Merge master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / drivers / md / linear.c
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  */
32 static 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          */
41         block >>= conf->preshift;
42         (void)sector_div(block, conf->hash_spacing);
43         hash = conf->hash_table[block];
44
45         while ((sector>>1) >= (hash->size + hash->offset))
46                 hash++;
47         return hash;
48 }
49
50 /**
51  *      linear_mergeable_bvec -- tell bio layer if two requests can be merged
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  */
58 static 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
82 static 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
95 static 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
114 static int linear_run (mddev_t *mddev)
115 {
116         linear_conf_t *conf;
117         dev_info_t **table;
118         mdk_rdev_t *rdev;
119         int i, nb_zone, cnt;
120         sector_t min_spacing;
121         sector_t curr_offset;
122         struct list_head *tmp;
123
124         conf = kzalloc (sizeof (*conf) + mddev->raid_disks*sizeof(dev_info_t),
125                         GFP_KERNEL);
126         if (!conf)
127                 goto out;
128         mddev->private = conf;
129
130         cnt = 0;
131         mddev->array_size = 0;
132
133         ITERATE_RDEV(mddev,rdev,tmp) {
134                 int j = rdev->raid_disk;
135                 dev_info_t *disk = conf->disks + j;
136
137                 if (j < 0 || j > mddev->raid_disks || disk->rdev) {
138                         printk("linear: disk numbering problem. Aborting!\n");
139                         goto out;
140                 }
141
142                 disk->rdev = rdev;
143
144                 blk_queue_stack_limits(mddev->queue,
145                                        rdev->bdev->bd_disk->queue);
146                 /* as we don't honour merge_bvec_fn, we must never risk
147                  * violating it, so limit ->max_sector to one PAGE, as
148                  * a one page request is never in violation.
149                  */
150                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
151                     mddev->queue->max_sectors > (PAGE_SIZE>>9))
152                         blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
153
154                 disk->size = rdev->size;
155                 mddev->array_size += rdev->size;
156
157                 cnt++;
158         }
159         if (cnt != mddev->raid_disks) {
160                 printk("linear: not enough drives present. Aborting!\n");
161                 goto out;
162         }
163
164         min_spacing = mddev->array_size;
165         sector_div(min_spacing, PAGE_SIZE/sizeof(struct dev_info *));
166
167         /* min_spacing is the minimum spacing that will fit the hash
168          * table in one PAGE.  This may be much smaller than needed.
169          * We find the smallest non-terminal set of consecutive devices
170          * that is larger than min_spacing as use the size of that as
171          * the actual spacing
172          */
173         conf->hash_spacing = mddev->array_size;
174         for (i=0; i < cnt-1 ; i++) {
175                 sector_t sz = 0;
176                 int j;
177                 for (j=i; i<cnt-1 && sz < min_spacing ; j++)
178                         sz += conf->disks[j].size;
179                 if (sz >= min_spacing && sz < conf->hash_spacing)
180                         conf->hash_spacing = sz;
181         }
182
183         /* hash_spacing may be too large for sector_div to work with,
184          * so we might need to pre-shift
185          */
186         conf->preshift = 0;
187         if (sizeof(sector_t) > sizeof(u32)) {
188                 sector_t space = conf->hash_spacing;
189                 while (space > (sector_t)(~(u32)0)) {
190                         space >>= 1;
191                         conf->preshift++;
192                 }
193         }
194         /*
195          * This code was restructured to work around a gcc-2.95.3 internal
196          * compiler error.  Alter it with care.
197          */
198         {
199                 sector_t sz;
200                 unsigned round;
201                 unsigned long base;
202
203                 sz = mddev->array_size >> conf->preshift;
204                 sz += 1; /* force round-up */
205                 base = conf->hash_spacing >> conf->preshift;
206                 round = sector_div(sz, base);
207                 nb_zone = sz + (round ? 1 : 0);
208         }
209         BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
210
211         conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
212                                         GFP_KERNEL);
213         if (!conf->hash_table)
214                 goto out;
215
216         /*
217          * Here we generate the linear hash table
218          * First calculate the device offsets.
219          */
220         conf->disks[0].offset = 0;
221         for (i=1; i<mddev->raid_disks; i++)
222                 conf->disks[i].offset =
223                         conf->disks[i-1].offset +
224                         conf->disks[i-1].size;
225
226         table = conf->hash_table;
227         curr_offset = 0;
228         i = 0;
229         for (curr_offset = 0;
230              curr_offset < mddev->array_size;
231              curr_offset += conf->hash_spacing) {
232
233                 while (i < mddev->raid_disks-1 &&
234                        curr_offset >= conf->disks[i+1].offset)
235                         i++;
236
237                 *table ++ = conf->disks + i;
238         }
239
240         if (conf->preshift) {
241                 conf->hash_spacing >>= conf->preshift;
242                 /* round hash_spacing up so that when we divide by it,
243                  * we err on the side of "too-low", which is safest.
244                  */
245                 conf->hash_spacing++;
246         }
247
248         BUG_ON(table - conf->hash_table > nb_zone);
249
250         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
251         mddev->queue->unplug_fn = linear_unplug;
252         mddev->queue->issue_flush_fn = linear_issue_flush;
253         return 0;
254
255 out:
256         kfree(conf);
257         return 1;
258 }
259
260 static int linear_stop (mddev_t *mddev)
261 {
262         linear_conf_t *conf = mddev_to_conf(mddev);
263   
264         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
265         kfree(conf->hash_table);
266         kfree(conf);
267
268         return 0;
269 }
270
271 static int linear_make_request (request_queue_t *q, struct bio *bio)
272 {
273         const int rw = bio_data_dir(bio);
274         mddev_t *mddev = q->queuedata;
275         dev_info_t *tmp_dev;
276         sector_t block;
277
278         if (unlikely(bio_barrier(bio))) {
279                 bio_endio(bio, bio->bi_size, -EOPNOTSUPP);
280                 return 0;
281         }
282
283         disk_stat_inc(mddev->gendisk, ios[rw]);
284         disk_stat_add(mddev->gendisk, sectors[rw], bio_sectors(bio));
285
286         tmp_dev = which_dev(mddev, bio->bi_sector);
287         block = bio->bi_sector >> 1;
288     
289         if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
290                      || block < tmp_dev->offset)) {
291                 char b[BDEVNAME_SIZE];
292
293                 printk("linear_make_request: Block %llu out of bounds on "
294                         "dev %s size %llu offset %llu\n",
295                         (unsigned long long)block,
296                         bdevname(tmp_dev->rdev->bdev, b),
297                         (unsigned long long)tmp_dev->size,
298                         (unsigned long long)tmp_dev->offset);
299                 bio_io_error(bio, bio->bi_size);
300                 return 0;
301         }
302         if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
303                      (tmp_dev->offset + tmp_dev->size)<<1)) {
304                 /* This bio crosses a device boundary, so we have to
305                  * split it.
306                  */
307                 struct bio_pair *bp;
308                 bp = bio_split(bio, bio_split_pool,
309                                ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector);
310                 if (linear_make_request(q, &bp->bio1))
311                         generic_make_request(&bp->bio1);
312                 if (linear_make_request(q, &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 - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset;
320
321         return 1;
322 }
323
324 static void linear_status (struct seq_file *seq, mddev_t *mddev)
325 {
326
327 #undef MD_DEBUG
328 #ifdef MD_DEBUG
329         int j;
330         linear_conf_t *conf = mddev_to_conf(mddev);
331         sector_t s = 0;
332   
333         seq_printf(seq, "      ");
334         for (j = 0; j < mddev->raid_disks; j++)
335         {
336                 char b[BDEVNAME_SIZE];
337                 s += conf->smallest_size;
338                 seq_printf(seq, "[%s",
339                            bdevname(conf->hash_table[j][0].rdev->bdev,b));
340
341                 while (s > conf->hash_table[j][0].offset +
342                            conf->hash_table[j][0].size)
343                         seq_printf(seq, "/%s] ",
344                                    bdevname(conf->hash_table[j][1].rdev->bdev,b));
345                 else
346                         seq_printf(seq, "] ");
347         }
348         seq_printf(seq, "\n");
349 #endif
350         seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
351 }
352
353
354 static struct mdk_personality linear_personality =
355 {
356         .name           = "linear",
357         .level          = LEVEL_LINEAR,
358         .owner          = THIS_MODULE,
359         .make_request   = linear_make_request,
360         .run            = linear_run,
361         .stop           = linear_stop,
362         .status         = linear_status,
363 };
364
365 static int __init linear_init (void)
366 {
367         return register_md_personality (&linear_personality);
368 }
369
370 static void linear_exit (void)
371 {
372         unregister_md_personality (&linear_personality);
373 }
374
375
376 module_init(linear_init);
377 module_exit(linear_exit);
378 MODULE_LICENSE("GPL");
379 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
380 MODULE_ALIAS("md-linear");
381 MODULE_ALIAS("md-level--1");