Merge master.kernel.org:/home/rmk/linux-2.6-drvmodel
[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 = kmalloc (sizeof (*conf) + mddev->raid_disks*sizeof(dev_info_t),
125                         GFP_KERNEL);
126         if (!conf)
127                 goto out;
128         memset(conf, 0, sizeof(*conf) + mddev->raid_disks*sizeof(dev_info_t));
129         mddev->private = conf;
130
131         cnt = 0;
132         mddev->array_size = 0;
133
134         ITERATE_RDEV(mddev,rdev,tmp) {
135                 int j = rdev->raid_disk;
136                 dev_info_t *disk = conf->disks + j;
137
138                 if (j < 0 || j > mddev->raid_disks || disk->rdev) {
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;
156                 mddev->array_size += rdev->size;
157
158                 cnt++;
159         }
160         if (cnt != mddev->raid_disks) {
161                 printk("linear: not enough drives present. Aborting!\n");
162                 goto out;
163         }
164
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         }
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
204                 sz = mddev->array_size >> conf->preshift;
205                 sz += 1; /* force round-up */
206                 base = conf->hash_spacing >> conf->preshift;
207                 round = sector_div(sz, base);
208                 nb_zone = sz + (round ? 1 : 0);
209         }
210         BUG_ON(nb_zone > PAGE_SIZE / sizeof(struct dev_info *));
211
212         conf->hash_table = kmalloc (sizeof (struct dev_info *) * nb_zone,
213                                         GFP_KERNEL);
214         if (!conf->hash_table)
215                 goto out;
216
217         /*
218          * Here we generate the linear hash table
219          * First calculate the device offsets.
220          */
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
227         table = conf->hash_table;
228         curr_offset = 0;
229         i = 0;
230         for (curr_offset = 0;
231              curr_offset < mddev->array_size;
232              curr_offset += conf->hash_spacing) {
233
234                 while (i < mddev->raid_disks-1 &&
235                        curr_offset >= conf->disks[i+1].offset)
236                         i++;
237
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.
245                  */
246                 conf->hash_spacing++;
247         }
248
249         BUG_ON(table - conf->hash_table > nb_zone);
250
251         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
252         mddev->queue->unplug_fn = linear_unplug;
253         mddev->queue->issue_flush_fn = linear_issue_flush;
254         return 0;
255
256 out:
257         kfree(conf);
258         return 1;
259 }
260
261 static int linear_stop (mddev_t *mddev)
262 {
263         linear_conf_t *conf = mddev_to_conf(mddev);
264   
265         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
266         kfree(conf->hash_table);
267         kfree(conf);
268
269         return 0;
270 }
271
272 static int linear_make_request (request_queue_t *q, struct bio *bio)
273 {
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         if (bio_data_dir(bio)==WRITE) {
284                 disk_stat_inc(mddev->gendisk, writes);
285                 disk_stat_add(mddev->gendisk, write_sectors, bio_sectors(bio));
286         } else {
287                 disk_stat_inc(mddev->gendisk, reads);
288                 disk_stat_add(mddev->gendisk, read_sectors, bio_sectors(bio));
289         }
290
291         tmp_dev = which_dev(mddev, bio->bi_sector);
292         block = bio->bi_sector >> 1;
293     
294         if (unlikely(block >= (tmp_dev->size + tmp_dev->offset)
295                      || block < tmp_dev->offset)) {
296                 char b[BDEVNAME_SIZE];
297
298                 printk("linear_make_request: Block %llu out of bounds on "
299                         "dev %s size %llu offset %llu\n",
300                         (unsigned long long)block,
301                         bdevname(tmp_dev->rdev->bdev, b),
302                         (unsigned long long)tmp_dev->size,
303                         (unsigned long long)tmp_dev->offset);
304                 bio_io_error(bio, bio->bi_size);
305                 return 0;
306         }
307         if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
308                      (tmp_dev->offset + tmp_dev->size)<<1)) {
309                 /* This bio crosses a device boundary, so we have to
310                  * split it.
311                  */
312                 struct bio_pair *bp;
313                 bp = bio_split(bio, bio_split_pool,
314                                ((tmp_dev->offset + tmp_dev->size)<<1) - bio->bi_sector);
315                 if (linear_make_request(q, &bp->bio1))
316                         generic_make_request(&bp->bio1);
317                 if (linear_make_request(q, &bp->bio2))
318                         generic_make_request(&bp->bio2);
319                 bio_pair_release(bp);
320                 return 0;
321         }
322                     
323         bio->bi_bdev = tmp_dev->rdev->bdev;
324         bio->bi_sector = bio->bi_sector - (tmp_dev->offset << 1) + tmp_dev->rdev->data_offset;
325
326         return 1;
327 }
328
329 static void linear_status (struct seq_file *seq, mddev_t *mddev)
330 {
331
332 #undef MD_DEBUG
333 #ifdef MD_DEBUG
334         int j;
335         linear_conf_t *conf = mddev_to_conf(mddev);
336         sector_t s = 0;
337   
338         seq_printf(seq, "      ");
339         for (j = 0; j < mddev->raid_disks; j++)
340         {
341                 char b[BDEVNAME_SIZE];
342                 s += conf->smallest_size;
343                 seq_printf(seq, "[%s",
344                            bdevname(conf->hash_table[j][0].rdev->bdev,b));
345
346                 while (s > conf->hash_table[j][0].offset +
347                            conf->hash_table[j][0].size)
348                         seq_printf(seq, "/%s] ",
349                                    bdevname(conf->hash_table[j][1].rdev->bdev,b));
350                 else
351                         seq_printf(seq, "] ");
352         }
353         seq_printf(seq, "\n");
354 #endif
355         seq_printf(seq, " %dk rounding", mddev->chunk_size/1024);
356 }
357
358
359 static mdk_personality_t linear_personality=
360 {
361         .name           = "linear",
362         .owner          = THIS_MODULE,
363         .make_request   = linear_make_request,
364         .run            = linear_run,
365         .stop           = linear_stop,
366         .status         = linear_status,
367 };
368
369 static int __init linear_init (void)
370 {
371         return register_md_personality (LINEAR, &linear_personality);
372 }
373
374 static void linear_exit (void)
375 {
376         unregister_md_personality (LINEAR);
377 }
378
379
380 module_init(linear_init);
381 module_exit(linear_exit);
382 MODULE_LICENSE("GPL");
383 MODULE_ALIAS("md-personality-1"); /* LINEAR */