Merge commit '3ff195b011d7decf501a4d55aeed312731094796' into for-linus
[linux-2.6.git] / 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/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  */
29 static 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  */
62 static 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
90 static void linear_unplug(struct request_queue *q)
91 {
92         mddev_t *mddev = q->queuedata;
93         linear_conf_t *conf;
94         int i;
95
96         rcu_read_lock();
97         conf = rcu_dereference(mddev->private);
98
99         for (i=0; i < mddev->raid_disks; i++) {
100                 struct request_queue *r_queue = bdev_get_queue(conf->disks[i].rdev->bdev);
101                 blk_unplug(r_queue);
102         }
103         rcu_read_unlock();
104 }
105
106 static int linear_congested(void *data, int bits)
107 {
108         mddev_t *mddev = data;
109         linear_conf_t *conf;
110         int i, ret = 0;
111
112         if (mddev_congested(mddev, bits))
113                 return 1;
114
115         rcu_read_lock();
116         conf = rcu_dereference(mddev->private);
117
118         for (i = 0; i < mddev->raid_disks && !ret ; i++) {
119                 struct request_queue *q = bdev_get_queue(conf->disks[i].rdev->bdev);
120                 ret |= bdi_congested(&q->backing_dev_info, bits);
121         }
122
123         rcu_read_unlock();
124         return ret;
125 }
126
127 static sector_t linear_size(mddev_t *mddev, sector_t sectors, int raid_disks)
128 {
129         linear_conf_t *conf;
130         sector_t array_sectors;
131
132         rcu_read_lock();
133         conf = rcu_dereference(mddev->private);
134         WARN_ONCE(sectors || raid_disks,
135                   "%s does not support generic reshape\n", __func__);
136         array_sectors = conf->array_sectors;
137         rcu_read_unlock();
138
139         return array_sectors;
140 }
141
142 static linear_conf_t *linear_conf(mddev_t *mddev, int raid_disks)
143 {
144         linear_conf_t *conf;
145         mdk_rdev_t *rdev;
146         int i, cnt;
147
148         conf = kzalloc (sizeof (*conf) + raid_disks*sizeof(dev_info_t),
149                         GFP_KERNEL);
150         if (!conf)
151                 return NULL;
152
153         cnt = 0;
154         conf->array_sectors = 0;
155
156         list_for_each_entry(rdev, &mddev->disks, same_set) {
157                 int j = rdev->raid_disk;
158                 dev_info_t *disk = conf->disks + j;
159                 sector_t sectors;
160
161                 if (j < 0 || j >= raid_disks || disk->rdev) {
162                         printk(KERN_ERR "md/linear:%s: disk numbering problem. Aborting!\n",
163                                mdname(mddev));
164                         goto out;
165                 }
166
167                 disk->rdev = rdev;
168                 if (mddev->chunk_sectors) {
169                         sectors = rdev->sectors;
170                         sector_div(sectors, mddev->chunk_sectors);
171                         rdev->sectors = sectors * mddev->chunk_sectors;
172                 }
173
174                 disk_stack_limits(mddev->gendisk, rdev->bdev,
175                                   rdev->data_offset << 9);
176                 /* as we don't honour merge_bvec_fn, we must never risk
177                  * violating it, so limit max_segments to 1 lying within
178                  * a single page.
179                  */
180                 if (rdev->bdev->bd_disk->queue->merge_bvec_fn) {
181                         blk_queue_max_segments(mddev->queue, 1);
182                         blk_queue_segment_boundary(mddev->queue,
183                                                    PAGE_CACHE_SIZE - 1);
184                 }
185
186                 conf->array_sectors += rdev->sectors;
187                 cnt++;
188
189         }
190         if (cnt != raid_disks) {
191                 printk(KERN_ERR "md/linear:%s: not enough drives present. Aborting!\n",
192                        mdname(mddev));
193                 goto out;
194         }
195
196         /*
197          * Here we calculate the device offsets.
198          */
199         conf->disks[0].end_sector = conf->disks[0].rdev->sectors;
200
201         for (i = 1; i < raid_disks; i++)
202                 conf->disks[i].end_sector =
203                         conf->disks[i-1].end_sector +
204                         conf->disks[i].rdev->sectors;
205
206         return conf;
207
208 out:
209         kfree(conf);
210         return NULL;
211 }
212
213 static int linear_run (mddev_t *mddev)
214 {
215         linear_conf_t *conf;
216
217         if (md_check_no_bitmap(mddev))
218                 return -EINVAL;
219         mddev->queue->queue_lock = &mddev->queue->__queue_lock;
220         conf = linear_conf(mddev, mddev->raid_disks);
221
222         if (!conf)
223                 return 1;
224         mddev->private = conf;
225         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
226
227         blk_queue_merge_bvec(mddev->queue, linear_mergeable_bvec);
228         mddev->queue->unplug_fn = linear_unplug;
229         mddev->queue->backing_dev_info.congested_fn = linear_congested;
230         mddev->queue->backing_dev_info.congested_data = mddev;
231         md_integrity_register(mddev);
232         return 0;
233 }
234
235 static void free_conf(struct rcu_head *head)
236 {
237         linear_conf_t *conf = container_of(head, linear_conf_t, rcu);
238         kfree(conf);
239 }
240
241 static int linear_add(mddev_t *mddev, mdk_rdev_t *rdev)
242 {
243         /* Adding a drive to a linear array allows the array to grow.
244          * It is permitted if the new drive has a matching superblock
245          * already on it, with raid_disk equal to raid_disks.
246          * It is achieved by creating a new linear_private_data structure
247          * and swapping it in in-place of the current one.
248          * The current one is never freed until the array is stopped.
249          * This avoids races.
250          */
251         linear_conf_t *newconf, *oldconf;
252
253         if (rdev->saved_raid_disk != mddev->raid_disks)
254                 return -EINVAL;
255
256         rdev->raid_disk = rdev->saved_raid_disk;
257
258         newconf = linear_conf(mddev,mddev->raid_disks+1);
259
260         if (!newconf)
261                 return -ENOMEM;
262
263         oldconf = rcu_dereference(mddev->private);
264         mddev->raid_disks++;
265         rcu_assign_pointer(mddev->private, newconf);
266         md_set_array_sectors(mddev, linear_size(mddev, 0, 0));
267         set_capacity(mddev->gendisk, mddev->array_sectors);
268         revalidate_disk(mddev->gendisk);
269         call_rcu(&oldconf->rcu, free_conf);
270         return 0;
271 }
272
273 static int linear_stop (mddev_t *mddev)
274 {
275         linear_conf_t *conf = mddev->private;
276
277         /*
278          * We do not require rcu protection here since
279          * we hold reconfig_mutex for both linear_add and
280          * linear_stop, so they cannot race.
281          * We should make sure any old 'conf's are properly
282          * freed though.
283          */
284         rcu_barrier();
285         blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
286         kfree(conf);
287         mddev->private = NULL;
288
289         return 0;
290 }
291
292 static int linear_make_request (mddev_t *mddev, struct bio *bio)
293 {
294         dev_info_t *tmp_dev;
295         sector_t start_sector;
296
297         if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) {
298                 md_barrier_request(mddev, bio);
299                 return 0;
300         }
301
302         rcu_read_lock();
303         tmp_dev = which_dev(mddev, bio->bi_sector);
304         start_sector = tmp_dev->end_sector - tmp_dev->rdev->sectors;
305
306
307         if (unlikely(bio->bi_sector >= (tmp_dev->end_sector)
308                      || (bio->bi_sector < start_sector))) {
309                 char b[BDEVNAME_SIZE];
310
311                 printk(KERN_ERR
312                        "md/linear:%s: make_request: Sector %llu out of bounds on "
313                        "dev %s: %llu sectors, offset %llu\n",
314                        mdname(mddev),
315                        (unsigned long long)bio->bi_sector,
316                        bdevname(tmp_dev->rdev->bdev, b),
317                        (unsigned long long)tmp_dev->rdev->sectors,
318                        (unsigned long long)start_sector);
319                 rcu_read_unlock();
320                 bio_io_error(bio);
321                 return 0;
322         }
323         if (unlikely(bio->bi_sector + (bio->bi_size >> 9) >
324                      tmp_dev->end_sector)) {
325                 /* This bio crosses a device boundary, so we have to
326                  * split it.
327                  */
328                 struct bio_pair *bp;
329                 sector_t end_sector = tmp_dev->end_sector;
330
331                 rcu_read_unlock();
332
333                 bp = bio_split(bio, end_sector - bio->bi_sector);
334
335                 if (linear_make_request(mddev, &bp->bio1))
336                         generic_make_request(&bp->bio1);
337                 if (linear_make_request(mddev, &bp->bio2))
338                         generic_make_request(&bp->bio2);
339                 bio_pair_release(bp);
340                 return 0;
341         }
342                     
343         bio->bi_bdev = tmp_dev->rdev->bdev;
344         bio->bi_sector = bio->bi_sector - start_sector
345                 + tmp_dev->rdev->data_offset;
346         rcu_read_unlock();
347
348         return 1;
349 }
350
351 static void linear_status (struct seq_file *seq, mddev_t *mddev)
352 {
353
354         seq_printf(seq, " %dk rounding", mddev->chunk_sectors / 2);
355 }
356
357
358 static struct mdk_personality linear_personality =
359 {
360         .name           = "linear",
361         .level          = LEVEL_LINEAR,
362         .owner          = THIS_MODULE,
363         .make_request   = linear_make_request,
364         .run            = linear_run,
365         .stop           = linear_stop,
366         .status         = linear_status,
367         .hot_add_disk   = linear_add,
368         .size           = linear_size,
369 };
370
371 static int __init linear_init (void)
372 {
373         return register_md_personality (&linear_personality);
374 }
375
376 static void linear_exit (void)
377 {
378         unregister_md_personality (&linear_personality);
379 }
380
381
382 module_init(linear_init);
383 module_exit(linear_exit);
384 MODULE_LICENSE("GPL");
385 MODULE_DESCRIPTION("Linear device concatenation personality for MD");
386 MODULE_ALIAS("md-personality-1"); /* LINEAR - deprecated*/
387 MODULE_ALIAS("md-linear");
388 MODULE_ALIAS("md-level--1");