear_run,
408 .stop = linear_stop,
409 .status = linear_status,
7c7546cc 410 .hot_add_disk = linear_add,
1da177e4
LT
411};
412
413static int __init linear_init (void)
414{
2604b703 415 return register_md_personality (&linear_personality);
1da177e4
LT
416}
417
418static void linear_exit (void)
419{
2604b703 420 unregister_md_personality (&linear_personality);
1da177e4
LT
421}
422
423
424module_init(linear_init);
425module_exit(linear_exit);
426MODULE_LICENSE("GPL");
d9d166c2
N
427MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
428MODULE_ALIAS("md-linear");
2604b703 429MODULE_ALIAS("md-level--1");