blob: e9a73bb242b0935eefae2ce0744ed97965a2143d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Copyright (C) 2001 Sistina Software (UK) Limited.
Mikulas Patockad5816872009-01-06 03:05:10 +00003 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
5 * This file is released under the GPL.
6 */
7
8#include "dm.h"
9
10#include <linux/module.h>
11#include <linux/vmalloc.h>
12#include <linux/blkdev.h>
13#include <linux/namei.h>
14#include <linux/ctype.h>
15#include <linux/slab.h>
16#include <linux/interrupt.h>
Arjan van de Ven48c9c272006-03-27 01:18:20 -080017#include <linux/mutex.h>
Mikulas Patockad5816872009-01-06 03:05:10 +000018#include <linux/delay.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <asm/atomic.h>
20
Alasdair G Kergon72d94862006-06-26 00:27:35 -070021#define DM_MSG_PREFIX "table"
22
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#define MAX_DEPTH 16
24#define NODE_SIZE L1_CACHE_BYTES
25#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
26#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
27
Mikulas Patockad5816872009-01-06 03:05:10 +000028/*
29 * The table has always exactly one reference from either mapped_device->map
30 * or hash_cell->new_map. This reference is not counted in table->holders.
31 * A pair of dm_create_table/dm_destroy_table functions is used for table
32 * creation/destruction.
33 *
34 * Temporary references from the other code increase table->holders. A pair
35 * of dm_table_get/dm_table_put functions is used to manipulate it.
36 *
37 * When the table is about to be destroyed, we wait for table->holders to
38 * drop to zero.
39 */
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041struct dm_table {
Mike Anderson1134e5a2006-03-27 01:17:54 -080042 struct mapped_device *md;
Linus Torvalds1da177e2005-04-16 15:20:36 -070043 atomic_t holders;
44
45 /* btree table */
46 unsigned int depth;
47 unsigned int counts[MAX_DEPTH]; /* in nodes */
48 sector_t *index[MAX_DEPTH];
49
50 unsigned int num_targets;
51 unsigned int num_allocated;
52 sector_t *highs;
53 struct dm_target *targets;
54
55 /*
56 * Indicates the rw permissions for the new logical
57 * device. This should be a combination of FMODE_READ
58 * and FMODE_WRITE.
59 */
Al Viroaeb5d722008-09-02 15:28:45 -040060 fmode_t mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
62 /* a list of devices used by this table */
63 struct list_head devices;
64
65 /*
66 * These are optimistic limits taken from all the
67 * targets, some targets will need smaller limits.
68 */
69 struct io_restrictions limits;
70
71 /* events get handed up using this callback */
72 void (*event_fn)(void *);
73 void *event_context;
74};
75
76/*
77 * Similar to ceiling(log_size(n))
78 */
79static unsigned int int_log(unsigned int n, unsigned int base)
80{
81 int result = 0;
82
83 while (n > 1) {
84 n = dm_div_up(n, base);
85 result++;
86 }
87
88 return result;
89}
90
91/*
92 * Returns the minimum that is _not_ zero, unless both are zero.
93 */
94#define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
95
96/*
97 * Combine two io_restrictions, always taking the lower value.
98 */
99static void combine_restrictions_low(struct io_restrictions *lhs,
100 struct io_restrictions *rhs)
101{
102 lhs->max_sectors =
103 min_not_zero(lhs->max_sectors, rhs->max_sectors);
104
105 lhs->max_phys_segments =
106 min_not_zero(lhs->max_phys_segments, rhs->max_phys_segments);
107
108 lhs->max_hw_segments =
109 min_not_zero(lhs->max_hw_segments, rhs->max_hw_segments);
110
Martin K. Petersene1defc42009-05-22 17:17:49 -0400111 lhs->logical_block_size = max(lhs->logical_block_size,
112 rhs->logical_block_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
114 lhs->max_segment_size =
115 min_not_zero(lhs->max_segment_size, rhs->max_segment_size);
116
Neil Brown91212502007-12-13 14:16:04 +0000117 lhs->max_hw_sectors =
118 min_not_zero(lhs->max_hw_sectors, rhs->max_hw_sectors);
119
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 lhs->seg_boundary_mask =
121 min_not_zero(lhs->seg_boundary_mask, rhs->seg_boundary_mask);
NeilBrown969429b2006-03-27 01:17:49 -0800122
Vasily Averin5ec140e2007-10-31 08:33:24 +0100123 lhs->bounce_pfn = min_not_zero(lhs->bounce_pfn, rhs->bounce_pfn);
124
NeilBrown969429b2006-03-27 01:17:49 -0800125 lhs->no_cluster |= rhs->no_cluster;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126}
127
128/*
129 * Calculate the index of the child node of the n'th node k'th key.
130 */
131static inline unsigned int get_child(unsigned int n, unsigned int k)
132{
133 return (n * CHILDREN_PER_NODE) + k;
134}
135
136/*
137 * Return the n'th node of level l from table t.
138 */
139static inline sector_t *get_node(struct dm_table *t,
140 unsigned int l, unsigned int n)
141{
142 return t->index[l] + (n * KEYS_PER_NODE);
143}
144
145/*
146 * Return the highest key that you could lookup from the n'th
147 * node on level l of the btree.
148 */
149static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
150{
151 for (; l < t->depth - 1; l++)
152 n = get_child(n, CHILDREN_PER_NODE - 1);
153
154 if (n >= t->counts[l])
155 return (sector_t) - 1;
156
157 return get_node(t, l, n)[KEYS_PER_NODE - 1];
158}
159
160/*
161 * Fills in a level of the btree based on the highs of the level
162 * below it.
163 */
164static int setup_btree_index(unsigned int l, struct dm_table *t)
165{
166 unsigned int n, k;
167 sector_t *node;
168
169 for (n = 0U; n < t->counts[l]; n++) {
170 node = get_node(t, l, n);
171
172 for (k = 0U; k < KEYS_PER_NODE; k++)
173 node[k] = high(t, l + 1, get_child(n, k));
174 }
175
176 return 0;
177}
178
179void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
180{
181 unsigned long size;
182 void *addr;
183
184 /*
185 * Check that we're not going to overflow.
186 */
187 if (nmemb > (ULONG_MAX / elem_size))
188 return NULL;
189
190 size = nmemb * elem_size;
191 addr = vmalloc(size);
192 if (addr)
193 memset(addr, 0, size);
194
195 return addr;
196}
197
198/*
199 * highs, and targets are managed as dynamic arrays during a
200 * table load.
201 */
202static int alloc_targets(struct dm_table *t, unsigned int num)
203{
204 sector_t *n_highs;
205 struct dm_target *n_targets;
206 int n = t->num_targets;
207
208 /*
209 * Allocate both the target array and offset array at once.
Jun'ichi Nomura512875b2007-12-13 14:15:25 +0000210 * Append an empty entry to catch sectors beyond the end of
211 * the device.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 */
Jun'ichi Nomura512875b2007-12-13 14:15:25 +0000213 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 sizeof(sector_t));
215 if (!n_highs)
216 return -ENOMEM;
217
218 n_targets = (struct dm_target *) (n_highs + num);
219
220 if (n) {
221 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
222 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
223 }
224
225 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
226 vfree(t->highs);
227
228 t->num_allocated = num;
229 t->highs = n_highs;
230 t->targets = n_targets;
231
232 return 0;
233}
234
Al Viroaeb5d722008-09-02 15:28:45 -0400235int dm_table_create(struct dm_table **result, fmode_t mode,
Mike Anderson1134e5a2006-03-27 01:17:54 -0800236 unsigned num_targets, struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Dmitry Monakhov094262d2007-10-19 22:38:51 +0100238 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
240 if (!t)
241 return -ENOMEM;
242
Linus Torvalds1da177e2005-04-16 15:20:36 -0700243 INIT_LIST_HEAD(&t->devices);
Mikulas Patockad5816872009-01-06 03:05:10 +0000244 atomic_set(&t->holders, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245
246 if (!num_targets)
247 num_targets = KEYS_PER_NODE;
248
249 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
250
251 if (alloc_targets(t, num_targets)) {
252 kfree(t);
253 t = NULL;
254 return -ENOMEM;
255 }
256
257 t->mode = mode;
Mike Anderson1134e5a2006-03-27 01:17:54 -0800258 t->md = md;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 *result = t;
260 return 0;
261}
262
263static void free_devices(struct list_head *devices)
264{
265 struct list_head *tmp, *next;
266
Paul Jimenezafb24522008-02-08 02:09:59 +0000267 list_for_each_safe(tmp, next, devices) {
Mikulas Patocka82b15192008-10-10 13:37:09 +0100268 struct dm_dev_internal *dd =
269 list_entry(tmp, struct dm_dev_internal, list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 kfree(dd);
271 }
272}
273
Mikulas Patockad5816872009-01-06 03:05:10 +0000274void dm_table_destroy(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275{
276 unsigned int i;
277
Mikulas Patockad5816872009-01-06 03:05:10 +0000278 while (atomic_read(&t->holders))
279 msleep(1);
280 smp_mb();
281
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 /* free the indexes (see dm_table_complete) */
283 if (t->depth >= 2)
284 vfree(t->index[t->depth - 2]);
285
286 /* free the targets */
287 for (i = 0; i < t->num_targets; i++) {
288 struct dm_target *tgt = t->targets + i;
289
290 if (tgt->type->dtr)
291 tgt->type->dtr(tgt);
292
293 dm_put_target_type(tgt->type);
294 }
295
296 vfree(t->highs);
297
298 /* free the device list */
299 if (t->devices.next != &t->devices) {
300 DMWARN("devices still present during destroy: "
301 "dm_table_remove_device calls missing");
302
303 free_devices(&t->devices);
304 }
305
306 kfree(t);
307}
308
309void dm_table_get(struct dm_table *t)
310{
311 atomic_inc(&t->holders);
312}
313
314void dm_table_put(struct dm_table *t)
315{
316 if (!t)
317 return;
318
Mikulas Patockad5816872009-01-06 03:05:10 +0000319 smp_mb__before_atomic_dec();
320 atomic_dec(&t->holders);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
323/*
324 * Checks to see if we need to extend highs or targets.
325 */
326static inline int check_space(struct dm_table *t)
327{
328 if (t->num_targets >= t->num_allocated)
329 return alloc_targets(t, t->num_allocated * 2);
330
331 return 0;
332}
333
334/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 * See if we've already got a device in the list.
336 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100337static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338{
Mikulas Patocka82b15192008-10-10 13:37:09 +0100339 struct dm_dev_internal *dd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
341 list_for_each_entry (dd, l, list)
Mikulas Patocka82b15192008-10-10 13:37:09 +0100342 if (dd->dm_dev.bdev->bd_dev == dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 return dd;
344
345 return NULL;
346}
347
348/*
349 * Open a device so we can use it as a map destination.
350 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100351static int open_dev(struct dm_dev_internal *d, dev_t dev,
352 struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353{
354 static char *_claim_ptr = "I belong to device-mapper";
355 struct block_device *bdev;
356
357 int r;
358
Mikulas Patocka82b15192008-10-10 13:37:09 +0100359 BUG_ON(d->dm_dev.bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
Mikulas Patocka82b15192008-10-10 13:37:09 +0100361 bdev = open_by_devnum(dev, d->dm_dev.mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 if (IS_ERR(bdev))
363 return PTR_ERR(bdev);
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800364 r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 if (r)
Al Viro9a1c3542008-02-22 20:40:24 -0500366 blkdev_put(bdev, d->dm_dev.mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 else
Mikulas Patocka82b15192008-10-10 13:37:09 +0100368 d->dm_dev.bdev = bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 return r;
370}
371
372/*
373 * Close a device that we've been using.
374 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100375static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376{
Mikulas Patocka82b15192008-10-10 13:37:09 +0100377 if (!d->dm_dev.bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 return;
379
Mikulas Patocka82b15192008-10-10 13:37:09 +0100380 bd_release_from_disk(d->dm_dev.bdev, dm_disk(md));
Al Viro9a1c3542008-02-22 20:40:24 -0500381 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode);
Mikulas Patocka82b15192008-10-10 13:37:09 +0100382 d->dm_dev.bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
384
385/*
Mike Anderson2cd54d92007-05-09 02:32:57 -0700386 * If possible, this checks an area of a destination device is valid.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100388static int check_device_area(struct dm_dev_internal *dd, sector_t start,
389 sector_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390{
Mikulas Patocka82b15192008-10-10 13:37:09 +0100391 sector_t dev_size = dd->dm_dev.bdev->bd_inode->i_size >> SECTOR_SHIFT;
Mike Anderson2cd54d92007-05-09 02:32:57 -0700392
393 if (!dev_size)
394 return 1;
395
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 return ((start < dev_size) && (len <= (dev_size - start)));
397}
398
399/*
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100400 * This upgrades the mode on an already open dm_dev, being
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401 * careful to leave things as they were if we fail to reopen the
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100402 * device and not to touch the existing bdev field in case
403 * it is accessed concurrently inside dm_table_any_congested().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 */
Al Viroaeb5d722008-09-02 15:28:45 -0400405static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
Mikulas Patocka82b15192008-10-10 13:37:09 +0100406 struct mapped_device *md)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407{
408 int r;
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100409 struct dm_dev_internal dd_new, dd_old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100411 dd_new = dd_old = *dd;
412
413 dd_new.dm_dev.mode |= new_mode;
414 dd_new.dm_dev.bdev = NULL;
415
416 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
417 if (r)
418 return r;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419
Mikulas Patocka82b15192008-10-10 13:37:09 +0100420 dd->dm_dev.mode |= new_mode;
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100421 close_dev(&dd_old, md);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422
Alasdair G Kergon570b9d92009-04-02 19:55:28 +0100423 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425
426/*
427 * Add a device to the list, or just increment the usage count if
428 * it's already present.
429 */
430static int __table_get_device(struct dm_table *t, struct dm_target *ti,
431 const char *path, sector_t start, sector_t len,
Al Viroaeb5d722008-09-02 15:28:45 -0400432 fmode_t mode, struct dm_dev **result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433{
434 int r;
Andrew Morton69a2ce72008-02-08 02:10:14 +0000435 dev_t uninitialized_var(dev);
Mikulas Patocka82b15192008-10-10 13:37:09 +0100436 struct dm_dev_internal *dd;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 unsigned int major, minor;
438
Eric Sesterhenn547bc922006-03-26 18:22:50 +0200439 BUG_ON(!t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
441 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
442 /* Extract the major/minor numbers */
443 dev = MKDEV(major, minor);
444 if (MAJOR(dev) != major || MINOR(dev) != minor)
445 return -EOVERFLOW;
446 } else {
447 /* convert the path to a device */
Christoph Hellwig72e82642008-08-11 00:24:08 +0200448 struct block_device *bdev = lookup_bdev(path);
449
450 if (IS_ERR(bdev))
451 return PTR_ERR(bdev);
452 dev = bdev->bd_dev;
453 bdput(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 }
455
456 dd = find_device(&t->devices, dev);
457 if (!dd) {
458 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
459 if (!dd)
460 return -ENOMEM;
461
Mikulas Patocka82b15192008-10-10 13:37:09 +0100462 dd->dm_dev.mode = mode;
463 dd->dm_dev.bdev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800465 if ((r = open_dev(dd, dev, t->md))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 kfree(dd);
467 return r;
468 }
469
Mikulas Patocka82b15192008-10-10 13:37:09 +0100470 format_dev_t(dd->dm_dev.name, dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
472 atomic_set(&dd->count, 0);
473 list_add(&dd->list, &t->devices);
474
Mikulas Patocka82b15192008-10-10 13:37:09 +0100475 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800476 r = upgrade_mode(dd, mode, t->md);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 if (r)
478 return r;
479 }
480 atomic_inc(&dd->count);
481
482 if (!check_device_area(dd, start, len)) {
483 DMWARN("device %s too small for target", path);
Mikulas Patocka82b15192008-10-10 13:37:09 +0100484 dm_put_device(ti, &dd->dm_dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 return -EINVAL;
486 }
487
Mikulas Patocka82b15192008-10-10 13:37:09 +0100488 *result = &dd->dm_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
490 return 0;
491}
492
Bryn Reeves3cb40212006-10-03 01:15:42 -0700493void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev)
494{
Jens Axboe165125e2007-07-24 09:28:11 +0200495 struct request_queue *q = bdev_get_queue(bdev);
Bryn Reeves3cb40212006-10-03 01:15:42 -0700496 struct io_restrictions *rs = &ti->limits;
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +0100497 char b[BDEVNAME_SIZE];
498
499 if (unlikely(!q)) {
500 DMWARN("%s: Cannot set limits for nonexistent device %s",
501 dm_device_name(ti->table->md), bdevname(bdev, b));
502 return;
503 }
Bryn Reeves3cb40212006-10-03 01:15:42 -0700504
505 /*
506 * Combine the device limits low.
507 *
508 * FIXME: if we move an io_restriction struct
509 * into q this would just be a call to
510 * combine_restrictions_low()
511 */
512 rs->max_sectors =
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400513 min_not_zero(rs->max_sectors, queue_max_sectors(q));
Bryn Reeves3cb40212006-10-03 01:15:42 -0700514
Milan Broz9980c632008-07-21 12:00:39 +0100515 /*
516 * Check if merge fn is supported.
517 * If not we'll force DM to use PAGE_SIZE or
518 * smaller I/O, just to be safe.
Bryn Reeves3cb40212006-10-03 01:15:42 -0700519 */
Milan Broz9980c632008-07-21 12:00:39 +0100520
521 if (q->merge_bvec_fn && !ti->type->merge)
Bryn Reeves3cb40212006-10-03 01:15:42 -0700522 rs->max_sectors =
523 min_not_zero(rs->max_sectors,
524 (unsigned int) (PAGE_SIZE >> 9));
525
526 rs->max_phys_segments =
527 min_not_zero(rs->max_phys_segments,
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400528 queue_max_phys_segments(q));
Bryn Reeves3cb40212006-10-03 01:15:42 -0700529
530 rs->max_hw_segments =
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400531 min_not_zero(rs->max_hw_segments, queue_max_hw_segments(q));
Bryn Reeves3cb40212006-10-03 01:15:42 -0700532
Martin K. Petersene1defc42009-05-22 17:17:49 -0400533 rs->logical_block_size = max(rs->logical_block_size,
534 queue_logical_block_size(q));
Bryn Reeves3cb40212006-10-03 01:15:42 -0700535
536 rs->max_segment_size =
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400537 min_not_zero(rs->max_segment_size, queue_max_segment_size(q));
Bryn Reeves3cb40212006-10-03 01:15:42 -0700538
Neil Brown91212502007-12-13 14:16:04 +0000539 rs->max_hw_sectors =
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400540 min_not_zero(rs->max_hw_sectors, queue_max_hw_sectors(q));
Neil Brown91212502007-12-13 14:16:04 +0000541
Bryn Reeves3cb40212006-10-03 01:15:42 -0700542 rs->seg_boundary_mask =
543 min_not_zero(rs->seg_boundary_mask,
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400544 queue_segment_boundary(q));
Bryn Reeves3cb40212006-10-03 01:15:42 -0700545
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400546 rs->bounce_pfn = min_not_zero(rs->bounce_pfn, queue_bounce_pfn(q));
Vasily Averin5ec140e2007-10-31 08:33:24 +0100547
Bryn Reeves3cb40212006-10-03 01:15:42 -0700548 rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
549}
550EXPORT_SYMBOL_GPL(dm_set_device_limits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551
552int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
Al Viroaeb5d722008-09-02 15:28:45 -0400553 sector_t len, fmode_t mode, struct dm_dev **result)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
555 int r = __table_get_device(ti->table, ti, path,
556 start, len, mode, result);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557
Bryn Reeves3cb40212006-10-03 01:15:42 -0700558 if (!r)
559 dm_set_device_limits(ti, (*result)->bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561 return r;
562}
563
564/*
565 * Decrement a devices use count and remove it if necessary.
566 */
Mikulas Patocka82b15192008-10-10 13:37:09 +0100567void dm_put_device(struct dm_target *ti, struct dm_dev *d)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
Mikulas Patocka82b15192008-10-10 13:37:09 +0100569 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
570 dm_dev);
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 if (atomic_dec_and_test(&dd->count)) {
Jun'ichi Nomuraf1659212006-03-27 01:17:59 -0800573 close_dev(dd, ti->table->md);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 list_del(&dd->list);
575 kfree(dd);
576 }
577}
578
579/*
580 * Checks to see if the target joins onto the end of the table.
581 */
582static int adjoin(struct dm_table *table, struct dm_target *ti)
583{
584 struct dm_target *prev;
585
586 if (!table->num_targets)
587 return !ti->begin;
588
589 prev = &table->targets[table->num_targets - 1];
590 return (ti->begin == (prev->begin + prev->len));
591}
592
593/*
594 * Used to dynamically allocate the arg array.
595 */
596static char **realloc_argv(unsigned *array_size, char **old_argv)
597{
598 char **argv;
599 unsigned new_size;
600
601 new_size = *array_size ? *array_size * 2 : 64;
602 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
603 if (argv) {
604 memcpy(argv, old_argv, *array_size * sizeof(*argv));
605 *array_size = new_size;
606 }
607
608 kfree(old_argv);
609 return argv;
610}
611
612/*
613 * Destructively splits up the argument list to pass to ctr.
614 */
615int dm_split_args(int *argc, char ***argvp, char *input)
616{
617 char *start, *end = input, *out, **argv = NULL;
618 unsigned array_size = 0;
619
620 *argc = 0;
David Teigland814d6862006-06-26 00:27:31 -0700621
622 if (!input) {
623 *argvp = NULL;
624 return 0;
625 }
626
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 argv = realloc_argv(&array_size, argv);
628 if (!argv)
629 return -ENOMEM;
630
631 while (1) {
632 start = end;
633
634 /* Skip whitespace */
635 while (*start && isspace(*start))
636 start++;
637
638 if (!*start)
639 break; /* success, we hit the end */
640
641 /* 'out' is used to remove any back-quotes */
642 end = out = start;
643 while (*end) {
644 /* Everything apart from '\0' can be quoted */
645 if (*end == '\\' && *(end + 1)) {
646 *out++ = *(end + 1);
647 end += 2;
648 continue;
649 }
650
651 if (isspace(*end))
652 break; /* end of token */
653
654 *out++ = *end++;
655 }
656
657 /* have we already filled the array ? */
658 if ((*argc + 1) > array_size) {
659 argv = realloc_argv(&array_size, argv);
660 if (!argv)
661 return -ENOMEM;
662 }
663
664 /* we know this is whitespace */
665 if (*end)
666 end++;
667
668 /* terminate the string and put it in the array */
669 *out = '\0';
670 argv[*argc] = start;
671 (*argc)++;
672 }
673
674 *argvp = argv;
675 return 0;
676}
677
678static void check_for_valid_limits(struct io_restrictions *rs)
679{
680 if (!rs->max_sectors)
Mike Christiedefd94b2005-12-05 02:37:06 -0600681 rs->max_sectors = SAFE_MAX_SECTORS;
Neil Brown91212502007-12-13 14:16:04 +0000682 if (!rs->max_hw_sectors)
683 rs->max_hw_sectors = SAFE_MAX_SECTORS;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 if (!rs->max_phys_segments)
685 rs->max_phys_segments = MAX_PHYS_SEGMENTS;
686 if (!rs->max_hw_segments)
687 rs->max_hw_segments = MAX_HW_SEGMENTS;
Martin K. Petersene1defc42009-05-22 17:17:49 -0400688 if (!rs->logical_block_size)
689 rs->logical_block_size = 1 << SECTOR_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 if (!rs->max_segment_size)
691 rs->max_segment_size = MAX_SEGMENT_SIZE;
692 if (!rs->seg_boundary_mask)
Milan Broz0e435ac2008-12-03 12:55:08 +0100693 rs->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
Vasily Averin5ec140e2007-10-31 08:33:24 +0100694 if (!rs->bounce_pfn)
695 rs->bounce_pfn = -1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696}
697
698int dm_table_add_target(struct dm_table *t, const char *type,
699 sector_t start, sector_t len, char *params)
700{
701 int r = -EINVAL, argc;
702 char **argv;
703 struct dm_target *tgt;
704
705 if ((r = check_space(t)))
706 return r;
707
708 tgt = t->targets + t->num_targets;
709 memset(tgt, 0, sizeof(*tgt));
710
711 if (!len) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700712 DMERR("%s: zero-length target", dm_device_name(t->md));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 return -EINVAL;
714 }
715
716 tgt->type = dm_get_target_type(type);
717 if (!tgt->type) {
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700718 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
719 type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 return -EINVAL;
721 }
722
723 tgt->table = t;
724 tgt->begin = start;
725 tgt->len = len;
726 tgt->error = "Unknown error";
727
728 /*
729 * Does this target adjoin the previous one ?
730 */
731 if (!adjoin(t, tgt)) {
732 tgt->error = "Gap in table";
733 r = -EINVAL;
734 goto bad;
735 }
736
737 r = dm_split_args(&argc, &argv, params);
738 if (r) {
739 tgt->error = "couldn't split parameters (insufficient memory)";
740 goto bad;
741 }
742
743 r = tgt->type->ctr(tgt, argc, argv);
744 kfree(argv);
745 if (r)
746 goto bad;
747
748 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
749
750 /* FIXME: the plan is to combine high here and then have
751 * the merge fn apply the target level restrictions. */
752 combine_restrictions_low(&t->limits, &tgt->limits);
753 return 0;
754
755 bad:
Alasdair G Kergon72d94862006-06-26 00:27:35 -0700756 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757 dm_put_target_type(tgt->type);
758 return r;
759}
760
761static int setup_indexes(struct dm_table *t)
762{
763 int i;
764 unsigned int total = 0;
765 sector_t *indexes;
766
767 /* allocate the space for *all* the indexes */
768 for (i = t->depth - 2; i >= 0; i--) {
769 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
770 total += t->counts[i];
771 }
772
773 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
774 if (!indexes)
775 return -ENOMEM;
776
777 /* set up internal nodes, bottom-up */
Jun'ichi Nomura82d601d2008-02-08 02:10:04 +0000778 for (i = t->depth - 2; i >= 0; i--) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 t->index[i] = indexes;
780 indexes += (KEYS_PER_NODE * t->counts[i]);
781 setup_btree_index(i, t);
782 }
783
784 return 0;
785}
786
787/*
788 * Builds the btree to index the map.
789 */
790int dm_table_complete(struct dm_table *t)
791{
792 int r = 0;
793 unsigned int leaf_nodes;
794
795 check_for_valid_limits(&t->limits);
796
797 /* how many indexes will the btree have ? */
798 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
799 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
800
801 /* leaf layer has already been set up */
802 t->counts[t->depth - 1] = leaf_nodes;
803 t->index[t->depth - 1] = t->highs;
804
805 if (t->depth >= 2)
806 r = setup_indexes(t);
807
808 return r;
809}
810
Arjan van de Ven48c9c272006-03-27 01:18:20 -0800811static DEFINE_MUTEX(_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812void dm_table_event_callback(struct dm_table *t,
813 void (*fn)(void *), void *context)
814{
Arjan van de Ven48c9c272006-03-27 01:18:20 -0800815 mutex_lock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 t->event_fn = fn;
817 t->event_context = context;
Arjan van de Ven48c9c272006-03-27 01:18:20 -0800818 mutex_unlock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819}
820
821void dm_table_event(struct dm_table *t)
822{
823 /*
824 * You can no longer call dm_table_event() from interrupt
825 * context, use a bottom half instead.
826 */
827 BUG_ON(in_interrupt());
828
Arjan van de Ven48c9c272006-03-27 01:18:20 -0800829 mutex_lock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 if (t->event_fn)
831 t->event_fn(t->event_context);
Arjan van de Ven48c9c272006-03-27 01:18:20 -0800832 mutex_unlock(&_event_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833}
834
835sector_t dm_table_get_size(struct dm_table *t)
836{
837 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
838}
839
840struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
841{
Milan Broz14353532006-06-26 00:27:27 -0700842 if (index >= t->num_targets)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 return NULL;
844
845 return t->targets + index;
846}
847
848/*
849 * Search the btree for the correct target.
Jun'ichi Nomura512875b2007-12-13 14:15:25 +0000850 *
851 * Caller should check returned pointer with dm_target_is_valid()
852 * to trap I/O beyond end of device.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 */
854struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
855{
856 unsigned int l, n = 0, k = 0;
857 sector_t *node;
858
859 for (l = 0; l < t->depth; l++) {
860 n = get_child(n, k);
861 node = get_node(t, l, n);
862
863 for (k = 0; k < KEYS_PER_NODE; k++)
864 if (node[k] >= sector)
865 break;
866 }
867
868 return &t->targets[(KEYS_PER_NODE * n) + k];
869}
870
Martin K. Petersen9c470082009-04-09 00:27:12 +0100871/*
872 * Set the integrity profile for this device if all devices used have
873 * matching profiles.
874 */
875static void dm_table_set_integrity(struct dm_table *t)
876{
877 struct list_head *devices = dm_table_get_devices(t);
878 struct dm_dev_internal *prev = NULL, *dd = NULL;
879
880 if (!blk_get_integrity(dm_disk(t->md)))
881 return;
882
883 list_for_each_entry(dd, devices, list) {
884 if (prev &&
885 blk_integrity_compare(prev->dm_dev.bdev->bd_disk,
886 dd->dm_dev.bdev->bd_disk) < 0) {
887 DMWARN("%s: integrity not set: %s and %s mismatch",
888 dm_device_name(t->md),
889 prev->dm_dev.bdev->bd_disk->disk_name,
890 dd->dm_dev.bdev->bd_disk->disk_name);
891 goto no_integrity;
892 }
893 prev = dd;
894 }
895
896 if (!prev || !bdev_get_integrity(prev->dm_dev.bdev))
897 goto no_integrity;
898
899 blk_integrity_register(dm_disk(t->md),
900 bdev_get_integrity(prev->dm_dev.bdev));
901
902 return;
903
904no_integrity:
905 blk_integrity_register(dm_disk(t->md), NULL);
906
907 return;
908}
909
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
911{
912 /*
913 * Make sure we obey the optimistic sub devices
914 * restrictions.
915 */
916 blk_queue_max_sectors(q, t->limits.max_sectors);
Martin K. Petersenae03bf62009-05-22 17:17:50 -0400917 blk_queue_max_phys_segments(q, t->limits.max_phys_segments);
918 blk_queue_max_hw_segments(q, t->limits.max_hw_segments);
919 blk_queue_logical_block_size(q, t->limits.logical_block_size);
920 blk_queue_max_segment_size(q, t->limits.max_segment_size);
921 blk_queue_max_hw_sectors(q, t->limits.max_hw_sectors);
922 blk_queue_segment_boundary(q, t->limits.seg_boundary_mask);
Jens Axboe9df1bb92009-06-09 06:22:57 +0200923 blk_queue_bounce_limit(q, t->limits.bounce_pfn);
Jens Axboec9a3f6d2008-04-29 19:12:35 +0200924
NeilBrown969429b2006-03-27 01:17:49 -0800925 if (t->limits.no_cluster)
Jens Axboec9a3f6d2008-04-29 19:12:35 +0200926 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
NeilBrown969429b2006-03-27 01:17:49 -0800927 else
Jens Axboec9a3f6d2008-04-29 19:12:35 +0200928 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, q);
NeilBrown969429b2006-03-27 01:17:49 -0800929
Martin K. Petersen9c470082009-04-09 00:27:12 +0100930 dm_table_set_integrity(t);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931}
932
933unsigned int dm_table_get_num_targets(struct dm_table *t)
934{
935 return t->num_targets;
936}
937
938struct list_head *dm_table_get_devices(struct dm_table *t)
939{
940 return &t->devices;
941}
942
Al Viroaeb5d722008-09-02 15:28:45 -0400943fmode_t dm_table_get_mode(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944{
945 return t->mode;
946}
947
948static void suspend_targets(struct dm_table *t, unsigned postsuspend)
949{
950 int i = t->num_targets;
951 struct dm_target *ti = t->targets;
952
953 while (i--) {
954 if (postsuspend) {
955 if (ti->type->postsuspend)
956 ti->type->postsuspend(ti);
957 } else if (ti->type->presuspend)
958 ti->type->presuspend(ti);
959
960 ti++;
961 }
962}
963
964void dm_table_presuspend_targets(struct dm_table *t)
965{
Alasdair G Kergoncf222b32005-07-28 21:15:57 -0700966 if (!t)
967 return;
968
Adrian Bunke8488d02008-04-24 22:10:51 +0100969 suspend_targets(t, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970}
971
972void dm_table_postsuspend_targets(struct dm_table *t)
973{
Alasdair G Kergoncf222b32005-07-28 21:15:57 -0700974 if (!t)
975 return;
976
Adrian Bunke8488d02008-04-24 22:10:51 +0100977 suspend_targets(t, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978}
979
Milan Broz8757b772006-10-03 01:15:36 -0700980int dm_table_resume_targets(struct dm_table *t)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981{
Milan Broz8757b772006-10-03 01:15:36 -0700982 int i, r = 0;
983
984 for (i = 0; i < t->num_targets; i++) {
985 struct dm_target *ti = t->targets + i;
986
987 if (!ti->type->preresume)
988 continue;
989
990 r = ti->type->preresume(ti);
991 if (r)
992 return r;
993 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
995 for (i = 0; i < t->num_targets; i++) {
996 struct dm_target *ti = t->targets + i;
997
998 if (ti->type->resume)
999 ti->type->resume(ti);
1000 }
Milan Broz8757b772006-10-03 01:15:36 -07001001
1002 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003}
1004
1005int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1006{
Mikulas Patocka82b15192008-10-10 13:37:09 +01001007 struct dm_dev_internal *dd;
Paul Jimenezafb24522008-02-08 02:09:59 +00001008 struct list_head *devices = dm_table_get_devices(t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 int r = 0;
1010
Paul Jimenezafb24522008-02-08 02:09:59 +00001011 list_for_each_entry(dd, devices, list) {
Mikulas Patocka82b15192008-10-10 13:37:09 +01001012 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +01001013 char b[BDEVNAME_SIZE];
1014
1015 if (likely(q))
1016 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1017 else
1018 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1019 dm_device_name(t->md),
1020 bdevname(dd->dm_dev.bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 }
1022
1023 return r;
1024}
1025
1026void dm_table_unplug_all(struct dm_table *t)
1027{
Mikulas Patocka82b15192008-10-10 13:37:09 +01001028 struct dm_dev_internal *dd;
Paul Jimenezafb24522008-02-08 02:09:59 +00001029 struct list_head *devices = dm_table_get_devices(t);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
Paul Jimenezafb24522008-02-08 02:09:59 +00001031 list_for_each_entry(dd, devices, list) {
Mikulas Patocka82b15192008-10-10 13:37:09 +01001032 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +01001033 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034
Alasdair G Kergon0c2322e2008-10-10 13:37:13 +01001035 if (likely(q))
1036 blk_unplug(q);
1037 else
1038 DMWARN_LIMIT("%s: Cannot unplug nonexistent device %s",
1039 dm_device_name(t->md),
1040 bdevname(dd->dm_dev.bdev, b));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 }
1042}
1043
Mike Anderson1134e5a2006-03-27 01:17:54 -08001044struct mapped_device *dm_table_get_md(struct dm_table *t)
1045{
1046 dm_get(t->md);
1047
1048 return t->md;
1049}
1050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051EXPORT_SYMBOL(dm_vcalloc);
1052EXPORT_SYMBOL(dm_get_device);
1053EXPORT_SYMBOL(dm_put_device);
1054EXPORT_SYMBOL(dm_table_event);
Alasdair G Kergond5e404c2005-07-12 15:53:05 -07001055EXPORT_SYMBOL(dm_table_get_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056EXPORT_SYMBOL(dm_table_get_mode);
Mike Anderson1134e5a2006-03-27 01:17:54 -08001057EXPORT_SYMBOL(dm_table_get_md);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058EXPORT_SYMBOL(dm_table_put);
1059EXPORT_SYMBOL(dm_table_get);
1060EXPORT_SYMBOL(dm_table_unplug_all);