]>
Commit | Line | Data |
---|---|---|
32a7627c N |
1 | /* |
2 | * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003 | |
3 | * | |
4 | * bitmap_create - sets up the bitmap structure | |
5 | * bitmap_destroy - destroys the bitmap structure | |
6 | * | |
7 | * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.: | |
8 | * - added disk storage for bitmap | |
9 | * - changes to allow various bitmap chunk sizes | |
32a7627c N |
10 | */ |
11 | ||
12 | /* | |
13 | * Still to do: | |
14 | * | |
15 | * flush after percent set rather than just time based. (maybe both). | |
32a7627c N |
16 | */ |
17 | ||
bff61975 | 18 | #include <linux/blkdev.h> |
32a7627c | 19 | #include <linux/module.h> |
32a7627c N |
20 | #include <linux/errno.h> |
21 | #include <linux/slab.h> | |
22 | #include <linux/init.h> | |
32a7627c N |
23 | #include <linux/timer.h> |
24 | #include <linux/sched.h> | |
25 | #include <linux/list.h> | |
26 | #include <linux/file.h> | |
27 | #include <linux/mount.h> | |
28 | #include <linux/buffer_head.h> | |
57148964 | 29 | #include <linux/seq_file.h> |
43b2e5d8 | 30 | #include "md.h" |
ef740c37 | 31 | #include "bitmap.h" |
32a7627c | 32 | |
ac2f40be | 33 | static inline char *bmname(struct bitmap *bitmap) |
32a7627c N |
34 | { |
35 | return bitmap->mddev ? mdname(bitmap->mddev) : "mdX"; | |
36 | } | |
37 | ||
32a7627c N |
38 | /* |
39 | * check a page and, if necessary, allocate it (or hijack it if the alloc fails) | |
40 | * | |
41 | * 1) check to see if this page is allocated, if it's not then try to alloc | |
42 | * 2) if the alloc fails, set the page's hijacked flag so we'll use the | |
43 | * page pointer directly as a counter | |
44 | * | |
45 | * if we find our page, we increment the page's refcount so that it stays | |
46 | * allocated while we're using it | |
47 | */ | |
ac2f40be N |
48 | static int bitmap_checkpage(struct bitmap *bitmap, |
49 | unsigned long page, int create) | |
ee305ace N |
50 | __releases(bitmap->lock) |
51 | __acquires(bitmap->lock) | |
32a7627c N |
52 | { |
53 | unsigned char *mappage; | |
54 | ||
55 | if (page >= bitmap->pages) { | |
1187cf0a N |
56 | /* This can happen if bitmap_start_sync goes beyond |
57 | * End-of-device while looking for a whole page. | |
58 | * It is harmless. | |
59 | */ | |
32a7627c N |
60 | return -EINVAL; |
61 | } | |
62 | ||
32a7627c N |
63 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ |
64 | return 0; | |
65 | ||
66 | if (bitmap->bp[page].map) /* page is already allocated, just return */ | |
67 | return 0; | |
68 | ||
69 | if (!create) | |
70 | return -ENOENT; | |
71 | ||
32a7627c N |
72 | /* this page has not been allocated yet */ |
73 | ||
ac2f40be | 74 | spin_unlock_irq(&bitmap->lock); |
792a1d4b | 75 | mappage = kzalloc(PAGE_SIZE, GFP_NOIO); |
ac2f40be N |
76 | spin_lock_irq(&bitmap->lock); |
77 | ||
78 | if (mappage == NULL) { | |
36a4e1fe N |
79 | pr_debug("%s: bitmap map page allocation failed, hijacking\n", |
80 | bmname(bitmap)); | |
32a7627c N |
81 | /* failed - set the hijacked flag so that we can use the |
82 | * pointer as a counter */ | |
32a7627c N |
83 | if (!bitmap->bp[page].map) |
84 | bitmap->bp[page].hijacked = 1; | |
ac2f40be N |
85 | } else if (bitmap->bp[page].map || |
86 | bitmap->bp[page].hijacked) { | |
32a7627c | 87 | /* somebody beat us to getting the page */ |
792a1d4b | 88 | kfree(mappage); |
32a7627c | 89 | return 0; |
ac2f40be | 90 | } else { |
32a7627c | 91 | |
ac2f40be | 92 | /* no page was in place and we have one, so install it */ |
32a7627c | 93 | |
ac2f40be N |
94 | bitmap->bp[page].map = mappage; |
95 | bitmap->missing_pages--; | |
96 | } | |
32a7627c N |
97 | return 0; |
98 | } | |
99 | ||
32a7627c N |
100 | /* if page is completely empty, put it back on the free list, or dealloc it */ |
101 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
102 | /* Note: lock should be held when calling this */ | |
858119e1 | 103 | static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page) |
32a7627c N |
104 | { |
105 | char *ptr; | |
106 | ||
107 | if (bitmap->bp[page].count) /* page is still busy */ | |
108 | return; | |
109 | ||
110 | /* page is no longer in use, it can be released */ | |
111 | ||
112 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
113 | bitmap->bp[page].hijacked = 0; | |
114 | bitmap->bp[page].map = NULL; | |
ac2f40be N |
115 | } else { |
116 | /* normal case, free the page */ | |
117 | ptr = bitmap->bp[page].map; | |
118 | bitmap->bp[page].map = NULL; | |
119 | bitmap->missing_pages++; | |
792a1d4b | 120 | kfree(ptr); |
32a7627c | 121 | } |
32a7627c N |
122 | } |
123 | ||
32a7627c N |
124 | /* |
125 | * bitmap file handling - read and write the bitmap file and its superblock | |
126 | */ | |
127 | ||
32a7627c N |
128 | /* |
129 | * basic page I/O operations | |
130 | */ | |
131 | ||
a654b9d8 | 132 | /* IO operations when bitmap is stored near all superblocks */ |
27581e5a N |
133 | static int read_sb_page(struct mddev *mddev, loff_t offset, |
134 | struct page *page, | |
135 | unsigned long index, int size) | |
a654b9d8 N |
136 | { |
137 | /* choose a good rdev and read the page from there */ | |
138 | ||
3cb03002 | 139 | struct md_rdev *rdev; |
a654b9d8 | 140 | sector_t target; |
a654b9d8 | 141 | |
dafb20fa | 142 | rdev_for_each(rdev, mddev) { |
b2d444d7 N |
143 | if (! test_bit(In_sync, &rdev->flags) |
144 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
145 | continue; |
146 | ||
ccebd4c4 | 147 | target = offset + index * (PAGE_SIZE/512); |
a654b9d8 | 148 | |
2b193363 | 149 | if (sync_page_io(rdev, target, |
e1defc4f | 150 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
ccebd4c4 | 151 | page, READ, true)) { |
ab904d63 | 152 | page->index = index; |
27581e5a | 153 | return 0; |
ab904d63 N |
154 | } |
155 | } | |
27581e5a | 156 | return -EIO; |
a654b9d8 N |
157 | } |
158 | ||
fd01b88c | 159 | static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) |
b2d2c4ce N |
160 | { |
161 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
162 | * linked list, and raising the refcount of devices we return to ensure | |
163 | * they don't disappear while in use. | |
164 | * As devices are only added or removed when raid_disk is < 0 and | |
165 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
166 | * still be in the same position on the list when we re-enter | |
167 | * list_for_each_continue_rcu. | |
168 | */ | |
169 | struct list_head *pos; | |
170 | rcu_read_lock(); | |
171 | if (rdev == NULL) | |
172 | /* start at the beginning */ | |
173 | pos = &mddev->disks; | |
174 | else { | |
175 | /* release the previous rdev and start from there. */ | |
176 | rdev_dec_pending(rdev, mddev); | |
177 | pos = &rdev->same_set; | |
178 | } | |
179 | list_for_each_continue_rcu(pos, &mddev->disks) { | |
3cb03002 | 180 | rdev = list_entry(pos, struct md_rdev, same_set); |
b2d2c4ce | 181 | if (rdev->raid_disk >= 0 && |
b2d2c4ce N |
182 | !test_bit(Faulty, &rdev->flags)) { |
183 | /* this is a usable devices */ | |
184 | atomic_inc(&rdev->nr_pending); | |
185 | rcu_read_unlock(); | |
186 | return rdev; | |
187 | } | |
188 | } | |
189 | rcu_read_unlock(); | |
190 | return NULL; | |
191 | } | |
192 | ||
ab6085c7 | 193 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 194 | { |
3cb03002 | 195 | struct md_rdev *rdev = NULL; |
a6ff7e08 | 196 | struct block_device *bdev; |
fd01b88c | 197 | struct mddev *mddev = bitmap->mddev; |
1ec885cd | 198 | struct bitmap_storage *store = &bitmap->storage; |
a654b9d8 | 199 | |
b2d2c4ce | 200 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ac2f40be N |
201 | int size = PAGE_SIZE; |
202 | loff_t offset = mddev->bitmap_info.offset; | |
a6ff7e08 JB |
203 | |
204 | bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; | |
205 | ||
1ec885cd N |
206 | if (page->index == store->file_pages-1) |
207 | size = roundup(store->last_page_size, | |
a6ff7e08 | 208 | bdev_logical_block_size(bdev)); |
ac2f40be N |
209 | /* Just make sure we aren't corrupting data or |
210 | * metadata | |
211 | */ | |
212 | if (mddev->external) { | |
213 | /* Bitmap could be anywhere. */ | |
214 | if (rdev->sb_start + offset + (page->index | |
215 | * (PAGE_SIZE/512)) | |
216 | > rdev->data_offset | |
217 | && | |
218 | rdev->sb_start + offset | |
219 | < (rdev->data_offset + mddev->dev_sectors | |
220 | + (PAGE_SIZE/512))) | |
221 | goto bad_alignment; | |
222 | } else if (offset < 0) { | |
223 | /* DATA BITMAP METADATA */ | |
224 | if (offset | |
225 | + (long)(page->index * (PAGE_SIZE/512)) | |
226 | + size/512 > 0) | |
227 | /* bitmap runs in to metadata */ | |
228 | goto bad_alignment; | |
229 | if (rdev->data_offset + mddev->dev_sectors | |
230 | > rdev->sb_start + offset) | |
231 | /* data runs in to bitmap */ | |
232 | goto bad_alignment; | |
233 | } else if (rdev->sb_start < rdev->data_offset) { | |
234 | /* METADATA BITMAP DATA */ | |
235 | if (rdev->sb_start | |
236 | + offset | |
237 | + page->index*(PAGE_SIZE/512) + size/512 | |
238 | > rdev->data_offset) | |
239 | /* bitmap runs in to data */ | |
240 | goto bad_alignment; | |
241 | } else { | |
242 | /* DATA METADATA BITMAP - no problems */ | |
243 | } | |
244 | md_super_write(mddev, rdev, | |
245 | rdev->sb_start + offset | |
246 | + page->index * (PAGE_SIZE/512), | |
247 | size, | |
248 | page); | |
b2d2c4ce | 249 | } |
a654b9d8 N |
250 | |
251 | if (wait) | |
a9701a30 | 252 | md_super_wait(mddev); |
a654b9d8 | 253 | return 0; |
4b80991c N |
254 | |
255 | bad_alignment: | |
4b80991c | 256 | return -EINVAL; |
a654b9d8 N |
257 | } |
258 | ||
4ad13663 | 259 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 260 | /* |
a654b9d8 | 261 | * write out a page to a file |
32a7627c | 262 | */ |
4ad13663 | 263 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 264 | { |
d785a06a | 265 | struct buffer_head *bh; |
32a7627c | 266 | |
1ec885cd | 267 | if (bitmap->storage.file == NULL) { |
f0d76d70 N |
268 | switch (write_sb_page(bitmap, page, wait)) { |
269 | case -EINVAL: | |
270 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
f0d76d70 | 271 | } |
4ad13663 | 272 | } else { |
a654b9d8 | 273 | |
4ad13663 | 274 | bh = page_buffers(page); |
c708443c | 275 | |
4ad13663 N |
276 | while (bh && bh->b_blocknr) { |
277 | atomic_inc(&bitmap->pending_writes); | |
278 | set_buffer_locked(bh); | |
279 | set_buffer_mapped(bh); | |
721a9602 | 280 | submit_bh(WRITE | REQ_SYNC, bh); |
4ad13663 N |
281 | bh = bh->b_this_page; |
282 | } | |
d785a06a | 283 | |
ac2f40be | 284 | if (wait) |
4ad13663 N |
285 | wait_event(bitmap->write_wait, |
286 | atomic_read(&bitmap->pending_writes)==0); | |
8a5e9cf1 | 287 | } |
4ad13663 N |
288 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
289 | bitmap_file_kick(bitmap); | |
d785a06a N |
290 | } |
291 | ||
292 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
293 | { | |
294 | struct bitmap *bitmap = bh->b_private; | |
295 | unsigned long flags; | |
32a7627c | 296 | |
d785a06a N |
297 | if (!uptodate) { |
298 | spin_lock_irqsave(&bitmap->lock, flags); | |
299 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
300 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
32a7627c | 301 | } |
d785a06a N |
302 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
303 | wake_up(&bitmap->write_wait); | |
304 | } | |
32a7627c | 305 | |
d785a06a N |
306 | /* copied from buffer.c */ |
307 | static void | |
308 | __clear_page_buffers(struct page *page) | |
309 | { | |
310 | ClearPagePrivate(page); | |
311 | set_page_private(page, 0); | |
312 | page_cache_release(page); | |
313 | } | |
314 | static void free_buffers(struct page *page) | |
315 | { | |
27581e5a | 316 | struct buffer_head *bh; |
77ad4bc7 | 317 | |
27581e5a N |
318 | if (!PagePrivate(page)) |
319 | return; | |
320 | ||
321 | bh = page_buffers(page); | |
d785a06a N |
322 | while (bh) { |
323 | struct buffer_head *next = bh->b_this_page; | |
324 | free_buffer_head(bh); | |
325 | bh = next; | |
77ad4bc7 | 326 | } |
d785a06a N |
327 | __clear_page_buffers(page); |
328 | put_page(page); | |
32a7627c N |
329 | } |
330 | ||
d785a06a N |
331 | /* read a page from a file. |
332 | * We both read the page, and attach buffers to the page to record the | |
333 | * address of each block (using bmap). These addresses will be used | |
334 | * to write the block later, completely bypassing the filesystem. | |
335 | * This usage is similar to how swap files are handled, and allows us | |
336 | * to write to a file with no concerns of memory allocation failing. | |
337 | */ | |
27581e5a N |
338 | static int read_page(struct file *file, unsigned long index, |
339 | struct bitmap *bitmap, | |
340 | unsigned long count, | |
341 | struct page *page) | |
32a7627c | 342 | { |
27581e5a | 343 | int ret = 0; |
c649bb9c | 344 | struct inode *inode = file->f_path.dentry->d_inode; |
d785a06a N |
345 | struct buffer_head *bh; |
346 | sector_t block; | |
32a7627c | 347 | |
36a4e1fe N |
348 | pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
349 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 350 | |
d785a06a N |
351 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
352 | if (!bh) { | |
27581e5a | 353 | ret = -ENOMEM; |
32a7627c N |
354 | goto out; |
355 | } | |
d785a06a N |
356 | attach_page_buffers(page, bh); |
357 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
358 | while (bh) { | |
359 | if (count == 0) | |
360 | bh->b_blocknr = 0; | |
361 | else { | |
362 | bh->b_blocknr = bmap(inode, block); | |
363 | if (bh->b_blocknr == 0) { | |
364 | /* Cannot use this file! */ | |
27581e5a | 365 | ret = -EINVAL; |
d785a06a N |
366 | goto out; |
367 | } | |
368 | bh->b_bdev = inode->i_sb->s_bdev; | |
369 | if (count < (1<<inode->i_blkbits)) | |
370 | count = 0; | |
371 | else | |
372 | count -= (1<<inode->i_blkbits); | |
373 | ||
374 | bh->b_end_io = end_bitmap_write; | |
375 | bh->b_private = bitmap; | |
ce25c31b N |
376 | atomic_inc(&bitmap->pending_writes); |
377 | set_buffer_locked(bh); | |
378 | set_buffer_mapped(bh); | |
379 | submit_bh(READ, bh); | |
d785a06a N |
380 | } |
381 | block++; | |
382 | bh = bh->b_this_page; | |
383 | } | |
d785a06a | 384 | page->index = index; |
ce25c31b N |
385 | |
386 | wait_event(bitmap->write_wait, | |
387 | atomic_read(&bitmap->pending_writes)==0); | |
27581e5a N |
388 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
389 | ret = -EIO; | |
32a7627c | 390 | out: |
27581e5a N |
391 | if (ret) |
392 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n", | |
2d1f3b5d N |
393 | (int)PAGE_SIZE, |
394 | (unsigned long long)index << PAGE_SHIFT, | |
27581e5a N |
395 | ret); |
396 | return ret; | |
32a7627c N |
397 | } |
398 | ||
399 | /* | |
400 | * bitmap file superblock operations | |
401 | */ | |
402 | ||
403 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 404 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
405 | { |
406 | bitmap_super_t *sb; | |
32a7627c N |
407 | |
408 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 409 | return; |
ece5cff0 N |
410 | if (bitmap->mddev->bitmap_info.external) |
411 | return; | |
1ec885cd | 412 | if (!bitmap->storage.sb_page) /* no superblock */ |
4ad13663 | 413 | return; |
1ec885cd | 414 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 415 | sb->events = cpu_to_le64(bitmap->mddev->events); |
8258c532 | 416 | if (bitmap->mddev->events < bitmap->events_cleared) |
a0da84f3 NB |
417 | /* rocking back to read-only */ |
418 | bitmap->events_cleared = bitmap->mddev->events; | |
8258c532 N |
419 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); |
420 | sb->state = cpu_to_le32(bitmap->flags); | |
43a70507 N |
421 | /* Just in case these have been changed via sysfs: */ |
422 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
423 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
b2f46e68 | 424 | kunmap_atomic(sb); |
1ec885cd | 425 | write_page(bitmap, bitmap->storage.sb_page, 1); |
32a7627c N |
426 | } |
427 | ||
428 | /* print out the bitmap file superblock */ | |
429 | void bitmap_print_sb(struct bitmap *bitmap) | |
430 | { | |
431 | bitmap_super_t *sb; | |
432 | ||
1ec885cd | 433 | if (!bitmap || !bitmap->storage.sb_page) |
32a7627c | 434 | return; |
1ec885cd | 435 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 436 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
437 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
438 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
439 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
440 | *(__u32 *)(sb->uuid+0), |
441 | *(__u32 *)(sb->uuid+4), | |
442 | *(__u32 *)(sb->uuid+8), | |
443 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 444 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 445 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 446 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 447 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
448 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
449 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
450 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
451 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
452 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 453 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
b2f46e68 | 454 | kunmap_atomic(sb); |
32a7627c N |
455 | } |
456 | ||
9c81075f JB |
457 | /* |
458 | * bitmap_new_disk_sb | |
459 | * @bitmap | |
460 | * | |
461 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb | |
462 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. | |
463 | * This function verifies 'bitmap_info' and populates the on-disk bitmap | |
464 | * structure, which is to be written to disk. | |
465 | * | |
466 | * Returns: 0 on success, -Exxx on error | |
467 | */ | |
468 | static int bitmap_new_disk_sb(struct bitmap *bitmap) | |
469 | { | |
470 | bitmap_super_t *sb; | |
471 | unsigned long chunksize, daemon_sleep, write_behind; | |
472 | int err = -EINVAL; | |
473 | ||
1ec885cd N |
474 | bitmap->storage.sb_page = alloc_page(GFP_KERNEL); |
475 | if (IS_ERR(bitmap->storage.sb_page)) { | |
476 | err = PTR_ERR(bitmap->storage.sb_page); | |
477 | bitmap->storage.sb_page = NULL; | |
9c81075f JB |
478 | return err; |
479 | } | |
1ec885cd | 480 | bitmap->storage.sb_page->index = 0; |
9c81075f | 481 | |
1ec885cd | 482 | sb = kmap_atomic(bitmap->storage.sb_page); |
9c81075f JB |
483 | |
484 | sb->magic = cpu_to_le32(BITMAP_MAGIC); | |
485 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); | |
486 | ||
487 | chunksize = bitmap->mddev->bitmap_info.chunksize; | |
488 | BUG_ON(!chunksize); | |
489 | if (!is_power_of_2(chunksize)) { | |
b2f46e68 | 490 | kunmap_atomic(sb); |
9c81075f JB |
491 | printk(KERN_ERR "bitmap chunksize not a power of 2\n"); |
492 | return -EINVAL; | |
493 | } | |
494 | sb->chunksize = cpu_to_le32(chunksize); | |
495 | ||
496 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; | |
497 | if (!daemon_sleep || | |
498 | (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { | |
499 | printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n"); | |
500 | daemon_sleep = 5 * HZ; | |
501 | } | |
502 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); | |
503 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
504 | ||
505 | /* | |
506 | * FIXME: write_behind for RAID1. If not specified, what | |
507 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. | |
508 | */ | |
509 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; | |
510 | if (write_behind > COUNTER_MAX) | |
511 | write_behind = COUNTER_MAX / 2; | |
512 | sb->write_behind = cpu_to_le32(write_behind); | |
513 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
514 | ||
515 | /* keep the array size field of the bitmap superblock up to date */ | |
516 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
517 | ||
518 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); | |
519 | ||
520 | bitmap->flags |= BITMAP_STALE; | |
521 | sb->state |= cpu_to_le32(BITMAP_STALE); | |
522 | bitmap->events_cleared = bitmap->mddev->events; | |
523 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
524 | ||
b2f46e68 | 525 | kunmap_atomic(sb); |
9c81075f JB |
526 | |
527 | return 0; | |
528 | } | |
529 | ||
32a7627c N |
530 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
531 | static int bitmap_read_sb(struct bitmap *bitmap) | |
532 | { | |
533 | char *reason = NULL; | |
534 | bitmap_super_t *sb; | |
4b6d287f | 535 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c N |
536 | unsigned long long events; |
537 | int err = -EINVAL; | |
27581e5a | 538 | struct page *sb_page; |
32a7627c | 539 | |
1ec885cd | 540 | if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { |
ef99bf48 N |
541 | chunksize = 128 * 1024 * 1024; |
542 | daemon_sleep = 5 * HZ; | |
543 | write_behind = 0; | |
544 | bitmap->flags = BITMAP_STALE; | |
545 | err = 0; | |
546 | goto out_no_sb; | |
547 | } | |
32a7627c | 548 | /* page 0 is the superblock, read it... */ |
27581e5a N |
549 | sb_page = alloc_page(GFP_KERNEL); |
550 | if (!sb_page) | |
551 | return -ENOMEM; | |
1ec885cd | 552 | bitmap->storage.sb_page = sb_page; |
27581e5a | 553 | |
1ec885cd N |
554 | if (bitmap->storage.file) { |
555 | loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); | |
f49d5e62 N |
556 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; |
557 | ||
1ec885cd | 558 | err = read_page(bitmap->storage.file, 0, |
27581e5a | 559 | bitmap, bytes, sb_page); |
f49d5e62 | 560 | } else { |
27581e5a N |
561 | err = read_sb_page(bitmap->mddev, |
562 | bitmap->mddev->bitmap_info.offset, | |
563 | sb_page, | |
564 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 565 | } |
27581e5a | 566 | if (err) |
32a7627c | 567 | return err; |
32a7627c | 568 | |
27581e5a | 569 | sb = kmap_atomic(sb_page); |
32a7627c | 570 | |
32a7627c | 571 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 572 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 573 | write_behind = le32_to_cpu(sb->write_behind); |
32a7627c N |
574 | |
575 | /* verify that the bitmap-specific fields are valid */ | |
576 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
577 | reason = "bad magic"; | |
bd926c63 N |
578 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
579 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 580 | reason = "unrecognized superblock version"; |
1187cf0a | 581 | else if (chunksize < 512) |
7dd5d34c | 582 | reason = "bitmap chunksize too small"; |
d744540c | 583 | else if (!is_power_of_2(chunksize)) |
32a7627c | 584 | reason = "bitmap chunksize not a power of 2"; |
1b04be96 | 585 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 586 | reason = "daemon sleep period out of range"; |
4b6d287f N |
587 | else if (write_behind > COUNTER_MAX) |
588 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
589 | if (reason) { |
590 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
591 | bmname(bitmap), reason); | |
592 | goto out; | |
593 | } | |
594 | ||
595 | /* keep the array size field of the bitmap superblock up to date */ | |
596 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
597 | ||
278c1ca2 N |
598 | if (bitmap->mddev->persistent) { |
599 | /* | |
600 | * We have a persistent array superblock, so compare the | |
601 | * bitmap's UUID and event counter to the mddev's | |
602 | */ | |
603 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
604 | printk(KERN_INFO | |
605 | "%s: bitmap superblock UUID mismatch\n", | |
606 | bmname(bitmap)); | |
607 | goto out; | |
608 | } | |
609 | events = le64_to_cpu(sb->events); | |
610 | if (events < bitmap->mddev->events) { | |
611 | printk(KERN_INFO | |
612 | "%s: bitmap file is out of date (%llu < %llu) " | |
613 | "-- forcing full recovery\n", | |
614 | bmname(bitmap), events, | |
615 | (unsigned long long) bitmap->mddev->events); | |
616 | sb->state |= cpu_to_le32(BITMAP_STALE); | |
617 | } | |
32a7627c | 618 | } |
278c1ca2 | 619 | |
32a7627c | 620 | /* assign fields using values from superblock */ |
4f2e639a | 621 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 N |
622 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
623 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
32a7627c N |
624 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
625 | err = 0; | |
626 | out: | |
b2f46e68 | 627 | kunmap_atomic(sb); |
ef99bf48 N |
628 | out_no_sb: |
629 | if (bitmap->flags & BITMAP_STALE) | |
630 | bitmap->events_cleared = bitmap->mddev->events; | |
631 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
632 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
633 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
32a7627c N |
634 | if (err) |
635 | bitmap_print_sb(bitmap); | |
636 | return err; | |
637 | } | |
638 | ||
639 | enum bitmap_mask_op { | |
640 | MASK_SET, | |
641 | MASK_UNSET | |
642 | }; | |
643 | ||
4ad13663 N |
644 | /* record the state of the bitmap in the superblock. Return the old value */ |
645 | static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits, | |
646 | enum bitmap_mask_op op) | |
32a7627c N |
647 | { |
648 | bitmap_super_t *sb; | |
4ad13663 | 649 | int old; |
32a7627c | 650 | |
1ec885cd | 651 | if (!bitmap->storage.sb_page) /* can't set the state */ |
4ad13663 | 652 | return 0; |
1ec885cd | 653 | sb = kmap_atomic(bitmap->storage.sb_page); |
4ad13663 | 654 | old = le32_to_cpu(sb->state) & bits; |
32a7627c | 655 | switch (op) { |
ac2f40be N |
656 | case MASK_SET: |
657 | sb->state |= cpu_to_le32(bits); | |
8258c532 | 658 | bitmap->flags |= bits; |
ac2f40be N |
659 | break; |
660 | case MASK_UNSET: | |
661 | sb->state &= cpu_to_le32(~bits); | |
8258c532 | 662 | bitmap->flags &= ~bits; |
ac2f40be N |
663 | break; |
664 | default: | |
665 | BUG(); | |
32a7627c | 666 | } |
b2f46e68 | 667 | kunmap_atomic(sb); |
4ad13663 | 668 | return old; |
32a7627c N |
669 | } |
670 | ||
671 | /* | |
672 | * general bitmap file operations | |
673 | */ | |
674 | ||
ece5cff0 N |
675 | /* |
676 | * on-disk bitmap: | |
677 | * | |
678 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
679 | * file a page at a time. There's a superblock at the start of the file. | |
680 | */ | |
32a7627c | 681 | /* calculate the index of the page that contains this bit */ |
1ec885cd N |
682 | static inline unsigned long file_page_index(struct bitmap_storage *store, |
683 | unsigned long chunk) | |
32a7627c | 684 | { |
1ec885cd | 685 | if (store->sb_page) |
ece5cff0 N |
686 | chunk += sizeof(bitmap_super_t) << 3; |
687 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
688 | } |
689 | ||
690 | /* calculate the (bit) offset of this bit within a page */ | |
1ec885cd N |
691 | static inline unsigned long file_page_offset(struct bitmap_storage *store, |
692 | unsigned long chunk) | |
32a7627c | 693 | { |
1ec885cd | 694 | if (store->sb_page) |
ece5cff0 N |
695 | chunk += sizeof(bitmap_super_t) << 3; |
696 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
697 | } |
698 | ||
699 | /* | |
700 | * return a pointer to the page in the filemap that contains the given bit | |
701 | * | |
702 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
703 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
704 | * 0 or page 1 | |
705 | */ | |
1ec885cd | 706 | static inline struct page *filemap_get_page(struct bitmap_storage *store, |
3520fa4d | 707 | unsigned long chunk) |
32a7627c | 708 | { |
1ec885cd | 709 | if (file_page_index(store, chunk) >= store->file_pages) |
ac2f40be | 710 | return NULL; |
1ec885cd N |
711 | return store->filemap[file_page_index(store, chunk) |
712 | - file_page_index(store, 0)]; | |
32a7627c N |
713 | } |
714 | ||
32a7627c N |
715 | static void bitmap_file_unmap(struct bitmap *bitmap) |
716 | { | |
717 | struct page **map, *sb_page; | |
718 | unsigned long *attr; | |
719 | int pages; | |
720 | unsigned long flags; | |
1ec885cd | 721 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c N |
722 | |
723 | spin_lock_irqsave(&bitmap->lock, flags); | |
1ec885cd N |
724 | map = store->filemap; |
725 | store->filemap = NULL; | |
726 | attr = store->filemap_attr; | |
727 | store->filemap_attr = NULL; | |
728 | pages = store->file_pages; | |
729 | store->file_pages = 0; | |
730 | sb_page = store->sb_page; | |
731 | store->sb_page = NULL; | |
32a7627c N |
732 | spin_unlock_irqrestore(&bitmap->lock, flags); |
733 | ||
734 | while (pages--) | |
ece5cff0 | 735 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 736 | free_buffers(map[pages]); |
32a7627c N |
737 | kfree(map); |
738 | kfree(attr); | |
739 | ||
d785a06a N |
740 | if (sb_page) |
741 | free_buffers(sb_page); | |
32a7627c N |
742 | } |
743 | ||
744 | static void bitmap_file_put(struct bitmap *bitmap) | |
745 | { | |
746 | struct file *file; | |
32a7627c N |
747 | unsigned long flags; |
748 | ||
749 | spin_lock_irqsave(&bitmap->lock, flags); | |
1ec885cd N |
750 | file = bitmap->storage.file; |
751 | bitmap->storage.file = NULL; | |
32a7627c N |
752 | spin_unlock_irqrestore(&bitmap->lock, flags); |
753 | ||
d785a06a N |
754 | if (file) |
755 | wait_event(bitmap->write_wait, | |
756 | atomic_read(&bitmap->pending_writes)==0); | |
32a7627c N |
757 | bitmap_file_unmap(bitmap); |
758 | ||
d785a06a | 759 | if (file) { |
c649bb9c | 760 | struct inode *inode = file->f_path.dentry->d_inode; |
fc0ecff6 | 761 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 762 | fput(file); |
d785a06a | 763 | } |
32a7627c N |
764 | } |
765 | ||
32a7627c N |
766 | /* |
767 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
768 | * then it is no longer reliable, so we stop using it and we mark the file | |
769 | * as failed in the superblock | |
770 | */ | |
771 | static void bitmap_file_kick(struct bitmap *bitmap) | |
772 | { | |
773 | char *path, *ptr = NULL; | |
774 | ||
4ad13663 N |
775 | if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) { |
776 | bitmap_update_sb(bitmap); | |
32a7627c | 777 | |
1ec885cd | 778 | if (bitmap->storage.file) { |
4ad13663 N |
779 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); |
780 | if (path) | |
1ec885cd N |
781 | ptr = d_path(&bitmap->storage.file->f_path, |
782 | path, PAGE_SIZE); | |
6bcfd601 | 783 | |
4ad13663 N |
784 | printk(KERN_ALERT |
785 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 786 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 787 | |
4ad13663 N |
788 | kfree(path); |
789 | } else | |
790 | printk(KERN_ALERT | |
791 | "%s: disabling internal bitmap due to errors\n", | |
792 | bmname(bitmap)); | |
a654b9d8 | 793 | } |
32a7627c N |
794 | |
795 | bitmap_file_put(bitmap); | |
796 | ||
797 | return; | |
798 | } | |
799 | ||
800 | enum bitmap_page_attr { | |
ac2f40be | 801 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
5a537df4 N |
802 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
803 | * i.e. counter is 1 or 2. */ | |
ac2f40be | 804 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
32a7627c N |
805 | }; |
806 | ||
d189122d N |
807 | static inline void set_page_attr(struct bitmap *bitmap, int pnum, |
808 | enum bitmap_page_attr attr) | |
32a7627c | 809 | { |
1ec885cd | 810 | __set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
811 | } |
812 | ||
d189122d N |
813 | static inline void clear_page_attr(struct bitmap *bitmap, int pnum, |
814 | enum bitmap_page_attr attr) | |
32a7627c | 815 | { |
1ec885cd | 816 | __clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
817 | } |
818 | ||
d189122d | 819 | static inline unsigned long test_page_attr(struct bitmap *bitmap, int pnum, |
ec7a3197 | 820 | enum bitmap_page_attr attr) |
32a7627c | 821 | { |
1ec885cd | 822 | return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
823 | } |
824 | ||
825 | /* | |
826 | * bitmap_file_set_bit -- called before performing a write to the md device | |
827 | * to set (and eventually sync) a particular bit in the bitmap file | |
828 | * | |
829 | * we set the bit immediately, then we record the page number so that | |
830 | * when an unplug occurs, we can flush the dirty pages out to disk | |
831 | */ | |
832 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
833 | { | |
834 | unsigned long bit; | |
3520fa4d | 835 | struct page *page; |
32a7627c | 836 | void *kaddr; |
61a0d80c | 837 | unsigned long chunk = block >> bitmap->chunkshift; |
32a7627c | 838 | |
1ec885cd | 839 | page = filemap_get_page(&bitmap->storage, chunk); |
3520fa4d JB |
840 | if (!page) |
841 | return; | |
1ec885cd | 842 | bit = file_page_offset(&bitmap->storage, chunk); |
32a7627c | 843 | |
3520fa4d | 844 | /* set the bit */ |
b2f46e68 | 845 | kaddr = kmap_atomic(page); |
3520fa4d JB |
846 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
847 | set_bit(bit, kaddr); | |
848 | else | |
849 | __set_bit_le(bit, kaddr); | |
b2f46e68 | 850 | kunmap_atomic(kaddr); |
36a4e1fe | 851 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
32a7627c | 852 | /* record page number so it gets flushed to disk when unplug occurs */ |
d189122d | 853 | set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY); |
32a7627c N |
854 | } |
855 | ||
ef99bf48 N |
856 | static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) |
857 | { | |
858 | unsigned long bit; | |
859 | struct page *page; | |
860 | void *paddr; | |
861 | unsigned long chunk = block >> bitmap->chunkshift; | |
862 | ||
1ec885cd | 863 | page = filemap_get_page(&bitmap->storage, chunk); |
ef99bf48 N |
864 | if (!page) |
865 | return; | |
1ec885cd | 866 | bit = file_page_offset(&bitmap->storage, chunk); |
ef99bf48 N |
867 | paddr = kmap_atomic(page); |
868 | if (bitmap->flags & BITMAP_HOSTENDIAN) | |
869 | clear_bit(bit, paddr); | |
870 | else | |
871 | __clear_bit_le(bit, paddr); | |
872 | kunmap_atomic(paddr); | |
d189122d N |
873 | if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) { |
874 | set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING); | |
ef99bf48 N |
875 | bitmap->allclean = 0; |
876 | } | |
877 | } | |
878 | ||
32a7627c N |
879 | /* this gets called when the md device is ready to unplug its underlying |
880 | * (slave) device queues -- before we let any writes go down, we need to | |
881 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 882 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 883 | { |
ec7a3197 N |
884 | unsigned long i, flags; |
885 | int dirty, need_write; | |
32a7627c N |
886 | int wait = 0; |
887 | ||
1ec885cd | 888 | if (!bitmap || !bitmap->storage.filemap) |
4ad13663 | 889 | return; |
32a7627c N |
890 | |
891 | /* look at each page to see if there are any set bits that need to be | |
892 | * flushed out to disk */ | |
1ec885cd | 893 | for (i = 0; i < bitmap->storage.file_pages; i++) { |
32a7627c | 894 | spin_lock_irqsave(&bitmap->lock, flags); |
1ec885cd | 895 | if (!bitmap->storage.filemap) { |
32a7627c | 896 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 897 | return; |
32a7627c | 898 | } |
d189122d N |
899 | dirty = test_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
900 | need_write = test_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE); | |
901 | clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); | |
902 | clear_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE); | |
bf07bb7d | 903 | if (dirty || need_write) |
d189122d | 904 | clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); |
ec7a3197 | 905 | if (dirty) |
32a7627c N |
906 | wait = 1; |
907 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
908 | ||
ac2f40be | 909 | if (dirty || need_write) |
1ec885cd | 910 | write_page(bitmap, bitmap->storage.filemap[i], 0); |
32a7627c N |
911 | } |
912 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
1ec885cd | 913 | if (bitmap->storage.file) |
d785a06a N |
914 | wait_event(bitmap->write_wait, |
915 | atomic_read(&bitmap->pending_writes)==0); | |
0b79ccf0 | 916 | else |
a9701a30 | 917 | md_super_wait(bitmap->mddev); |
32a7627c | 918 | } |
d785a06a N |
919 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
920 | bitmap_file_kick(bitmap); | |
32a7627c | 921 | } |
ac2f40be | 922 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 923 | |
6a07997f | 924 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
925 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
926 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
927 | * memory mapping of the bitmap file | |
928 | * Special cases: | |
929 | * if there's no bitmap file, or if the bitmap file had been | |
930 | * previously kicked from the array, we mark all the bits as | |
931 | * 1's in order to cause a full resync. | |
6a07997f N |
932 | * |
933 | * We ignore all bits for sectors that end earlier than 'start'. | |
934 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 935 | */ |
6a07997f | 936 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
937 | { |
938 | unsigned long i, chunks, index, oldindex, bit; | |
27581e5a N |
939 | int pnum; |
940 | struct page *page = NULL; | |
32a7627c N |
941 | unsigned long num_pages, bit_cnt = 0; |
942 | struct file *file; | |
d785a06a | 943 | unsigned long bytes, offset; |
32a7627c N |
944 | int outofdate; |
945 | int ret = -ENOSPC; | |
ea03aff9 | 946 | void *paddr; |
1ec885cd | 947 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c N |
948 | |
949 | chunks = bitmap->chunks; | |
1ec885cd | 950 | file = store->file; |
32a7627c | 951 | |
ef99bf48 N |
952 | if (!file && !bitmap->mddev->bitmap_info.offset) { |
953 | /* No permanent bitmap - fill with '1s'. */ | |
1ec885cd N |
954 | store->filemap = NULL; |
955 | store->file_pages = 0; | |
ef99bf48 N |
956 | for (i = 0; i < chunks ; i++) { |
957 | /* if the disk bit is set, set the memory bit */ | |
958 | int needed = ((sector_t)(i+1) << (bitmap->chunkshift) | |
959 | >= start); | |
960 | bitmap_set_memory_bits(bitmap, | |
961 | (sector_t)i << bitmap->chunkshift, | |
962 | needed); | |
963 | } | |
964 | return 0; | |
965 | } | |
32a7627c | 966 | |
32a7627c | 967 | outofdate = bitmap->flags & BITMAP_STALE; |
32a7627c N |
968 | if (outofdate) |
969 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
970 | "recovery\n", bmname(bitmap)); | |
971 | ||
e384e585 | 972 | bytes = DIV_ROUND_UP(bitmap->chunks, 8); |
ece5cff0 N |
973 | if (!bitmap->mddev->bitmap_info.external) |
974 | bytes += sizeof(bitmap_super_t); | |
bc7f77de | 975 | |
e384e585 | 976 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); |
bc7f77de | 977 | |
ece5cff0 | 978 | if (file && i_size_read(file->f_mapping->host) < bytes) { |
32a7627c N |
979 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
980 | bmname(bitmap), | |
981 | (unsigned long) i_size_read(file->f_mapping->host), | |
ece5cff0 | 982 | bytes); |
4ad13663 | 983 | goto err; |
32a7627c | 984 | } |
bc7f77de N |
985 | |
986 | ret = -ENOMEM; | |
987 | ||
1ec885cd N |
988 | store->filemap = kmalloc(sizeof(struct page *) |
989 | * num_pages, GFP_KERNEL); | |
990 | if (!store->filemap) | |
4ad13663 | 991 | goto err; |
32a7627c | 992 | |
27581e5a N |
993 | pnum = 0; |
994 | offset = 0; | |
1ec885cd N |
995 | if (store->sb_page) { |
996 | store->filemap[0] = store->sb_page; | |
27581e5a N |
997 | pnum = 1; |
998 | offset = sizeof(bitmap_super_t); | |
999 | } | |
1000 | for ( ; pnum < num_pages; pnum++) { | |
1ec885cd N |
1001 | store->filemap[pnum] = alloc_page(GFP_KERNEL); |
1002 | if (!store->filemap[pnum]) { | |
1003 | store->file_pages = pnum; | |
27581e5a N |
1004 | goto err; |
1005 | } | |
1006 | } | |
1ec885cd | 1007 | store->file_pages = pnum; |
27581e5a | 1008 | |
e16b68b6 | 1009 | /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */ |
1ec885cd | 1010 | store->filemap_attr = kzalloc( |
ac2f40be | 1011 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), |
e16b68b6 | 1012 | GFP_KERNEL); |
1ec885cd | 1013 | if (!store->filemap_attr) |
4ad13663 | 1014 | goto err; |
32a7627c | 1015 | |
32a7627c N |
1016 | oldindex = ~0L; |
1017 | ||
1018 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 1019 | int b; |
1ec885cd N |
1020 | index = file_page_index(&bitmap->storage, i); |
1021 | bit = file_page_offset(&bitmap->storage, i); | |
32a7627c | 1022 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 1023 | int count; |
32a7627c | 1024 | /* unmap the old page, we're done with it */ |
d785a06a | 1025 | if (index == num_pages-1) |
ece5cff0 | 1026 | count = bytes - index * PAGE_SIZE; |
d785a06a N |
1027 | else |
1028 | count = PAGE_SIZE; | |
1ec885cd | 1029 | page = store->filemap[index]; |
27581e5a N |
1030 | if (file) |
1031 | ret = read_page(file, index, bitmap, | |
1032 | count, page); | |
1033 | else | |
1034 | ret = read_sb_page( | |
1035 | bitmap->mddev, | |
1036 | bitmap->mddev->bitmap_info.offset, | |
1037 | page, | |
1038 | index, count); | |
1039 | ||
1040 | if (ret) | |
4ad13663 | 1041 | goto err; |
a654b9d8 | 1042 | |
32a7627c | 1043 | oldindex = index; |
32a7627c | 1044 | |
1ec885cd | 1045 | store->last_page_size = count; |
b74fd282 | 1046 | |
32a7627c N |
1047 | if (outofdate) { |
1048 | /* | |
1049 | * if bitmap is out of date, dirty the | |
ac2f40be | 1050 | * whole page and write it out |
32a7627c | 1051 | */ |
b2f46e68 | 1052 | paddr = kmap_atomic(page); |
ea03aff9 | 1053 | memset(paddr + offset, 0xff, |
6a07997f | 1054 | PAGE_SIZE - offset); |
b2f46e68 | 1055 | kunmap_atomic(paddr); |
4ad13663 N |
1056 | write_page(bitmap, page, 1); |
1057 | ||
1058 | ret = -EIO; | |
b74fd282 | 1059 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
4ad13663 | 1060 | goto err; |
32a7627c | 1061 | } |
32a7627c | 1062 | } |
b2f46e68 | 1063 | paddr = kmap_atomic(page); |
bd926c63 | 1064 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1065 | b = test_bit(bit, paddr); |
bd926c63 | 1066 | else |
6b33aff3 | 1067 | b = test_bit_le(bit, paddr); |
b2f46e68 | 1068 | kunmap_atomic(paddr); |
bd926c63 | 1069 | if (b) { |
32a7627c | 1070 | /* if the disk bit is set, set the memory bit */ |
61a0d80c | 1071 | int needed = ((sector_t)(i+1) << bitmap->chunkshift |
db305e50 N |
1072 | >= start); |
1073 | bitmap_set_memory_bits(bitmap, | |
61a0d80c | 1074 | (sector_t)i << bitmap->chunkshift, |
db305e50 | 1075 | needed); |
32a7627c N |
1076 | bit_cnt++; |
1077 | } | |
27581e5a | 1078 | offset = 0; |
32a7627c N |
1079 | } |
1080 | ||
32a7627c | 1081 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
9c81075f | 1082 | "read %lu/%lu pages, set %lu of %lu bits\n", |
1ec885cd N |
1083 | bmname(bitmap), store->file_pages, |
1084 | num_pages, bit_cnt, chunks); | |
4ad13663 N |
1085 | |
1086 | return 0; | |
32a7627c | 1087 | |
4ad13663 N |
1088 | err: |
1089 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1090 | bmname(bitmap), ret); | |
32a7627c N |
1091 | return ret; |
1092 | } | |
1093 | ||
a654b9d8 N |
1094 | void bitmap_write_all(struct bitmap *bitmap) |
1095 | { | |
1096 | /* We don't actually write all bitmap blocks here, | |
1097 | * just flag them as needing to be written | |
1098 | */ | |
ec7a3197 | 1099 | int i; |
a654b9d8 | 1100 | |
1ec885cd | 1101 | if (!bitmap || !bitmap->storage.filemap) |
ef99bf48 | 1102 | return; |
1ec885cd | 1103 | if (bitmap->storage.file) |
ef99bf48 N |
1104 | /* Only one copy, so nothing needed */ |
1105 | return; | |
1106 | ||
7c8f4247 | 1107 | spin_lock_irq(&bitmap->lock); |
1ec885cd | 1108 | for (i = 0; i < bitmap->storage.file_pages; i++) |
d189122d | 1109 | set_page_attr(bitmap, i, |
ec7a3197 | 1110 | BITMAP_PAGE_NEEDWRITE); |
2585f3ef | 1111 | bitmap->allclean = 0; |
7c8f4247 | 1112 | spin_unlock_irq(&bitmap->lock); |
a654b9d8 N |
1113 | } |
1114 | ||
32a7627c N |
1115 | static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) |
1116 | { | |
61a0d80c | 1117 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1118 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1119 | bitmap->bp[page].count += inc; | |
32a7627c N |
1120 | bitmap_checkfree(bitmap, page); |
1121 | } | |
bf07bb7d N |
1122 | |
1123 | static void bitmap_set_pending(struct bitmap *bitmap, sector_t offset) | |
1124 | { | |
1125 | sector_t chunk = offset >> bitmap->chunkshift; | |
1126 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1127 | struct bitmap_page *bp = &bitmap->bp[page]; | |
1128 | ||
1129 | if (!bp->pending) | |
1130 | bp->pending = 1; | |
1131 | } | |
1132 | ||
32a7627c | 1133 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
57dab0bd | 1134 | sector_t offset, sector_t *blocks, |
32a7627c N |
1135 | int create); |
1136 | ||
1137 | /* | |
1138 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1139 | * out to disk | |
1140 | */ | |
1141 | ||
fd01b88c | 1142 | void bitmap_daemon_work(struct mddev *mddev) |
32a7627c | 1143 | { |
aa5cbd10 | 1144 | struct bitmap *bitmap; |
aa3163f8 | 1145 | unsigned long j; |
bf07bb7d | 1146 | unsigned long nextpage; |
32a7627c | 1147 | unsigned long flags; |
57dab0bd | 1148 | sector_t blocks; |
32a7627c | 1149 | |
aa5cbd10 N |
1150 | /* Use a mutex to guard daemon_work against |
1151 | * bitmap_destroy. | |
1152 | */ | |
c3d9714e | 1153 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1154 | bitmap = mddev->bitmap; |
1155 | if (bitmap == NULL) { | |
c3d9714e | 1156 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1157 | return; |
aa5cbd10 | 1158 | } |
42a04b50 | 1159 | if (time_before(jiffies, bitmap->daemon_lastrun |
2e61ebbc | 1160 | + mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1161 | goto done; |
1162 | ||
32a7627c | 1163 | bitmap->daemon_lastrun = jiffies; |
8311c29d | 1164 | if (bitmap->allclean) { |
2e61ebbc | 1165 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
aa5cbd10 | 1166 | goto done; |
8311c29d N |
1167 | } |
1168 | bitmap->allclean = 1; | |
32a7627c | 1169 | |
bf07bb7d N |
1170 | /* Any file-page which is PENDING now needs to be written. |
1171 | * So set NEEDWRITE now, then after we make any last-minute changes | |
1172 | * we will write it. | |
1173 | */ | |
be512691 | 1174 | spin_lock_irqsave(&bitmap->lock, flags); |
1ec885cd | 1175 | for (j = 0; j < bitmap->storage.file_pages; j++) |
d189122d | 1176 | if (test_page_attr(bitmap, j, |
bf07bb7d | 1177 | BITMAP_PAGE_PENDING)) { |
d189122d | 1178 | set_page_attr(bitmap, j, |
bf07bb7d | 1179 | BITMAP_PAGE_NEEDWRITE); |
d189122d | 1180 | clear_page_attr(bitmap, j, |
bf07bb7d N |
1181 | BITMAP_PAGE_PENDING); |
1182 | } | |
1183 | ||
1184 | if (bitmap->need_sync && | |
1185 | mddev->bitmap_info.external == 0) { | |
1186 | /* Arrange for superblock update as well as | |
1187 | * other changes */ | |
1188 | bitmap_super_t *sb; | |
1189 | bitmap->need_sync = 0; | |
1ec885cd N |
1190 | if (bitmap->storage.filemap) { |
1191 | sb = kmap_atomic(bitmap->storage.sb_page); | |
ef99bf48 N |
1192 | sb->events_cleared = |
1193 | cpu_to_le64(bitmap->events_cleared); | |
1194 | kunmap_atomic(sb); | |
d189122d | 1195 | set_page_attr(bitmap, 0, |
ef99bf48 N |
1196 | BITMAP_PAGE_NEEDWRITE); |
1197 | } | |
bf07bb7d N |
1198 | } |
1199 | /* Now look at the bitmap counters and if any are '2' or '1', | |
1200 | * decrement and handle accordingly. | |
1201 | */ | |
1202 | nextpage = 0; | |
32a7627c N |
1203 | for (j = 0; j < bitmap->chunks; j++) { |
1204 | bitmap_counter_t *bmc; | |
ef99bf48 | 1205 | sector_t block = (sector_t)j << bitmap->chunkshift; |
3520fa4d | 1206 | |
bf07bb7d N |
1207 | if (j == nextpage) { |
1208 | nextpage += PAGE_COUNTER_RATIO; | |
1209 | if (!bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending) { | |
1210 | j |= PAGE_COUNTER_MASK; | |
aa3163f8 N |
1211 | continue; |
1212 | } | |
bf07bb7d | 1213 | bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; |
32a7627c | 1214 | } |
db305e50 | 1215 | bmc = bitmap_get_counter(bitmap, |
ef99bf48 | 1216 | block, |
db305e50 | 1217 | &blocks, 0); |
bf07bb7d N |
1218 | |
1219 | if (!bmc) { | |
5a537df4 | 1220 | j |= PAGE_COUNTER_MASK; |
bf07bb7d N |
1221 | continue; |
1222 | } | |
1223 | if (*bmc == 1 && !bitmap->need_sync) { | |
1224 | /* We can clear the bit */ | |
bf07bb7d | 1225 | *bmc = 0; |
ef99bf48 N |
1226 | bitmap_count_page(bitmap, block, -1); |
1227 | bitmap_file_clear_bit(bitmap, block); | |
bf07bb7d N |
1228 | } else if (*bmc && *bmc <= 2) { |
1229 | *bmc = 1; | |
ef99bf48 | 1230 | bitmap_set_pending(bitmap, block); |
bf07bb7d | 1231 | bitmap->allclean = 0; |
5a537df4 | 1232 | } |
32a7627c N |
1233 | } |
1234 | ||
bf07bb7d N |
1235 | /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. |
1236 | * DIRTY pages need to be written by bitmap_unplug so it can wait | |
1237 | * for them. | |
1238 | * If we find any DIRTY page we stop there and let bitmap_unplug | |
1239 | * handle all the rest. This is important in the case where | |
1240 | * the first blocking holds the superblock and it has been updated. | |
1241 | * We mustn't write any other blocks before the superblock. | |
1242 | */ | |
1ec885cd | 1243 | for (j = 0; j < bitmap->storage.file_pages; j++) { |
bf07bb7d | 1244 | |
d189122d | 1245 | if (test_page_attr(bitmap, j, |
bf07bb7d N |
1246 | BITMAP_PAGE_DIRTY)) |
1247 | /* bitmap_unplug will handle the rest */ | |
1248 | break; | |
d189122d | 1249 | if (test_page_attr(bitmap, j, |
bf07bb7d | 1250 | BITMAP_PAGE_NEEDWRITE)) { |
d189122d | 1251 | clear_page_attr(bitmap, j, |
bf07bb7d | 1252 | BITMAP_PAGE_NEEDWRITE); |
32a7627c | 1253 | spin_unlock_irqrestore(&bitmap->lock, flags); |
1ec885cd | 1254 | write_page(bitmap, bitmap->storage.filemap[j], 0); |
bf07bb7d | 1255 | spin_lock_irqsave(&bitmap->lock, flags); |
1ec885cd | 1256 | if (!bitmap->storage.filemap) |
bf07bb7d | 1257 | break; |
32a7627c | 1258 | } |
32a7627c | 1259 | } |
bf07bb7d | 1260 | spin_unlock_irqrestore(&bitmap->lock, flags); |
32a7627c | 1261 | |
7be3dfec | 1262 | done: |
8311c29d | 1263 | if (bitmap->allclean == 0) |
2e61ebbc N |
1264 | mddev->thread->timeout = |
1265 | mddev->bitmap_info.daemon_sleep; | |
c3d9714e | 1266 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1267 | } |
1268 | ||
32a7627c | 1269 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
57dab0bd | 1270 | sector_t offset, sector_t *blocks, |
32a7627c | 1271 | int create) |
ee305ace N |
1272 | __releases(bitmap->lock) |
1273 | __acquires(bitmap->lock) | |
32a7627c N |
1274 | { |
1275 | /* If 'create', we might release the lock and reclaim it. | |
1276 | * The lock must have been taken with interrupts enabled. | |
1277 | * If !create, we don't release the lock. | |
1278 | */ | |
61a0d80c | 1279 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1280 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1281 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1282 | sector_t csize; | |
ef425673 | 1283 | int err; |
32a7627c | 1284 | |
ef425673 N |
1285 | err = bitmap_checkpage(bitmap, page, create); |
1286 | ||
1287 | if (bitmap->bp[page].hijacked || | |
1288 | bitmap->bp[page].map == NULL) | |
61a0d80c | 1289 | csize = ((sector_t)1) << (bitmap->chunkshift + |
ef425673 N |
1290 | PAGE_COUNTER_SHIFT - 1); |
1291 | else | |
61a0d80c | 1292 | csize = ((sector_t)1) << bitmap->chunkshift; |
ef425673 N |
1293 | *blocks = csize - (offset & (csize - 1)); |
1294 | ||
1295 | if (err < 0) | |
32a7627c | 1296 | return NULL; |
ef425673 | 1297 | |
32a7627c N |
1298 | /* now locked ... */ |
1299 | ||
1300 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1301 | /* should we use the first or second counter field | |
1302 | * of the hijacked pointer? */ | |
1303 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1304 | return &((bitmap_counter_t *) |
1305 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1306 | } else /* page is allocated */ |
32a7627c N |
1307 | return (bitmap_counter_t *) |
1308 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1309 | } |
1310 | ||
4b6d287f | 1311 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1312 | { |
ac2f40be N |
1313 | if (!bitmap) |
1314 | return 0; | |
4b6d287f N |
1315 | |
1316 | if (behind) { | |
696fcd53 | 1317 | int bw; |
4b6d287f | 1318 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1319 | bw = atomic_read(&bitmap->behind_writes); |
1320 | if (bw > bitmap->behind_writes_used) | |
1321 | bitmap->behind_writes_used = bw; | |
1322 | ||
36a4e1fe N |
1323 | pr_debug("inc write-behind count %d/%lu\n", |
1324 | bw, bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1325 | } |
1326 | ||
32a7627c | 1327 | while (sectors) { |
57dab0bd | 1328 | sector_t blocks; |
32a7627c N |
1329 | bitmap_counter_t *bmc; |
1330 | ||
1331 | spin_lock_irq(&bitmap->lock); | |
1332 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); | |
1333 | if (!bmc) { | |
1334 | spin_unlock_irq(&bitmap->lock); | |
1335 | return 0; | |
1336 | } | |
1337 | ||
27d5ea04 | 1338 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
da6e1a32 NB |
1339 | DEFINE_WAIT(__wait); |
1340 | /* note that it is safe to do the prepare_to_wait | |
1341 | * after the test as long as we do it before dropping | |
1342 | * the spinlock. | |
1343 | */ | |
1344 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1345 | TASK_UNINTERRUPTIBLE); | |
1346 | spin_unlock_irq(&bitmap->lock); | |
7eaceacc | 1347 | io_schedule(); |
da6e1a32 NB |
1348 | finish_wait(&bitmap->overflow_wait, &__wait); |
1349 | continue; | |
1350 | } | |
1351 | ||
ac2f40be | 1352 | switch (*bmc) { |
32a7627c N |
1353 | case 0: |
1354 | bitmap_file_set_bit(bitmap, offset); | |
ac2f40be | 1355 | bitmap_count_page(bitmap, offset, 1); |
32a7627c N |
1356 | /* fall through */ |
1357 | case 1: | |
1358 | *bmc = 2; | |
1359 | } | |
da6e1a32 | 1360 | |
32a7627c N |
1361 | (*bmc)++; |
1362 | ||
1363 | spin_unlock_irq(&bitmap->lock); | |
1364 | ||
1365 | offset += blocks; | |
1366 | if (sectors > blocks) | |
1367 | sectors -= blocks; | |
ac2f40be N |
1368 | else |
1369 | sectors = 0; | |
32a7627c N |
1370 | } |
1371 | return 0; | |
1372 | } | |
ac2f40be | 1373 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1374 | |
1375 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1376 | int success, int behind) |
32a7627c | 1377 | { |
ac2f40be N |
1378 | if (!bitmap) |
1379 | return; | |
4b6d287f | 1380 | if (behind) { |
e555190d N |
1381 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1382 | wake_up(&bitmap->behind_wait); | |
36a4e1fe N |
1383 | pr_debug("dec write-behind count %d/%lu\n", |
1384 | atomic_read(&bitmap->behind_writes), | |
1385 | bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1386 | } |
1387 | ||
32a7627c | 1388 | while (sectors) { |
57dab0bd | 1389 | sector_t blocks; |
32a7627c N |
1390 | unsigned long flags; |
1391 | bitmap_counter_t *bmc; | |
1392 | ||
1393 | spin_lock_irqsave(&bitmap->lock, flags); | |
1394 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); | |
1395 | if (!bmc) { | |
1396 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1397 | return; | |
1398 | } | |
1399 | ||
961902c0 | 1400 | if (success && !bitmap->mddev->degraded && |
a0da84f3 NB |
1401 | bitmap->events_cleared < bitmap->mddev->events) { |
1402 | bitmap->events_cleared = bitmap->mddev->events; | |
1403 | bitmap->need_sync = 1; | |
5ff5afff | 1404 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1405 | } |
1406 | ||
27d5ea04 | 1407 | if (!success && !NEEDED(*bmc)) |
32a7627c N |
1408 | *bmc |= NEEDED_MASK; |
1409 | ||
27d5ea04 | 1410 | if (COUNTER(*bmc) == COUNTER_MAX) |
da6e1a32 NB |
1411 | wake_up(&bitmap->overflow_wait); |
1412 | ||
32a7627c | 1413 | (*bmc)--; |
2585f3ef | 1414 | if (*bmc <= 2) { |
bf07bb7d | 1415 | bitmap_set_pending(bitmap, offset); |
2585f3ef N |
1416 | bitmap->allclean = 0; |
1417 | } | |
32a7627c N |
1418 | spin_unlock_irqrestore(&bitmap->lock, flags); |
1419 | offset += blocks; | |
1420 | if (sectors > blocks) | |
1421 | sectors -= blocks; | |
ac2f40be N |
1422 | else |
1423 | sectors = 0; | |
32a7627c N |
1424 | } |
1425 | } | |
ac2f40be | 1426 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1427 | |
57dab0bd | 1428 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1429 | int degraded) |
32a7627c N |
1430 | { |
1431 | bitmap_counter_t *bmc; | |
1432 | int rv; | |
1433 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1434 | *blocks = 1024; | |
1435 | return 1; /* always resync if no bitmap */ | |
1436 | } | |
1437 | spin_lock_irq(&bitmap->lock); | |
1438 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1439 | rv = 0; | |
1440 | if (bmc) { | |
1441 | /* locked */ | |
1442 | if (RESYNC(*bmc)) | |
1443 | rv = 1; | |
1444 | else if (NEEDED(*bmc)) { | |
1445 | rv = 1; | |
6a806c51 N |
1446 | if (!degraded) { /* don't set/clear bits if degraded */ |
1447 | *bmc |= RESYNC_MASK; | |
1448 | *bmc &= ~NEEDED_MASK; | |
1449 | } | |
32a7627c N |
1450 | } |
1451 | } | |
1452 | spin_unlock_irq(&bitmap->lock); | |
1453 | return rv; | |
1454 | } | |
1455 | ||
57dab0bd | 1456 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a N |
1457 | int degraded) |
1458 | { | |
1459 | /* bitmap_start_sync must always report on multiples of whole | |
1460 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1461 | * get confused. | |
1462 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1463 | * At least PAGE_SIZE>>9 blocks are covered. | |
1464 | * Return the 'or' of the result. | |
1465 | */ | |
1466 | int rv = 0; | |
57dab0bd | 1467 | sector_t blocks1; |
1187cf0a N |
1468 | |
1469 | *blocks = 0; | |
1470 | while (*blocks < (PAGE_SIZE>>9)) { | |
1471 | rv |= __bitmap_start_sync(bitmap, offset, | |
1472 | &blocks1, degraded); | |
1473 | offset += blocks1; | |
1474 | *blocks += blocks1; | |
1475 | } | |
1476 | return rv; | |
1477 | } | |
ac2f40be | 1478 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1479 | |
57dab0bd | 1480 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1481 | { |
1482 | bitmap_counter_t *bmc; | |
1483 | unsigned long flags; | |
ac2f40be N |
1484 | |
1485 | if (bitmap == NULL) { | |
32a7627c N |
1486 | *blocks = 1024; |
1487 | return; | |
1488 | } | |
1489 | spin_lock_irqsave(&bitmap->lock, flags); | |
1490 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1491 | if (bmc == NULL) | |
1492 | goto unlock; | |
1493 | /* locked */ | |
32a7627c N |
1494 | if (RESYNC(*bmc)) { |
1495 | *bmc &= ~RESYNC_MASK; | |
1496 | ||
1497 | if (!NEEDED(*bmc) && aborted) | |
1498 | *bmc |= NEEDED_MASK; | |
1499 | else { | |
2585f3ef | 1500 | if (*bmc <= 2) { |
bf07bb7d | 1501 | bitmap_set_pending(bitmap, offset); |
2585f3ef N |
1502 | bitmap->allclean = 0; |
1503 | } | |
32a7627c N |
1504 | } |
1505 | } | |
1506 | unlock: | |
1507 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1508 | } | |
ac2f40be | 1509 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1510 | |
1511 | void bitmap_close_sync(struct bitmap *bitmap) | |
1512 | { | |
1513 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1514 | * properly have been aborted. It remains to us to clear the | |
1515 | * RESYNC bit wherever it is still on | |
1516 | */ | |
1517 | sector_t sector = 0; | |
57dab0bd | 1518 | sector_t blocks; |
b47490c9 N |
1519 | if (!bitmap) |
1520 | return; | |
32a7627c N |
1521 | while (sector < bitmap->mddev->resync_max_sectors) { |
1522 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1523 | sector += blocks; |
1524 | } | |
1525 | } | |
ac2f40be | 1526 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 N |
1527 | |
1528 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1529 | { | |
1530 | sector_t s = 0; | |
57dab0bd | 1531 | sector_t blocks; |
b47490c9 N |
1532 | |
1533 | if (!bitmap) | |
1534 | return; | |
1535 | if (sector == 0) { | |
1536 | bitmap->last_end_sync = jiffies; | |
1537 | return; | |
1538 | } | |
1539 | if (time_before(jiffies, (bitmap->last_end_sync | |
1b04be96 | 1540 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1541 | return; |
1542 | wait_event(bitmap->mddev->recovery_wait, | |
1543 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1544 | ||
75d3da43 | 1545 | bitmap->mddev->curr_resync_completed = sector; |
070dc6dd | 1546 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
61a0d80c | 1547 | sector &= ~((1ULL << bitmap->chunkshift) - 1); |
b47490c9 N |
1548 | s = 0; |
1549 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1550 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1551 | s += blocks; | |
32a7627c | 1552 | } |
b47490c9 | 1553 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1554 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1555 | } |
ac2f40be | 1556 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1557 | |
6a07997f | 1558 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1559 | { |
1560 | /* For each chunk covered by any of these sectors, set the | |
ef99bf48 | 1561 | * counter to 2 and possibly set resync_needed. They should all |
32a7627c N |
1562 | * be 0 at this point |
1563 | */ | |
193f1c93 | 1564 | |
57dab0bd | 1565 | sector_t secs; |
193f1c93 N |
1566 | bitmap_counter_t *bmc; |
1567 | spin_lock_irq(&bitmap->lock); | |
1568 | bmc = bitmap_get_counter(bitmap, offset, &secs, 1); | |
1569 | if (!bmc) { | |
32a7627c | 1570 | spin_unlock_irq(&bitmap->lock); |
193f1c93 | 1571 | return; |
32a7627c | 1572 | } |
ac2f40be | 1573 | if (!*bmc) { |
915c420d | 1574 | *bmc = 2 | (needed ? NEEDED_MASK : 0); |
193f1c93 | 1575 | bitmap_count_page(bitmap, offset, 1); |
bf07bb7d | 1576 | bitmap_set_pending(bitmap, offset); |
2585f3ef | 1577 | bitmap->allclean = 0; |
193f1c93 N |
1578 | } |
1579 | spin_unlock_irq(&bitmap->lock); | |
32a7627c N |
1580 | } |
1581 | ||
9b1d1dac PC |
1582 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1583 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1584 | { | |
1585 | unsigned long chunk; | |
1586 | ||
1587 | for (chunk = s; chunk <= e; chunk++) { | |
61a0d80c | 1588 | sector_t sec = (sector_t)chunk << bitmap->chunkshift; |
9b1d1dac | 1589 | bitmap_set_memory_bits(bitmap, sec, 1); |
7c8f4247 | 1590 | spin_lock_irq(&bitmap->lock); |
9b1d1dac | 1591 | bitmap_file_set_bit(bitmap, sec); |
7c8f4247 | 1592 | spin_unlock_irq(&bitmap->lock); |
ffa23322 N |
1593 | if (sec < bitmap->mddev->recovery_cp) |
1594 | /* We are asserting that the array is dirty, | |
1595 | * so move the recovery_cp address back so | |
1596 | * that it is obvious that it is dirty | |
1597 | */ | |
1598 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1599 | } |
1600 | } | |
1601 | ||
6b8b3e8a N |
1602 | /* |
1603 | * flush out any pending updates | |
1604 | */ | |
fd01b88c | 1605 | void bitmap_flush(struct mddev *mddev) |
6b8b3e8a N |
1606 | { |
1607 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1608 | long sleep; |
6b8b3e8a N |
1609 | |
1610 | if (!bitmap) /* there was no bitmap */ | |
1611 | return; | |
1612 | ||
1613 | /* run the daemon_work three time to ensure everything is flushed | |
1614 | * that can be | |
1615 | */ | |
1b04be96 | 1616 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1617 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1618 | bitmap_daemon_work(mddev); |
42a04b50 | 1619 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1620 | bitmap_daemon_work(mddev); |
42a04b50 | 1621 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1622 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1623 | bitmap_update_sb(bitmap); |
1624 | } | |
1625 | ||
32a7627c N |
1626 | /* |
1627 | * free memory that was allocated | |
1628 | */ | |
3178b0db | 1629 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1630 | { |
1631 | unsigned long k, pages; | |
1632 | struct bitmap_page *bp; | |
32a7627c N |
1633 | |
1634 | if (!bitmap) /* there was no bitmap */ | |
1635 | return; | |
1636 | ||
32a7627c N |
1637 | /* release the bitmap file and kill the daemon */ |
1638 | bitmap_file_put(bitmap); | |
1639 | ||
1640 | bp = bitmap->bp; | |
1641 | pages = bitmap->pages; | |
1642 | ||
1643 | /* free all allocated memory */ | |
1644 | ||
32a7627c N |
1645 | if (bp) /* deallocate the page memory */ |
1646 | for (k = 0; k < pages; k++) | |
1647 | if (bp[k].map && !bp[k].hijacked) | |
1648 | kfree(bp[k].map); | |
1649 | kfree(bp); | |
1650 | kfree(bitmap); | |
1651 | } | |
aa5cbd10 | 1652 | |
fd01b88c | 1653 | void bitmap_destroy(struct mddev *mddev) |
3178b0db N |
1654 | { |
1655 | struct bitmap *bitmap = mddev->bitmap; | |
1656 | ||
1657 | if (!bitmap) /* there was no bitmap */ | |
1658 | return; | |
1659 | ||
c3d9714e | 1660 | mutex_lock(&mddev->bitmap_info.mutex); |
3178b0db | 1661 | mddev->bitmap = NULL; /* disconnect from the md device */ |
c3d9714e | 1662 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1663 | if (mddev->thread) |
1664 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db | 1665 | |
ece5cff0 N |
1666 | if (bitmap->sysfs_can_clear) |
1667 | sysfs_put(bitmap->sysfs_can_clear); | |
1668 | ||
3178b0db N |
1669 | bitmap_free(bitmap); |
1670 | } | |
32a7627c N |
1671 | |
1672 | /* | |
1673 | * initialize the bitmap structure | |
1674 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1675 | */ | |
fd01b88c | 1676 | int bitmap_create(struct mddev *mddev) |
32a7627c N |
1677 | { |
1678 | struct bitmap *bitmap; | |
1f593903 | 1679 | sector_t blocks = mddev->resync_max_sectors; |
32a7627c N |
1680 | unsigned long chunks; |
1681 | unsigned long pages; | |
c3d9714e | 1682 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1683 | int err; |
5ff5afff | 1684 | struct sysfs_dirent *bm = NULL; |
32a7627c | 1685 | |
5f6e3c83 | 1686 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1687 | |
c3d9714e | 1688 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1689 | |
9ffae0cf | 1690 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1691 | if (!bitmap) |
1692 | return -ENOMEM; | |
1693 | ||
32a7627c | 1694 | spin_lock_init(&bitmap->lock); |
ce25c31b N |
1695 | atomic_set(&bitmap->pending_writes, 0); |
1696 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1697 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1698 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1699 | |
32a7627c | 1700 | bitmap->mddev = mddev; |
32a7627c | 1701 | |
5ff5afff N |
1702 | if (mddev->kobj.sd) |
1703 | bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap"); | |
ece5cff0 | 1704 | if (bm) { |
3ff195b0 | 1705 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear"); |
ece5cff0 N |
1706 | sysfs_put(bm); |
1707 | } else | |
1708 | bitmap->sysfs_can_clear = NULL; | |
1709 | ||
1ec885cd | 1710 | bitmap->storage.file = file; |
ce25c31b N |
1711 | if (file) { |
1712 | get_file(file); | |
ae8fa283 N |
1713 | /* As future accesses to this file will use bmap, |
1714 | * and bypass the page cache, we must sync the file | |
1715 | * first. | |
1716 | */ | |
8018ab05 | 1717 | vfs_fsync(file, 1); |
ce25c31b | 1718 | } |
42a04b50 | 1719 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
9c81075f JB |
1720 | if (!mddev->bitmap_info.external) { |
1721 | /* | |
1722 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is | |
1723 | * instructing us to create a new on-disk bitmap instance. | |
1724 | */ | |
1725 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) | |
1726 | err = bitmap_new_disk_sb(bitmap); | |
1727 | else | |
1728 | err = bitmap_read_sb(bitmap); | |
1729 | } else { | |
ece5cff0 N |
1730 | err = 0; |
1731 | if (mddev->bitmap_info.chunksize == 0 || | |
1732 | mddev->bitmap_info.daemon_sleep == 0) | |
1733 | /* chunksize and time_base need to be | |
1734 | * set first. */ | |
1735 | err = -EINVAL; | |
1736 | } | |
32a7627c | 1737 | if (err) |
3178b0db | 1738 | goto error; |
32a7627c | 1739 | |
624ce4f5 | 1740 | bitmap->daemon_lastrun = jiffies; |
61a0d80c N |
1741 | bitmap->chunkshift = (ffz(~mddev->bitmap_info.chunksize) |
1742 | - BITMAP_BLOCK_SHIFT); | |
32a7627c | 1743 | |
b16b1b6c | 1744 | chunks = (blocks + (1 << bitmap->chunkshift) - 1) >> |
61a0d80c | 1745 | bitmap->chunkshift; |
ac2f40be | 1746 | pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; |
32a7627c N |
1747 | |
1748 | BUG_ON(!pages); | |
1749 | ||
1750 | bitmap->chunks = chunks; | |
1751 | bitmap->pages = pages; | |
1752 | bitmap->missing_pages = pages; | |
32a7627c | 1753 | |
9ffae0cf | 1754 | bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); |
29d3247e | 1755 | |
3178b0db | 1756 | err = -ENOMEM; |
32a7627c | 1757 | if (!bitmap->bp) |
3178b0db | 1758 | goto error; |
32a7627c | 1759 | |
69e51b44 N |
1760 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", |
1761 | pages, bmname(bitmap)); | |
1762 | ||
1763 | mddev->bitmap = bitmap; | |
1764 | ||
1765 | ||
1766 | return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0; | |
1767 | ||
1768 | error: | |
1769 | bitmap_free(bitmap); | |
1770 | return err; | |
1771 | } | |
1772 | ||
fd01b88c | 1773 | int bitmap_load(struct mddev *mddev) |
69e51b44 N |
1774 | { |
1775 | int err = 0; | |
3520fa4d | 1776 | sector_t start = 0; |
69e51b44 N |
1777 | sector_t sector = 0; |
1778 | struct bitmap *bitmap = mddev->bitmap; | |
1779 | ||
1780 | if (!bitmap) | |
1781 | goto out; | |
1782 | ||
1783 | /* Clear out old bitmap info first: Either there is none, or we | |
1784 | * are resuming after someone else has possibly changed things, | |
1785 | * so we should forget old cached info. | |
1786 | * All chunks should be clean, but some might need_sync. | |
1787 | */ | |
1788 | while (sector < mddev->resync_max_sectors) { | |
57dab0bd | 1789 | sector_t blocks; |
69e51b44 N |
1790 | bitmap_start_sync(bitmap, sector, &blocks, 0); |
1791 | sector += blocks; | |
1792 | } | |
1793 | bitmap_close_sync(bitmap); | |
1794 | ||
3520fa4d JB |
1795 | if (mddev->degraded == 0 |
1796 | || bitmap->events_cleared == mddev->events) | |
1797 | /* no need to keep dirty bits to optimise a | |
1798 | * re-add of a missing device */ | |
1799 | start = mddev->recovery_cp; | |
1800 | ||
afbaa90b | 1801 | mutex_lock(&mddev->bitmap_info.mutex); |
3520fa4d | 1802 | err = bitmap_init_from_disk(bitmap, start); |
afbaa90b | 1803 | mutex_unlock(&mddev->bitmap_info.mutex); |
3520fa4d | 1804 | |
32a7627c | 1805 | if (err) |
69e51b44 | 1806 | goto out; |
ef99bf48 N |
1807 | bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET); |
1808 | ||
1809 | /* Kick recovery in case any bits were set */ | |
1810 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
3178b0db | 1811 | |
1b04be96 | 1812 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1813 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1814 | |
4ad13663 N |
1815 | bitmap_update_sb(bitmap); |
1816 | ||
69e51b44 N |
1817 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
1818 | err = -EIO; | |
1819 | out: | |
3178b0db | 1820 | return err; |
32a7627c | 1821 | } |
69e51b44 | 1822 | EXPORT_SYMBOL_GPL(bitmap_load); |
32a7627c | 1823 | |
57148964 N |
1824 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap) |
1825 | { | |
1826 | unsigned long chunk_kb; | |
1827 | unsigned long flags; | |
1828 | ||
1829 | if (!bitmap) | |
1830 | return; | |
1831 | ||
1832 | spin_lock_irqsave(&bitmap->lock, flags); | |
1833 | chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; | |
1834 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
1835 | "%lu%s chunk", | |
1836 | bitmap->pages - bitmap->missing_pages, | |
1837 | bitmap->pages, | |
1838 | (bitmap->pages - bitmap->missing_pages) | |
1839 | << (PAGE_SHIFT - 10), | |
1840 | chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, | |
1841 | chunk_kb ? "KB" : "B"); | |
1ec885cd | 1842 | if (bitmap->storage.file) { |
57148964 | 1843 | seq_printf(seq, ", file: "); |
1ec885cd | 1844 | seq_path(seq, &bitmap->storage.file->f_path, " \t\n"); |
57148964 N |
1845 | } |
1846 | ||
1847 | seq_printf(seq, "\n"); | |
1848 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1849 | } | |
1850 | ||
43a70507 | 1851 | static ssize_t |
fd01b88c | 1852 | location_show(struct mddev *mddev, char *page) |
43a70507 N |
1853 | { |
1854 | ssize_t len; | |
ac2f40be | 1855 | if (mddev->bitmap_info.file) |
43a70507 | 1856 | len = sprintf(page, "file"); |
ac2f40be | 1857 | else if (mddev->bitmap_info.offset) |
43a70507 | 1858 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 1859 | else |
43a70507 N |
1860 | len = sprintf(page, "none"); |
1861 | len += sprintf(page+len, "\n"); | |
1862 | return len; | |
1863 | } | |
1864 | ||
1865 | static ssize_t | |
fd01b88c | 1866 | location_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
1867 | { |
1868 | ||
1869 | if (mddev->pers) { | |
1870 | if (!mddev->pers->quiesce) | |
1871 | return -EBUSY; | |
1872 | if (mddev->recovery || mddev->sync_thread) | |
1873 | return -EBUSY; | |
1874 | } | |
1875 | ||
1876 | if (mddev->bitmap || mddev->bitmap_info.file || | |
1877 | mddev->bitmap_info.offset) { | |
1878 | /* bitmap already configured. Only option is to clear it */ | |
1879 | if (strncmp(buf, "none", 4) != 0) | |
1880 | return -EBUSY; | |
1881 | if (mddev->pers) { | |
1882 | mddev->pers->quiesce(mddev, 1); | |
1883 | bitmap_destroy(mddev); | |
1884 | mddev->pers->quiesce(mddev, 0); | |
1885 | } | |
1886 | mddev->bitmap_info.offset = 0; | |
1887 | if (mddev->bitmap_info.file) { | |
1888 | struct file *f = mddev->bitmap_info.file; | |
1889 | mddev->bitmap_info.file = NULL; | |
1890 | restore_bitmap_write_access(f); | |
1891 | fput(f); | |
1892 | } | |
1893 | } else { | |
1894 | /* No bitmap, OK to set a location */ | |
1895 | long long offset; | |
1896 | if (strncmp(buf, "none", 4) == 0) | |
1897 | /* nothing to be done */; | |
1898 | else if (strncmp(buf, "file:", 5) == 0) { | |
1899 | /* Not supported yet */ | |
1900 | return -EINVAL; | |
1901 | } else { | |
1902 | int rv; | |
1903 | if (buf[0] == '+') | |
1904 | rv = strict_strtoll(buf+1, 10, &offset); | |
1905 | else | |
1906 | rv = strict_strtoll(buf, 10, &offset); | |
1907 | if (rv) | |
1908 | return rv; | |
1909 | if (offset == 0) | |
1910 | return -EINVAL; | |
ece5cff0 N |
1911 | if (mddev->bitmap_info.external == 0 && |
1912 | mddev->major_version == 0 && | |
43a70507 N |
1913 | offset != mddev->bitmap_info.default_offset) |
1914 | return -EINVAL; | |
1915 | mddev->bitmap_info.offset = offset; | |
1916 | if (mddev->pers) { | |
1917 | mddev->pers->quiesce(mddev, 1); | |
1918 | rv = bitmap_create(mddev); | |
4474ca42 N |
1919 | if (!rv) |
1920 | rv = bitmap_load(mddev); | |
43a70507 N |
1921 | if (rv) { |
1922 | bitmap_destroy(mddev); | |
1923 | mddev->bitmap_info.offset = 0; | |
1924 | } | |
1925 | mddev->pers->quiesce(mddev, 0); | |
1926 | if (rv) | |
1927 | return rv; | |
1928 | } | |
1929 | } | |
1930 | } | |
1931 | if (!mddev->external) { | |
1932 | /* Ensure new bitmap info is stored in | |
1933 | * metadata promptly. | |
1934 | */ | |
1935 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1936 | md_wakeup_thread(mddev->thread); | |
1937 | } | |
1938 | return len; | |
1939 | } | |
1940 | ||
1941 | static struct md_sysfs_entry bitmap_location = | |
1942 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
1943 | ||
6409bb05 N |
1944 | /* 'bitmap/space' is the space available at 'location' for the |
1945 | * bitmap. This allows the kernel to know when it is safe to | |
1946 | * resize the bitmap to match a resized array. | |
1947 | */ | |
1948 | static ssize_t | |
1949 | space_show(struct mddev *mddev, char *page) | |
1950 | { | |
1951 | return sprintf(page, "%lu\n", mddev->bitmap_info.space); | |
1952 | } | |
1953 | ||
1954 | static ssize_t | |
1955 | space_store(struct mddev *mddev, const char *buf, size_t len) | |
1956 | { | |
1957 | unsigned long sectors; | |
1958 | int rv; | |
1959 | ||
1960 | rv = kstrtoul(buf, 10, §ors); | |
1961 | if (rv) | |
1962 | return rv; | |
1963 | ||
1964 | if (sectors == 0) | |
1965 | return -EINVAL; | |
1966 | ||
1967 | if (mddev->bitmap && | |
1ec885cd N |
1968 | sectors < ((mddev->bitmap->storage.file_pages - 1) |
1969 | * PAGE_SIZE | |
1970 | + mddev->bitmap->storage.last_page_size + 511) >> 9) | |
6409bb05 N |
1971 | return -EFBIG; /* Bitmap is too big for this small space */ |
1972 | ||
1973 | /* could make sure it isn't too big, but that isn't really | |
1974 | * needed - user-space should be careful. | |
1975 | */ | |
1976 | mddev->bitmap_info.space = sectors; | |
1977 | return len; | |
1978 | } | |
1979 | ||
1980 | static struct md_sysfs_entry bitmap_space = | |
1981 | __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); | |
1982 | ||
43a70507 | 1983 | static ssize_t |
fd01b88c | 1984 | timeout_show(struct mddev *mddev, char *page) |
43a70507 N |
1985 | { |
1986 | ssize_t len; | |
1987 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
1988 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 1989 | |
43a70507 N |
1990 | len = sprintf(page, "%lu", secs); |
1991 | if (jifs) | |
1992 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
1993 | len += sprintf(page+len, "\n"); | |
1994 | return len; | |
1995 | } | |
1996 | ||
1997 | static ssize_t | |
fd01b88c | 1998 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
1999 | { |
2000 | /* timeout can be set at any time */ | |
2001 | unsigned long timeout; | |
2002 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
2003 | if (rv) | |
2004 | return rv; | |
2005 | ||
2006 | /* just to make sure we don't overflow... */ | |
2007 | if (timeout >= LONG_MAX / HZ) | |
2008 | return -EINVAL; | |
2009 | ||
2010 | timeout = timeout * HZ / 10000; | |
2011 | ||
2012 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
2013 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
2014 | if (timeout < 1) | |
2015 | timeout = 1; | |
2016 | mddev->bitmap_info.daemon_sleep = timeout; | |
2017 | if (mddev->thread) { | |
2018 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
2019 | * the bitmap is all clean and we don't need to | |
2020 | * adjust the timeout right now | |
2021 | */ | |
2022 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2023 | mddev->thread->timeout = timeout; | |
2024 | md_wakeup_thread(mddev->thread); | |
2025 | } | |
2026 | } | |
2027 | return len; | |
2028 | } | |
2029 | ||
2030 | static struct md_sysfs_entry bitmap_timeout = | |
2031 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2032 | ||
2033 | static ssize_t | |
fd01b88c | 2034 | backlog_show(struct mddev *mddev, char *page) |
43a70507 N |
2035 | { |
2036 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2037 | } | |
2038 | ||
2039 | static ssize_t | |
fd01b88c | 2040 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2041 | { |
2042 | unsigned long backlog; | |
2043 | int rv = strict_strtoul(buf, 10, &backlog); | |
2044 | if (rv) | |
2045 | return rv; | |
2046 | if (backlog > COUNTER_MAX) | |
2047 | return -EINVAL; | |
2048 | mddev->bitmap_info.max_write_behind = backlog; | |
2049 | return len; | |
2050 | } | |
2051 | ||
2052 | static struct md_sysfs_entry bitmap_backlog = | |
2053 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2054 | ||
2055 | static ssize_t | |
fd01b88c | 2056 | chunksize_show(struct mddev *mddev, char *page) |
43a70507 N |
2057 | { |
2058 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2059 | } | |
2060 | ||
2061 | static ssize_t | |
fd01b88c | 2062 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2063 | { |
2064 | /* Can only be changed when no bitmap is active */ | |
2065 | int rv; | |
2066 | unsigned long csize; | |
2067 | if (mddev->bitmap) | |
2068 | return -EBUSY; | |
2069 | rv = strict_strtoul(buf, 10, &csize); | |
2070 | if (rv) | |
2071 | return rv; | |
2072 | if (csize < 512 || | |
2073 | !is_power_of_2(csize)) | |
2074 | return -EINVAL; | |
2075 | mddev->bitmap_info.chunksize = csize; | |
2076 | return len; | |
2077 | } | |
2078 | ||
2079 | static struct md_sysfs_entry bitmap_chunksize = | |
2080 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2081 | ||
fd01b88c | 2082 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2083 | { |
2084 | return sprintf(page, "%s\n", (mddev->bitmap_info.external | |
2085 | ? "external" : "internal")); | |
2086 | } | |
2087 | ||
fd01b88c | 2088 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2089 | { |
2090 | if (mddev->bitmap || | |
2091 | mddev->bitmap_info.file || | |
2092 | mddev->bitmap_info.offset) | |
2093 | return -EBUSY; | |
2094 | if (strncmp(buf, "external", 8) == 0) | |
2095 | mddev->bitmap_info.external = 1; | |
2096 | else if (strncmp(buf, "internal", 8) == 0) | |
2097 | mddev->bitmap_info.external = 0; | |
2098 | else | |
2099 | return -EINVAL; | |
2100 | return len; | |
2101 | } | |
2102 | ||
2103 | static struct md_sysfs_entry bitmap_metadata = | |
2104 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2105 | ||
fd01b88c | 2106 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2107 | { |
2108 | int len; | |
2109 | if (mddev->bitmap) | |
2110 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2111 | "false" : "true")); | |
2112 | else | |
2113 | len = sprintf(page, "\n"); | |
2114 | return len; | |
2115 | } | |
2116 | ||
fd01b88c | 2117 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2118 | { |
2119 | if (mddev->bitmap == NULL) | |
2120 | return -ENOENT; | |
2121 | if (strncmp(buf, "false", 5) == 0) | |
2122 | mddev->bitmap->need_sync = 1; | |
2123 | else if (strncmp(buf, "true", 4) == 0) { | |
2124 | if (mddev->degraded) | |
2125 | return -EBUSY; | |
2126 | mddev->bitmap->need_sync = 0; | |
2127 | } else | |
2128 | return -EINVAL; | |
2129 | return len; | |
2130 | } | |
2131 | ||
2132 | static struct md_sysfs_entry bitmap_can_clear = | |
2133 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2134 | ||
696fcd53 | 2135 | static ssize_t |
fd01b88c | 2136 | behind_writes_used_show(struct mddev *mddev, char *page) |
696fcd53 PC |
2137 | { |
2138 | if (mddev->bitmap == NULL) | |
2139 | return sprintf(page, "0\n"); | |
2140 | return sprintf(page, "%lu\n", | |
2141 | mddev->bitmap->behind_writes_used); | |
2142 | } | |
2143 | ||
2144 | static ssize_t | |
fd01b88c | 2145 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
696fcd53 PC |
2146 | { |
2147 | if (mddev->bitmap) | |
2148 | mddev->bitmap->behind_writes_used = 0; | |
2149 | return len; | |
2150 | } | |
2151 | ||
2152 | static struct md_sysfs_entry max_backlog_used = | |
2153 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2154 | behind_writes_used_show, behind_writes_used_reset); | |
2155 | ||
43a70507 N |
2156 | static struct attribute *md_bitmap_attrs[] = { |
2157 | &bitmap_location.attr, | |
6409bb05 | 2158 | &bitmap_space.attr, |
43a70507 N |
2159 | &bitmap_timeout.attr, |
2160 | &bitmap_backlog.attr, | |
2161 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2162 | &bitmap_metadata.attr, |
2163 | &bitmap_can_clear.attr, | |
696fcd53 | 2164 | &max_backlog_used.attr, |
43a70507 N |
2165 | NULL |
2166 | }; | |
2167 | struct attribute_group md_bitmap_group = { | |
2168 | .name = "bitmap", | |
2169 | .attrs = md_bitmap_attrs, | |
2170 | }; | |
2171 |