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 | */ | |
40cffcc0 | 48 | static int bitmap_checkpage(struct bitmap_counts *bitmap, |
ac2f40be | 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) { | |
40cffcc0 | 79 | pr_debug("md/bitmap: map page allocation failed, hijacking\n"); |
32a7627c N |
80 | /* failed - set the hijacked flag so that we can use the |
81 | * pointer as a counter */ | |
32a7627c N |
82 | if (!bitmap->bp[page].map) |
83 | bitmap->bp[page].hijacked = 1; | |
ac2f40be N |
84 | } else if (bitmap->bp[page].map || |
85 | bitmap->bp[page].hijacked) { | |
32a7627c | 86 | /* somebody beat us to getting the page */ |
792a1d4b | 87 | kfree(mappage); |
32a7627c | 88 | return 0; |
ac2f40be | 89 | } else { |
32a7627c | 90 | |
ac2f40be | 91 | /* no page was in place and we have one, so install it */ |
32a7627c | 92 | |
ac2f40be N |
93 | bitmap->bp[page].map = mappage; |
94 | bitmap->missing_pages--; | |
95 | } | |
32a7627c N |
96 | return 0; |
97 | } | |
98 | ||
32a7627c N |
99 | /* if page is completely empty, put it back on the free list, or dealloc it */ |
100 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
101 | /* Note: lock should be held when calling this */ | |
40cffcc0 | 102 | static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page) |
32a7627c N |
103 | { |
104 | char *ptr; | |
105 | ||
106 | if (bitmap->bp[page].count) /* page is still busy */ | |
107 | return; | |
108 | ||
109 | /* page is no longer in use, it can be released */ | |
110 | ||
111 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
112 | bitmap->bp[page].hijacked = 0; | |
113 | bitmap->bp[page].map = NULL; | |
ac2f40be N |
114 | } else { |
115 | /* normal case, free the page */ | |
116 | ptr = bitmap->bp[page].map; | |
117 | bitmap->bp[page].map = NULL; | |
118 | bitmap->missing_pages++; | |
792a1d4b | 119 | kfree(ptr); |
32a7627c | 120 | } |
32a7627c N |
121 | } |
122 | ||
32a7627c N |
123 | /* |
124 | * bitmap file handling - read and write the bitmap file and its superblock | |
125 | */ | |
126 | ||
32a7627c N |
127 | /* |
128 | * basic page I/O operations | |
129 | */ | |
130 | ||
a654b9d8 | 131 | /* IO operations when bitmap is stored near all superblocks */ |
27581e5a N |
132 | static int read_sb_page(struct mddev *mddev, loff_t offset, |
133 | struct page *page, | |
134 | unsigned long index, int size) | |
a654b9d8 N |
135 | { |
136 | /* choose a good rdev and read the page from there */ | |
137 | ||
3cb03002 | 138 | struct md_rdev *rdev; |
a654b9d8 | 139 | sector_t target; |
a654b9d8 | 140 | |
dafb20fa | 141 | rdev_for_each(rdev, mddev) { |
b2d444d7 N |
142 | if (! test_bit(In_sync, &rdev->flags) |
143 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
144 | continue; |
145 | ||
ccebd4c4 | 146 | target = offset + index * (PAGE_SIZE/512); |
a654b9d8 | 147 | |
2b193363 | 148 | if (sync_page_io(rdev, target, |
e1defc4f | 149 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
ccebd4c4 | 150 | page, READ, true)) { |
ab904d63 | 151 | page->index = index; |
27581e5a | 152 | return 0; |
ab904d63 N |
153 | } |
154 | } | |
27581e5a | 155 | return -EIO; |
a654b9d8 N |
156 | } |
157 | ||
fd01b88c | 158 | static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) |
b2d2c4ce N |
159 | { |
160 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
161 | * linked list, and raising the refcount of devices we return to ensure | |
162 | * they don't disappear while in use. | |
163 | * As devices are only added or removed when raid_disk is < 0 and | |
164 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
165 | * still be in the same position on the list when we re-enter | |
166 | * list_for_each_continue_rcu. | |
167 | */ | |
168 | struct list_head *pos; | |
169 | rcu_read_lock(); | |
170 | if (rdev == NULL) | |
171 | /* start at the beginning */ | |
172 | pos = &mddev->disks; | |
173 | else { | |
174 | /* release the previous rdev and start from there. */ | |
175 | rdev_dec_pending(rdev, mddev); | |
176 | pos = &rdev->same_set; | |
177 | } | |
178 | list_for_each_continue_rcu(pos, &mddev->disks) { | |
3cb03002 | 179 | rdev = list_entry(pos, struct md_rdev, same_set); |
b2d2c4ce | 180 | if (rdev->raid_disk >= 0 && |
b2d2c4ce N |
181 | !test_bit(Faulty, &rdev->flags)) { |
182 | /* this is a usable devices */ | |
183 | atomic_inc(&rdev->nr_pending); | |
184 | rcu_read_unlock(); | |
185 | return rdev; | |
186 | } | |
187 | } | |
188 | rcu_read_unlock(); | |
189 | return NULL; | |
190 | } | |
191 | ||
ab6085c7 | 192 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 193 | { |
3cb03002 | 194 | struct md_rdev *rdev = NULL; |
a6ff7e08 | 195 | struct block_device *bdev; |
fd01b88c | 196 | struct mddev *mddev = bitmap->mddev; |
1ec885cd | 197 | struct bitmap_storage *store = &bitmap->storage; |
a654b9d8 | 198 | |
b2d2c4ce | 199 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ac2f40be N |
200 | int size = PAGE_SIZE; |
201 | loff_t offset = mddev->bitmap_info.offset; | |
a6ff7e08 JB |
202 | |
203 | bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; | |
204 | ||
9b1215c1 N |
205 | if (page->index == store->file_pages-1) { |
206 | int last_page_size = store->bytes & (PAGE_SIZE-1); | |
207 | if (last_page_size == 0) | |
208 | last_page_size = PAGE_SIZE; | |
209 | size = roundup(last_page_size, | |
a6ff7e08 | 210 | bdev_logical_block_size(bdev)); |
9b1215c1 | 211 | } |
ac2f40be N |
212 | /* Just make sure we aren't corrupting data or |
213 | * metadata | |
214 | */ | |
215 | if (mddev->external) { | |
216 | /* Bitmap could be anywhere. */ | |
217 | if (rdev->sb_start + offset + (page->index | |
218 | * (PAGE_SIZE/512)) | |
219 | > rdev->data_offset | |
220 | && | |
221 | rdev->sb_start + offset | |
222 | < (rdev->data_offset + mddev->dev_sectors | |
223 | + (PAGE_SIZE/512))) | |
224 | goto bad_alignment; | |
225 | } else if (offset < 0) { | |
226 | /* DATA BITMAP METADATA */ | |
227 | if (offset | |
228 | + (long)(page->index * (PAGE_SIZE/512)) | |
229 | + size/512 > 0) | |
230 | /* bitmap runs in to metadata */ | |
231 | goto bad_alignment; | |
232 | if (rdev->data_offset + mddev->dev_sectors | |
233 | > rdev->sb_start + offset) | |
234 | /* data runs in to bitmap */ | |
235 | goto bad_alignment; | |
236 | } else if (rdev->sb_start < rdev->data_offset) { | |
237 | /* METADATA BITMAP DATA */ | |
238 | if (rdev->sb_start | |
239 | + offset | |
240 | + page->index*(PAGE_SIZE/512) + size/512 | |
241 | > rdev->data_offset) | |
242 | /* bitmap runs in to data */ | |
243 | goto bad_alignment; | |
244 | } else { | |
245 | /* DATA METADATA BITMAP - no problems */ | |
246 | } | |
247 | md_super_write(mddev, rdev, | |
248 | rdev->sb_start + offset | |
249 | + page->index * (PAGE_SIZE/512), | |
250 | size, | |
251 | page); | |
b2d2c4ce | 252 | } |
a654b9d8 N |
253 | |
254 | if (wait) | |
a9701a30 | 255 | md_super_wait(mddev); |
a654b9d8 | 256 | return 0; |
4b80991c N |
257 | |
258 | bad_alignment: | |
4b80991c | 259 | return -EINVAL; |
a654b9d8 N |
260 | } |
261 | ||
4ad13663 | 262 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 263 | /* |
a654b9d8 | 264 | * write out a page to a file |
32a7627c | 265 | */ |
4ad13663 | 266 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 267 | { |
d785a06a | 268 | struct buffer_head *bh; |
32a7627c | 269 | |
1ec885cd | 270 | if (bitmap->storage.file == NULL) { |
f0d76d70 N |
271 | switch (write_sb_page(bitmap, page, wait)) { |
272 | case -EINVAL: | |
b405fe91 | 273 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); |
f0d76d70 | 274 | } |
4ad13663 | 275 | } else { |
a654b9d8 | 276 | |
4ad13663 | 277 | bh = page_buffers(page); |
c708443c | 278 | |
4ad13663 N |
279 | while (bh && bh->b_blocknr) { |
280 | atomic_inc(&bitmap->pending_writes); | |
281 | set_buffer_locked(bh); | |
282 | set_buffer_mapped(bh); | |
721a9602 | 283 | submit_bh(WRITE | REQ_SYNC, bh); |
4ad13663 N |
284 | bh = bh->b_this_page; |
285 | } | |
d785a06a | 286 | |
ac2f40be | 287 | if (wait) |
4ad13663 N |
288 | wait_event(bitmap->write_wait, |
289 | atomic_read(&bitmap->pending_writes)==0); | |
8a5e9cf1 | 290 | } |
b405fe91 | 291 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
4ad13663 | 292 | bitmap_file_kick(bitmap); |
d785a06a N |
293 | } |
294 | ||
295 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
296 | { | |
297 | struct bitmap *bitmap = bh->b_private; | |
32a7627c | 298 | |
b405fe91 N |
299 | if (!uptodate) |
300 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); | |
d785a06a N |
301 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
302 | wake_up(&bitmap->write_wait); | |
303 | } | |
32a7627c | 304 | |
d785a06a N |
305 | /* copied from buffer.c */ |
306 | static void | |
307 | __clear_page_buffers(struct page *page) | |
308 | { | |
309 | ClearPagePrivate(page); | |
310 | set_page_private(page, 0); | |
311 | page_cache_release(page); | |
312 | } | |
313 | static void free_buffers(struct page *page) | |
314 | { | |
27581e5a | 315 | struct buffer_head *bh; |
77ad4bc7 | 316 | |
27581e5a N |
317 | if (!PagePrivate(page)) |
318 | return; | |
319 | ||
320 | bh = page_buffers(page); | |
d785a06a N |
321 | while (bh) { |
322 | struct buffer_head *next = bh->b_this_page; | |
323 | free_buffer_head(bh); | |
324 | bh = next; | |
77ad4bc7 | 325 | } |
d785a06a N |
326 | __clear_page_buffers(page); |
327 | put_page(page); | |
32a7627c N |
328 | } |
329 | ||
d785a06a N |
330 | /* read a page from a file. |
331 | * We both read the page, and attach buffers to the page to record the | |
332 | * address of each block (using bmap). These addresses will be used | |
333 | * to write the block later, completely bypassing the filesystem. | |
334 | * This usage is similar to how swap files are handled, and allows us | |
335 | * to write to a file with no concerns of memory allocation failing. | |
336 | */ | |
27581e5a N |
337 | static int read_page(struct file *file, unsigned long index, |
338 | struct bitmap *bitmap, | |
339 | unsigned long count, | |
340 | struct page *page) | |
32a7627c | 341 | { |
27581e5a | 342 | int ret = 0; |
c649bb9c | 343 | struct inode *inode = file->f_path.dentry->d_inode; |
d785a06a N |
344 | struct buffer_head *bh; |
345 | sector_t block; | |
32a7627c | 346 | |
36a4e1fe N |
347 | pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
348 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 349 | |
d785a06a N |
350 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
351 | if (!bh) { | |
27581e5a | 352 | ret = -ENOMEM; |
32a7627c N |
353 | goto out; |
354 | } | |
d785a06a N |
355 | attach_page_buffers(page, bh); |
356 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
357 | while (bh) { | |
358 | if (count == 0) | |
359 | bh->b_blocknr = 0; | |
360 | else { | |
361 | bh->b_blocknr = bmap(inode, block); | |
362 | if (bh->b_blocknr == 0) { | |
363 | /* Cannot use this file! */ | |
27581e5a | 364 | ret = -EINVAL; |
d785a06a N |
365 | goto out; |
366 | } | |
367 | bh->b_bdev = inode->i_sb->s_bdev; | |
368 | if (count < (1<<inode->i_blkbits)) | |
369 | count = 0; | |
370 | else | |
371 | count -= (1<<inode->i_blkbits); | |
372 | ||
373 | bh->b_end_io = end_bitmap_write; | |
374 | bh->b_private = bitmap; | |
ce25c31b N |
375 | atomic_inc(&bitmap->pending_writes); |
376 | set_buffer_locked(bh); | |
377 | set_buffer_mapped(bh); | |
378 | submit_bh(READ, bh); | |
d785a06a N |
379 | } |
380 | block++; | |
381 | bh = bh->b_this_page; | |
382 | } | |
d785a06a | 383 | page->index = index; |
ce25c31b N |
384 | |
385 | wait_event(bitmap->write_wait, | |
386 | atomic_read(&bitmap->pending_writes)==0); | |
b405fe91 | 387 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
27581e5a | 388 | ret = -EIO; |
32a7627c | 389 | out: |
27581e5a N |
390 | if (ret) |
391 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n", | |
2d1f3b5d N |
392 | (int)PAGE_SIZE, |
393 | (unsigned long long)index << PAGE_SHIFT, | |
27581e5a N |
394 | ret); |
395 | return ret; | |
32a7627c N |
396 | } |
397 | ||
398 | /* | |
399 | * bitmap file superblock operations | |
400 | */ | |
401 | ||
402 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 403 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
404 | { |
405 | bitmap_super_t *sb; | |
32a7627c N |
406 | |
407 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 408 | return; |
ece5cff0 N |
409 | if (bitmap->mddev->bitmap_info.external) |
410 | return; | |
1ec885cd | 411 | if (!bitmap->storage.sb_page) /* no superblock */ |
4ad13663 | 412 | return; |
1ec885cd | 413 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 414 | sb->events = cpu_to_le64(bitmap->mddev->events); |
8258c532 | 415 | if (bitmap->mddev->events < bitmap->events_cleared) |
a0da84f3 NB |
416 | /* rocking back to read-only */ |
417 | bitmap->events_cleared = bitmap->mddev->events; | |
8258c532 N |
418 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); |
419 | sb->state = cpu_to_le32(bitmap->flags); | |
43a70507 N |
420 | /* Just in case these have been changed via sysfs: */ |
421 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
422 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
b81a0404 N |
423 | /* This might have been changed by a reshape */ |
424 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
425 | sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize); | |
1dff2b87 N |
426 | sb->sectors_reserved = cpu_to_le32(bitmap->mddev-> |
427 | bitmap_info.space); | |
b2f46e68 | 428 | kunmap_atomic(sb); |
1ec885cd | 429 | write_page(bitmap, bitmap->storage.sb_page, 1); |
32a7627c N |
430 | } |
431 | ||
432 | /* print out the bitmap file superblock */ | |
433 | void bitmap_print_sb(struct bitmap *bitmap) | |
434 | { | |
435 | bitmap_super_t *sb; | |
436 | ||
1ec885cd | 437 | if (!bitmap || !bitmap->storage.sb_page) |
32a7627c | 438 | return; |
1ec885cd | 439 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 440 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
441 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
442 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
443 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
444 | *(__u32 *)(sb->uuid+0), |
445 | *(__u32 *)(sb->uuid+4), | |
446 | *(__u32 *)(sb->uuid+8), | |
447 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 448 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 449 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 450 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 451 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
452 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
453 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
454 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
455 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
456 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 457 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
b2f46e68 | 458 | kunmap_atomic(sb); |
32a7627c N |
459 | } |
460 | ||
9c81075f JB |
461 | /* |
462 | * bitmap_new_disk_sb | |
463 | * @bitmap | |
464 | * | |
465 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb | |
466 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. | |
467 | * This function verifies 'bitmap_info' and populates the on-disk bitmap | |
468 | * structure, which is to be written to disk. | |
469 | * | |
470 | * Returns: 0 on success, -Exxx on error | |
471 | */ | |
472 | static int bitmap_new_disk_sb(struct bitmap *bitmap) | |
473 | { | |
474 | bitmap_super_t *sb; | |
475 | unsigned long chunksize, daemon_sleep, write_behind; | |
476 | int err = -EINVAL; | |
477 | ||
1ec885cd N |
478 | bitmap->storage.sb_page = alloc_page(GFP_KERNEL); |
479 | if (IS_ERR(bitmap->storage.sb_page)) { | |
480 | err = PTR_ERR(bitmap->storage.sb_page); | |
481 | bitmap->storage.sb_page = NULL; | |
9c81075f JB |
482 | return err; |
483 | } | |
1ec885cd | 484 | bitmap->storage.sb_page->index = 0; |
9c81075f | 485 | |
1ec885cd | 486 | sb = kmap_atomic(bitmap->storage.sb_page); |
9c81075f JB |
487 | |
488 | sb->magic = cpu_to_le32(BITMAP_MAGIC); | |
489 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); | |
490 | ||
491 | chunksize = bitmap->mddev->bitmap_info.chunksize; | |
492 | BUG_ON(!chunksize); | |
493 | if (!is_power_of_2(chunksize)) { | |
b2f46e68 | 494 | kunmap_atomic(sb); |
9c81075f JB |
495 | printk(KERN_ERR "bitmap chunksize not a power of 2\n"); |
496 | return -EINVAL; | |
497 | } | |
498 | sb->chunksize = cpu_to_le32(chunksize); | |
499 | ||
500 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; | |
501 | if (!daemon_sleep || | |
502 | (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { | |
503 | printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n"); | |
504 | daemon_sleep = 5 * HZ; | |
505 | } | |
506 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); | |
507 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
508 | ||
509 | /* | |
510 | * FIXME: write_behind for RAID1. If not specified, what | |
511 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. | |
512 | */ | |
513 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; | |
514 | if (write_behind > COUNTER_MAX) | |
515 | write_behind = COUNTER_MAX / 2; | |
516 | sb->write_behind = cpu_to_le32(write_behind); | |
517 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
518 | ||
519 | /* keep the array size field of the bitmap superblock up to date */ | |
520 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
521 | ||
522 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); | |
523 | ||
b405fe91 | 524 | set_bit(BITMAP_STALE, &bitmap->flags); |
84e92345 | 525 | sb->state = cpu_to_le32(bitmap->flags); |
9c81075f JB |
526 | bitmap->events_cleared = bitmap->mddev->events; |
527 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
528 | ||
b2f46e68 | 529 | kunmap_atomic(sb); |
9c81075f JB |
530 | |
531 | return 0; | |
532 | } | |
533 | ||
32a7627c N |
534 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
535 | static int bitmap_read_sb(struct bitmap *bitmap) | |
536 | { | |
537 | char *reason = NULL; | |
538 | bitmap_super_t *sb; | |
4b6d287f | 539 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c | 540 | unsigned long long events; |
1dff2b87 | 541 | unsigned long sectors_reserved = 0; |
32a7627c | 542 | int err = -EINVAL; |
27581e5a | 543 | struct page *sb_page; |
32a7627c | 544 | |
1ec885cd | 545 | if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { |
ef99bf48 N |
546 | chunksize = 128 * 1024 * 1024; |
547 | daemon_sleep = 5 * HZ; | |
548 | write_behind = 0; | |
b405fe91 | 549 | set_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
550 | err = 0; |
551 | goto out_no_sb; | |
552 | } | |
32a7627c | 553 | /* page 0 is the superblock, read it... */ |
27581e5a N |
554 | sb_page = alloc_page(GFP_KERNEL); |
555 | if (!sb_page) | |
556 | return -ENOMEM; | |
1ec885cd | 557 | bitmap->storage.sb_page = sb_page; |
27581e5a | 558 | |
1ec885cd N |
559 | if (bitmap->storage.file) { |
560 | loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); | |
f49d5e62 N |
561 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; |
562 | ||
1ec885cd | 563 | err = read_page(bitmap->storage.file, 0, |
27581e5a | 564 | bitmap, bytes, sb_page); |
f49d5e62 | 565 | } else { |
27581e5a N |
566 | err = read_sb_page(bitmap->mddev, |
567 | bitmap->mddev->bitmap_info.offset, | |
568 | sb_page, | |
569 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 570 | } |
27581e5a | 571 | if (err) |
32a7627c | 572 | return err; |
32a7627c | 573 | |
27581e5a | 574 | sb = kmap_atomic(sb_page); |
32a7627c | 575 | |
32a7627c | 576 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 577 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 578 | write_behind = le32_to_cpu(sb->write_behind); |
1dff2b87 | 579 | sectors_reserved = le32_to_cpu(sb->sectors_reserved); |
32a7627c N |
580 | |
581 | /* verify that the bitmap-specific fields are valid */ | |
582 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
583 | reason = "bad magic"; | |
bd926c63 N |
584 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
585 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 586 | reason = "unrecognized superblock version"; |
1187cf0a | 587 | else if (chunksize < 512) |
7dd5d34c | 588 | reason = "bitmap chunksize too small"; |
d744540c | 589 | else if (!is_power_of_2(chunksize)) |
32a7627c | 590 | reason = "bitmap chunksize not a power of 2"; |
1b04be96 | 591 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 592 | reason = "daemon sleep period out of range"; |
4b6d287f N |
593 | else if (write_behind > COUNTER_MAX) |
594 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
595 | if (reason) { |
596 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
597 | bmname(bitmap), reason); | |
598 | goto out; | |
599 | } | |
600 | ||
601 | /* keep the array size field of the bitmap superblock up to date */ | |
602 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
603 | ||
278c1ca2 N |
604 | if (bitmap->mddev->persistent) { |
605 | /* | |
606 | * We have a persistent array superblock, so compare the | |
607 | * bitmap's UUID and event counter to the mddev's | |
608 | */ | |
609 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
610 | printk(KERN_INFO | |
611 | "%s: bitmap superblock UUID mismatch\n", | |
612 | bmname(bitmap)); | |
613 | goto out; | |
614 | } | |
615 | events = le64_to_cpu(sb->events); | |
616 | if (events < bitmap->mddev->events) { | |
617 | printk(KERN_INFO | |
618 | "%s: bitmap file is out of date (%llu < %llu) " | |
619 | "-- forcing full recovery\n", | |
620 | bmname(bitmap), events, | |
621 | (unsigned long long) bitmap->mddev->events); | |
b405fe91 | 622 | set_bit(BITMAP_STALE, &bitmap->flags); |
278c1ca2 | 623 | } |
32a7627c | 624 | } |
278c1ca2 | 625 | |
32a7627c | 626 | /* assign fields using values from superblock */ |
4f2e639a | 627 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 | 628 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
b405fe91 | 629 | set_bit(BITMAP_HOSTENDIAN, &bitmap->flags); |
32a7627c N |
630 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
631 | err = 0; | |
632 | out: | |
b2f46e68 | 633 | kunmap_atomic(sb); |
ef99bf48 | 634 | out_no_sb: |
b405fe91 | 635 | if (test_bit(BITMAP_STALE, &bitmap->flags)) |
ef99bf48 N |
636 | bitmap->events_cleared = bitmap->mddev->events; |
637 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
638 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
639 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
1dff2b87 N |
640 | if (bitmap->mddev->bitmap_info.space == 0 || |
641 | bitmap->mddev->bitmap_info.space > sectors_reserved) | |
642 | bitmap->mddev->bitmap_info.space = sectors_reserved; | |
32a7627c N |
643 | if (err) |
644 | bitmap_print_sb(bitmap); | |
645 | return err; | |
646 | } | |
647 | ||
32a7627c N |
648 | /* |
649 | * general bitmap file operations | |
650 | */ | |
651 | ||
ece5cff0 N |
652 | /* |
653 | * on-disk bitmap: | |
654 | * | |
655 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
656 | * file a page at a time. There's a superblock at the start of the file. | |
657 | */ | |
32a7627c | 658 | /* calculate the index of the page that contains this bit */ |
1ec885cd N |
659 | static inline unsigned long file_page_index(struct bitmap_storage *store, |
660 | unsigned long chunk) | |
32a7627c | 661 | { |
1ec885cd | 662 | if (store->sb_page) |
ece5cff0 N |
663 | chunk += sizeof(bitmap_super_t) << 3; |
664 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
665 | } |
666 | ||
667 | /* calculate the (bit) offset of this bit within a page */ | |
1ec885cd N |
668 | static inline unsigned long file_page_offset(struct bitmap_storage *store, |
669 | unsigned long chunk) | |
32a7627c | 670 | { |
1ec885cd | 671 | if (store->sb_page) |
ece5cff0 N |
672 | chunk += sizeof(bitmap_super_t) << 3; |
673 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
674 | } |
675 | ||
676 | /* | |
677 | * return a pointer to the page in the filemap that contains the given bit | |
678 | * | |
679 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
680 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
681 | * 0 or page 1 | |
682 | */ | |
1ec885cd | 683 | static inline struct page *filemap_get_page(struct bitmap_storage *store, |
3520fa4d | 684 | unsigned long chunk) |
32a7627c | 685 | { |
1ec885cd | 686 | if (file_page_index(store, chunk) >= store->file_pages) |
ac2f40be | 687 | return NULL; |
1ec885cd N |
688 | return store->filemap[file_page_index(store, chunk) |
689 | - file_page_index(store, 0)]; | |
32a7627c N |
690 | } |
691 | ||
d1244cb0 N |
692 | static int bitmap_storage_alloc(struct bitmap_storage *store, |
693 | unsigned long chunks, int with_super) | |
694 | { | |
695 | int pnum; | |
696 | unsigned long num_pages; | |
697 | unsigned long bytes; | |
698 | ||
699 | bytes = DIV_ROUND_UP(chunks, 8); | |
700 | if (with_super) | |
701 | bytes += sizeof(bitmap_super_t); | |
702 | ||
703 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); | |
704 | ||
705 | store->filemap = kmalloc(sizeof(struct page *) | |
706 | * num_pages, GFP_KERNEL); | |
707 | if (!store->filemap) | |
708 | return -ENOMEM; | |
709 | ||
710 | if (with_super && !store->sb_page) { | |
d60b479d | 711 | store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
712 | if (store->sb_page == NULL) |
713 | return -ENOMEM; | |
714 | store->sb_page->index = 0; | |
715 | } | |
716 | pnum = 0; | |
717 | if (store->sb_page) { | |
718 | store->filemap[0] = store->sb_page; | |
719 | pnum = 1; | |
720 | } | |
721 | for ( ; pnum < num_pages; pnum++) { | |
d60b479d | 722 | store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
723 | if (!store->filemap[pnum]) { |
724 | store->file_pages = pnum; | |
725 | return -ENOMEM; | |
726 | } | |
727 | store->filemap[pnum]->index = pnum; | |
728 | } | |
729 | store->file_pages = pnum; | |
730 | ||
731 | /* We need 4 bits per page, rounded up to a multiple | |
732 | * of sizeof(unsigned long) */ | |
733 | store->filemap_attr = kzalloc( | |
734 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), | |
735 | GFP_KERNEL); | |
736 | if (!store->filemap_attr) | |
737 | return -ENOMEM; | |
738 | ||
739 | store->bytes = bytes; | |
740 | ||
741 | return 0; | |
742 | } | |
743 | ||
fae7d326 | 744 | static void bitmap_file_unmap(struct bitmap_storage *store) |
32a7627c N |
745 | { |
746 | struct page **map, *sb_page; | |
32a7627c | 747 | int pages; |
fae7d326 | 748 | struct file *file; |
32a7627c | 749 | |
fae7d326 | 750 | file = store->file; |
1ec885cd | 751 | map = store->filemap; |
1ec885cd | 752 | pages = store->file_pages; |
1ec885cd | 753 | sb_page = store->sb_page; |
32a7627c N |
754 | |
755 | while (pages--) | |
ece5cff0 | 756 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 757 | free_buffers(map[pages]); |
32a7627c | 758 | kfree(map); |
fae7d326 | 759 | kfree(store->filemap_attr); |
32a7627c | 760 | |
d785a06a N |
761 | if (sb_page) |
762 | free_buffers(sb_page); | |
32a7627c | 763 | |
d785a06a | 764 | if (file) { |
c649bb9c | 765 | struct inode *inode = file->f_path.dentry->d_inode; |
fc0ecff6 | 766 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 767 | fput(file); |
d785a06a | 768 | } |
32a7627c N |
769 | } |
770 | ||
32a7627c N |
771 | /* |
772 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
773 | * then it is no longer reliable, so we stop using it and we mark the file | |
774 | * as failed in the superblock | |
775 | */ | |
776 | static void bitmap_file_kick(struct bitmap *bitmap) | |
777 | { | |
778 | char *path, *ptr = NULL; | |
779 | ||
b405fe91 | 780 | if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) { |
4ad13663 | 781 | bitmap_update_sb(bitmap); |
32a7627c | 782 | |
1ec885cd | 783 | if (bitmap->storage.file) { |
4ad13663 N |
784 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); |
785 | if (path) | |
1ec885cd N |
786 | ptr = d_path(&bitmap->storage.file->f_path, |
787 | path, PAGE_SIZE); | |
6bcfd601 | 788 | |
4ad13663 N |
789 | printk(KERN_ALERT |
790 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 791 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 792 | |
4ad13663 N |
793 | kfree(path); |
794 | } else | |
795 | printk(KERN_ALERT | |
796 | "%s: disabling internal bitmap due to errors\n", | |
797 | bmname(bitmap)); | |
a654b9d8 | 798 | } |
32a7627c N |
799 | } |
800 | ||
801 | enum bitmap_page_attr { | |
ac2f40be | 802 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
5a537df4 N |
803 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
804 | * i.e. counter is 1 or 2. */ | |
ac2f40be | 805 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
32a7627c N |
806 | }; |
807 | ||
d189122d N |
808 | static inline void set_page_attr(struct bitmap *bitmap, int pnum, |
809 | enum bitmap_page_attr attr) | |
32a7627c | 810 | { |
bdfd1140 | 811 | set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
812 | } |
813 | ||
d189122d N |
814 | static inline void clear_page_attr(struct bitmap *bitmap, int pnum, |
815 | enum bitmap_page_attr attr) | |
32a7627c | 816 | { |
bdfd1140 | 817 | clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
818 | } |
819 | ||
bdfd1140 N |
820 | static inline int test_page_attr(struct bitmap *bitmap, int pnum, |
821 | enum bitmap_page_attr attr) | |
32a7627c | 822 | { |
1ec885cd | 823 | return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
824 | } |
825 | ||
bdfd1140 N |
826 | static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum, |
827 | enum bitmap_page_attr attr) | |
828 | { | |
829 | return test_and_clear_bit((pnum<<2) + attr, | |
830 | bitmap->storage.filemap_attr); | |
831 | } | |
32a7627c N |
832 | /* |
833 | * bitmap_file_set_bit -- called before performing a write to the md device | |
834 | * to set (and eventually sync) a particular bit in the bitmap file | |
835 | * | |
836 | * we set the bit immediately, then we record the page number so that | |
837 | * when an unplug occurs, we can flush the dirty pages out to disk | |
838 | */ | |
839 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
840 | { | |
841 | unsigned long bit; | |
3520fa4d | 842 | struct page *page; |
32a7627c | 843 | void *kaddr; |
40cffcc0 | 844 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
32a7627c | 845 | |
1ec885cd | 846 | page = filemap_get_page(&bitmap->storage, chunk); |
3520fa4d JB |
847 | if (!page) |
848 | return; | |
1ec885cd | 849 | bit = file_page_offset(&bitmap->storage, chunk); |
32a7627c | 850 | |
3520fa4d | 851 | /* set the bit */ |
b2f46e68 | 852 | kaddr = kmap_atomic(page); |
b405fe91 | 853 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
3520fa4d JB |
854 | set_bit(bit, kaddr); |
855 | else | |
63c68268 | 856 | test_and_set_bit_le(bit, kaddr); |
b2f46e68 | 857 | kunmap_atomic(kaddr); |
36a4e1fe | 858 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
32a7627c | 859 | /* record page number so it gets flushed to disk when unplug occurs */ |
d189122d | 860 | set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY); |
32a7627c N |
861 | } |
862 | ||
ef99bf48 N |
863 | static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) |
864 | { | |
865 | unsigned long bit; | |
866 | struct page *page; | |
867 | void *paddr; | |
40cffcc0 | 868 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
ef99bf48 | 869 | |
1ec885cd | 870 | page = filemap_get_page(&bitmap->storage, chunk); |
ef99bf48 N |
871 | if (!page) |
872 | return; | |
1ec885cd | 873 | bit = file_page_offset(&bitmap->storage, chunk); |
ef99bf48 | 874 | paddr = kmap_atomic(page); |
b405fe91 | 875 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ef99bf48 N |
876 | clear_bit(bit, paddr); |
877 | else | |
63c68268 | 878 | test_and_clear_bit_le(bit, paddr); |
ef99bf48 | 879 | kunmap_atomic(paddr); |
d189122d N |
880 | if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) { |
881 | set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING); | |
ef99bf48 N |
882 | bitmap->allclean = 0; |
883 | } | |
884 | } | |
885 | ||
32a7627c N |
886 | /* this gets called when the md device is ready to unplug its underlying |
887 | * (slave) device queues -- before we let any writes go down, we need to | |
888 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 889 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 890 | { |
74667123 | 891 | unsigned long i; |
ec7a3197 | 892 | int dirty, need_write; |
32a7627c N |
893 | int wait = 0; |
894 | ||
62f82faa N |
895 | if (!bitmap || !bitmap->storage.filemap || |
896 | test_bit(BITMAP_STALE, &bitmap->flags)) | |
4ad13663 | 897 | return; |
32a7627c N |
898 | |
899 | /* look at each page to see if there are any set bits that need to be | |
900 | * flushed out to disk */ | |
1ec885cd | 901 | for (i = 0; i < bitmap->storage.file_pages; i++) { |
bdfd1140 | 902 | if (!bitmap->storage.filemap) |
4ad13663 | 903 | return; |
bdfd1140 N |
904 | dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
905 | need_write = test_and_clear_page_attr(bitmap, i, | |
906 | BITMAP_PAGE_NEEDWRITE); | |
907 | if (dirty || need_write) { | |
d189122d | 908 | clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); |
bdfd1140 N |
909 | write_page(bitmap, bitmap->storage.filemap[i], 0); |
910 | } | |
ec7a3197 | 911 | if (dirty) |
32a7627c | 912 | wait = 1; |
32a7627c N |
913 | } |
914 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
1ec885cd | 915 | if (bitmap->storage.file) |
d785a06a N |
916 | wait_event(bitmap->write_wait, |
917 | atomic_read(&bitmap->pending_writes)==0); | |
0b79ccf0 | 918 | else |
a9701a30 | 919 | md_super_wait(bitmap->mddev); |
32a7627c | 920 | } |
b405fe91 | 921 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
d785a06a | 922 | bitmap_file_kick(bitmap); |
32a7627c | 923 | } |
ac2f40be | 924 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 925 | |
6a07997f | 926 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
927 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
928 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
929 | * memory mapping of the bitmap file | |
930 | * Special cases: | |
931 | * if there's no bitmap file, or if the bitmap file had been | |
932 | * previously kicked from the array, we mark all the bits as | |
933 | * 1's in order to cause a full resync. | |
6a07997f N |
934 | * |
935 | * We ignore all bits for sectors that end earlier than 'start'. | |
936 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 937 | */ |
6a07997f | 938 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
939 | { |
940 | unsigned long i, chunks, index, oldindex, bit; | |
27581e5a | 941 | struct page *page = NULL; |
d1244cb0 | 942 | unsigned long bit_cnt = 0; |
32a7627c | 943 | struct file *file; |
d1244cb0 | 944 | unsigned long offset; |
32a7627c N |
945 | int outofdate; |
946 | int ret = -ENOSPC; | |
ea03aff9 | 947 | void *paddr; |
1ec885cd | 948 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c | 949 | |
40cffcc0 | 950 | chunks = bitmap->counts.chunks; |
1ec885cd | 951 | file = store->file; |
32a7627c | 952 | |
ef99bf48 N |
953 | if (!file && !bitmap->mddev->bitmap_info.offset) { |
954 | /* No permanent bitmap - fill with '1s'. */ | |
1ec885cd N |
955 | store->filemap = NULL; |
956 | store->file_pages = 0; | |
ef99bf48 N |
957 | for (i = 0; i < chunks ; i++) { |
958 | /* if the disk bit is set, set the memory bit */ | |
40cffcc0 | 959 | int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift) |
ef99bf48 N |
960 | >= start); |
961 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 962 | (sector_t)i << bitmap->counts.chunkshift, |
ef99bf48 N |
963 | needed); |
964 | } | |
965 | return 0; | |
966 | } | |
32a7627c | 967 | |
b405fe91 | 968 | outofdate = test_bit(BITMAP_STALE, &bitmap->flags); |
32a7627c N |
969 | if (outofdate) |
970 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
971 | "recovery\n", bmname(bitmap)); | |
972 | ||
d1244cb0 | 973 | if (file && i_size_read(file->f_mapping->host) < store->bytes) { |
32a7627c | 974 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
d1244cb0 N |
975 | bmname(bitmap), |
976 | (unsigned long) i_size_read(file->f_mapping->host), | |
977 | store->bytes); | |
4ad13663 | 978 | goto err; |
32a7627c | 979 | } |
bc7f77de | 980 | |
d1244cb0 | 981 | oldindex = ~0L; |
27581e5a | 982 | offset = 0; |
d1244cb0 | 983 | if (!bitmap->mddev->bitmap_info.external) |
27581e5a | 984 | offset = sizeof(bitmap_super_t); |
32a7627c N |
985 | |
986 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 987 | int b; |
1ec885cd N |
988 | index = file_page_index(&bitmap->storage, i); |
989 | bit = file_page_offset(&bitmap->storage, i); | |
32a7627c | 990 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 991 | int count; |
32a7627c | 992 | /* unmap the old page, we're done with it */ |
d1244cb0 N |
993 | if (index == store->file_pages-1) |
994 | count = store->bytes - index * PAGE_SIZE; | |
d785a06a N |
995 | else |
996 | count = PAGE_SIZE; | |
1ec885cd | 997 | page = store->filemap[index]; |
27581e5a N |
998 | if (file) |
999 | ret = read_page(file, index, bitmap, | |
1000 | count, page); | |
1001 | else | |
1002 | ret = read_sb_page( | |
1003 | bitmap->mddev, | |
1004 | bitmap->mddev->bitmap_info.offset, | |
1005 | page, | |
1006 | index, count); | |
1007 | ||
1008 | if (ret) | |
4ad13663 | 1009 | goto err; |
a654b9d8 | 1010 | |
32a7627c | 1011 | oldindex = index; |
32a7627c N |
1012 | |
1013 | if (outofdate) { | |
1014 | /* | |
1015 | * if bitmap is out of date, dirty the | |
ac2f40be | 1016 | * whole page and write it out |
32a7627c | 1017 | */ |
b2f46e68 | 1018 | paddr = kmap_atomic(page); |
ea03aff9 | 1019 | memset(paddr + offset, 0xff, |
6a07997f | 1020 | PAGE_SIZE - offset); |
b2f46e68 | 1021 | kunmap_atomic(paddr); |
4ad13663 N |
1022 | write_page(bitmap, page, 1); |
1023 | ||
1024 | ret = -EIO; | |
b405fe91 N |
1025 | if (test_bit(BITMAP_WRITE_ERROR, |
1026 | &bitmap->flags)) | |
4ad13663 | 1027 | goto err; |
32a7627c | 1028 | } |
32a7627c | 1029 | } |
b2f46e68 | 1030 | paddr = kmap_atomic(page); |
b405fe91 | 1031 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ea03aff9 | 1032 | b = test_bit(bit, paddr); |
bd926c63 | 1033 | else |
6b33aff3 | 1034 | b = test_bit_le(bit, paddr); |
b2f46e68 | 1035 | kunmap_atomic(paddr); |
bd926c63 | 1036 | if (b) { |
32a7627c | 1037 | /* if the disk bit is set, set the memory bit */ |
40cffcc0 | 1038 | int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift |
db305e50 N |
1039 | >= start); |
1040 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 1041 | (sector_t)i << bitmap->counts.chunkshift, |
db305e50 | 1042 | needed); |
32a7627c N |
1043 | bit_cnt++; |
1044 | } | |
27581e5a | 1045 | offset = 0; |
32a7627c N |
1046 | } |
1047 | ||
32a7627c | 1048 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
d1244cb0 | 1049 | "read %lu pages, set %lu of %lu bits\n", |
1ec885cd | 1050 | bmname(bitmap), store->file_pages, |
d1244cb0 | 1051 | bit_cnt, chunks); |
4ad13663 N |
1052 | |
1053 | return 0; | |
32a7627c | 1054 | |
4ad13663 N |
1055 | err: |
1056 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1057 | bmname(bitmap), ret); | |
32a7627c N |
1058 | return ret; |
1059 | } | |
1060 | ||
a654b9d8 N |
1061 | void bitmap_write_all(struct bitmap *bitmap) |
1062 | { | |
1063 | /* We don't actually write all bitmap blocks here, | |
1064 | * just flag them as needing to be written | |
1065 | */ | |
ec7a3197 | 1066 | int i; |
a654b9d8 | 1067 | |
1ec885cd | 1068 | if (!bitmap || !bitmap->storage.filemap) |
ef99bf48 | 1069 | return; |
1ec885cd | 1070 | if (bitmap->storage.file) |
ef99bf48 N |
1071 | /* Only one copy, so nothing needed */ |
1072 | return; | |
1073 | ||
1ec885cd | 1074 | for (i = 0; i < bitmap->storage.file_pages; i++) |
d189122d | 1075 | set_page_attr(bitmap, i, |
ec7a3197 | 1076 | BITMAP_PAGE_NEEDWRITE); |
2585f3ef | 1077 | bitmap->allclean = 0; |
a654b9d8 N |
1078 | } |
1079 | ||
40cffcc0 N |
1080 | static void bitmap_count_page(struct bitmap_counts *bitmap, |
1081 | sector_t offset, int inc) | |
32a7627c | 1082 | { |
61a0d80c | 1083 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1084 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1085 | bitmap->bp[page].count += inc; | |
32a7627c N |
1086 | bitmap_checkfree(bitmap, page); |
1087 | } | |
bf07bb7d | 1088 | |
40cffcc0 | 1089 | static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset) |
bf07bb7d N |
1090 | { |
1091 | sector_t chunk = offset >> bitmap->chunkshift; | |
1092 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1093 | struct bitmap_page *bp = &bitmap->bp[page]; | |
1094 | ||
1095 | if (!bp->pending) | |
1096 | bp->pending = 1; | |
1097 | } | |
1098 | ||
40cffcc0 | 1099 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1100 | sector_t offset, sector_t *blocks, |
32a7627c N |
1101 | int create); |
1102 | ||
1103 | /* | |
1104 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1105 | * out to disk | |
1106 | */ | |
1107 | ||
fd01b88c | 1108 | void bitmap_daemon_work(struct mddev *mddev) |
32a7627c | 1109 | { |
aa5cbd10 | 1110 | struct bitmap *bitmap; |
aa3163f8 | 1111 | unsigned long j; |
bf07bb7d | 1112 | unsigned long nextpage; |
57dab0bd | 1113 | sector_t blocks; |
40cffcc0 | 1114 | struct bitmap_counts *counts; |
32a7627c | 1115 | |
aa5cbd10 N |
1116 | /* Use a mutex to guard daemon_work against |
1117 | * bitmap_destroy. | |
1118 | */ | |
c3d9714e | 1119 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1120 | bitmap = mddev->bitmap; |
1121 | if (bitmap == NULL) { | |
c3d9714e | 1122 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1123 | return; |
aa5cbd10 | 1124 | } |
42a04b50 | 1125 | if (time_before(jiffies, bitmap->daemon_lastrun |
2e61ebbc | 1126 | + mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1127 | goto done; |
1128 | ||
32a7627c | 1129 | bitmap->daemon_lastrun = jiffies; |
8311c29d | 1130 | if (bitmap->allclean) { |
2e61ebbc | 1131 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
aa5cbd10 | 1132 | goto done; |
8311c29d N |
1133 | } |
1134 | bitmap->allclean = 1; | |
32a7627c | 1135 | |
bf07bb7d N |
1136 | /* Any file-page which is PENDING now needs to be written. |
1137 | * So set NEEDWRITE now, then after we make any last-minute changes | |
1138 | * we will write it. | |
1139 | */ | |
1ec885cd | 1140 | for (j = 0; j < bitmap->storage.file_pages; j++) |
bdfd1140 N |
1141 | if (test_and_clear_page_attr(bitmap, j, |
1142 | BITMAP_PAGE_PENDING)) | |
d189122d | 1143 | set_page_attr(bitmap, j, |
bf07bb7d | 1144 | BITMAP_PAGE_NEEDWRITE); |
bf07bb7d N |
1145 | |
1146 | if (bitmap->need_sync && | |
1147 | mddev->bitmap_info.external == 0) { | |
1148 | /* Arrange for superblock update as well as | |
1149 | * other changes */ | |
1150 | bitmap_super_t *sb; | |
1151 | bitmap->need_sync = 0; | |
1ec885cd N |
1152 | if (bitmap->storage.filemap) { |
1153 | sb = kmap_atomic(bitmap->storage.sb_page); | |
ef99bf48 N |
1154 | sb->events_cleared = |
1155 | cpu_to_le64(bitmap->events_cleared); | |
1156 | kunmap_atomic(sb); | |
d189122d | 1157 | set_page_attr(bitmap, 0, |
ef99bf48 N |
1158 | BITMAP_PAGE_NEEDWRITE); |
1159 | } | |
bf07bb7d N |
1160 | } |
1161 | /* Now look at the bitmap counters and if any are '2' or '1', | |
1162 | * decrement and handle accordingly. | |
1163 | */ | |
40cffcc0 N |
1164 | counts = &bitmap->counts; |
1165 | spin_lock_irq(&counts->lock); | |
bf07bb7d | 1166 | nextpage = 0; |
40cffcc0 | 1167 | for (j = 0; j < counts->chunks; j++) { |
32a7627c | 1168 | bitmap_counter_t *bmc; |
40cffcc0 | 1169 | sector_t block = (sector_t)j << counts->chunkshift; |
3520fa4d | 1170 | |
bf07bb7d N |
1171 | if (j == nextpage) { |
1172 | nextpage += PAGE_COUNTER_RATIO; | |
40cffcc0 | 1173 | if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) { |
bf07bb7d | 1174 | j |= PAGE_COUNTER_MASK; |
aa3163f8 N |
1175 | continue; |
1176 | } | |
40cffcc0 | 1177 | counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; |
32a7627c | 1178 | } |
40cffcc0 | 1179 | bmc = bitmap_get_counter(counts, |
ef99bf48 | 1180 | block, |
db305e50 | 1181 | &blocks, 0); |
bf07bb7d N |
1182 | |
1183 | if (!bmc) { | |
5a537df4 | 1184 | j |= PAGE_COUNTER_MASK; |
bf07bb7d N |
1185 | continue; |
1186 | } | |
1187 | if (*bmc == 1 && !bitmap->need_sync) { | |
1188 | /* We can clear the bit */ | |
bf07bb7d | 1189 | *bmc = 0; |
40cffcc0 | 1190 | bitmap_count_page(counts, block, -1); |
ef99bf48 | 1191 | bitmap_file_clear_bit(bitmap, block); |
bf07bb7d N |
1192 | } else if (*bmc && *bmc <= 2) { |
1193 | *bmc = 1; | |
40cffcc0 | 1194 | bitmap_set_pending(counts, block); |
bf07bb7d | 1195 | bitmap->allclean = 0; |
5a537df4 | 1196 | } |
32a7627c | 1197 | } |
40cffcc0 | 1198 | spin_unlock_irq(&counts->lock); |
32a7627c | 1199 | |
bf07bb7d N |
1200 | /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. |
1201 | * DIRTY pages need to be written by bitmap_unplug so it can wait | |
1202 | * for them. | |
1203 | * If we find any DIRTY page we stop there and let bitmap_unplug | |
1204 | * handle all the rest. This is important in the case where | |
1205 | * the first blocking holds the superblock and it has been updated. | |
1206 | * We mustn't write any other blocks before the superblock. | |
1207 | */ | |
62f82faa N |
1208 | for (j = 0; |
1209 | j < bitmap->storage.file_pages | |
1210 | && !test_bit(BITMAP_STALE, &bitmap->flags); | |
1211 | j++) { | |
bf07bb7d | 1212 | |
d189122d | 1213 | if (test_page_attr(bitmap, j, |
bf07bb7d N |
1214 | BITMAP_PAGE_DIRTY)) |
1215 | /* bitmap_unplug will handle the rest */ | |
1216 | break; | |
bdfd1140 N |
1217 | if (test_and_clear_page_attr(bitmap, j, |
1218 | BITMAP_PAGE_NEEDWRITE)) { | |
1ec885cd | 1219 | write_page(bitmap, bitmap->storage.filemap[j], 0); |
32a7627c | 1220 | } |
32a7627c N |
1221 | } |
1222 | ||
7be3dfec | 1223 | done: |
8311c29d | 1224 | if (bitmap->allclean == 0) |
2e61ebbc N |
1225 | mddev->thread->timeout = |
1226 | mddev->bitmap_info.daemon_sleep; | |
c3d9714e | 1227 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1228 | } |
1229 | ||
40cffcc0 | 1230 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1231 | sector_t offset, sector_t *blocks, |
32a7627c | 1232 | int create) |
ee305ace N |
1233 | __releases(bitmap->lock) |
1234 | __acquires(bitmap->lock) | |
32a7627c N |
1235 | { |
1236 | /* If 'create', we might release the lock and reclaim it. | |
1237 | * The lock must have been taken with interrupts enabled. | |
1238 | * If !create, we don't release the lock. | |
1239 | */ | |
61a0d80c | 1240 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1241 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1242 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1243 | sector_t csize; | |
ef425673 | 1244 | int err; |
32a7627c | 1245 | |
ef425673 N |
1246 | err = bitmap_checkpage(bitmap, page, create); |
1247 | ||
1248 | if (bitmap->bp[page].hijacked || | |
1249 | bitmap->bp[page].map == NULL) | |
61a0d80c | 1250 | csize = ((sector_t)1) << (bitmap->chunkshift + |
ef425673 N |
1251 | PAGE_COUNTER_SHIFT - 1); |
1252 | else | |
61a0d80c | 1253 | csize = ((sector_t)1) << bitmap->chunkshift; |
ef425673 N |
1254 | *blocks = csize - (offset & (csize - 1)); |
1255 | ||
1256 | if (err < 0) | |
32a7627c | 1257 | return NULL; |
ef425673 | 1258 | |
32a7627c N |
1259 | /* now locked ... */ |
1260 | ||
1261 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1262 | /* should we use the first or second counter field | |
1263 | * of the hijacked pointer? */ | |
1264 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1265 | return &((bitmap_counter_t *) |
1266 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1267 | } else /* page is allocated */ |
32a7627c N |
1268 | return (bitmap_counter_t *) |
1269 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1270 | } |
1271 | ||
4b6d287f | 1272 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1273 | { |
ac2f40be N |
1274 | if (!bitmap) |
1275 | return 0; | |
4b6d287f N |
1276 | |
1277 | if (behind) { | |
696fcd53 | 1278 | int bw; |
4b6d287f | 1279 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1280 | bw = atomic_read(&bitmap->behind_writes); |
1281 | if (bw > bitmap->behind_writes_used) | |
1282 | bitmap->behind_writes_used = bw; | |
1283 | ||
36a4e1fe N |
1284 | pr_debug("inc write-behind count %d/%lu\n", |
1285 | bw, bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1286 | } |
1287 | ||
32a7627c | 1288 | while (sectors) { |
57dab0bd | 1289 | sector_t blocks; |
32a7627c N |
1290 | bitmap_counter_t *bmc; |
1291 | ||
40cffcc0 N |
1292 | spin_lock_irq(&bitmap->counts.lock); |
1293 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1); | |
32a7627c | 1294 | if (!bmc) { |
40cffcc0 | 1295 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1296 | return 0; |
1297 | } | |
1298 | ||
27d5ea04 | 1299 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
da6e1a32 NB |
1300 | DEFINE_WAIT(__wait); |
1301 | /* note that it is safe to do the prepare_to_wait | |
1302 | * after the test as long as we do it before dropping | |
1303 | * the spinlock. | |
1304 | */ | |
1305 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1306 | TASK_UNINTERRUPTIBLE); | |
40cffcc0 | 1307 | spin_unlock_irq(&bitmap->counts.lock); |
7eaceacc | 1308 | io_schedule(); |
da6e1a32 NB |
1309 | finish_wait(&bitmap->overflow_wait, &__wait); |
1310 | continue; | |
1311 | } | |
1312 | ||
ac2f40be | 1313 | switch (*bmc) { |
32a7627c N |
1314 | case 0: |
1315 | bitmap_file_set_bit(bitmap, offset); | |
40cffcc0 | 1316 | bitmap_count_page(&bitmap->counts, offset, 1); |
32a7627c N |
1317 | /* fall through */ |
1318 | case 1: | |
1319 | *bmc = 2; | |
1320 | } | |
da6e1a32 | 1321 | |
32a7627c N |
1322 | (*bmc)++; |
1323 | ||
40cffcc0 | 1324 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1325 | |
1326 | offset += blocks; | |
1327 | if (sectors > blocks) | |
1328 | sectors -= blocks; | |
ac2f40be N |
1329 | else |
1330 | sectors = 0; | |
32a7627c N |
1331 | } |
1332 | return 0; | |
1333 | } | |
ac2f40be | 1334 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1335 | |
1336 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1337 | int success, int behind) |
32a7627c | 1338 | { |
ac2f40be N |
1339 | if (!bitmap) |
1340 | return; | |
4b6d287f | 1341 | if (behind) { |
e555190d N |
1342 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1343 | wake_up(&bitmap->behind_wait); | |
36a4e1fe N |
1344 | pr_debug("dec write-behind count %d/%lu\n", |
1345 | atomic_read(&bitmap->behind_writes), | |
1346 | bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1347 | } |
1348 | ||
32a7627c | 1349 | while (sectors) { |
57dab0bd | 1350 | sector_t blocks; |
32a7627c N |
1351 | unsigned long flags; |
1352 | bitmap_counter_t *bmc; | |
1353 | ||
40cffcc0 N |
1354 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1355 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0); | |
32a7627c | 1356 | if (!bmc) { |
40cffcc0 | 1357 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1358 | return; |
1359 | } | |
1360 | ||
961902c0 | 1361 | if (success && !bitmap->mddev->degraded && |
a0da84f3 NB |
1362 | bitmap->events_cleared < bitmap->mddev->events) { |
1363 | bitmap->events_cleared = bitmap->mddev->events; | |
1364 | bitmap->need_sync = 1; | |
5ff5afff | 1365 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1366 | } |
1367 | ||
27d5ea04 | 1368 | if (!success && !NEEDED(*bmc)) |
32a7627c N |
1369 | *bmc |= NEEDED_MASK; |
1370 | ||
27d5ea04 | 1371 | if (COUNTER(*bmc) == COUNTER_MAX) |
da6e1a32 NB |
1372 | wake_up(&bitmap->overflow_wait); |
1373 | ||
32a7627c | 1374 | (*bmc)--; |
2585f3ef | 1375 | if (*bmc <= 2) { |
40cffcc0 | 1376 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1377 | bitmap->allclean = 0; |
1378 | } | |
40cffcc0 | 1379 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1380 | offset += blocks; |
1381 | if (sectors > blocks) | |
1382 | sectors -= blocks; | |
ac2f40be N |
1383 | else |
1384 | sectors = 0; | |
32a7627c N |
1385 | } |
1386 | } | |
ac2f40be | 1387 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1388 | |
57dab0bd | 1389 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1390 | int degraded) |
32a7627c N |
1391 | { |
1392 | bitmap_counter_t *bmc; | |
1393 | int rv; | |
1394 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1395 | *blocks = 1024; | |
1396 | return 1; /* always resync if no bitmap */ | |
1397 | } | |
40cffcc0 N |
1398 | spin_lock_irq(&bitmap->counts.lock); |
1399 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1400 | rv = 0; |
1401 | if (bmc) { | |
1402 | /* locked */ | |
1403 | if (RESYNC(*bmc)) | |
1404 | rv = 1; | |
1405 | else if (NEEDED(*bmc)) { | |
1406 | rv = 1; | |
6a806c51 N |
1407 | if (!degraded) { /* don't set/clear bits if degraded */ |
1408 | *bmc |= RESYNC_MASK; | |
1409 | *bmc &= ~NEEDED_MASK; | |
1410 | } | |
32a7627c N |
1411 | } |
1412 | } | |
40cffcc0 | 1413 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1414 | return rv; |
1415 | } | |
1416 | ||
57dab0bd | 1417 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a N |
1418 | int degraded) |
1419 | { | |
1420 | /* bitmap_start_sync must always report on multiples of whole | |
1421 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1422 | * get confused. | |
1423 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1424 | * At least PAGE_SIZE>>9 blocks are covered. | |
1425 | * Return the 'or' of the result. | |
1426 | */ | |
1427 | int rv = 0; | |
57dab0bd | 1428 | sector_t blocks1; |
1187cf0a N |
1429 | |
1430 | *blocks = 0; | |
1431 | while (*blocks < (PAGE_SIZE>>9)) { | |
1432 | rv |= __bitmap_start_sync(bitmap, offset, | |
1433 | &blocks1, degraded); | |
1434 | offset += blocks1; | |
1435 | *blocks += blocks1; | |
1436 | } | |
1437 | return rv; | |
1438 | } | |
ac2f40be | 1439 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1440 | |
57dab0bd | 1441 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1442 | { |
1443 | bitmap_counter_t *bmc; | |
1444 | unsigned long flags; | |
ac2f40be N |
1445 | |
1446 | if (bitmap == NULL) { | |
32a7627c N |
1447 | *blocks = 1024; |
1448 | return; | |
1449 | } | |
40cffcc0 N |
1450 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1451 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1452 | if (bmc == NULL) |
1453 | goto unlock; | |
1454 | /* locked */ | |
32a7627c N |
1455 | if (RESYNC(*bmc)) { |
1456 | *bmc &= ~RESYNC_MASK; | |
1457 | ||
1458 | if (!NEEDED(*bmc) && aborted) | |
1459 | *bmc |= NEEDED_MASK; | |
1460 | else { | |
2585f3ef | 1461 | if (*bmc <= 2) { |
40cffcc0 | 1462 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1463 | bitmap->allclean = 0; |
1464 | } | |
32a7627c N |
1465 | } |
1466 | } | |
1467 | unlock: | |
40cffcc0 | 1468 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c | 1469 | } |
ac2f40be | 1470 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1471 | |
1472 | void bitmap_close_sync(struct bitmap *bitmap) | |
1473 | { | |
1474 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1475 | * properly have been aborted. It remains to us to clear the | |
1476 | * RESYNC bit wherever it is still on | |
1477 | */ | |
1478 | sector_t sector = 0; | |
57dab0bd | 1479 | sector_t blocks; |
b47490c9 N |
1480 | if (!bitmap) |
1481 | return; | |
32a7627c N |
1482 | while (sector < bitmap->mddev->resync_max_sectors) { |
1483 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1484 | sector += blocks; |
1485 | } | |
1486 | } | |
ac2f40be | 1487 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 N |
1488 | |
1489 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1490 | { | |
1491 | sector_t s = 0; | |
57dab0bd | 1492 | sector_t blocks; |
b47490c9 N |
1493 | |
1494 | if (!bitmap) | |
1495 | return; | |
1496 | if (sector == 0) { | |
1497 | bitmap->last_end_sync = jiffies; | |
1498 | return; | |
1499 | } | |
1500 | if (time_before(jiffies, (bitmap->last_end_sync | |
1b04be96 | 1501 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1502 | return; |
1503 | wait_event(bitmap->mddev->recovery_wait, | |
1504 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1505 | ||
75d3da43 | 1506 | bitmap->mddev->curr_resync_completed = sector; |
070dc6dd | 1507 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
40cffcc0 | 1508 | sector &= ~((1ULL << bitmap->counts.chunkshift) - 1); |
b47490c9 N |
1509 | s = 0; |
1510 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1511 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1512 | s += blocks; | |
32a7627c | 1513 | } |
b47490c9 | 1514 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1515 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1516 | } |
ac2f40be | 1517 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1518 | |
6a07997f | 1519 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1520 | { |
1521 | /* For each chunk covered by any of these sectors, set the | |
ef99bf48 | 1522 | * counter to 2 and possibly set resync_needed. They should all |
32a7627c N |
1523 | * be 0 at this point |
1524 | */ | |
193f1c93 | 1525 | |
57dab0bd | 1526 | sector_t secs; |
193f1c93 | 1527 | bitmap_counter_t *bmc; |
40cffcc0 N |
1528 | spin_lock_irq(&bitmap->counts.lock); |
1529 | bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1); | |
193f1c93 | 1530 | if (!bmc) { |
40cffcc0 | 1531 | spin_unlock_irq(&bitmap->counts.lock); |
193f1c93 | 1532 | return; |
32a7627c | 1533 | } |
ac2f40be | 1534 | if (!*bmc) { |
915c420d | 1535 | *bmc = 2 | (needed ? NEEDED_MASK : 0); |
40cffcc0 N |
1536 | bitmap_count_page(&bitmap->counts, offset, 1); |
1537 | bitmap_set_pending(&bitmap->counts, offset); | |
2585f3ef | 1538 | bitmap->allclean = 0; |
193f1c93 | 1539 | } |
40cffcc0 | 1540 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1541 | } |
1542 | ||
9b1d1dac PC |
1543 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1544 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1545 | { | |
1546 | unsigned long chunk; | |
1547 | ||
1548 | for (chunk = s; chunk <= e; chunk++) { | |
40cffcc0 | 1549 | sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift; |
9b1d1dac PC |
1550 | bitmap_set_memory_bits(bitmap, sec, 1); |
1551 | bitmap_file_set_bit(bitmap, sec); | |
ffa23322 N |
1552 | if (sec < bitmap->mddev->recovery_cp) |
1553 | /* We are asserting that the array is dirty, | |
1554 | * so move the recovery_cp address back so | |
1555 | * that it is obvious that it is dirty | |
1556 | */ | |
1557 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1558 | } |
1559 | } | |
1560 | ||
6b8b3e8a N |
1561 | /* |
1562 | * flush out any pending updates | |
1563 | */ | |
fd01b88c | 1564 | void bitmap_flush(struct mddev *mddev) |
6b8b3e8a N |
1565 | { |
1566 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1567 | long sleep; |
6b8b3e8a N |
1568 | |
1569 | if (!bitmap) /* there was no bitmap */ | |
1570 | return; | |
1571 | ||
1572 | /* run the daemon_work three time to ensure everything is flushed | |
1573 | * that can be | |
1574 | */ | |
1b04be96 | 1575 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1576 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1577 | bitmap_daemon_work(mddev); |
42a04b50 | 1578 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1579 | bitmap_daemon_work(mddev); |
42a04b50 | 1580 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1581 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1582 | bitmap_update_sb(bitmap); |
1583 | } | |
1584 | ||
32a7627c N |
1585 | /* |
1586 | * free memory that was allocated | |
1587 | */ | |
3178b0db | 1588 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1589 | { |
1590 | unsigned long k, pages; | |
1591 | struct bitmap_page *bp; | |
32a7627c N |
1592 | |
1593 | if (!bitmap) /* there was no bitmap */ | |
1594 | return; | |
1595 | ||
fae7d326 N |
1596 | /* Shouldn't be needed - but just in case.... */ |
1597 | wait_event(bitmap->write_wait, | |
1598 | atomic_read(&bitmap->pending_writes) == 0); | |
1599 | ||
1600 | /* release the bitmap file */ | |
1601 | bitmap_file_unmap(&bitmap->storage); | |
32a7627c | 1602 | |
40cffcc0 N |
1603 | bp = bitmap->counts.bp; |
1604 | pages = bitmap->counts.pages; | |
32a7627c N |
1605 | |
1606 | /* free all allocated memory */ | |
1607 | ||
32a7627c N |
1608 | if (bp) /* deallocate the page memory */ |
1609 | for (k = 0; k < pages; k++) | |
1610 | if (bp[k].map && !bp[k].hijacked) | |
1611 | kfree(bp[k].map); | |
1612 | kfree(bp); | |
1613 | kfree(bitmap); | |
1614 | } | |
aa5cbd10 | 1615 | |
fd01b88c | 1616 | void bitmap_destroy(struct mddev *mddev) |
3178b0db N |
1617 | { |
1618 | struct bitmap *bitmap = mddev->bitmap; | |
1619 | ||
1620 | if (!bitmap) /* there was no bitmap */ | |
1621 | return; | |
1622 | ||
c3d9714e | 1623 | mutex_lock(&mddev->bitmap_info.mutex); |
3178b0db | 1624 | mddev->bitmap = NULL; /* disconnect from the md device */ |
c3d9714e | 1625 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1626 | if (mddev->thread) |
1627 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db | 1628 | |
ece5cff0 N |
1629 | if (bitmap->sysfs_can_clear) |
1630 | sysfs_put(bitmap->sysfs_can_clear); | |
1631 | ||
3178b0db N |
1632 | bitmap_free(bitmap); |
1633 | } | |
32a7627c N |
1634 | |
1635 | /* | |
1636 | * initialize the bitmap structure | |
1637 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1638 | */ | |
fd01b88c | 1639 | int bitmap_create(struct mddev *mddev) |
32a7627c N |
1640 | { |
1641 | struct bitmap *bitmap; | |
1f593903 | 1642 | sector_t blocks = mddev->resync_max_sectors; |
c3d9714e | 1643 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1644 | int err; |
5ff5afff | 1645 | struct sysfs_dirent *bm = NULL; |
32a7627c | 1646 | |
5f6e3c83 | 1647 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1648 | |
c3d9714e | 1649 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1650 | |
9ffae0cf | 1651 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1652 | if (!bitmap) |
1653 | return -ENOMEM; | |
1654 | ||
40cffcc0 | 1655 | spin_lock_init(&bitmap->counts.lock); |
ce25c31b N |
1656 | atomic_set(&bitmap->pending_writes, 0); |
1657 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1658 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1659 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1660 | |
32a7627c | 1661 | bitmap->mddev = mddev; |
32a7627c | 1662 | |
5ff5afff N |
1663 | if (mddev->kobj.sd) |
1664 | bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap"); | |
ece5cff0 | 1665 | if (bm) { |
3ff195b0 | 1666 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear"); |
ece5cff0 N |
1667 | sysfs_put(bm); |
1668 | } else | |
1669 | bitmap->sysfs_can_clear = NULL; | |
1670 | ||
1ec885cd | 1671 | bitmap->storage.file = file; |
ce25c31b N |
1672 | if (file) { |
1673 | get_file(file); | |
ae8fa283 N |
1674 | /* As future accesses to this file will use bmap, |
1675 | * and bypass the page cache, we must sync the file | |
1676 | * first. | |
1677 | */ | |
8018ab05 | 1678 | vfs_fsync(file, 1); |
ce25c31b | 1679 | } |
42a04b50 | 1680 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
9c81075f JB |
1681 | if (!mddev->bitmap_info.external) { |
1682 | /* | |
1683 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is | |
1684 | * instructing us to create a new on-disk bitmap instance. | |
1685 | */ | |
1686 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) | |
1687 | err = bitmap_new_disk_sb(bitmap); | |
1688 | else | |
1689 | err = bitmap_read_sb(bitmap); | |
1690 | } else { | |
ece5cff0 N |
1691 | err = 0; |
1692 | if (mddev->bitmap_info.chunksize == 0 || | |
1693 | mddev->bitmap_info.daemon_sleep == 0) | |
1694 | /* chunksize and time_base need to be | |
1695 | * set first. */ | |
1696 | err = -EINVAL; | |
1697 | } | |
32a7627c | 1698 | if (err) |
3178b0db | 1699 | goto error; |
32a7627c | 1700 | |
624ce4f5 | 1701 | bitmap->daemon_lastrun = jiffies; |
d60b479d N |
1702 | err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1); |
1703 | if (err) | |
3178b0db | 1704 | goto error; |
32a7627c | 1705 | |
69e51b44 | 1706 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", |
d60b479d | 1707 | bitmap->counts.pages, bmname(bitmap)); |
69e51b44 N |
1708 | |
1709 | mddev->bitmap = bitmap; | |
b405fe91 | 1710 | return test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0; |
69e51b44 N |
1711 | |
1712 | error: | |
1713 | bitmap_free(bitmap); | |
1714 | return err; | |
1715 | } | |
1716 | ||
fd01b88c | 1717 | int bitmap_load(struct mddev *mddev) |
69e51b44 N |
1718 | { |
1719 | int err = 0; | |
3520fa4d | 1720 | sector_t start = 0; |
69e51b44 N |
1721 | sector_t sector = 0; |
1722 | struct bitmap *bitmap = mddev->bitmap; | |
1723 | ||
1724 | if (!bitmap) | |
1725 | goto out; | |
1726 | ||
1727 | /* Clear out old bitmap info first: Either there is none, or we | |
1728 | * are resuming after someone else has possibly changed things, | |
1729 | * so we should forget old cached info. | |
1730 | * All chunks should be clean, but some might need_sync. | |
1731 | */ | |
1732 | while (sector < mddev->resync_max_sectors) { | |
57dab0bd | 1733 | sector_t blocks; |
69e51b44 N |
1734 | bitmap_start_sync(bitmap, sector, &blocks, 0); |
1735 | sector += blocks; | |
1736 | } | |
1737 | bitmap_close_sync(bitmap); | |
1738 | ||
3520fa4d JB |
1739 | if (mddev->degraded == 0 |
1740 | || bitmap->events_cleared == mddev->events) | |
1741 | /* no need to keep dirty bits to optimise a | |
1742 | * re-add of a missing device */ | |
1743 | start = mddev->recovery_cp; | |
1744 | ||
afbaa90b | 1745 | mutex_lock(&mddev->bitmap_info.mutex); |
3520fa4d | 1746 | err = bitmap_init_from_disk(bitmap, start); |
afbaa90b | 1747 | mutex_unlock(&mddev->bitmap_info.mutex); |
3520fa4d | 1748 | |
32a7627c | 1749 | if (err) |
69e51b44 | 1750 | goto out; |
b405fe91 | 1751 | clear_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
1752 | |
1753 | /* Kick recovery in case any bits were set */ | |
1754 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
3178b0db | 1755 | |
1b04be96 | 1756 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1757 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1758 | |
4ad13663 N |
1759 | bitmap_update_sb(bitmap); |
1760 | ||
b405fe91 | 1761 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
69e51b44 N |
1762 | err = -EIO; |
1763 | out: | |
3178b0db | 1764 | return err; |
32a7627c | 1765 | } |
69e51b44 | 1766 | EXPORT_SYMBOL_GPL(bitmap_load); |
32a7627c | 1767 | |
57148964 N |
1768 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap) |
1769 | { | |
1770 | unsigned long chunk_kb; | |
40cffcc0 | 1771 | struct bitmap_counts *counts; |
57148964 N |
1772 | |
1773 | if (!bitmap) | |
1774 | return; | |
1775 | ||
40cffcc0 N |
1776 | counts = &bitmap->counts; |
1777 | ||
57148964 N |
1778 | chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; |
1779 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
1780 | "%lu%s chunk", | |
40cffcc0 N |
1781 | counts->pages - counts->missing_pages, |
1782 | counts->pages, | |
1783 | (counts->pages - counts->missing_pages) | |
57148964 N |
1784 | << (PAGE_SHIFT - 10), |
1785 | chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, | |
1786 | chunk_kb ? "KB" : "B"); | |
1ec885cd | 1787 | if (bitmap->storage.file) { |
57148964 | 1788 | seq_printf(seq, ", file: "); |
1ec885cd | 1789 | seq_path(seq, &bitmap->storage.file->f_path, " \t\n"); |
57148964 N |
1790 | } |
1791 | ||
1792 | seq_printf(seq, "\n"); | |
57148964 N |
1793 | } |
1794 | ||
d60b479d N |
1795 | int bitmap_resize(struct bitmap *bitmap, sector_t blocks, |
1796 | int chunksize, int init) | |
1797 | { | |
1798 | /* If chunk_size is 0, choose an appropriate chunk size. | |
1799 | * Then possibly allocate new storage space. | |
1800 | * Then quiesce, copy bits, replace bitmap, and re-start | |
1801 | * | |
1802 | * This function is called both to set up the initial bitmap | |
1803 | * and to resize the bitmap while the array is active. | |
1804 | * If this happens as a result of the array being resized, | |
1805 | * chunksize will be zero, and we need to choose a suitable | |
1806 | * chunksize, otherwise we use what we are given. | |
1807 | */ | |
1808 | struct bitmap_storage store; | |
1809 | struct bitmap_counts old_counts; | |
1810 | unsigned long chunks; | |
1811 | sector_t block; | |
1812 | sector_t old_blocks, new_blocks; | |
1813 | int chunkshift; | |
1814 | int ret = 0; | |
1815 | long pages; | |
1816 | struct bitmap_page *new_bp; | |
1817 | ||
1818 | if (chunksize == 0) { | |
1819 | /* If there is enough space, leave the chunk size unchanged, | |
1820 | * else increase by factor of two until there is enough space. | |
1821 | */ | |
1822 | long bytes; | |
1823 | long space = bitmap->mddev->bitmap_info.space; | |
1824 | ||
1825 | if (space == 0) { | |
1826 | /* We don't know how much space there is, so limit | |
1827 | * to current size - in sectors. | |
1828 | */ | |
1829 | bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8); | |
1830 | if (!bitmap->mddev->bitmap_info.external) | |
1831 | bytes += sizeof(bitmap_super_t); | |
1832 | space = DIV_ROUND_UP(bytes, 512); | |
1833 | bitmap->mddev->bitmap_info.space = space; | |
1834 | } | |
1835 | chunkshift = bitmap->counts.chunkshift; | |
1836 | chunkshift--; | |
1837 | do { | |
1838 | /* 'chunkshift' is shift from block size to chunk size */ | |
1839 | chunkshift++; | |
1840 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
1841 | bytes = DIV_ROUND_UP(chunks, 8); | |
1842 | if (!bitmap->mddev->bitmap_info.external) | |
1843 | bytes += sizeof(bitmap_super_t); | |
1844 | } while (bytes > (space << 9)); | |
1845 | } else | |
1846 | chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT; | |
1847 | ||
1848 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
1849 | memset(&store, 0, sizeof(store)); | |
1850 | if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file) | |
1851 | ret = bitmap_storage_alloc(&store, chunks, | |
1852 | !bitmap->mddev->bitmap_info.external); | |
1853 | if (ret) | |
1854 | goto err; | |
1855 | ||
1856 | pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO); | |
1857 | ||
1858 | new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL); | |
1859 | ret = -ENOMEM; | |
1860 | if (!new_bp) { | |
1861 | bitmap_file_unmap(&store); | |
1862 | goto err; | |
1863 | } | |
1864 | ||
1865 | if (!init) | |
1866 | bitmap->mddev->pers->quiesce(bitmap->mddev, 1); | |
1867 | ||
1868 | store.file = bitmap->storage.file; | |
1869 | bitmap->storage.file = NULL; | |
1870 | ||
1871 | if (store.sb_page && bitmap->storage.sb_page) | |
1872 | memcpy(page_address(store.sb_page), | |
1873 | page_address(bitmap->storage.sb_page), | |
1874 | sizeof(bitmap_super_t)); | |
1875 | bitmap_file_unmap(&bitmap->storage); | |
1876 | bitmap->storage = store; | |
1877 | ||
1878 | old_counts = bitmap->counts; | |
1879 | bitmap->counts.bp = new_bp; | |
1880 | bitmap->counts.pages = pages; | |
1881 | bitmap->counts.missing_pages = pages; | |
1882 | bitmap->counts.chunkshift = chunkshift; | |
1883 | bitmap->counts.chunks = chunks; | |
1884 | bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift + | |
1885 | BITMAP_BLOCK_SHIFT); | |
1886 | ||
1887 | blocks = min(old_counts.chunks << old_counts.chunkshift, | |
1888 | chunks << chunkshift); | |
1889 | ||
1890 | spin_lock_irq(&bitmap->counts.lock); | |
1891 | for (block = 0; block < blocks; ) { | |
1892 | bitmap_counter_t *bmc_old, *bmc_new; | |
1893 | int set; | |
1894 | ||
1895 | bmc_old = bitmap_get_counter(&old_counts, block, | |
1896 | &old_blocks, 0); | |
1897 | set = bmc_old && NEEDED(*bmc_old); | |
1898 | ||
1899 | if (set) { | |
1900 | bmc_new = bitmap_get_counter(&bitmap->counts, block, | |
1901 | &new_blocks, 1); | |
1902 | if (*bmc_new == 0) { | |
1903 | /* need to set on-disk bits too. */ | |
1904 | sector_t end = block + new_blocks; | |
1905 | sector_t start = block >> chunkshift; | |
1906 | start <<= chunkshift; | |
1907 | while (start < end) { | |
1908 | bitmap_file_set_bit(bitmap, block); | |
1909 | start += 1 << chunkshift; | |
1910 | } | |
1911 | *bmc_new = 2; | |
1912 | bitmap_count_page(&bitmap->counts, | |
1913 | block, 1); | |
1914 | bitmap_set_pending(&bitmap->counts, | |
1915 | block); | |
1916 | } | |
1917 | *bmc_new |= NEEDED_MASK; | |
1918 | if (new_blocks < old_blocks) | |
1919 | old_blocks = new_blocks; | |
1920 | } | |
1921 | block += old_blocks; | |
1922 | } | |
1923 | ||
1924 | if (!init) { | |
1925 | int i; | |
1926 | while (block < (chunks << chunkshift)) { | |
1927 | bitmap_counter_t *bmc; | |
1928 | bmc = bitmap_get_counter(&bitmap->counts, block, | |
1929 | &new_blocks, 1); | |
1930 | if (bmc) { | |
1931 | /* new space. It needs to be resynced, so | |
1932 | * we set NEEDED_MASK. | |
1933 | */ | |
1934 | if (*bmc == 0) { | |
1935 | *bmc = NEEDED_MASK | 2; | |
1936 | bitmap_count_page(&bitmap->counts, | |
1937 | block, 1); | |
1938 | bitmap_set_pending(&bitmap->counts, | |
1939 | block); | |
1940 | } | |
1941 | } | |
1942 | block += new_blocks; | |
1943 | } | |
1944 | for (i = 0; i < bitmap->storage.file_pages; i++) | |
1945 | set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); | |
1946 | } | |
1947 | spin_unlock_irq(&bitmap->counts.lock); | |
1948 | ||
1949 | if (!init) { | |
1950 | bitmap_unplug(bitmap); | |
1951 | bitmap->mddev->pers->quiesce(bitmap->mddev, 0); | |
1952 | } | |
1953 | ret = 0; | |
1954 | err: | |
1955 | return ret; | |
1956 | } | |
1957 | EXPORT_SYMBOL_GPL(bitmap_resize); | |
1958 | ||
43a70507 | 1959 | static ssize_t |
fd01b88c | 1960 | location_show(struct mddev *mddev, char *page) |
43a70507 N |
1961 | { |
1962 | ssize_t len; | |
ac2f40be | 1963 | if (mddev->bitmap_info.file) |
43a70507 | 1964 | len = sprintf(page, "file"); |
ac2f40be | 1965 | else if (mddev->bitmap_info.offset) |
43a70507 | 1966 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 1967 | else |
43a70507 N |
1968 | len = sprintf(page, "none"); |
1969 | len += sprintf(page+len, "\n"); | |
1970 | return len; | |
1971 | } | |
1972 | ||
1973 | static ssize_t | |
fd01b88c | 1974 | location_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
1975 | { |
1976 | ||
1977 | if (mddev->pers) { | |
1978 | if (!mddev->pers->quiesce) | |
1979 | return -EBUSY; | |
1980 | if (mddev->recovery || mddev->sync_thread) | |
1981 | return -EBUSY; | |
1982 | } | |
1983 | ||
1984 | if (mddev->bitmap || mddev->bitmap_info.file || | |
1985 | mddev->bitmap_info.offset) { | |
1986 | /* bitmap already configured. Only option is to clear it */ | |
1987 | if (strncmp(buf, "none", 4) != 0) | |
1988 | return -EBUSY; | |
1989 | if (mddev->pers) { | |
1990 | mddev->pers->quiesce(mddev, 1); | |
1991 | bitmap_destroy(mddev); | |
1992 | mddev->pers->quiesce(mddev, 0); | |
1993 | } | |
1994 | mddev->bitmap_info.offset = 0; | |
1995 | if (mddev->bitmap_info.file) { | |
1996 | struct file *f = mddev->bitmap_info.file; | |
1997 | mddev->bitmap_info.file = NULL; | |
1998 | restore_bitmap_write_access(f); | |
1999 | fput(f); | |
2000 | } | |
2001 | } else { | |
2002 | /* No bitmap, OK to set a location */ | |
2003 | long long offset; | |
2004 | if (strncmp(buf, "none", 4) == 0) | |
2005 | /* nothing to be done */; | |
2006 | else if (strncmp(buf, "file:", 5) == 0) { | |
2007 | /* Not supported yet */ | |
2008 | return -EINVAL; | |
2009 | } else { | |
2010 | int rv; | |
2011 | if (buf[0] == '+') | |
2012 | rv = strict_strtoll(buf+1, 10, &offset); | |
2013 | else | |
2014 | rv = strict_strtoll(buf, 10, &offset); | |
2015 | if (rv) | |
2016 | return rv; | |
2017 | if (offset == 0) | |
2018 | return -EINVAL; | |
ece5cff0 N |
2019 | if (mddev->bitmap_info.external == 0 && |
2020 | mddev->major_version == 0 && | |
43a70507 N |
2021 | offset != mddev->bitmap_info.default_offset) |
2022 | return -EINVAL; | |
2023 | mddev->bitmap_info.offset = offset; | |
2024 | if (mddev->pers) { | |
2025 | mddev->pers->quiesce(mddev, 1); | |
2026 | rv = bitmap_create(mddev); | |
4474ca42 N |
2027 | if (!rv) |
2028 | rv = bitmap_load(mddev); | |
43a70507 N |
2029 | if (rv) { |
2030 | bitmap_destroy(mddev); | |
2031 | mddev->bitmap_info.offset = 0; | |
2032 | } | |
2033 | mddev->pers->quiesce(mddev, 0); | |
2034 | if (rv) | |
2035 | return rv; | |
2036 | } | |
2037 | } | |
2038 | } | |
2039 | if (!mddev->external) { | |
2040 | /* Ensure new bitmap info is stored in | |
2041 | * metadata promptly. | |
2042 | */ | |
2043 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2044 | md_wakeup_thread(mddev->thread); | |
2045 | } | |
2046 | return len; | |
2047 | } | |
2048 | ||
2049 | static struct md_sysfs_entry bitmap_location = | |
2050 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
2051 | ||
6409bb05 N |
2052 | /* 'bitmap/space' is the space available at 'location' for the |
2053 | * bitmap. This allows the kernel to know when it is safe to | |
2054 | * resize the bitmap to match a resized array. | |
2055 | */ | |
2056 | static ssize_t | |
2057 | space_show(struct mddev *mddev, char *page) | |
2058 | { | |
2059 | return sprintf(page, "%lu\n", mddev->bitmap_info.space); | |
2060 | } | |
2061 | ||
2062 | static ssize_t | |
2063 | space_store(struct mddev *mddev, const char *buf, size_t len) | |
2064 | { | |
2065 | unsigned long sectors; | |
2066 | int rv; | |
2067 | ||
2068 | rv = kstrtoul(buf, 10, §ors); | |
2069 | if (rv) | |
2070 | return rv; | |
2071 | ||
2072 | if (sectors == 0) | |
2073 | return -EINVAL; | |
2074 | ||
2075 | if (mddev->bitmap && | |
9b1215c1 | 2076 | sectors < (mddev->bitmap->storage.bytes + 511) >> 9) |
6409bb05 N |
2077 | return -EFBIG; /* Bitmap is too big for this small space */ |
2078 | ||
2079 | /* could make sure it isn't too big, but that isn't really | |
2080 | * needed - user-space should be careful. | |
2081 | */ | |
2082 | mddev->bitmap_info.space = sectors; | |
2083 | return len; | |
2084 | } | |
2085 | ||
2086 | static struct md_sysfs_entry bitmap_space = | |
2087 | __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); | |
2088 | ||
43a70507 | 2089 | static ssize_t |
fd01b88c | 2090 | timeout_show(struct mddev *mddev, char *page) |
43a70507 N |
2091 | { |
2092 | ssize_t len; | |
2093 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
2094 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 2095 | |
43a70507 N |
2096 | len = sprintf(page, "%lu", secs); |
2097 | if (jifs) | |
2098 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
2099 | len += sprintf(page+len, "\n"); | |
2100 | return len; | |
2101 | } | |
2102 | ||
2103 | static ssize_t | |
fd01b88c | 2104 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2105 | { |
2106 | /* timeout can be set at any time */ | |
2107 | unsigned long timeout; | |
2108 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
2109 | if (rv) | |
2110 | return rv; | |
2111 | ||
2112 | /* just to make sure we don't overflow... */ | |
2113 | if (timeout >= LONG_MAX / HZ) | |
2114 | return -EINVAL; | |
2115 | ||
2116 | timeout = timeout * HZ / 10000; | |
2117 | ||
2118 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
2119 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
2120 | if (timeout < 1) | |
2121 | timeout = 1; | |
2122 | mddev->bitmap_info.daemon_sleep = timeout; | |
2123 | if (mddev->thread) { | |
2124 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
2125 | * the bitmap is all clean and we don't need to | |
2126 | * adjust the timeout right now | |
2127 | */ | |
2128 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2129 | mddev->thread->timeout = timeout; | |
2130 | md_wakeup_thread(mddev->thread); | |
2131 | } | |
2132 | } | |
2133 | return len; | |
2134 | } | |
2135 | ||
2136 | static struct md_sysfs_entry bitmap_timeout = | |
2137 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2138 | ||
2139 | static ssize_t | |
fd01b88c | 2140 | backlog_show(struct mddev *mddev, char *page) |
43a70507 N |
2141 | { |
2142 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2143 | } | |
2144 | ||
2145 | static ssize_t | |
fd01b88c | 2146 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2147 | { |
2148 | unsigned long backlog; | |
2149 | int rv = strict_strtoul(buf, 10, &backlog); | |
2150 | if (rv) | |
2151 | return rv; | |
2152 | if (backlog > COUNTER_MAX) | |
2153 | return -EINVAL; | |
2154 | mddev->bitmap_info.max_write_behind = backlog; | |
2155 | return len; | |
2156 | } | |
2157 | ||
2158 | static struct md_sysfs_entry bitmap_backlog = | |
2159 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2160 | ||
2161 | static ssize_t | |
fd01b88c | 2162 | chunksize_show(struct mddev *mddev, char *page) |
43a70507 N |
2163 | { |
2164 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2165 | } | |
2166 | ||
2167 | static ssize_t | |
fd01b88c | 2168 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2169 | { |
2170 | /* Can only be changed when no bitmap is active */ | |
2171 | int rv; | |
2172 | unsigned long csize; | |
2173 | if (mddev->bitmap) | |
2174 | return -EBUSY; | |
2175 | rv = strict_strtoul(buf, 10, &csize); | |
2176 | if (rv) | |
2177 | return rv; | |
2178 | if (csize < 512 || | |
2179 | !is_power_of_2(csize)) | |
2180 | return -EINVAL; | |
2181 | mddev->bitmap_info.chunksize = csize; | |
2182 | return len; | |
2183 | } | |
2184 | ||
2185 | static struct md_sysfs_entry bitmap_chunksize = | |
2186 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2187 | ||
fd01b88c | 2188 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2189 | { |
2190 | return sprintf(page, "%s\n", (mddev->bitmap_info.external | |
2191 | ? "external" : "internal")); | |
2192 | } | |
2193 | ||
fd01b88c | 2194 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2195 | { |
2196 | if (mddev->bitmap || | |
2197 | mddev->bitmap_info.file || | |
2198 | mddev->bitmap_info.offset) | |
2199 | return -EBUSY; | |
2200 | if (strncmp(buf, "external", 8) == 0) | |
2201 | mddev->bitmap_info.external = 1; | |
2202 | else if (strncmp(buf, "internal", 8) == 0) | |
2203 | mddev->bitmap_info.external = 0; | |
2204 | else | |
2205 | return -EINVAL; | |
2206 | return len; | |
2207 | } | |
2208 | ||
2209 | static struct md_sysfs_entry bitmap_metadata = | |
2210 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2211 | ||
fd01b88c | 2212 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2213 | { |
2214 | int len; | |
2215 | if (mddev->bitmap) | |
2216 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2217 | "false" : "true")); | |
2218 | else | |
2219 | len = sprintf(page, "\n"); | |
2220 | return len; | |
2221 | } | |
2222 | ||
fd01b88c | 2223 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2224 | { |
2225 | if (mddev->bitmap == NULL) | |
2226 | return -ENOENT; | |
2227 | if (strncmp(buf, "false", 5) == 0) | |
2228 | mddev->bitmap->need_sync = 1; | |
2229 | else if (strncmp(buf, "true", 4) == 0) { | |
2230 | if (mddev->degraded) | |
2231 | return -EBUSY; | |
2232 | mddev->bitmap->need_sync = 0; | |
2233 | } else | |
2234 | return -EINVAL; | |
2235 | return len; | |
2236 | } | |
2237 | ||
2238 | static struct md_sysfs_entry bitmap_can_clear = | |
2239 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2240 | ||
696fcd53 | 2241 | static ssize_t |
fd01b88c | 2242 | behind_writes_used_show(struct mddev *mddev, char *page) |
696fcd53 PC |
2243 | { |
2244 | if (mddev->bitmap == NULL) | |
2245 | return sprintf(page, "0\n"); | |
2246 | return sprintf(page, "%lu\n", | |
2247 | mddev->bitmap->behind_writes_used); | |
2248 | } | |
2249 | ||
2250 | static ssize_t | |
fd01b88c | 2251 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
696fcd53 PC |
2252 | { |
2253 | if (mddev->bitmap) | |
2254 | mddev->bitmap->behind_writes_used = 0; | |
2255 | return len; | |
2256 | } | |
2257 | ||
2258 | static struct md_sysfs_entry max_backlog_used = | |
2259 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2260 | behind_writes_used_show, behind_writes_used_reset); | |
2261 | ||
43a70507 N |
2262 | static struct attribute *md_bitmap_attrs[] = { |
2263 | &bitmap_location.attr, | |
6409bb05 | 2264 | &bitmap_space.attr, |
43a70507 N |
2265 | &bitmap_timeout.attr, |
2266 | &bitmap_backlog.attr, | |
2267 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2268 | &bitmap_metadata.attr, |
2269 | &bitmap_can_clear.attr, | |
696fcd53 | 2270 | &max_backlog_used.attr, |
43a70507 N |
2271 | NULL |
2272 | }; | |
2273 | struct attribute_group md_bitmap_group = { | |
2274 | .name = "bitmap", | |
2275 | .attrs = md_bitmap_attrs, | |
2276 | }; | |
2277 |