]>
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); |
d9590143 N |
75 | /* It is possible that this is being called inside a |
76 | * prepare_to_wait/finish_wait loop from raid5c:make_request(). | |
77 | * In general it is not permitted to sleep in that context as it | |
78 | * can cause the loop to spin freely. | |
79 | * That doesn't apply here as we can only reach this point | |
80 | * once with any loop. | |
81 | * When this function completes, either bp[page].map or | |
82 | * bp[page].hijacked. In either case, this function will | |
83 | * abort before getting to this point again. So there is | |
84 | * no risk of a free-spin, and so it is safe to assert | |
85 | * that sleeping here is allowed. | |
86 | */ | |
87 | sched_annotate_sleep(); | |
792a1d4b | 88 | mappage = kzalloc(PAGE_SIZE, GFP_NOIO); |
ac2f40be N |
89 | spin_lock_irq(&bitmap->lock); |
90 | ||
91 | if (mappage == NULL) { | |
40cffcc0 | 92 | pr_debug("md/bitmap: map page allocation failed, hijacking\n"); |
32a7627c N |
93 | /* failed - set the hijacked flag so that we can use the |
94 | * pointer as a counter */ | |
32a7627c N |
95 | if (!bitmap->bp[page].map) |
96 | bitmap->bp[page].hijacked = 1; | |
ac2f40be N |
97 | } else if (bitmap->bp[page].map || |
98 | bitmap->bp[page].hijacked) { | |
32a7627c | 99 | /* somebody beat us to getting the page */ |
792a1d4b | 100 | kfree(mappage); |
32a7627c | 101 | return 0; |
ac2f40be | 102 | } else { |
32a7627c | 103 | |
ac2f40be | 104 | /* no page was in place and we have one, so install it */ |
32a7627c | 105 | |
ac2f40be N |
106 | bitmap->bp[page].map = mappage; |
107 | bitmap->missing_pages--; | |
108 | } | |
32a7627c N |
109 | return 0; |
110 | } | |
111 | ||
32a7627c N |
112 | /* if page is completely empty, put it back on the free list, or dealloc it */ |
113 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
114 | /* Note: lock should be held when calling this */ | |
40cffcc0 | 115 | static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page) |
32a7627c N |
116 | { |
117 | char *ptr; | |
118 | ||
119 | if (bitmap->bp[page].count) /* page is still busy */ | |
120 | return; | |
121 | ||
122 | /* page is no longer in use, it can be released */ | |
123 | ||
124 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
125 | bitmap->bp[page].hijacked = 0; | |
126 | bitmap->bp[page].map = NULL; | |
ac2f40be N |
127 | } else { |
128 | /* normal case, free the page */ | |
129 | ptr = bitmap->bp[page].map; | |
130 | bitmap->bp[page].map = NULL; | |
131 | bitmap->missing_pages++; | |
792a1d4b | 132 | kfree(ptr); |
32a7627c | 133 | } |
32a7627c N |
134 | } |
135 | ||
32a7627c N |
136 | /* |
137 | * bitmap file handling - read and write the bitmap file and its superblock | |
138 | */ | |
139 | ||
32a7627c N |
140 | /* |
141 | * basic page I/O operations | |
142 | */ | |
143 | ||
a654b9d8 | 144 | /* IO operations when bitmap is stored near all superblocks */ |
27581e5a N |
145 | static int read_sb_page(struct mddev *mddev, loff_t offset, |
146 | struct page *page, | |
147 | unsigned long index, int size) | |
a654b9d8 N |
148 | { |
149 | /* choose a good rdev and read the page from there */ | |
150 | ||
3cb03002 | 151 | struct md_rdev *rdev; |
a654b9d8 | 152 | sector_t target; |
a654b9d8 | 153 | |
dafb20fa | 154 | rdev_for_each(rdev, mddev) { |
b2d444d7 N |
155 | if (! test_bit(In_sync, &rdev->flags) |
156 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
157 | continue; |
158 | ||
ccebd4c4 | 159 | target = offset + index * (PAGE_SIZE/512); |
a654b9d8 | 160 | |
2b193363 | 161 | if (sync_page_io(rdev, target, |
e1defc4f | 162 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
ccebd4c4 | 163 | page, READ, true)) { |
ab904d63 | 164 | page->index = index; |
27581e5a | 165 | return 0; |
ab904d63 N |
166 | } |
167 | } | |
27581e5a | 168 | return -EIO; |
a654b9d8 N |
169 | } |
170 | ||
fd01b88c | 171 | static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) |
b2d2c4ce N |
172 | { |
173 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
174 | * linked list, and raising the refcount of devices we return to ensure | |
175 | * they don't disappear while in use. | |
176 | * As devices are only added or removed when raid_disk is < 0 and | |
177 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
178 | * still be in the same position on the list when we re-enter | |
fd177481 | 179 | * list_for_each_entry_continue_rcu. |
b2d2c4ce | 180 | */ |
b2d2c4ce N |
181 | rcu_read_lock(); |
182 | if (rdev == NULL) | |
183 | /* start at the beginning */ | |
fd177481 | 184 | rdev = list_entry_rcu(&mddev->disks, struct md_rdev, same_set); |
b2d2c4ce N |
185 | else { |
186 | /* release the previous rdev and start from there. */ | |
187 | rdev_dec_pending(rdev, mddev); | |
b2d2c4ce | 188 | } |
fd177481 | 189 | list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) { |
b2d2c4ce | 190 | if (rdev->raid_disk >= 0 && |
b2d2c4ce N |
191 | !test_bit(Faulty, &rdev->flags)) { |
192 | /* this is a usable devices */ | |
193 | atomic_inc(&rdev->nr_pending); | |
194 | rcu_read_unlock(); | |
195 | return rdev; | |
196 | } | |
197 | } | |
198 | rcu_read_unlock(); | |
199 | return NULL; | |
200 | } | |
201 | ||
ab6085c7 | 202 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 203 | { |
3cb03002 | 204 | struct md_rdev *rdev = NULL; |
a6ff7e08 | 205 | struct block_device *bdev; |
fd01b88c | 206 | struct mddev *mddev = bitmap->mddev; |
1ec885cd | 207 | struct bitmap_storage *store = &bitmap->storage; |
a654b9d8 | 208 | |
b2d2c4ce | 209 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ac2f40be N |
210 | int size = PAGE_SIZE; |
211 | loff_t offset = mddev->bitmap_info.offset; | |
a6ff7e08 JB |
212 | |
213 | bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; | |
214 | ||
9b1215c1 N |
215 | if (page->index == store->file_pages-1) { |
216 | int last_page_size = store->bytes & (PAGE_SIZE-1); | |
217 | if (last_page_size == 0) | |
218 | last_page_size = PAGE_SIZE; | |
219 | size = roundup(last_page_size, | |
a6ff7e08 | 220 | bdev_logical_block_size(bdev)); |
9b1215c1 | 221 | } |
ac2f40be N |
222 | /* Just make sure we aren't corrupting data or |
223 | * metadata | |
224 | */ | |
225 | if (mddev->external) { | |
226 | /* Bitmap could be anywhere. */ | |
227 | if (rdev->sb_start + offset + (page->index | |
228 | * (PAGE_SIZE/512)) | |
229 | > rdev->data_offset | |
230 | && | |
231 | rdev->sb_start + offset | |
232 | < (rdev->data_offset + mddev->dev_sectors | |
233 | + (PAGE_SIZE/512))) | |
234 | goto bad_alignment; | |
235 | } else if (offset < 0) { | |
236 | /* DATA BITMAP METADATA */ | |
237 | if (offset | |
238 | + (long)(page->index * (PAGE_SIZE/512)) | |
239 | + size/512 > 0) | |
240 | /* bitmap runs in to metadata */ | |
241 | goto bad_alignment; | |
242 | if (rdev->data_offset + mddev->dev_sectors | |
243 | > rdev->sb_start + offset) | |
244 | /* data runs in to bitmap */ | |
245 | goto bad_alignment; | |
246 | } else if (rdev->sb_start < rdev->data_offset) { | |
247 | /* METADATA BITMAP DATA */ | |
248 | if (rdev->sb_start | |
249 | + offset | |
250 | + page->index*(PAGE_SIZE/512) + size/512 | |
251 | > rdev->data_offset) | |
252 | /* bitmap runs in to data */ | |
253 | goto bad_alignment; | |
254 | } else { | |
255 | /* DATA METADATA BITMAP - no problems */ | |
256 | } | |
257 | md_super_write(mddev, rdev, | |
258 | rdev->sb_start + offset | |
259 | + page->index * (PAGE_SIZE/512), | |
260 | size, | |
261 | page); | |
b2d2c4ce | 262 | } |
a654b9d8 N |
263 | |
264 | if (wait) | |
a9701a30 | 265 | md_super_wait(mddev); |
a654b9d8 | 266 | return 0; |
4b80991c N |
267 | |
268 | bad_alignment: | |
4b80991c | 269 | return -EINVAL; |
a654b9d8 N |
270 | } |
271 | ||
4ad13663 | 272 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 273 | /* |
a654b9d8 | 274 | * write out a page to a file |
32a7627c | 275 | */ |
4ad13663 | 276 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 277 | { |
d785a06a | 278 | struct buffer_head *bh; |
32a7627c | 279 | |
1ec885cd | 280 | if (bitmap->storage.file == NULL) { |
f0d76d70 N |
281 | switch (write_sb_page(bitmap, page, wait)) { |
282 | case -EINVAL: | |
b405fe91 | 283 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); |
f0d76d70 | 284 | } |
4ad13663 | 285 | } else { |
a654b9d8 | 286 | |
4ad13663 | 287 | bh = page_buffers(page); |
c708443c | 288 | |
4ad13663 N |
289 | while (bh && bh->b_blocknr) { |
290 | atomic_inc(&bitmap->pending_writes); | |
291 | set_buffer_locked(bh); | |
292 | set_buffer_mapped(bh); | |
721a9602 | 293 | submit_bh(WRITE | REQ_SYNC, bh); |
4ad13663 N |
294 | bh = bh->b_this_page; |
295 | } | |
d785a06a | 296 | |
ac2f40be | 297 | if (wait) |
4ad13663 N |
298 | wait_event(bitmap->write_wait, |
299 | atomic_read(&bitmap->pending_writes)==0); | |
8a5e9cf1 | 300 | } |
b405fe91 | 301 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
4ad13663 | 302 | bitmap_file_kick(bitmap); |
d785a06a N |
303 | } |
304 | ||
305 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
306 | { | |
307 | struct bitmap *bitmap = bh->b_private; | |
32a7627c | 308 | |
b405fe91 N |
309 | if (!uptodate) |
310 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); | |
d785a06a N |
311 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
312 | wake_up(&bitmap->write_wait); | |
313 | } | |
32a7627c | 314 | |
d785a06a N |
315 | /* copied from buffer.c */ |
316 | static void | |
317 | __clear_page_buffers(struct page *page) | |
318 | { | |
319 | ClearPagePrivate(page); | |
320 | set_page_private(page, 0); | |
321 | page_cache_release(page); | |
322 | } | |
323 | static void free_buffers(struct page *page) | |
324 | { | |
27581e5a | 325 | struct buffer_head *bh; |
77ad4bc7 | 326 | |
27581e5a N |
327 | if (!PagePrivate(page)) |
328 | return; | |
329 | ||
330 | bh = page_buffers(page); | |
d785a06a N |
331 | while (bh) { |
332 | struct buffer_head *next = bh->b_this_page; | |
333 | free_buffer_head(bh); | |
334 | bh = next; | |
77ad4bc7 | 335 | } |
d785a06a N |
336 | __clear_page_buffers(page); |
337 | put_page(page); | |
32a7627c N |
338 | } |
339 | ||
d785a06a N |
340 | /* read a page from a file. |
341 | * We both read the page, and attach buffers to the page to record the | |
342 | * address of each block (using bmap). These addresses will be used | |
343 | * to write the block later, completely bypassing the filesystem. | |
344 | * This usage is similar to how swap files are handled, and allows us | |
345 | * to write to a file with no concerns of memory allocation failing. | |
346 | */ | |
27581e5a N |
347 | static int read_page(struct file *file, unsigned long index, |
348 | struct bitmap *bitmap, | |
349 | unsigned long count, | |
350 | struct page *page) | |
32a7627c | 351 | { |
27581e5a | 352 | int ret = 0; |
496ad9aa | 353 | struct inode *inode = file_inode(file); |
d785a06a N |
354 | struct buffer_head *bh; |
355 | sector_t block; | |
32a7627c | 356 | |
36a4e1fe N |
357 | pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
358 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 359 | |
d785a06a N |
360 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
361 | if (!bh) { | |
27581e5a | 362 | ret = -ENOMEM; |
32a7627c N |
363 | goto out; |
364 | } | |
d785a06a N |
365 | attach_page_buffers(page, bh); |
366 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
367 | while (bh) { | |
368 | if (count == 0) | |
369 | bh->b_blocknr = 0; | |
370 | else { | |
371 | bh->b_blocknr = bmap(inode, block); | |
372 | if (bh->b_blocknr == 0) { | |
373 | /* Cannot use this file! */ | |
27581e5a | 374 | ret = -EINVAL; |
d785a06a N |
375 | goto out; |
376 | } | |
377 | bh->b_bdev = inode->i_sb->s_bdev; | |
378 | if (count < (1<<inode->i_blkbits)) | |
379 | count = 0; | |
380 | else | |
381 | count -= (1<<inode->i_blkbits); | |
382 | ||
383 | bh->b_end_io = end_bitmap_write; | |
384 | bh->b_private = bitmap; | |
ce25c31b N |
385 | atomic_inc(&bitmap->pending_writes); |
386 | set_buffer_locked(bh); | |
387 | set_buffer_mapped(bh); | |
388 | submit_bh(READ, bh); | |
d785a06a N |
389 | } |
390 | block++; | |
391 | bh = bh->b_this_page; | |
392 | } | |
d785a06a | 393 | page->index = index; |
ce25c31b N |
394 | |
395 | wait_event(bitmap->write_wait, | |
396 | atomic_read(&bitmap->pending_writes)==0); | |
b405fe91 | 397 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
27581e5a | 398 | ret = -EIO; |
32a7627c | 399 | out: |
27581e5a N |
400 | if (ret) |
401 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n", | |
2d1f3b5d N |
402 | (int)PAGE_SIZE, |
403 | (unsigned long long)index << PAGE_SHIFT, | |
27581e5a N |
404 | ret); |
405 | return ret; | |
32a7627c N |
406 | } |
407 | ||
408 | /* | |
409 | * bitmap file superblock operations | |
410 | */ | |
411 | ||
412 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 413 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
414 | { |
415 | bitmap_super_t *sb; | |
32a7627c N |
416 | |
417 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 418 | return; |
ece5cff0 N |
419 | if (bitmap->mddev->bitmap_info.external) |
420 | return; | |
1ec885cd | 421 | if (!bitmap->storage.sb_page) /* no superblock */ |
4ad13663 | 422 | return; |
1ec885cd | 423 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 424 | sb->events = cpu_to_le64(bitmap->mddev->events); |
8258c532 | 425 | if (bitmap->mddev->events < bitmap->events_cleared) |
a0da84f3 NB |
426 | /* rocking back to read-only */ |
427 | bitmap->events_cleared = bitmap->mddev->events; | |
8258c532 N |
428 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); |
429 | sb->state = cpu_to_le32(bitmap->flags); | |
43a70507 N |
430 | /* Just in case these have been changed via sysfs: */ |
431 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
432 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
b81a0404 N |
433 | /* This might have been changed by a reshape */ |
434 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
435 | sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize); | |
c4ce867f | 436 | sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes); |
1dff2b87 N |
437 | sb->sectors_reserved = cpu_to_le32(bitmap->mddev-> |
438 | bitmap_info.space); | |
b2f46e68 | 439 | kunmap_atomic(sb); |
1ec885cd | 440 | write_page(bitmap, bitmap->storage.sb_page, 1); |
32a7627c N |
441 | } |
442 | ||
443 | /* print out the bitmap file superblock */ | |
444 | void bitmap_print_sb(struct bitmap *bitmap) | |
445 | { | |
446 | bitmap_super_t *sb; | |
447 | ||
1ec885cd | 448 | if (!bitmap || !bitmap->storage.sb_page) |
32a7627c | 449 | return; |
1ec885cd | 450 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 451 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
452 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
453 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
454 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
455 | *(__u32 *)(sb->uuid+0), |
456 | *(__u32 *)(sb->uuid+4), | |
457 | *(__u32 *)(sb->uuid+8), | |
458 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 459 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 460 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 461 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 462 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
463 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
464 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
465 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
466 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
467 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 468 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
b2f46e68 | 469 | kunmap_atomic(sb); |
32a7627c N |
470 | } |
471 | ||
9c81075f JB |
472 | /* |
473 | * bitmap_new_disk_sb | |
474 | * @bitmap | |
475 | * | |
476 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb | |
477 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. | |
478 | * This function verifies 'bitmap_info' and populates the on-disk bitmap | |
479 | * structure, which is to be written to disk. | |
480 | * | |
481 | * Returns: 0 on success, -Exxx on error | |
482 | */ | |
483 | static int bitmap_new_disk_sb(struct bitmap *bitmap) | |
484 | { | |
485 | bitmap_super_t *sb; | |
486 | unsigned long chunksize, daemon_sleep, write_behind; | |
9c81075f | 487 | |
1ec885cd | 488 | bitmap->storage.sb_page = alloc_page(GFP_KERNEL); |
582e2e05 JM |
489 | if (bitmap->storage.sb_page == NULL) |
490 | return -ENOMEM; | |
1ec885cd | 491 | bitmap->storage.sb_page->index = 0; |
9c81075f | 492 | |
1ec885cd | 493 | sb = kmap_atomic(bitmap->storage.sb_page); |
9c81075f JB |
494 | |
495 | sb->magic = cpu_to_le32(BITMAP_MAGIC); | |
496 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); | |
497 | ||
498 | chunksize = bitmap->mddev->bitmap_info.chunksize; | |
499 | BUG_ON(!chunksize); | |
500 | if (!is_power_of_2(chunksize)) { | |
b2f46e68 | 501 | kunmap_atomic(sb); |
9c81075f JB |
502 | printk(KERN_ERR "bitmap chunksize not a power of 2\n"); |
503 | return -EINVAL; | |
504 | } | |
505 | sb->chunksize = cpu_to_le32(chunksize); | |
506 | ||
507 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; | |
508 | if (!daemon_sleep || | |
509 | (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { | |
510 | printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n"); | |
511 | daemon_sleep = 5 * HZ; | |
512 | } | |
513 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); | |
514 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
515 | ||
516 | /* | |
517 | * FIXME: write_behind for RAID1. If not specified, what | |
518 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. | |
519 | */ | |
520 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; | |
521 | if (write_behind > COUNTER_MAX) | |
522 | write_behind = COUNTER_MAX / 2; | |
523 | sb->write_behind = cpu_to_le32(write_behind); | |
524 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
525 | ||
526 | /* keep the array size field of the bitmap superblock up to date */ | |
527 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
528 | ||
529 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); | |
530 | ||
b405fe91 | 531 | set_bit(BITMAP_STALE, &bitmap->flags); |
84e92345 | 532 | sb->state = cpu_to_le32(bitmap->flags); |
9c81075f JB |
533 | bitmap->events_cleared = bitmap->mddev->events; |
534 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
535 | ||
b2f46e68 | 536 | kunmap_atomic(sb); |
9c81075f JB |
537 | |
538 | return 0; | |
539 | } | |
540 | ||
32a7627c N |
541 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
542 | static int bitmap_read_sb(struct bitmap *bitmap) | |
543 | { | |
544 | char *reason = NULL; | |
545 | bitmap_super_t *sb; | |
4b6d287f | 546 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c | 547 | unsigned long long events; |
c4ce867f | 548 | int nodes = 0; |
1dff2b87 | 549 | unsigned long sectors_reserved = 0; |
32a7627c | 550 | int err = -EINVAL; |
27581e5a | 551 | struct page *sb_page; |
32a7627c | 552 | |
1ec885cd | 553 | if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { |
ef99bf48 N |
554 | chunksize = 128 * 1024 * 1024; |
555 | daemon_sleep = 5 * HZ; | |
556 | write_behind = 0; | |
b405fe91 | 557 | set_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
558 | err = 0; |
559 | goto out_no_sb; | |
560 | } | |
32a7627c | 561 | /* page 0 is the superblock, read it... */ |
27581e5a N |
562 | sb_page = alloc_page(GFP_KERNEL); |
563 | if (!sb_page) | |
564 | return -ENOMEM; | |
1ec885cd | 565 | bitmap->storage.sb_page = sb_page; |
27581e5a | 566 | |
1ec885cd N |
567 | if (bitmap->storage.file) { |
568 | loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); | |
f49d5e62 N |
569 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; |
570 | ||
1ec885cd | 571 | err = read_page(bitmap->storage.file, 0, |
27581e5a | 572 | bitmap, bytes, sb_page); |
f49d5e62 | 573 | } else { |
27581e5a N |
574 | err = read_sb_page(bitmap->mddev, |
575 | bitmap->mddev->bitmap_info.offset, | |
576 | sb_page, | |
577 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 578 | } |
27581e5a | 579 | if (err) |
32a7627c | 580 | return err; |
32a7627c | 581 | |
27581e5a | 582 | sb = kmap_atomic(sb_page); |
32a7627c | 583 | |
32a7627c | 584 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 585 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 586 | write_behind = le32_to_cpu(sb->write_behind); |
1dff2b87 | 587 | sectors_reserved = le32_to_cpu(sb->sectors_reserved); |
c4ce867f | 588 | nodes = le32_to_cpu(sb->nodes); |
32a7627c N |
589 | |
590 | /* verify that the bitmap-specific fields are valid */ | |
591 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
592 | reason = "bad magic"; | |
bd926c63 N |
593 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
594 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 595 | reason = "unrecognized superblock version"; |
1187cf0a | 596 | else if (chunksize < 512) |
7dd5d34c | 597 | reason = "bitmap chunksize too small"; |
d744540c | 598 | else if (!is_power_of_2(chunksize)) |
32a7627c | 599 | reason = "bitmap chunksize not a power of 2"; |
1b04be96 | 600 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 601 | reason = "daemon sleep period out of range"; |
4b6d287f N |
602 | else if (write_behind > COUNTER_MAX) |
603 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
604 | if (reason) { |
605 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
606 | bmname(bitmap), reason); | |
607 | goto out; | |
608 | } | |
609 | ||
610 | /* keep the array size field of the bitmap superblock up to date */ | |
611 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
612 | ||
278c1ca2 N |
613 | if (bitmap->mddev->persistent) { |
614 | /* | |
615 | * We have a persistent array superblock, so compare the | |
616 | * bitmap's UUID and event counter to the mddev's | |
617 | */ | |
618 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
619 | printk(KERN_INFO | |
620 | "%s: bitmap superblock UUID mismatch\n", | |
621 | bmname(bitmap)); | |
622 | goto out; | |
623 | } | |
624 | events = le64_to_cpu(sb->events); | |
625 | if (events < bitmap->mddev->events) { | |
626 | printk(KERN_INFO | |
627 | "%s: bitmap file is out of date (%llu < %llu) " | |
628 | "-- forcing full recovery\n", | |
629 | bmname(bitmap), events, | |
630 | (unsigned long long) bitmap->mddev->events); | |
b405fe91 | 631 | set_bit(BITMAP_STALE, &bitmap->flags); |
278c1ca2 | 632 | } |
32a7627c | 633 | } |
278c1ca2 | 634 | |
32a7627c | 635 | /* assign fields using values from superblock */ |
4f2e639a | 636 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 | 637 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
b405fe91 | 638 | set_bit(BITMAP_HOSTENDIAN, &bitmap->flags); |
32a7627c | 639 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
cf921cc1 | 640 | strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64); |
32a7627c N |
641 | err = 0; |
642 | out: | |
b2f46e68 | 643 | kunmap_atomic(sb); |
ef99bf48 | 644 | out_no_sb: |
b405fe91 | 645 | if (test_bit(BITMAP_STALE, &bitmap->flags)) |
ef99bf48 N |
646 | bitmap->events_cleared = bitmap->mddev->events; |
647 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
648 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
649 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
c4ce867f | 650 | bitmap->mddev->bitmap_info.nodes = nodes; |
1dff2b87 N |
651 | if (bitmap->mddev->bitmap_info.space == 0 || |
652 | bitmap->mddev->bitmap_info.space > sectors_reserved) | |
653 | bitmap->mddev->bitmap_info.space = sectors_reserved; | |
32a7627c N |
654 | if (err) |
655 | bitmap_print_sb(bitmap); | |
656 | return err; | |
657 | } | |
658 | ||
32a7627c N |
659 | /* |
660 | * general bitmap file operations | |
661 | */ | |
662 | ||
ece5cff0 N |
663 | /* |
664 | * on-disk bitmap: | |
665 | * | |
666 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
667 | * file a page at a time. There's a superblock at the start of the file. | |
668 | */ | |
32a7627c | 669 | /* calculate the index of the page that contains this bit */ |
1ec885cd N |
670 | static inline unsigned long file_page_index(struct bitmap_storage *store, |
671 | unsigned long chunk) | |
32a7627c | 672 | { |
1ec885cd | 673 | if (store->sb_page) |
ece5cff0 N |
674 | chunk += sizeof(bitmap_super_t) << 3; |
675 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
676 | } |
677 | ||
678 | /* calculate the (bit) offset of this bit within a page */ | |
1ec885cd N |
679 | static inline unsigned long file_page_offset(struct bitmap_storage *store, |
680 | unsigned long chunk) | |
32a7627c | 681 | { |
1ec885cd | 682 | if (store->sb_page) |
ece5cff0 N |
683 | chunk += sizeof(bitmap_super_t) << 3; |
684 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
685 | } |
686 | ||
687 | /* | |
688 | * return a pointer to the page in the filemap that contains the given bit | |
689 | * | |
32a7627c | 690 | */ |
1ec885cd | 691 | static inline struct page *filemap_get_page(struct bitmap_storage *store, |
3520fa4d | 692 | unsigned long chunk) |
32a7627c | 693 | { |
1ec885cd | 694 | if (file_page_index(store, chunk) >= store->file_pages) |
ac2f40be | 695 | return NULL; |
f2e06c58 | 696 | return store->filemap[file_page_index(store, chunk)]; |
32a7627c N |
697 | } |
698 | ||
d1244cb0 N |
699 | static int bitmap_storage_alloc(struct bitmap_storage *store, |
700 | unsigned long chunks, int with_super) | |
701 | { | |
702 | int pnum; | |
703 | unsigned long num_pages; | |
704 | unsigned long bytes; | |
705 | ||
706 | bytes = DIV_ROUND_UP(chunks, 8); | |
707 | if (with_super) | |
708 | bytes += sizeof(bitmap_super_t); | |
709 | ||
710 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); | |
711 | ||
712 | store->filemap = kmalloc(sizeof(struct page *) | |
713 | * num_pages, GFP_KERNEL); | |
714 | if (!store->filemap) | |
715 | return -ENOMEM; | |
716 | ||
717 | if (with_super && !store->sb_page) { | |
d60b479d | 718 | store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
719 | if (store->sb_page == NULL) |
720 | return -ENOMEM; | |
721 | store->sb_page->index = 0; | |
722 | } | |
723 | pnum = 0; | |
724 | if (store->sb_page) { | |
725 | store->filemap[0] = store->sb_page; | |
726 | pnum = 1; | |
727 | } | |
728 | for ( ; pnum < num_pages; pnum++) { | |
d60b479d | 729 | store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
730 | if (!store->filemap[pnum]) { |
731 | store->file_pages = pnum; | |
732 | return -ENOMEM; | |
733 | } | |
734 | store->filemap[pnum]->index = pnum; | |
735 | } | |
736 | store->file_pages = pnum; | |
737 | ||
738 | /* We need 4 bits per page, rounded up to a multiple | |
739 | * of sizeof(unsigned long) */ | |
740 | store->filemap_attr = kzalloc( | |
741 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), | |
742 | GFP_KERNEL); | |
743 | if (!store->filemap_attr) | |
744 | return -ENOMEM; | |
745 | ||
746 | store->bytes = bytes; | |
747 | ||
748 | return 0; | |
749 | } | |
750 | ||
fae7d326 | 751 | static void bitmap_file_unmap(struct bitmap_storage *store) |
32a7627c N |
752 | { |
753 | struct page **map, *sb_page; | |
32a7627c | 754 | int pages; |
fae7d326 | 755 | struct file *file; |
32a7627c | 756 | |
fae7d326 | 757 | file = store->file; |
1ec885cd | 758 | map = store->filemap; |
1ec885cd | 759 | pages = store->file_pages; |
1ec885cd | 760 | sb_page = store->sb_page; |
32a7627c N |
761 | |
762 | while (pages--) | |
ece5cff0 | 763 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 764 | free_buffers(map[pages]); |
32a7627c | 765 | kfree(map); |
fae7d326 | 766 | kfree(store->filemap_attr); |
32a7627c | 767 | |
d785a06a N |
768 | if (sb_page) |
769 | free_buffers(sb_page); | |
32a7627c | 770 | |
d785a06a | 771 | if (file) { |
496ad9aa | 772 | struct inode *inode = file_inode(file); |
fc0ecff6 | 773 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 774 | fput(file); |
d785a06a | 775 | } |
32a7627c N |
776 | } |
777 | ||
32a7627c N |
778 | /* |
779 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
780 | * then it is no longer reliable, so we stop using it and we mark the file | |
781 | * as failed in the superblock | |
782 | */ | |
783 | static void bitmap_file_kick(struct bitmap *bitmap) | |
784 | { | |
785 | char *path, *ptr = NULL; | |
786 | ||
b405fe91 | 787 | if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) { |
4ad13663 | 788 | bitmap_update_sb(bitmap); |
32a7627c | 789 | |
1ec885cd | 790 | if (bitmap->storage.file) { |
4ad13663 N |
791 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); |
792 | if (path) | |
1ec885cd N |
793 | ptr = d_path(&bitmap->storage.file->f_path, |
794 | path, PAGE_SIZE); | |
6bcfd601 | 795 | |
4ad13663 N |
796 | printk(KERN_ALERT |
797 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 798 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 799 | |
4ad13663 N |
800 | kfree(path); |
801 | } else | |
802 | printk(KERN_ALERT | |
803 | "%s: disabling internal bitmap due to errors\n", | |
804 | bmname(bitmap)); | |
a654b9d8 | 805 | } |
32a7627c N |
806 | } |
807 | ||
808 | enum bitmap_page_attr { | |
ac2f40be | 809 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
5a537df4 N |
810 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
811 | * i.e. counter is 1 or 2. */ | |
ac2f40be | 812 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
32a7627c N |
813 | }; |
814 | ||
d189122d N |
815 | static inline void set_page_attr(struct bitmap *bitmap, int pnum, |
816 | enum bitmap_page_attr attr) | |
32a7627c | 817 | { |
bdfd1140 | 818 | set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
819 | } |
820 | ||
d189122d N |
821 | static inline void clear_page_attr(struct bitmap *bitmap, int pnum, |
822 | enum bitmap_page_attr attr) | |
32a7627c | 823 | { |
bdfd1140 | 824 | clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
825 | } |
826 | ||
bdfd1140 N |
827 | static inline int test_page_attr(struct bitmap *bitmap, int pnum, |
828 | enum bitmap_page_attr attr) | |
32a7627c | 829 | { |
1ec885cd | 830 | return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
831 | } |
832 | ||
bdfd1140 N |
833 | static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum, |
834 | enum bitmap_page_attr attr) | |
835 | { | |
836 | return test_and_clear_bit((pnum<<2) + attr, | |
837 | bitmap->storage.filemap_attr); | |
838 | } | |
32a7627c N |
839 | /* |
840 | * bitmap_file_set_bit -- called before performing a write to the md device | |
841 | * to set (and eventually sync) a particular bit in the bitmap file | |
842 | * | |
843 | * we set the bit immediately, then we record the page number so that | |
844 | * when an unplug occurs, we can flush the dirty pages out to disk | |
845 | */ | |
846 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
847 | { | |
848 | unsigned long bit; | |
3520fa4d | 849 | struct page *page; |
32a7627c | 850 | void *kaddr; |
40cffcc0 | 851 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
32a7627c | 852 | |
1ec885cd | 853 | page = filemap_get_page(&bitmap->storage, chunk); |
3520fa4d JB |
854 | if (!page) |
855 | return; | |
1ec885cd | 856 | bit = file_page_offset(&bitmap->storage, chunk); |
32a7627c | 857 | |
3520fa4d | 858 | /* set the bit */ |
b2f46e68 | 859 | kaddr = kmap_atomic(page); |
b405fe91 | 860 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
3520fa4d JB |
861 | set_bit(bit, kaddr); |
862 | else | |
3f810b6c | 863 | set_bit_le(bit, kaddr); |
b2f46e68 | 864 | kunmap_atomic(kaddr); |
36a4e1fe | 865 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
32a7627c | 866 | /* record page number so it gets flushed to disk when unplug occurs */ |
d189122d | 867 | set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY); |
32a7627c N |
868 | } |
869 | ||
ef99bf48 N |
870 | static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) |
871 | { | |
872 | unsigned long bit; | |
873 | struct page *page; | |
874 | void *paddr; | |
40cffcc0 | 875 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
ef99bf48 | 876 | |
1ec885cd | 877 | page = filemap_get_page(&bitmap->storage, chunk); |
ef99bf48 N |
878 | if (!page) |
879 | return; | |
1ec885cd | 880 | bit = file_page_offset(&bitmap->storage, chunk); |
ef99bf48 | 881 | paddr = kmap_atomic(page); |
b405fe91 | 882 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ef99bf48 N |
883 | clear_bit(bit, paddr); |
884 | else | |
3f810b6c | 885 | clear_bit_le(bit, paddr); |
ef99bf48 | 886 | kunmap_atomic(paddr); |
d189122d N |
887 | if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) { |
888 | set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING); | |
ef99bf48 N |
889 | bitmap->allclean = 0; |
890 | } | |
891 | } | |
892 | ||
32a7627c N |
893 | /* this gets called when the md device is ready to unplug its underlying |
894 | * (slave) device queues -- before we let any writes go down, we need to | |
895 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 896 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 897 | { |
74667123 | 898 | unsigned long i; |
ec7a3197 | 899 | int dirty, need_write; |
32a7627c | 900 | |
62f82faa N |
901 | if (!bitmap || !bitmap->storage.filemap || |
902 | test_bit(BITMAP_STALE, &bitmap->flags)) | |
4ad13663 | 903 | return; |
32a7627c N |
904 | |
905 | /* look at each page to see if there are any set bits that need to be | |
906 | * flushed out to disk */ | |
1ec885cd | 907 | for (i = 0; i < bitmap->storage.file_pages; i++) { |
bdfd1140 | 908 | if (!bitmap->storage.filemap) |
4ad13663 | 909 | return; |
bdfd1140 N |
910 | dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
911 | need_write = test_and_clear_page_attr(bitmap, i, | |
912 | BITMAP_PAGE_NEEDWRITE); | |
913 | if (dirty || need_write) { | |
d189122d | 914 | clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); |
bdfd1140 N |
915 | write_page(bitmap, bitmap->storage.filemap[i], 0); |
916 | } | |
32a7627c | 917 | } |
4b5060dd N |
918 | if (bitmap->storage.file) |
919 | wait_event(bitmap->write_wait, | |
920 | atomic_read(&bitmap->pending_writes)==0); | |
921 | else | |
922 | md_super_wait(bitmap->mddev); | |
923 | ||
b405fe91 | 924 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
d785a06a | 925 | bitmap_file_kick(bitmap); |
32a7627c | 926 | } |
ac2f40be | 927 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 928 | |
6a07997f | 929 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
930 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
931 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
932 | * memory mapping of the bitmap file | |
933 | * Special cases: | |
934 | * if there's no bitmap file, or if the bitmap file had been | |
935 | * previously kicked from the array, we mark all the bits as | |
936 | * 1's in order to cause a full resync. | |
6a07997f N |
937 | * |
938 | * We ignore all bits for sectors that end earlier than 'start'. | |
939 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 940 | */ |
6a07997f | 941 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
942 | { |
943 | unsigned long i, chunks, index, oldindex, bit; | |
27581e5a | 944 | struct page *page = NULL; |
d1244cb0 | 945 | unsigned long bit_cnt = 0; |
32a7627c | 946 | struct file *file; |
d1244cb0 | 947 | unsigned long offset; |
32a7627c N |
948 | int outofdate; |
949 | int ret = -ENOSPC; | |
ea03aff9 | 950 | void *paddr; |
1ec885cd | 951 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c | 952 | |
40cffcc0 | 953 | chunks = bitmap->counts.chunks; |
1ec885cd | 954 | file = store->file; |
32a7627c | 955 | |
ef99bf48 N |
956 | if (!file && !bitmap->mddev->bitmap_info.offset) { |
957 | /* No permanent bitmap - fill with '1s'. */ | |
1ec885cd N |
958 | store->filemap = NULL; |
959 | store->file_pages = 0; | |
ef99bf48 N |
960 | for (i = 0; i < chunks ; i++) { |
961 | /* if the disk bit is set, set the memory bit */ | |
40cffcc0 | 962 | int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift) |
ef99bf48 N |
963 | >= start); |
964 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 965 | (sector_t)i << bitmap->counts.chunkshift, |
ef99bf48 N |
966 | needed); |
967 | } | |
968 | return 0; | |
969 | } | |
32a7627c | 970 | |
b405fe91 | 971 | outofdate = test_bit(BITMAP_STALE, &bitmap->flags); |
32a7627c N |
972 | if (outofdate) |
973 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
974 | "recovery\n", bmname(bitmap)); | |
975 | ||
d1244cb0 | 976 | if (file && i_size_read(file->f_mapping->host) < store->bytes) { |
32a7627c | 977 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
d1244cb0 N |
978 | bmname(bitmap), |
979 | (unsigned long) i_size_read(file->f_mapping->host), | |
980 | store->bytes); | |
4ad13663 | 981 | goto err; |
32a7627c | 982 | } |
bc7f77de | 983 | |
d1244cb0 | 984 | oldindex = ~0L; |
27581e5a | 985 | offset = 0; |
d1244cb0 | 986 | if (!bitmap->mddev->bitmap_info.external) |
27581e5a | 987 | offset = sizeof(bitmap_super_t); |
32a7627c N |
988 | |
989 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 990 | int b; |
1ec885cd N |
991 | index = file_page_index(&bitmap->storage, i); |
992 | bit = file_page_offset(&bitmap->storage, i); | |
32a7627c | 993 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 994 | int count; |
32a7627c | 995 | /* unmap the old page, we're done with it */ |
d1244cb0 N |
996 | if (index == store->file_pages-1) |
997 | count = store->bytes - index * PAGE_SIZE; | |
d785a06a N |
998 | else |
999 | count = PAGE_SIZE; | |
1ec885cd | 1000 | page = store->filemap[index]; |
27581e5a N |
1001 | if (file) |
1002 | ret = read_page(file, index, bitmap, | |
1003 | count, page); | |
1004 | else | |
1005 | ret = read_sb_page( | |
1006 | bitmap->mddev, | |
1007 | bitmap->mddev->bitmap_info.offset, | |
1008 | page, | |
1009 | index, count); | |
1010 | ||
1011 | if (ret) | |
4ad13663 | 1012 | goto err; |
a654b9d8 | 1013 | |
32a7627c | 1014 | oldindex = index; |
32a7627c N |
1015 | |
1016 | if (outofdate) { | |
1017 | /* | |
1018 | * if bitmap is out of date, dirty the | |
ac2f40be | 1019 | * whole page and write it out |
32a7627c | 1020 | */ |
b2f46e68 | 1021 | paddr = kmap_atomic(page); |
ea03aff9 | 1022 | memset(paddr + offset, 0xff, |
6a07997f | 1023 | PAGE_SIZE - offset); |
b2f46e68 | 1024 | kunmap_atomic(paddr); |
4ad13663 N |
1025 | write_page(bitmap, page, 1); |
1026 | ||
1027 | ret = -EIO; | |
b405fe91 N |
1028 | if (test_bit(BITMAP_WRITE_ERROR, |
1029 | &bitmap->flags)) | |
4ad13663 | 1030 | goto err; |
32a7627c | 1031 | } |
32a7627c | 1032 | } |
b2f46e68 | 1033 | paddr = kmap_atomic(page); |
b405fe91 | 1034 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ea03aff9 | 1035 | b = test_bit(bit, paddr); |
bd926c63 | 1036 | else |
6b33aff3 | 1037 | b = test_bit_le(bit, paddr); |
b2f46e68 | 1038 | kunmap_atomic(paddr); |
bd926c63 | 1039 | if (b) { |
32a7627c | 1040 | /* if the disk bit is set, set the memory bit */ |
40cffcc0 | 1041 | int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift |
db305e50 N |
1042 | >= start); |
1043 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 1044 | (sector_t)i << bitmap->counts.chunkshift, |
db305e50 | 1045 | needed); |
32a7627c N |
1046 | bit_cnt++; |
1047 | } | |
27581e5a | 1048 | offset = 0; |
32a7627c N |
1049 | } |
1050 | ||
32a7627c | 1051 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
d1244cb0 | 1052 | "read %lu pages, set %lu of %lu bits\n", |
1ec885cd | 1053 | bmname(bitmap), store->file_pages, |
d1244cb0 | 1054 | bit_cnt, chunks); |
4ad13663 N |
1055 | |
1056 | return 0; | |
32a7627c | 1057 | |
4ad13663 N |
1058 | err: |
1059 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1060 | bmname(bitmap), ret); | |
32a7627c N |
1061 | return ret; |
1062 | } | |
1063 | ||
a654b9d8 N |
1064 | void bitmap_write_all(struct bitmap *bitmap) |
1065 | { | |
1066 | /* We don't actually write all bitmap blocks here, | |
1067 | * just flag them as needing to be written | |
1068 | */ | |
ec7a3197 | 1069 | int i; |
a654b9d8 | 1070 | |
1ec885cd | 1071 | if (!bitmap || !bitmap->storage.filemap) |
ef99bf48 | 1072 | return; |
1ec885cd | 1073 | if (bitmap->storage.file) |
ef99bf48 N |
1074 | /* Only one copy, so nothing needed */ |
1075 | return; | |
1076 | ||
1ec885cd | 1077 | for (i = 0; i < bitmap->storage.file_pages; i++) |
d189122d | 1078 | set_page_attr(bitmap, i, |
ec7a3197 | 1079 | BITMAP_PAGE_NEEDWRITE); |
2585f3ef | 1080 | bitmap->allclean = 0; |
a654b9d8 N |
1081 | } |
1082 | ||
40cffcc0 N |
1083 | static void bitmap_count_page(struct bitmap_counts *bitmap, |
1084 | sector_t offset, int inc) | |
32a7627c | 1085 | { |
61a0d80c | 1086 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1087 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1088 | bitmap->bp[page].count += inc; | |
32a7627c N |
1089 | bitmap_checkfree(bitmap, page); |
1090 | } | |
bf07bb7d | 1091 | |
40cffcc0 | 1092 | static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset) |
bf07bb7d N |
1093 | { |
1094 | sector_t chunk = offset >> bitmap->chunkshift; | |
1095 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1096 | struct bitmap_page *bp = &bitmap->bp[page]; | |
1097 | ||
1098 | if (!bp->pending) | |
1099 | bp->pending = 1; | |
1100 | } | |
1101 | ||
40cffcc0 | 1102 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1103 | sector_t offset, sector_t *blocks, |
32a7627c N |
1104 | int create); |
1105 | ||
1106 | /* | |
1107 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1108 | * out to disk | |
1109 | */ | |
1110 | ||
fd01b88c | 1111 | void bitmap_daemon_work(struct mddev *mddev) |
32a7627c | 1112 | { |
aa5cbd10 | 1113 | struct bitmap *bitmap; |
aa3163f8 | 1114 | unsigned long j; |
bf07bb7d | 1115 | unsigned long nextpage; |
57dab0bd | 1116 | sector_t blocks; |
40cffcc0 | 1117 | struct bitmap_counts *counts; |
32a7627c | 1118 | |
aa5cbd10 N |
1119 | /* Use a mutex to guard daemon_work against |
1120 | * bitmap_destroy. | |
1121 | */ | |
c3d9714e | 1122 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1123 | bitmap = mddev->bitmap; |
1124 | if (bitmap == NULL) { | |
c3d9714e | 1125 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1126 | return; |
aa5cbd10 | 1127 | } |
42a04b50 | 1128 | if (time_before(jiffies, bitmap->daemon_lastrun |
2e61ebbc | 1129 | + mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1130 | goto done; |
1131 | ||
32a7627c | 1132 | bitmap->daemon_lastrun = jiffies; |
8311c29d | 1133 | if (bitmap->allclean) { |
2e61ebbc | 1134 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
aa5cbd10 | 1135 | goto done; |
8311c29d N |
1136 | } |
1137 | bitmap->allclean = 1; | |
32a7627c | 1138 | |
bf07bb7d N |
1139 | /* Any file-page which is PENDING now needs to be written. |
1140 | * So set NEEDWRITE now, then after we make any last-minute changes | |
1141 | * we will write it. | |
1142 | */ | |
1ec885cd | 1143 | for (j = 0; j < bitmap->storage.file_pages; j++) |
bdfd1140 N |
1144 | if (test_and_clear_page_attr(bitmap, j, |
1145 | BITMAP_PAGE_PENDING)) | |
d189122d | 1146 | set_page_attr(bitmap, j, |
bf07bb7d | 1147 | BITMAP_PAGE_NEEDWRITE); |
bf07bb7d N |
1148 | |
1149 | if (bitmap->need_sync && | |
1150 | mddev->bitmap_info.external == 0) { | |
1151 | /* Arrange for superblock update as well as | |
1152 | * other changes */ | |
1153 | bitmap_super_t *sb; | |
1154 | bitmap->need_sync = 0; | |
1ec885cd N |
1155 | if (bitmap->storage.filemap) { |
1156 | sb = kmap_atomic(bitmap->storage.sb_page); | |
ef99bf48 N |
1157 | sb->events_cleared = |
1158 | cpu_to_le64(bitmap->events_cleared); | |
1159 | kunmap_atomic(sb); | |
d189122d | 1160 | set_page_attr(bitmap, 0, |
ef99bf48 N |
1161 | BITMAP_PAGE_NEEDWRITE); |
1162 | } | |
bf07bb7d N |
1163 | } |
1164 | /* Now look at the bitmap counters and if any are '2' or '1', | |
1165 | * decrement and handle accordingly. | |
1166 | */ | |
40cffcc0 N |
1167 | counts = &bitmap->counts; |
1168 | spin_lock_irq(&counts->lock); | |
bf07bb7d | 1169 | nextpage = 0; |
40cffcc0 | 1170 | for (j = 0; j < counts->chunks; j++) { |
32a7627c | 1171 | bitmap_counter_t *bmc; |
40cffcc0 | 1172 | sector_t block = (sector_t)j << counts->chunkshift; |
3520fa4d | 1173 | |
bf07bb7d N |
1174 | if (j == nextpage) { |
1175 | nextpage += PAGE_COUNTER_RATIO; | |
40cffcc0 | 1176 | if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) { |
bf07bb7d | 1177 | j |= PAGE_COUNTER_MASK; |
aa3163f8 N |
1178 | continue; |
1179 | } | |
40cffcc0 | 1180 | counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; |
32a7627c | 1181 | } |
40cffcc0 | 1182 | bmc = bitmap_get_counter(counts, |
ef99bf48 | 1183 | block, |
db305e50 | 1184 | &blocks, 0); |
bf07bb7d N |
1185 | |
1186 | if (!bmc) { | |
5a537df4 | 1187 | j |= PAGE_COUNTER_MASK; |
bf07bb7d N |
1188 | continue; |
1189 | } | |
1190 | if (*bmc == 1 && !bitmap->need_sync) { | |
1191 | /* We can clear the bit */ | |
bf07bb7d | 1192 | *bmc = 0; |
40cffcc0 | 1193 | bitmap_count_page(counts, block, -1); |
ef99bf48 | 1194 | bitmap_file_clear_bit(bitmap, block); |
bf07bb7d N |
1195 | } else if (*bmc && *bmc <= 2) { |
1196 | *bmc = 1; | |
40cffcc0 | 1197 | bitmap_set_pending(counts, block); |
bf07bb7d | 1198 | bitmap->allclean = 0; |
5a537df4 | 1199 | } |
32a7627c | 1200 | } |
40cffcc0 | 1201 | spin_unlock_irq(&counts->lock); |
32a7627c | 1202 | |
bf07bb7d N |
1203 | /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. |
1204 | * DIRTY pages need to be written by bitmap_unplug so it can wait | |
1205 | * for them. | |
1206 | * If we find any DIRTY page we stop there and let bitmap_unplug | |
1207 | * handle all the rest. This is important in the case where | |
1208 | * the first blocking holds the superblock and it has been updated. | |
1209 | * We mustn't write any other blocks before the superblock. | |
1210 | */ | |
62f82faa N |
1211 | for (j = 0; |
1212 | j < bitmap->storage.file_pages | |
1213 | && !test_bit(BITMAP_STALE, &bitmap->flags); | |
1214 | j++) { | |
bf07bb7d | 1215 | |
d189122d | 1216 | if (test_page_attr(bitmap, j, |
bf07bb7d N |
1217 | BITMAP_PAGE_DIRTY)) |
1218 | /* bitmap_unplug will handle the rest */ | |
1219 | break; | |
bdfd1140 N |
1220 | if (test_and_clear_page_attr(bitmap, j, |
1221 | BITMAP_PAGE_NEEDWRITE)) { | |
1ec885cd | 1222 | write_page(bitmap, bitmap->storage.filemap[j], 0); |
32a7627c | 1223 | } |
32a7627c N |
1224 | } |
1225 | ||
7be3dfec | 1226 | done: |
8311c29d | 1227 | if (bitmap->allclean == 0) |
2e61ebbc N |
1228 | mddev->thread->timeout = |
1229 | mddev->bitmap_info.daemon_sleep; | |
c3d9714e | 1230 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1231 | } |
1232 | ||
40cffcc0 | 1233 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1234 | sector_t offset, sector_t *blocks, |
32a7627c | 1235 | int create) |
ee305ace N |
1236 | __releases(bitmap->lock) |
1237 | __acquires(bitmap->lock) | |
32a7627c N |
1238 | { |
1239 | /* If 'create', we might release the lock and reclaim it. | |
1240 | * The lock must have been taken with interrupts enabled. | |
1241 | * If !create, we don't release the lock. | |
1242 | */ | |
61a0d80c | 1243 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1244 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1245 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1246 | sector_t csize; | |
ef425673 | 1247 | int err; |
32a7627c | 1248 | |
ef425673 N |
1249 | err = bitmap_checkpage(bitmap, page, create); |
1250 | ||
1251 | if (bitmap->bp[page].hijacked || | |
1252 | bitmap->bp[page].map == NULL) | |
61a0d80c | 1253 | csize = ((sector_t)1) << (bitmap->chunkshift + |
ef425673 N |
1254 | PAGE_COUNTER_SHIFT - 1); |
1255 | else | |
61a0d80c | 1256 | csize = ((sector_t)1) << bitmap->chunkshift; |
ef425673 N |
1257 | *blocks = csize - (offset & (csize - 1)); |
1258 | ||
1259 | if (err < 0) | |
32a7627c | 1260 | return NULL; |
ef425673 | 1261 | |
32a7627c N |
1262 | /* now locked ... */ |
1263 | ||
1264 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1265 | /* should we use the first or second counter field | |
1266 | * of the hijacked pointer? */ | |
1267 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1268 | return &((bitmap_counter_t *) |
1269 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1270 | } else /* page is allocated */ |
32a7627c N |
1271 | return (bitmap_counter_t *) |
1272 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1273 | } |
1274 | ||
4b6d287f | 1275 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1276 | { |
ac2f40be N |
1277 | if (!bitmap) |
1278 | return 0; | |
4b6d287f N |
1279 | |
1280 | if (behind) { | |
696fcd53 | 1281 | int bw; |
4b6d287f | 1282 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1283 | bw = atomic_read(&bitmap->behind_writes); |
1284 | if (bw > bitmap->behind_writes_used) | |
1285 | bitmap->behind_writes_used = bw; | |
1286 | ||
36a4e1fe N |
1287 | pr_debug("inc write-behind count %d/%lu\n", |
1288 | bw, bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1289 | } |
1290 | ||
32a7627c | 1291 | while (sectors) { |
57dab0bd | 1292 | sector_t blocks; |
32a7627c N |
1293 | bitmap_counter_t *bmc; |
1294 | ||
40cffcc0 N |
1295 | spin_lock_irq(&bitmap->counts.lock); |
1296 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1); | |
32a7627c | 1297 | if (!bmc) { |
40cffcc0 | 1298 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1299 | return 0; |
1300 | } | |
1301 | ||
27d5ea04 | 1302 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
da6e1a32 NB |
1303 | DEFINE_WAIT(__wait); |
1304 | /* note that it is safe to do the prepare_to_wait | |
1305 | * after the test as long as we do it before dropping | |
1306 | * the spinlock. | |
1307 | */ | |
1308 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1309 | TASK_UNINTERRUPTIBLE); | |
40cffcc0 | 1310 | spin_unlock_irq(&bitmap->counts.lock); |
f54a9d0e | 1311 | schedule(); |
da6e1a32 NB |
1312 | finish_wait(&bitmap->overflow_wait, &__wait); |
1313 | continue; | |
1314 | } | |
1315 | ||
ac2f40be | 1316 | switch (*bmc) { |
32a7627c N |
1317 | case 0: |
1318 | bitmap_file_set_bit(bitmap, offset); | |
40cffcc0 | 1319 | bitmap_count_page(&bitmap->counts, offset, 1); |
32a7627c N |
1320 | /* fall through */ |
1321 | case 1: | |
1322 | *bmc = 2; | |
1323 | } | |
da6e1a32 | 1324 | |
32a7627c N |
1325 | (*bmc)++; |
1326 | ||
40cffcc0 | 1327 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1328 | |
1329 | offset += blocks; | |
1330 | if (sectors > blocks) | |
1331 | sectors -= blocks; | |
ac2f40be N |
1332 | else |
1333 | sectors = 0; | |
32a7627c N |
1334 | } |
1335 | return 0; | |
1336 | } | |
ac2f40be | 1337 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1338 | |
1339 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1340 | int success, int behind) |
32a7627c | 1341 | { |
ac2f40be N |
1342 | if (!bitmap) |
1343 | return; | |
4b6d287f | 1344 | if (behind) { |
e555190d N |
1345 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1346 | wake_up(&bitmap->behind_wait); | |
36a4e1fe N |
1347 | pr_debug("dec write-behind count %d/%lu\n", |
1348 | atomic_read(&bitmap->behind_writes), | |
1349 | bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1350 | } |
1351 | ||
32a7627c | 1352 | while (sectors) { |
57dab0bd | 1353 | sector_t blocks; |
32a7627c N |
1354 | unsigned long flags; |
1355 | bitmap_counter_t *bmc; | |
1356 | ||
40cffcc0 N |
1357 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1358 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0); | |
32a7627c | 1359 | if (!bmc) { |
40cffcc0 | 1360 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1361 | return; |
1362 | } | |
1363 | ||
961902c0 | 1364 | if (success && !bitmap->mddev->degraded && |
a0da84f3 NB |
1365 | bitmap->events_cleared < bitmap->mddev->events) { |
1366 | bitmap->events_cleared = bitmap->mddev->events; | |
1367 | bitmap->need_sync = 1; | |
5ff5afff | 1368 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1369 | } |
1370 | ||
27d5ea04 | 1371 | if (!success && !NEEDED(*bmc)) |
32a7627c N |
1372 | *bmc |= NEEDED_MASK; |
1373 | ||
27d5ea04 | 1374 | if (COUNTER(*bmc) == COUNTER_MAX) |
da6e1a32 NB |
1375 | wake_up(&bitmap->overflow_wait); |
1376 | ||
32a7627c | 1377 | (*bmc)--; |
2585f3ef | 1378 | if (*bmc <= 2) { |
40cffcc0 | 1379 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1380 | bitmap->allclean = 0; |
1381 | } | |
40cffcc0 | 1382 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1383 | offset += blocks; |
1384 | if (sectors > blocks) | |
1385 | sectors -= blocks; | |
ac2f40be N |
1386 | else |
1387 | sectors = 0; | |
32a7627c N |
1388 | } |
1389 | } | |
ac2f40be | 1390 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1391 | |
57dab0bd | 1392 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1393 | int degraded) |
32a7627c N |
1394 | { |
1395 | bitmap_counter_t *bmc; | |
1396 | int rv; | |
1397 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1398 | *blocks = 1024; | |
1399 | return 1; /* always resync if no bitmap */ | |
1400 | } | |
40cffcc0 N |
1401 | spin_lock_irq(&bitmap->counts.lock); |
1402 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1403 | rv = 0; |
1404 | if (bmc) { | |
1405 | /* locked */ | |
1406 | if (RESYNC(*bmc)) | |
1407 | rv = 1; | |
1408 | else if (NEEDED(*bmc)) { | |
1409 | rv = 1; | |
6a806c51 N |
1410 | if (!degraded) { /* don't set/clear bits if degraded */ |
1411 | *bmc |= RESYNC_MASK; | |
1412 | *bmc &= ~NEEDED_MASK; | |
1413 | } | |
32a7627c N |
1414 | } |
1415 | } | |
40cffcc0 | 1416 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1417 | return rv; |
1418 | } | |
1419 | ||
57dab0bd | 1420 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a N |
1421 | int degraded) |
1422 | { | |
1423 | /* bitmap_start_sync must always report on multiples of whole | |
1424 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1425 | * get confused. | |
1426 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1427 | * At least PAGE_SIZE>>9 blocks are covered. | |
1428 | * Return the 'or' of the result. | |
1429 | */ | |
1430 | int rv = 0; | |
57dab0bd | 1431 | sector_t blocks1; |
1187cf0a N |
1432 | |
1433 | *blocks = 0; | |
1434 | while (*blocks < (PAGE_SIZE>>9)) { | |
1435 | rv |= __bitmap_start_sync(bitmap, offset, | |
1436 | &blocks1, degraded); | |
1437 | offset += blocks1; | |
1438 | *blocks += blocks1; | |
1439 | } | |
1440 | return rv; | |
1441 | } | |
ac2f40be | 1442 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1443 | |
57dab0bd | 1444 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1445 | { |
1446 | bitmap_counter_t *bmc; | |
1447 | unsigned long flags; | |
ac2f40be N |
1448 | |
1449 | if (bitmap == NULL) { | |
32a7627c N |
1450 | *blocks = 1024; |
1451 | return; | |
1452 | } | |
40cffcc0 N |
1453 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1454 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1455 | if (bmc == NULL) |
1456 | goto unlock; | |
1457 | /* locked */ | |
32a7627c N |
1458 | if (RESYNC(*bmc)) { |
1459 | *bmc &= ~RESYNC_MASK; | |
1460 | ||
1461 | if (!NEEDED(*bmc) && aborted) | |
1462 | *bmc |= NEEDED_MASK; | |
1463 | else { | |
2585f3ef | 1464 | if (*bmc <= 2) { |
40cffcc0 | 1465 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1466 | bitmap->allclean = 0; |
1467 | } | |
32a7627c N |
1468 | } |
1469 | } | |
1470 | unlock: | |
40cffcc0 | 1471 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c | 1472 | } |
ac2f40be | 1473 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1474 | |
1475 | void bitmap_close_sync(struct bitmap *bitmap) | |
1476 | { | |
1477 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1478 | * properly have been aborted. It remains to us to clear the | |
1479 | * RESYNC bit wherever it is still on | |
1480 | */ | |
1481 | sector_t sector = 0; | |
57dab0bd | 1482 | sector_t blocks; |
b47490c9 N |
1483 | if (!bitmap) |
1484 | return; | |
32a7627c N |
1485 | while (sector < bitmap->mddev->resync_max_sectors) { |
1486 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1487 | sector += blocks; |
1488 | } | |
1489 | } | |
ac2f40be | 1490 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 N |
1491 | |
1492 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1493 | { | |
1494 | sector_t s = 0; | |
57dab0bd | 1495 | sector_t blocks; |
b47490c9 N |
1496 | |
1497 | if (!bitmap) | |
1498 | return; | |
1499 | if (sector == 0) { | |
1500 | bitmap->last_end_sync = jiffies; | |
1501 | return; | |
1502 | } | |
1503 | if (time_before(jiffies, (bitmap->last_end_sync | |
1b04be96 | 1504 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1505 | return; |
1506 | wait_event(bitmap->mddev->recovery_wait, | |
1507 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1508 | ||
75d3da43 | 1509 | bitmap->mddev->curr_resync_completed = sector; |
070dc6dd | 1510 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
40cffcc0 | 1511 | sector &= ~((1ULL << bitmap->counts.chunkshift) - 1); |
b47490c9 N |
1512 | s = 0; |
1513 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1514 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1515 | s += blocks; | |
32a7627c | 1516 | } |
b47490c9 | 1517 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1518 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1519 | } |
ac2f40be | 1520 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1521 | |
6a07997f | 1522 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1523 | { |
1524 | /* For each chunk covered by any of these sectors, set the | |
ef99bf48 | 1525 | * counter to 2 and possibly set resync_needed. They should all |
32a7627c N |
1526 | * be 0 at this point |
1527 | */ | |
193f1c93 | 1528 | |
57dab0bd | 1529 | sector_t secs; |
193f1c93 | 1530 | bitmap_counter_t *bmc; |
40cffcc0 N |
1531 | spin_lock_irq(&bitmap->counts.lock); |
1532 | bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1); | |
193f1c93 | 1533 | if (!bmc) { |
40cffcc0 | 1534 | spin_unlock_irq(&bitmap->counts.lock); |
193f1c93 | 1535 | return; |
32a7627c | 1536 | } |
ac2f40be | 1537 | if (!*bmc) { |
915c420d | 1538 | *bmc = 2 | (needed ? NEEDED_MASK : 0); |
40cffcc0 N |
1539 | bitmap_count_page(&bitmap->counts, offset, 1); |
1540 | bitmap_set_pending(&bitmap->counts, offset); | |
2585f3ef | 1541 | bitmap->allclean = 0; |
193f1c93 | 1542 | } |
40cffcc0 | 1543 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1544 | } |
1545 | ||
9b1d1dac PC |
1546 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1547 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1548 | { | |
1549 | unsigned long chunk; | |
1550 | ||
1551 | for (chunk = s; chunk <= e; chunk++) { | |
40cffcc0 | 1552 | sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift; |
9b1d1dac PC |
1553 | bitmap_set_memory_bits(bitmap, sec, 1); |
1554 | bitmap_file_set_bit(bitmap, sec); | |
ffa23322 N |
1555 | if (sec < bitmap->mddev->recovery_cp) |
1556 | /* We are asserting that the array is dirty, | |
1557 | * so move the recovery_cp address back so | |
1558 | * that it is obvious that it is dirty | |
1559 | */ | |
1560 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1561 | } |
1562 | } | |
1563 | ||
6b8b3e8a N |
1564 | /* |
1565 | * flush out any pending updates | |
1566 | */ | |
fd01b88c | 1567 | void bitmap_flush(struct mddev *mddev) |
6b8b3e8a N |
1568 | { |
1569 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1570 | long sleep; |
6b8b3e8a N |
1571 | |
1572 | if (!bitmap) /* there was no bitmap */ | |
1573 | return; | |
1574 | ||
1575 | /* run the daemon_work three time to ensure everything is flushed | |
1576 | * that can be | |
1577 | */ | |
1b04be96 | 1578 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1579 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1580 | bitmap_daemon_work(mddev); |
42a04b50 | 1581 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1582 | bitmap_daemon_work(mddev); |
42a04b50 | 1583 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1584 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1585 | bitmap_update_sb(bitmap); |
1586 | } | |
1587 | ||
32a7627c N |
1588 | /* |
1589 | * free memory that was allocated | |
1590 | */ | |
3178b0db | 1591 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1592 | { |
1593 | unsigned long k, pages; | |
1594 | struct bitmap_page *bp; | |
32a7627c N |
1595 | |
1596 | if (!bitmap) /* there was no bitmap */ | |
1597 | return; | |
1598 | ||
fae7d326 N |
1599 | /* Shouldn't be needed - but just in case.... */ |
1600 | wait_event(bitmap->write_wait, | |
1601 | atomic_read(&bitmap->pending_writes) == 0); | |
1602 | ||
1603 | /* release the bitmap file */ | |
1604 | bitmap_file_unmap(&bitmap->storage); | |
32a7627c | 1605 | |
40cffcc0 N |
1606 | bp = bitmap->counts.bp; |
1607 | pages = bitmap->counts.pages; | |
32a7627c N |
1608 | |
1609 | /* free all allocated memory */ | |
1610 | ||
32a7627c N |
1611 | if (bp) /* deallocate the page memory */ |
1612 | for (k = 0; k < pages; k++) | |
1613 | if (bp[k].map && !bp[k].hijacked) | |
1614 | kfree(bp[k].map); | |
1615 | kfree(bp); | |
1616 | kfree(bitmap); | |
1617 | } | |
aa5cbd10 | 1618 | |
fd01b88c | 1619 | void bitmap_destroy(struct mddev *mddev) |
3178b0db N |
1620 | { |
1621 | struct bitmap *bitmap = mddev->bitmap; | |
1622 | ||
1623 | if (!bitmap) /* there was no bitmap */ | |
1624 | return; | |
1625 | ||
c3d9714e | 1626 | mutex_lock(&mddev->bitmap_info.mutex); |
978a7a47 | 1627 | spin_lock(&mddev->lock); |
3178b0db | 1628 | mddev->bitmap = NULL; /* disconnect from the md device */ |
978a7a47 | 1629 | spin_unlock(&mddev->lock); |
c3d9714e | 1630 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1631 | if (mddev->thread) |
1632 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db | 1633 | |
ece5cff0 N |
1634 | if (bitmap->sysfs_can_clear) |
1635 | sysfs_put(bitmap->sysfs_can_clear); | |
1636 | ||
3178b0db N |
1637 | bitmap_free(bitmap); |
1638 | } | |
32a7627c N |
1639 | |
1640 | /* | |
1641 | * initialize the bitmap structure | |
1642 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1643 | */ | |
fd01b88c | 1644 | int bitmap_create(struct mddev *mddev) |
32a7627c N |
1645 | { |
1646 | struct bitmap *bitmap; | |
1f593903 | 1647 | sector_t blocks = mddev->resync_max_sectors; |
c3d9714e | 1648 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1649 | int err; |
324a56e1 | 1650 | struct kernfs_node *bm = NULL; |
32a7627c | 1651 | |
5f6e3c83 | 1652 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1653 | |
c3d9714e | 1654 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1655 | |
9ffae0cf | 1656 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1657 | if (!bitmap) |
1658 | return -ENOMEM; | |
1659 | ||
40cffcc0 | 1660 | spin_lock_init(&bitmap->counts.lock); |
ce25c31b N |
1661 | atomic_set(&bitmap->pending_writes, 0); |
1662 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1663 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1664 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1665 | |
32a7627c | 1666 | bitmap->mddev = mddev; |
32a7627c | 1667 | |
5ff5afff | 1668 | if (mddev->kobj.sd) |
388975cc | 1669 | bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap"); |
ece5cff0 | 1670 | if (bm) { |
388975cc | 1671 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear"); |
ece5cff0 N |
1672 | sysfs_put(bm); |
1673 | } else | |
1674 | bitmap->sysfs_can_clear = NULL; | |
1675 | ||
1ec885cd | 1676 | bitmap->storage.file = file; |
ce25c31b N |
1677 | if (file) { |
1678 | get_file(file); | |
ae8fa283 N |
1679 | /* As future accesses to this file will use bmap, |
1680 | * and bypass the page cache, we must sync the file | |
1681 | * first. | |
1682 | */ | |
8018ab05 | 1683 | vfs_fsync(file, 1); |
ce25c31b | 1684 | } |
42a04b50 | 1685 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
9c81075f JB |
1686 | if (!mddev->bitmap_info.external) { |
1687 | /* | |
1688 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is | |
1689 | * instructing us to create a new on-disk bitmap instance. | |
1690 | */ | |
1691 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) | |
1692 | err = bitmap_new_disk_sb(bitmap); | |
1693 | else | |
1694 | err = bitmap_read_sb(bitmap); | |
1695 | } else { | |
ece5cff0 N |
1696 | err = 0; |
1697 | if (mddev->bitmap_info.chunksize == 0 || | |
1698 | mddev->bitmap_info.daemon_sleep == 0) | |
1699 | /* chunksize and time_base need to be | |
1700 | * set first. */ | |
1701 | err = -EINVAL; | |
1702 | } | |
32a7627c | 1703 | if (err) |
3178b0db | 1704 | goto error; |
32a7627c | 1705 | |
624ce4f5 | 1706 | bitmap->daemon_lastrun = jiffies; |
d60b479d N |
1707 | err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1); |
1708 | if (err) | |
3178b0db | 1709 | goto error; |
32a7627c | 1710 | |
69e51b44 | 1711 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", |
d60b479d | 1712 | bitmap->counts.pages, bmname(bitmap)); |
69e51b44 N |
1713 | |
1714 | mddev->bitmap = bitmap; | |
b405fe91 | 1715 | return test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0; |
69e51b44 N |
1716 | |
1717 | error: | |
1718 | bitmap_free(bitmap); | |
1719 | return err; | |
1720 | } | |
1721 | ||
fd01b88c | 1722 | int bitmap_load(struct mddev *mddev) |
69e51b44 N |
1723 | { |
1724 | int err = 0; | |
3520fa4d | 1725 | sector_t start = 0; |
69e51b44 N |
1726 | sector_t sector = 0; |
1727 | struct bitmap *bitmap = mddev->bitmap; | |
1728 | ||
1729 | if (!bitmap) | |
1730 | goto out; | |
1731 | ||
1732 | /* Clear out old bitmap info first: Either there is none, or we | |
1733 | * are resuming after someone else has possibly changed things, | |
1734 | * so we should forget old cached info. | |
1735 | * All chunks should be clean, but some might need_sync. | |
1736 | */ | |
1737 | while (sector < mddev->resync_max_sectors) { | |
57dab0bd | 1738 | sector_t blocks; |
69e51b44 N |
1739 | bitmap_start_sync(bitmap, sector, &blocks, 0); |
1740 | sector += blocks; | |
1741 | } | |
1742 | bitmap_close_sync(bitmap); | |
1743 | ||
3520fa4d JB |
1744 | if (mddev->degraded == 0 |
1745 | || bitmap->events_cleared == mddev->events) | |
1746 | /* no need to keep dirty bits to optimise a | |
1747 | * re-add of a missing device */ | |
1748 | start = mddev->recovery_cp; | |
1749 | ||
afbaa90b | 1750 | mutex_lock(&mddev->bitmap_info.mutex); |
3520fa4d | 1751 | err = bitmap_init_from_disk(bitmap, start); |
afbaa90b | 1752 | mutex_unlock(&mddev->bitmap_info.mutex); |
3520fa4d | 1753 | |
32a7627c | 1754 | if (err) |
69e51b44 | 1755 | goto out; |
b405fe91 | 1756 | clear_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
1757 | |
1758 | /* Kick recovery in case any bits were set */ | |
1759 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
3178b0db | 1760 | |
1b04be96 | 1761 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1762 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1763 | |
4ad13663 N |
1764 | bitmap_update_sb(bitmap); |
1765 | ||
b405fe91 | 1766 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
69e51b44 N |
1767 | err = -EIO; |
1768 | out: | |
3178b0db | 1769 | return err; |
32a7627c | 1770 | } |
69e51b44 | 1771 | EXPORT_SYMBOL_GPL(bitmap_load); |
32a7627c | 1772 | |
57148964 N |
1773 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap) |
1774 | { | |
1775 | unsigned long chunk_kb; | |
40cffcc0 | 1776 | struct bitmap_counts *counts; |
57148964 N |
1777 | |
1778 | if (!bitmap) | |
1779 | return; | |
1780 | ||
40cffcc0 N |
1781 | counts = &bitmap->counts; |
1782 | ||
57148964 N |
1783 | chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; |
1784 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
1785 | "%lu%s chunk", | |
40cffcc0 N |
1786 | counts->pages - counts->missing_pages, |
1787 | counts->pages, | |
1788 | (counts->pages - counts->missing_pages) | |
57148964 N |
1789 | << (PAGE_SHIFT - 10), |
1790 | chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, | |
1791 | chunk_kb ? "KB" : "B"); | |
1ec885cd | 1792 | if (bitmap->storage.file) { |
57148964 | 1793 | seq_printf(seq, ", file: "); |
1ec885cd | 1794 | seq_path(seq, &bitmap->storage.file->f_path, " \t\n"); |
57148964 N |
1795 | } |
1796 | ||
1797 | seq_printf(seq, "\n"); | |
57148964 N |
1798 | } |
1799 | ||
d60b479d N |
1800 | int bitmap_resize(struct bitmap *bitmap, sector_t blocks, |
1801 | int chunksize, int init) | |
1802 | { | |
1803 | /* If chunk_size is 0, choose an appropriate chunk size. | |
1804 | * Then possibly allocate new storage space. | |
1805 | * Then quiesce, copy bits, replace bitmap, and re-start | |
1806 | * | |
1807 | * This function is called both to set up the initial bitmap | |
1808 | * and to resize the bitmap while the array is active. | |
1809 | * If this happens as a result of the array being resized, | |
1810 | * chunksize will be zero, and we need to choose a suitable | |
1811 | * chunksize, otherwise we use what we are given. | |
1812 | */ | |
1813 | struct bitmap_storage store; | |
1814 | struct bitmap_counts old_counts; | |
1815 | unsigned long chunks; | |
1816 | sector_t block; | |
1817 | sector_t old_blocks, new_blocks; | |
1818 | int chunkshift; | |
1819 | int ret = 0; | |
1820 | long pages; | |
1821 | struct bitmap_page *new_bp; | |
1822 | ||
1823 | if (chunksize == 0) { | |
1824 | /* If there is enough space, leave the chunk size unchanged, | |
1825 | * else increase by factor of two until there is enough space. | |
1826 | */ | |
1827 | long bytes; | |
1828 | long space = bitmap->mddev->bitmap_info.space; | |
1829 | ||
1830 | if (space == 0) { | |
1831 | /* We don't know how much space there is, so limit | |
1832 | * to current size - in sectors. | |
1833 | */ | |
1834 | bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8); | |
1835 | if (!bitmap->mddev->bitmap_info.external) | |
1836 | bytes += sizeof(bitmap_super_t); | |
1837 | space = DIV_ROUND_UP(bytes, 512); | |
1838 | bitmap->mddev->bitmap_info.space = space; | |
1839 | } | |
1840 | chunkshift = bitmap->counts.chunkshift; | |
1841 | chunkshift--; | |
1842 | do { | |
1843 | /* 'chunkshift' is shift from block size to chunk size */ | |
1844 | chunkshift++; | |
1845 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
1846 | bytes = DIV_ROUND_UP(chunks, 8); | |
1847 | if (!bitmap->mddev->bitmap_info.external) | |
1848 | bytes += sizeof(bitmap_super_t); | |
1849 | } while (bytes > (space << 9)); | |
1850 | } else | |
1851 | chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT; | |
1852 | ||
1853 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
1854 | memset(&store, 0, sizeof(store)); | |
1855 | if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file) | |
1856 | ret = bitmap_storage_alloc(&store, chunks, | |
1857 | !bitmap->mddev->bitmap_info.external); | |
1858 | if (ret) | |
1859 | goto err; | |
1860 | ||
1861 | pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO); | |
1862 | ||
1863 | new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL); | |
1864 | ret = -ENOMEM; | |
1865 | if (!new_bp) { | |
1866 | bitmap_file_unmap(&store); | |
1867 | goto err; | |
1868 | } | |
1869 | ||
1870 | if (!init) | |
1871 | bitmap->mddev->pers->quiesce(bitmap->mddev, 1); | |
1872 | ||
1873 | store.file = bitmap->storage.file; | |
1874 | bitmap->storage.file = NULL; | |
1875 | ||
1876 | if (store.sb_page && bitmap->storage.sb_page) | |
1877 | memcpy(page_address(store.sb_page), | |
1878 | page_address(bitmap->storage.sb_page), | |
1879 | sizeof(bitmap_super_t)); | |
1880 | bitmap_file_unmap(&bitmap->storage); | |
1881 | bitmap->storage = store; | |
1882 | ||
1883 | old_counts = bitmap->counts; | |
1884 | bitmap->counts.bp = new_bp; | |
1885 | bitmap->counts.pages = pages; | |
1886 | bitmap->counts.missing_pages = pages; | |
1887 | bitmap->counts.chunkshift = chunkshift; | |
1888 | bitmap->counts.chunks = chunks; | |
1889 | bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift + | |
1890 | BITMAP_BLOCK_SHIFT); | |
1891 | ||
1892 | blocks = min(old_counts.chunks << old_counts.chunkshift, | |
1893 | chunks << chunkshift); | |
1894 | ||
1895 | spin_lock_irq(&bitmap->counts.lock); | |
1896 | for (block = 0; block < blocks; ) { | |
1897 | bitmap_counter_t *bmc_old, *bmc_new; | |
1898 | int set; | |
1899 | ||
1900 | bmc_old = bitmap_get_counter(&old_counts, block, | |
1901 | &old_blocks, 0); | |
1902 | set = bmc_old && NEEDED(*bmc_old); | |
1903 | ||
1904 | if (set) { | |
1905 | bmc_new = bitmap_get_counter(&bitmap->counts, block, | |
1906 | &new_blocks, 1); | |
1907 | if (*bmc_new == 0) { | |
1908 | /* need to set on-disk bits too. */ | |
1909 | sector_t end = block + new_blocks; | |
1910 | sector_t start = block >> chunkshift; | |
1911 | start <<= chunkshift; | |
1912 | while (start < end) { | |
1913 | bitmap_file_set_bit(bitmap, block); | |
1914 | start += 1 << chunkshift; | |
1915 | } | |
1916 | *bmc_new = 2; | |
1917 | bitmap_count_page(&bitmap->counts, | |
1918 | block, 1); | |
1919 | bitmap_set_pending(&bitmap->counts, | |
1920 | block); | |
1921 | } | |
1922 | *bmc_new |= NEEDED_MASK; | |
1923 | if (new_blocks < old_blocks) | |
1924 | old_blocks = new_blocks; | |
1925 | } | |
1926 | block += old_blocks; | |
1927 | } | |
1928 | ||
1929 | if (!init) { | |
1930 | int i; | |
1931 | while (block < (chunks << chunkshift)) { | |
1932 | bitmap_counter_t *bmc; | |
1933 | bmc = bitmap_get_counter(&bitmap->counts, block, | |
1934 | &new_blocks, 1); | |
1935 | if (bmc) { | |
1936 | /* new space. It needs to be resynced, so | |
1937 | * we set NEEDED_MASK. | |
1938 | */ | |
1939 | if (*bmc == 0) { | |
1940 | *bmc = NEEDED_MASK | 2; | |
1941 | bitmap_count_page(&bitmap->counts, | |
1942 | block, 1); | |
1943 | bitmap_set_pending(&bitmap->counts, | |
1944 | block); | |
1945 | } | |
1946 | } | |
1947 | block += new_blocks; | |
1948 | } | |
1949 | for (i = 0; i < bitmap->storage.file_pages; i++) | |
1950 | set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); | |
1951 | } | |
1952 | spin_unlock_irq(&bitmap->counts.lock); | |
1953 | ||
1954 | if (!init) { | |
1955 | bitmap_unplug(bitmap); | |
1956 | bitmap->mddev->pers->quiesce(bitmap->mddev, 0); | |
1957 | } | |
1958 | ret = 0; | |
1959 | err: | |
1960 | return ret; | |
1961 | } | |
1962 | EXPORT_SYMBOL_GPL(bitmap_resize); | |
1963 | ||
43a70507 | 1964 | static ssize_t |
fd01b88c | 1965 | location_show(struct mddev *mddev, char *page) |
43a70507 N |
1966 | { |
1967 | ssize_t len; | |
ac2f40be | 1968 | if (mddev->bitmap_info.file) |
43a70507 | 1969 | len = sprintf(page, "file"); |
ac2f40be | 1970 | else if (mddev->bitmap_info.offset) |
43a70507 | 1971 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 1972 | else |
43a70507 N |
1973 | len = sprintf(page, "none"); |
1974 | len += sprintf(page+len, "\n"); | |
1975 | return len; | |
1976 | } | |
1977 | ||
1978 | static ssize_t | |
fd01b88c | 1979 | location_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
1980 | { |
1981 | ||
1982 | if (mddev->pers) { | |
1983 | if (!mddev->pers->quiesce) | |
1984 | return -EBUSY; | |
1985 | if (mddev->recovery || mddev->sync_thread) | |
1986 | return -EBUSY; | |
1987 | } | |
1988 | ||
1989 | if (mddev->bitmap || mddev->bitmap_info.file || | |
1990 | mddev->bitmap_info.offset) { | |
1991 | /* bitmap already configured. Only option is to clear it */ | |
1992 | if (strncmp(buf, "none", 4) != 0) | |
1993 | return -EBUSY; | |
1994 | if (mddev->pers) { | |
1995 | mddev->pers->quiesce(mddev, 1); | |
1996 | bitmap_destroy(mddev); | |
1997 | mddev->pers->quiesce(mddev, 0); | |
1998 | } | |
1999 | mddev->bitmap_info.offset = 0; | |
2000 | if (mddev->bitmap_info.file) { | |
2001 | struct file *f = mddev->bitmap_info.file; | |
2002 | mddev->bitmap_info.file = NULL; | |
43a70507 N |
2003 | fput(f); |
2004 | } | |
2005 | } else { | |
2006 | /* No bitmap, OK to set a location */ | |
2007 | long long offset; | |
2008 | if (strncmp(buf, "none", 4) == 0) | |
2009 | /* nothing to be done */; | |
2010 | else if (strncmp(buf, "file:", 5) == 0) { | |
2011 | /* Not supported yet */ | |
2012 | return -EINVAL; | |
2013 | } else { | |
2014 | int rv; | |
2015 | if (buf[0] == '+') | |
b29bebd6 | 2016 | rv = kstrtoll(buf+1, 10, &offset); |
43a70507 | 2017 | else |
b29bebd6 | 2018 | rv = kstrtoll(buf, 10, &offset); |
43a70507 N |
2019 | if (rv) |
2020 | return rv; | |
2021 | if (offset == 0) | |
2022 | return -EINVAL; | |
ece5cff0 N |
2023 | if (mddev->bitmap_info.external == 0 && |
2024 | mddev->major_version == 0 && | |
43a70507 N |
2025 | offset != mddev->bitmap_info.default_offset) |
2026 | return -EINVAL; | |
2027 | mddev->bitmap_info.offset = offset; | |
2028 | if (mddev->pers) { | |
2029 | mddev->pers->quiesce(mddev, 1); | |
2030 | rv = bitmap_create(mddev); | |
4474ca42 N |
2031 | if (!rv) |
2032 | rv = bitmap_load(mddev); | |
43a70507 N |
2033 | if (rv) { |
2034 | bitmap_destroy(mddev); | |
2035 | mddev->bitmap_info.offset = 0; | |
2036 | } | |
2037 | mddev->pers->quiesce(mddev, 0); | |
2038 | if (rv) | |
2039 | return rv; | |
2040 | } | |
2041 | } | |
2042 | } | |
2043 | if (!mddev->external) { | |
2044 | /* Ensure new bitmap info is stored in | |
2045 | * metadata promptly. | |
2046 | */ | |
2047 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2048 | md_wakeup_thread(mddev->thread); | |
2049 | } | |
2050 | return len; | |
2051 | } | |
2052 | ||
2053 | static struct md_sysfs_entry bitmap_location = | |
2054 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
2055 | ||
6409bb05 N |
2056 | /* 'bitmap/space' is the space available at 'location' for the |
2057 | * bitmap. This allows the kernel to know when it is safe to | |
2058 | * resize the bitmap to match a resized array. | |
2059 | */ | |
2060 | static ssize_t | |
2061 | space_show(struct mddev *mddev, char *page) | |
2062 | { | |
2063 | return sprintf(page, "%lu\n", mddev->bitmap_info.space); | |
2064 | } | |
2065 | ||
2066 | static ssize_t | |
2067 | space_store(struct mddev *mddev, const char *buf, size_t len) | |
2068 | { | |
2069 | unsigned long sectors; | |
2070 | int rv; | |
2071 | ||
2072 | rv = kstrtoul(buf, 10, §ors); | |
2073 | if (rv) | |
2074 | return rv; | |
2075 | ||
2076 | if (sectors == 0) | |
2077 | return -EINVAL; | |
2078 | ||
2079 | if (mddev->bitmap && | |
9b1215c1 | 2080 | sectors < (mddev->bitmap->storage.bytes + 511) >> 9) |
6409bb05 N |
2081 | return -EFBIG; /* Bitmap is too big for this small space */ |
2082 | ||
2083 | /* could make sure it isn't too big, but that isn't really | |
2084 | * needed - user-space should be careful. | |
2085 | */ | |
2086 | mddev->bitmap_info.space = sectors; | |
2087 | return len; | |
2088 | } | |
2089 | ||
2090 | static struct md_sysfs_entry bitmap_space = | |
2091 | __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); | |
2092 | ||
43a70507 | 2093 | static ssize_t |
fd01b88c | 2094 | timeout_show(struct mddev *mddev, char *page) |
43a70507 N |
2095 | { |
2096 | ssize_t len; | |
2097 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
2098 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 2099 | |
43a70507 N |
2100 | len = sprintf(page, "%lu", secs); |
2101 | if (jifs) | |
2102 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
2103 | len += sprintf(page+len, "\n"); | |
2104 | return len; | |
2105 | } | |
2106 | ||
2107 | static ssize_t | |
fd01b88c | 2108 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2109 | { |
2110 | /* timeout can be set at any time */ | |
2111 | unsigned long timeout; | |
2112 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
2113 | if (rv) | |
2114 | return rv; | |
2115 | ||
2116 | /* just to make sure we don't overflow... */ | |
2117 | if (timeout >= LONG_MAX / HZ) | |
2118 | return -EINVAL; | |
2119 | ||
2120 | timeout = timeout * HZ / 10000; | |
2121 | ||
2122 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
2123 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
2124 | if (timeout < 1) | |
2125 | timeout = 1; | |
2126 | mddev->bitmap_info.daemon_sleep = timeout; | |
2127 | if (mddev->thread) { | |
2128 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
2129 | * the bitmap is all clean and we don't need to | |
2130 | * adjust the timeout right now | |
2131 | */ | |
2132 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2133 | mddev->thread->timeout = timeout; | |
2134 | md_wakeup_thread(mddev->thread); | |
2135 | } | |
2136 | } | |
2137 | return len; | |
2138 | } | |
2139 | ||
2140 | static struct md_sysfs_entry bitmap_timeout = | |
2141 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2142 | ||
2143 | static ssize_t | |
fd01b88c | 2144 | backlog_show(struct mddev *mddev, char *page) |
43a70507 N |
2145 | { |
2146 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2147 | } | |
2148 | ||
2149 | static ssize_t | |
fd01b88c | 2150 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2151 | { |
2152 | unsigned long backlog; | |
b29bebd6 | 2153 | int rv = kstrtoul(buf, 10, &backlog); |
43a70507 N |
2154 | if (rv) |
2155 | return rv; | |
2156 | if (backlog > COUNTER_MAX) | |
2157 | return -EINVAL; | |
2158 | mddev->bitmap_info.max_write_behind = backlog; | |
2159 | return len; | |
2160 | } | |
2161 | ||
2162 | static struct md_sysfs_entry bitmap_backlog = | |
2163 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2164 | ||
2165 | static ssize_t | |
fd01b88c | 2166 | chunksize_show(struct mddev *mddev, char *page) |
43a70507 N |
2167 | { |
2168 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2169 | } | |
2170 | ||
2171 | static ssize_t | |
fd01b88c | 2172 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2173 | { |
2174 | /* Can only be changed when no bitmap is active */ | |
2175 | int rv; | |
2176 | unsigned long csize; | |
2177 | if (mddev->bitmap) | |
2178 | return -EBUSY; | |
b29bebd6 | 2179 | rv = kstrtoul(buf, 10, &csize); |
43a70507 N |
2180 | if (rv) |
2181 | return rv; | |
2182 | if (csize < 512 || | |
2183 | !is_power_of_2(csize)) | |
2184 | return -EINVAL; | |
2185 | mddev->bitmap_info.chunksize = csize; | |
2186 | return len; | |
2187 | } | |
2188 | ||
2189 | static struct md_sysfs_entry bitmap_chunksize = | |
2190 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2191 | ||
fd01b88c | 2192 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
ece5cff0 | 2193 | { |
c4ce867f GR |
2194 | if (mddev_is_clustered(mddev)) |
2195 | return sprintf(page, "clustered\n"); | |
ece5cff0 N |
2196 | return sprintf(page, "%s\n", (mddev->bitmap_info.external |
2197 | ? "external" : "internal")); | |
2198 | } | |
2199 | ||
fd01b88c | 2200 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2201 | { |
2202 | if (mddev->bitmap || | |
2203 | mddev->bitmap_info.file || | |
2204 | mddev->bitmap_info.offset) | |
2205 | return -EBUSY; | |
2206 | if (strncmp(buf, "external", 8) == 0) | |
2207 | mddev->bitmap_info.external = 1; | |
c4ce867f GR |
2208 | else if ((strncmp(buf, "internal", 8) == 0) || |
2209 | (strncmp(buf, "clustered", 9) == 0)) | |
ece5cff0 N |
2210 | mddev->bitmap_info.external = 0; |
2211 | else | |
2212 | return -EINVAL; | |
2213 | return len; | |
2214 | } | |
2215 | ||
2216 | static struct md_sysfs_entry bitmap_metadata = | |
2217 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2218 | ||
fd01b88c | 2219 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2220 | { |
2221 | int len; | |
b7b17c9b | 2222 | spin_lock(&mddev->lock); |
ece5cff0 N |
2223 | if (mddev->bitmap) |
2224 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2225 | "false" : "true")); | |
2226 | else | |
2227 | len = sprintf(page, "\n"); | |
b7b17c9b | 2228 | spin_unlock(&mddev->lock); |
ece5cff0 N |
2229 | return len; |
2230 | } | |
2231 | ||
fd01b88c | 2232 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2233 | { |
2234 | if (mddev->bitmap == NULL) | |
2235 | return -ENOENT; | |
2236 | if (strncmp(buf, "false", 5) == 0) | |
2237 | mddev->bitmap->need_sync = 1; | |
2238 | else if (strncmp(buf, "true", 4) == 0) { | |
2239 | if (mddev->degraded) | |
2240 | return -EBUSY; | |
2241 | mddev->bitmap->need_sync = 0; | |
2242 | } else | |
2243 | return -EINVAL; | |
2244 | return len; | |
2245 | } | |
2246 | ||
2247 | static struct md_sysfs_entry bitmap_can_clear = | |
2248 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2249 | ||
696fcd53 | 2250 | static ssize_t |
fd01b88c | 2251 | behind_writes_used_show(struct mddev *mddev, char *page) |
696fcd53 | 2252 | { |
b7b17c9b N |
2253 | ssize_t ret; |
2254 | spin_lock(&mddev->lock); | |
696fcd53 | 2255 | if (mddev->bitmap == NULL) |
b7b17c9b N |
2256 | ret = sprintf(page, "0\n"); |
2257 | else | |
2258 | ret = sprintf(page, "%lu\n", | |
2259 | mddev->bitmap->behind_writes_used); | |
2260 | spin_unlock(&mddev->lock); | |
2261 | return ret; | |
696fcd53 PC |
2262 | } |
2263 | ||
2264 | static ssize_t | |
fd01b88c | 2265 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
696fcd53 PC |
2266 | { |
2267 | if (mddev->bitmap) | |
2268 | mddev->bitmap->behind_writes_used = 0; | |
2269 | return len; | |
2270 | } | |
2271 | ||
2272 | static struct md_sysfs_entry max_backlog_used = | |
2273 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2274 | behind_writes_used_show, behind_writes_used_reset); | |
2275 | ||
43a70507 N |
2276 | static struct attribute *md_bitmap_attrs[] = { |
2277 | &bitmap_location.attr, | |
6409bb05 | 2278 | &bitmap_space.attr, |
43a70507 N |
2279 | &bitmap_timeout.attr, |
2280 | &bitmap_backlog.attr, | |
2281 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2282 | &bitmap_metadata.attr, |
2283 | &bitmap_can_clear.attr, | |
696fcd53 | 2284 | &max_backlog_used.attr, |
43a70507 N |
2285 | NULL |
2286 | }; | |
2287 | struct attribute_group md_bitmap_group = { | |
2288 | .name = "bitmap", | |
2289 | .attrs = md_bitmap_attrs, | |
2290 | }; | |
2291 |