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); | |
1dff2b87 N |
436 | sb->sectors_reserved = cpu_to_le32(bitmap->mddev-> |
437 | bitmap_info.space); | |
b2f46e68 | 438 | kunmap_atomic(sb); |
1ec885cd | 439 | write_page(bitmap, bitmap->storage.sb_page, 1); |
32a7627c N |
440 | } |
441 | ||
442 | /* print out the bitmap file superblock */ | |
443 | void bitmap_print_sb(struct bitmap *bitmap) | |
444 | { | |
445 | bitmap_super_t *sb; | |
446 | ||
1ec885cd | 447 | if (!bitmap || !bitmap->storage.sb_page) |
32a7627c | 448 | return; |
1ec885cd | 449 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 450 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
451 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
452 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
453 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
454 | *(__u32 *)(sb->uuid+0), |
455 | *(__u32 *)(sb->uuid+4), | |
456 | *(__u32 *)(sb->uuid+8), | |
457 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 458 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 459 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 460 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 461 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
462 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
463 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
464 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
465 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
466 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 467 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
b2f46e68 | 468 | kunmap_atomic(sb); |
32a7627c N |
469 | } |
470 | ||
9c81075f JB |
471 | /* |
472 | * bitmap_new_disk_sb | |
473 | * @bitmap | |
474 | * | |
475 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb | |
476 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. | |
477 | * This function verifies 'bitmap_info' and populates the on-disk bitmap | |
478 | * structure, which is to be written to disk. | |
479 | * | |
480 | * Returns: 0 on success, -Exxx on error | |
481 | */ | |
482 | static int bitmap_new_disk_sb(struct bitmap *bitmap) | |
483 | { | |
484 | bitmap_super_t *sb; | |
485 | unsigned long chunksize, daemon_sleep, write_behind; | |
9c81075f | 486 | |
1ec885cd | 487 | bitmap->storage.sb_page = alloc_page(GFP_KERNEL); |
582e2e05 JM |
488 | if (bitmap->storage.sb_page == NULL) |
489 | return -ENOMEM; | |
1ec885cd | 490 | bitmap->storage.sb_page->index = 0; |
9c81075f | 491 | |
1ec885cd | 492 | sb = kmap_atomic(bitmap->storage.sb_page); |
9c81075f JB |
493 | |
494 | sb->magic = cpu_to_le32(BITMAP_MAGIC); | |
495 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); | |
496 | ||
497 | chunksize = bitmap->mddev->bitmap_info.chunksize; | |
498 | BUG_ON(!chunksize); | |
499 | if (!is_power_of_2(chunksize)) { | |
b2f46e68 | 500 | kunmap_atomic(sb); |
9c81075f JB |
501 | printk(KERN_ERR "bitmap chunksize not a power of 2\n"); |
502 | return -EINVAL; | |
503 | } | |
504 | sb->chunksize = cpu_to_le32(chunksize); | |
505 | ||
506 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; | |
507 | if (!daemon_sleep || | |
508 | (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { | |
509 | printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n"); | |
510 | daemon_sleep = 5 * HZ; | |
511 | } | |
512 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); | |
513 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
514 | ||
515 | /* | |
516 | * FIXME: write_behind for RAID1. If not specified, what | |
517 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. | |
518 | */ | |
519 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; | |
520 | if (write_behind > COUNTER_MAX) | |
521 | write_behind = COUNTER_MAX / 2; | |
522 | sb->write_behind = cpu_to_le32(write_behind); | |
523 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
524 | ||
525 | /* keep the array size field of the bitmap superblock up to date */ | |
526 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
527 | ||
528 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); | |
529 | ||
b405fe91 | 530 | set_bit(BITMAP_STALE, &bitmap->flags); |
84e92345 | 531 | sb->state = cpu_to_le32(bitmap->flags); |
9c81075f JB |
532 | bitmap->events_cleared = bitmap->mddev->events; |
533 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
534 | ||
b2f46e68 | 535 | kunmap_atomic(sb); |
9c81075f JB |
536 | |
537 | return 0; | |
538 | } | |
539 | ||
32a7627c N |
540 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
541 | static int bitmap_read_sb(struct bitmap *bitmap) | |
542 | { | |
543 | char *reason = NULL; | |
544 | bitmap_super_t *sb; | |
4b6d287f | 545 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c | 546 | unsigned long long events; |
1dff2b87 | 547 | unsigned long sectors_reserved = 0; |
32a7627c | 548 | int err = -EINVAL; |
27581e5a | 549 | struct page *sb_page; |
32a7627c | 550 | |
1ec885cd | 551 | if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { |
ef99bf48 N |
552 | chunksize = 128 * 1024 * 1024; |
553 | daemon_sleep = 5 * HZ; | |
554 | write_behind = 0; | |
b405fe91 | 555 | set_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
556 | err = 0; |
557 | goto out_no_sb; | |
558 | } | |
32a7627c | 559 | /* page 0 is the superblock, read it... */ |
27581e5a N |
560 | sb_page = alloc_page(GFP_KERNEL); |
561 | if (!sb_page) | |
562 | return -ENOMEM; | |
1ec885cd | 563 | bitmap->storage.sb_page = sb_page; |
27581e5a | 564 | |
1ec885cd N |
565 | if (bitmap->storage.file) { |
566 | loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); | |
f49d5e62 N |
567 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; |
568 | ||
1ec885cd | 569 | err = read_page(bitmap->storage.file, 0, |
27581e5a | 570 | bitmap, bytes, sb_page); |
f49d5e62 | 571 | } else { |
27581e5a N |
572 | err = read_sb_page(bitmap->mddev, |
573 | bitmap->mddev->bitmap_info.offset, | |
574 | sb_page, | |
575 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 576 | } |
27581e5a | 577 | if (err) |
32a7627c | 578 | return err; |
32a7627c | 579 | |
27581e5a | 580 | sb = kmap_atomic(sb_page); |
32a7627c | 581 | |
32a7627c | 582 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 583 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 584 | write_behind = le32_to_cpu(sb->write_behind); |
1dff2b87 | 585 | sectors_reserved = le32_to_cpu(sb->sectors_reserved); |
32a7627c N |
586 | |
587 | /* verify that the bitmap-specific fields are valid */ | |
588 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
589 | reason = "bad magic"; | |
bd926c63 N |
590 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
591 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 592 | reason = "unrecognized superblock version"; |
1187cf0a | 593 | else if (chunksize < 512) |
7dd5d34c | 594 | reason = "bitmap chunksize too small"; |
d744540c | 595 | else if (!is_power_of_2(chunksize)) |
32a7627c | 596 | reason = "bitmap chunksize not a power of 2"; |
1b04be96 | 597 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 598 | reason = "daemon sleep period out of range"; |
4b6d287f N |
599 | else if (write_behind > COUNTER_MAX) |
600 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
601 | if (reason) { |
602 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
603 | bmname(bitmap), reason); | |
604 | goto out; | |
605 | } | |
606 | ||
607 | /* keep the array size field of the bitmap superblock up to date */ | |
608 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
609 | ||
278c1ca2 N |
610 | if (bitmap->mddev->persistent) { |
611 | /* | |
612 | * We have a persistent array superblock, so compare the | |
613 | * bitmap's UUID and event counter to the mddev's | |
614 | */ | |
615 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
616 | printk(KERN_INFO | |
617 | "%s: bitmap superblock UUID mismatch\n", | |
618 | bmname(bitmap)); | |
619 | goto out; | |
620 | } | |
621 | events = le64_to_cpu(sb->events); | |
622 | if (events < bitmap->mddev->events) { | |
623 | printk(KERN_INFO | |
624 | "%s: bitmap file is out of date (%llu < %llu) " | |
625 | "-- forcing full recovery\n", | |
626 | bmname(bitmap), events, | |
627 | (unsigned long long) bitmap->mddev->events); | |
b405fe91 | 628 | set_bit(BITMAP_STALE, &bitmap->flags); |
278c1ca2 | 629 | } |
32a7627c | 630 | } |
278c1ca2 | 631 | |
32a7627c | 632 | /* assign fields using values from superblock */ |
4f2e639a | 633 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 | 634 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
b405fe91 | 635 | set_bit(BITMAP_HOSTENDIAN, &bitmap->flags); |
32a7627c N |
636 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
637 | err = 0; | |
638 | out: | |
b2f46e68 | 639 | kunmap_atomic(sb); |
ef99bf48 | 640 | out_no_sb: |
b405fe91 | 641 | if (test_bit(BITMAP_STALE, &bitmap->flags)) |
ef99bf48 N |
642 | bitmap->events_cleared = bitmap->mddev->events; |
643 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
644 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
645 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
1dff2b87 N |
646 | if (bitmap->mddev->bitmap_info.space == 0 || |
647 | bitmap->mddev->bitmap_info.space > sectors_reserved) | |
648 | bitmap->mddev->bitmap_info.space = sectors_reserved; | |
32a7627c N |
649 | if (err) |
650 | bitmap_print_sb(bitmap); | |
651 | return err; | |
652 | } | |
653 | ||
32a7627c N |
654 | /* |
655 | * general bitmap file operations | |
656 | */ | |
657 | ||
ece5cff0 N |
658 | /* |
659 | * on-disk bitmap: | |
660 | * | |
661 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
662 | * file a page at a time. There's a superblock at the start of the file. | |
663 | */ | |
32a7627c | 664 | /* calculate the index of the page that contains this bit */ |
1ec885cd N |
665 | static inline unsigned long file_page_index(struct bitmap_storage *store, |
666 | unsigned long chunk) | |
32a7627c | 667 | { |
1ec885cd | 668 | if (store->sb_page) |
ece5cff0 N |
669 | chunk += sizeof(bitmap_super_t) << 3; |
670 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
671 | } |
672 | ||
673 | /* calculate the (bit) offset of this bit within a page */ | |
1ec885cd N |
674 | static inline unsigned long file_page_offset(struct bitmap_storage *store, |
675 | unsigned long chunk) | |
32a7627c | 676 | { |
1ec885cd | 677 | if (store->sb_page) |
ece5cff0 N |
678 | chunk += sizeof(bitmap_super_t) << 3; |
679 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
680 | } |
681 | ||
682 | /* | |
683 | * return a pointer to the page in the filemap that contains the given bit | |
684 | * | |
32a7627c | 685 | */ |
1ec885cd | 686 | static inline struct page *filemap_get_page(struct bitmap_storage *store, |
3520fa4d | 687 | unsigned long chunk) |
32a7627c | 688 | { |
1ec885cd | 689 | if (file_page_index(store, chunk) >= store->file_pages) |
ac2f40be | 690 | return NULL; |
f2e06c58 | 691 | return store->filemap[file_page_index(store, chunk)]; |
32a7627c N |
692 | } |
693 | ||
d1244cb0 N |
694 | static int bitmap_storage_alloc(struct bitmap_storage *store, |
695 | unsigned long chunks, int with_super) | |
696 | { | |
697 | int pnum; | |
698 | unsigned long num_pages; | |
699 | unsigned long bytes; | |
700 | ||
701 | bytes = DIV_ROUND_UP(chunks, 8); | |
702 | if (with_super) | |
703 | bytes += sizeof(bitmap_super_t); | |
704 | ||
705 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); | |
706 | ||
707 | store->filemap = kmalloc(sizeof(struct page *) | |
708 | * num_pages, GFP_KERNEL); | |
709 | if (!store->filemap) | |
710 | return -ENOMEM; | |
711 | ||
712 | if (with_super && !store->sb_page) { | |
d60b479d | 713 | store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
714 | if (store->sb_page == NULL) |
715 | return -ENOMEM; | |
716 | store->sb_page->index = 0; | |
717 | } | |
718 | pnum = 0; | |
719 | if (store->sb_page) { | |
720 | store->filemap[0] = store->sb_page; | |
721 | pnum = 1; | |
722 | } | |
723 | for ( ; pnum < num_pages; pnum++) { | |
d60b479d | 724 | store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
725 | if (!store->filemap[pnum]) { |
726 | store->file_pages = pnum; | |
727 | return -ENOMEM; | |
728 | } | |
729 | store->filemap[pnum]->index = pnum; | |
730 | } | |
731 | store->file_pages = pnum; | |
732 | ||
733 | /* We need 4 bits per page, rounded up to a multiple | |
734 | * of sizeof(unsigned long) */ | |
735 | store->filemap_attr = kzalloc( | |
736 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), | |
737 | GFP_KERNEL); | |
738 | if (!store->filemap_attr) | |
739 | return -ENOMEM; | |
740 | ||
741 | store->bytes = bytes; | |
742 | ||
743 | return 0; | |
744 | } | |
745 | ||
fae7d326 | 746 | static void bitmap_file_unmap(struct bitmap_storage *store) |
32a7627c N |
747 | { |
748 | struct page **map, *sb_page; | |
32a7627c | 749 | int pages; |
fae7d326 | 750 | struct file *file; |
32a7627c | 751 | |
fae7d326 | 752 | file = store->file; |
1ec885cd | 753 | map = store->filemap; |
1ec885cd | 754 | pages = store->file_pages; |
1ec885cd | 755 | sb_page = store->sb_page; |
32a7627c N |
756 | |
757 | while (pages--) | |
ece5cff0 | 758 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 759 | free_buffers(map[pages]); |
32a7627c | 760 | kfree(map); |
fae7d326 | 761 | kfree(store->filemap_attr); |
32a7627c | 762 | |
d785a06a N |
763 | if (sb_page) |
764 | free_buffers(sb_page); | |
32a7627c | 765 | |
d785a06a | 766 | if (file) { |
496ad9aa | 767 | struct inode *inode = file_inode(file); |
fc0ecff6 | 768 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 769 | fput(file); |
d785a06a | 770 | } |
32a7627c N |
771 | } |
772 | ||
32a7627c N |
773 | /* |
774 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
775 | * then it is no longer reliable, so we stop using it and we mark the file | |
776 | * as failed in the superblock | |
777 | */ | |
778 | static void bitmap_file_kick(struct bitmap *bitmap) | |
779 | { | |
780 | char *path, *ptr = NULL; | |
781 | ||
b405fe91 | 782 | if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) { |
4ad13663 | 783 | bitmap_update_sb(bitmap); |
32a7627c | 784 | |
1ec885cd | 785 | if (bitmap->storage.file) { |
4ad13663 N |
786 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); |
787 | if (path) | |
1ec885cd N |
788 | ptr = d_path(&bitmap->storage.file->f_path, |
789 | path, PAGE_SIZE); | |
6bcfd601 | 790 | |
4ad13663 N |
791 | printk(KERN_ALERT |
792 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 793 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 794 | |
4ad13663 N |
795 | kfree(path); |
796 | } else | |
797 | printk(KERN_ALERT | |
798 | "%s: disabling internal bitmap due to errors\n", | |
799 | bmname(bitmap)); | |
a654b9d8 | 800 | } |
32a7627c N |
801 | } |
802 | ||
803 | enum bitmap_page_attr { | |
ac2f40be | 804 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
5a537df4 N |
805 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
806 | * i.e. counter is 1 or 2. */ | |
ac2f40be | 807 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
32a7627c N |
808 | }; |
809 | ||
d189122d N |
810 | static inline void set_page_attr(struct bitmap *bitmap, int pnum, |
811 | enum bitmap_page_attr attr) | |
32a7627c | 812 | { |
bdfd1140 | 813 | set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
814 | } |
815 | ||
d189122d N |
816 | static inline void clear_page_attr(struct bitmap *bitmap, int pnum, |
817 | enum bitmap_page_attr attr) | |
32a7627c | 818 | { |
bdfd1140 | 819 | clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
820 | } |
821 | ||
bdfd1140 N |
822 | static inline int test_page_attr(struct bitmap *bitmap, int pnum, |
823 | enum bitmap_page_attr attr) | |
32a7627c | 824 | { |
1ec885cd | 825 | return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
826 | } |
827 | ||
bdfd1140 N |
828 | static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum, |
829 | enum bitmap_page_attr attr) | |
830 | { | |
831 | return test_and_clear_bit((pnum<<2) + attr, | |
832 | bitmap->storage.filemap_attr); | |
833 | } | |
32a7627c N |
834 | /* |
835 | * bitmap_file_set_bit -- called before performing a write to the md device | |
836 | * to set (and eventually sync) a particular bit in the bitmap file | |
837 | * | |
838 | * we set the bit immediately, then we record the page number so that | |
839 | * when an unplug occurs, we can flush the dirty pages out to disk | |
840 | */ | |
841 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
842 | { | |
843 | unsigned long bit; | |
3520fa4d | 844 | struct page *page; |
32a7627c | 845 | void *kaddr; |
40cffcc0 | 846 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
32a7627c | 847 | |
1ec885cd | 848 | page = filemap_get_page(&bitmap->storage, chunk); |
3520fa4d JB |
849 | if (!page) |
850 | return; | |
1ec885cd | 851 | bit = file_page_offset(&bitmap->storage, chunk); |
32a7627c | 852 | |
3520fa4d | 853 | /* set the bit */ |
b2f46e68 | 854 | kaddr = kmap_atomic(page); |
b405fe91 | 855 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
3520fa4d JB |
856 | set_bit(bit, kaddr); |
857 | else | |
3f810b6c | 858 | set_bit_le(bit, kaddr); |
b2f46e68 | 859 | kunmap_atomic(kaddr); |
36a4e1fe | 860 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
32a7627c | 861 | /* record page number so it gets flushed to disk when unplug occurs */ |
d189122d | 862 | set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY); |
32a7627c N |
863 | } |
864 | ||
ef99bf48 N |
865 | static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) |
866 | { | |
867 | unsigned long bit; | |
868 | struct page *page; | |
869 | void *paddr; | |
40cffcc0 | 870 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
ef99bf48 | 871 | |
1ec885cd | 872 | page = filemap_get_page(&bitmap->storage, chunk); |
ef99bf48 N |
873 | if (!page) |
874 | return; | |
1ec885cd | 875 | bit = file_page_offset(&bitmap->storage, chunk); |
ef99bf48 | 876 | paddr = kmap_atomic(page); |
b405fe91 | 877 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ef99bf48 N |
878 | clear_bit(bit, paddr); |
879 | else | |
3f810b6c | 880 | clear_bit_le(bit, paddr); |
ef99bf48 | 881 | kunmap_atomic(paddr); |
d189122d N |
882 | if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) { |
883 | set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING); | |
ef99bf48 N |
884 | bitmap->allclean = 0; |
885 | } | |
886 | } | |
887 | ||
32a7627c N |
888 | /* this gets called when the md device is ready to unplug its underlying |
889 | * (slave) device queues -- before we let any writes go down, we need to | |
890 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 891 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 892 | { |
74667123 | 893 | unsigned long i; |
ec7a3197 | 894 | int dirty, need_write; |
32a7627c | 895 | |
62f82faa N |
896 | if (!bitmap || !bitmap->storage.filemap || |
897 | test_bit(BITMAP_STALE, &bitmap->flags)) | |
4ad13663 | 898 | return; |
32a7627c N |
899 | |
900 | /* look at each page to see if there are any set bits that need to be | |
901 | * flushed out to disk */ | |
1ec885cd | 902 | for (i = 0; i < bitmap->storage.file_pages; i++) { |
bdfd1140 | 903 | if (!bitmap->storage.filemap) |
4ad13663 | 904 | return; |
bdfd1140 N |
905 | dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
906 | need_write = test_and_clear_page_attr(bitmap, i, | |
907 | BITMAP_PAGE_NEEDWRITE); | |
908 | if (dirty || need_write) { | |
d189122d | 909 | clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); |
bdfd1140 N |
910 | write_page(bitmap, bitmap->storage.filemap[i], 0); |
911 | } | |
32a7627c | 912 | } |
4b5060dd N |
913 | if (bitmap->storage.file) |
914 | wait_event(bitmap->write_wait, | |
915 | atomic_read(&bitmap->pending_writes)==0); | |
916 | else | |
917 | md_super_wait(bitmap->mddev); | |
918 | ||
b405fe91 | 919 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
d785a06a | 920 | bitmap_file_kick(bitmap); |
32a7627c | 921 | } |
ac2f40be | 922 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 923 | |
6a07997f | 924 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
925 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
926 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
927 | * memory mapping of the bitmap file | |
928 | * Special cases: | |
929 | * if there's no bitmap file, or if the bitmap file had been | |
930 | * previously kicked from the array, we mark all the bits as | |
931 | * 1's in order to cause a full resync. | |
6a07997f N |
932 | * |
933 | * We ignore all bits for sectors that end earlier than 'start'. | |
934 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 935 | */ |
6a07997f | 936 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
937 | { |
938 | unsigned long i, chunks, index, oldindex, bit; | |
27581e5a | 939 | struct page *page = NULL; |
d1244cb0 | 940 | unsigned long bit_cnt = 0; |
32a7627c | 941 | struct file *file; |
d1244cb0 | 942 | unsigned long offset; |
32a7627c N |
943 | int outofdate; |
944 | int ret = -ENOSPC; | |
ea03aff9 | 945 | void *paddr; |
1ec885cd | 946 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c | 947 | |
40cffcc0 | 948 | chunks = bitmap->counts.chunks; |
1ec885cd | 949 | file = store->file; |
32a7627c | 950 | |
ef99bf48 N |
951 | if (!file && !bitmap->mddev->bitmap_info.offset) { |
952 | /* No permanent bitmap - fill with '1s'. */ | |
1ec885cd N |
953 | store->filemap = NULL; |
954 | store->file_pages = 0; | |
ef99bf48 N |
955 | for (i = 0; i < chunks ; i++) { |
956 | /* if the disk bit is set, set the memory bit */ | |
40cffcc0 | 957 | int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift) |
ef99bf48 N |
958 | >= start); |
959 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 960 | (sector_t)i << bitmap->counts.chunkshift, |
ef99bf48 N |
961 | needed); |
962 | } | |
963 | return 0; | |
964 | } | |
32a7627c | 965 | |
b405fe91 | 966 | outofdate = test_bit(BITMAP_STALE, &bitmap->flags); |
32a7627c N |
967 | if (outofdate) |
968 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
969 | "recovery\n", bmname(bitmap)); | |
970 | ||
d1244cb0 | 971 | if (file && i_size_read(file->f_mapping->host) < store->bytes) { |
32a7627c | 972 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
d1244cb0 N |
973 | bmname(bitmap), |
974 | (unsigned long) i_size_read(file->f_mapping->host), | |
975 | store->bytes); | |
4ad13663 | 976 | goto err; |
32a7627c | 977 | } |
bc7f77de | 978 | |
d1244cb0 | 979 | oldindex = ~0L; |
27581e5a | 980 | offset = 0; |
d1244cb0 | 981 | if (!bitmap->mddev->bitmap_info.external) |
27581e5a | 982 | offset = sizeof(bitmap_super_t); |
32a7627c N |
983 | |
984 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 985 | int b; |
1ec885cd N |
986 | index = file_page_index(&bitmap->storage, i); |
987 | bit = file_page_offset(&bitmap->storage, i); | |
32a7627c | 988 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 989 | int count; |
32a7627c | 990 | /* unmap the old page, we're done with it */ |
d1244cb0 N |
991 | if (index == store->file_pages-1) |
992 | count = store->bytes - index * PAGE_SIZE; | |
d785a06a N |
993 | else |
994 | count = PAGE_SIZE; | |
1ec885cd | 995 | page = store->filemap[index]; |
27581e5a N |
996 | if (file) |
997 | ret = read_page(file, index, bitmap, | |
998 | count, page); | |
999 | else | |
1000 | ret = read_sb_page( | |
1001 | bitmap->mddev, | |
1002 | bitmap->mddev->bitmap_info.offset, | |
1003 | page, | |
1004 | index, count); | |
1005 | ||
1006 | if (ret) | |
4ad13663 | 1007 | goto err; |
a654b9d8 | 1008 | |
32a7627c | 1009 | oldindex = index; |
32a7627c N |
1010 | |
1011 | if (outofdate) { | |
1012 | /* | |
1013 | * if bitmap is out of date, dirty the | |
ac2f40be | 1014 | * whole page and write it out |
32a7627c | 1015 | */ |
b2f46e68 | 1016 | paddr = kmap_atomic(page); |
ea03aff9 | 1017 | memset(paddr + offset, 0xff, |
6a07997f | 1018 | PAGE_SIZE - offset); |
b2f46e68 | 1019 | kunmap_atomic(paddr); |
4ad13663 N |
1020 | write_page(bitmap, page, 1); |
1021 | ||
1022 | ret = -EIO; | |
b405fe91 N |
1023 | if (test_bit(BITMAP_WRITE_ERROR, |
1024 | &bitmap->flags)) | |
4ad13663 | 1025 | goto err; |
32a7627c | 1026 | } |
32a7627c | 1027 | } |
b2f46e68 | 1028 | paddr = kmap_atomic(page); |
b405fe91 | 1029 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ea03aff9 | 1030 | b = test_bit(bit, paddr); |
bd926c63 | 1031 | else |
6b33aff3 | 1032 | b = test_bit_le(bit, paddr); |
b2f46e68 | 1033 | kunmap_atomic(paddr); |
bd926c63 | 1034 | if (b) { |
32a7627c | 1035 | /* if the disk bit is set, set the memory bit */ |
40cffcc0 | 1036 | int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift |
db305e50 N |
1037 | >= start); |
1038 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 1039 | (sector_t)i << bitmap->counts.chunkshift, |
db305e50 | 1040 | needed); |
32a7627c N |
1041 | bit_cnt++; |
1042 | } | |
27581e5a | 1043 | offset = 0; |
32a7627c N |
1044 | } |
1045 | ||
32a7627c | 1046 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
d1244cb0 | 1047 | "read %lu pages, set %lu of %lu bits\n", |
1ec885cd | 1048 | bmname(bitmap), store->file_pages, |
d1244cb0 | 1049 | bit_cnt, chunks); |
4ad13663 N |
1050 | |
1051 | return 0; | |
32a7627c | 1052 | |
4ad13663 N |
1053 | err: |
1054 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1055 | bmname(bitmap), ret); | |
32a7627c N |
1056 | return ret; |
1057 | } | |
1058 | ||
a654b9d8 N |
1059 | void bitmap_write_all(struct bitmap *bitmap) |
1060 | { | |
1061 | /* We don't actually write all bitmap blocks here, | |
1062 | * just flag them as needing to be written | |
1063 | */ | |
ec7a3197 | 1064 | int i; |
a654b9d8 | 1065 | |
1ec885cd | 1066 | if (!bitmap || !bitmap->storage.filemap) |
ef99bf48 | 1067 | return; |
1ec885cd | 1068 | if (bitmap->storage.file) |
ef99bf48 N |
1069 | /* Only one copy, so nothing needed */ |
1070 | return; | |
1071 | ||
1ec885cd | 1072 | for (i = 0; i < bitmap->storage.file_pages; i++) |
d189122d | 1073 | set_page_attr(bitmap, i, |
ec7a3197 | 1074 | BITMAP_PAGE_NEEDWRITE); |
2585f3ef | 1075 | bitmap->allclean = 0; |
a654b9d8 N |
1076 | } |
1077 | ||
40cffcc0 N |
1078 | static void bitmap_count_page(struct bitmap_counts *bitmap, |
1079 | sector_t offset, int inc) | |
32a7627c | 1080 | { |
61a0d80c | 1081 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1082 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1083 | bitmap->bp[page].count += inc; | |
32a7627c N |
1084 | bitmap_checkfree(bitmap, page); |
1085 | } | |
bf07bb7d | 1086 | |
40cffcc0 | 1087 | static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset) |
bf07bb7d N |
1088 | { |
1089 | sector_t chunk = offset >> bitmap->chunkshift; | |
1090 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1091 | struct bitmap_page *bp = &bitmap->bp[page]; | |
1092 | ||
1093 | if (!bp->pending) | |
1094 | bp->pending = 1; | |
1095 | } | |
1096 | ||
40cffcc0 | 1097 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1098 | sector_t offset, sector_t *blocks, |
32a7627c N |
1099 | int create); |
1100 | ||
1101 | /* | |
1102 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1103 | * out to disk | |
1104 | */ | |
1105 | ||
fd01b88c | 1106 | void bitmap_daemon_work(struct mddev *mddev) |
32a7627c | 1107 | { |
aa5cbd10 | 1108 | struct bitmap *bitmap; |
aa3163f8 | 1109 | unsigned long j; |
bf07bb7d | 1110 | unsigned long nextpage; |
57dab0bd | 1111 | sector_t blocks; |
40cffcc0 | 1112 | struct bitmap_counts *counts; |
32a7627c | 1113 | |
aa5cbd10 N |
1114 | /* Use a mutex to guard daemon_work against |
1115 | * bitmap_destroy. | |
1116 | */ | |
c3d9714e | 1117 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1118 | bitmap = mddev->bitmap; |
1119 | if (bitmap == NULL) { | |
c3d9714e | 1120 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1121 | return; |
aa5cbd10 | 1122 | } |
42a04b50 | 1123 | if (time_before(jiffies, bitmap->daemon_lastrun |
2e61ebbc | 1124 | + mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1125 | goto done; |
1126 | ||
32a7627c | 1127 | bitmap->daemon_lastrun = jiffies; |
8311c29d | 1128 | if (bitmap->allclean) { |
2e61ebbc | 1129 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
aa5cbd10 | 1130 | goto done; |
8311c29d N |
1131 | } |
1132 | bitmap->allclean = 1; | |
32a7627c | 1133 | |
bf07bb7d N |
1134 | /* Any file-page which is PENDING now needs to be written. |
1135 | * So set NEEDWRITE now, then after we make any last-minute changes | |
1136 | * we will write it. | |
1137 | */ | |
1ec885cd | 1138 | for (j = 0; j < bitmap->storage.file_pages; j++) |
bdfd1140 N |
1139 | if (test_and_clear_page_attr(bitmap, j, |
1140 | BITMAP_PAGE_PENDING)) | |
d189122d | 1141 | set_page_attr(bitmap, j, |
bf07bb7d | 1142 | BITMAP_PAGE_NEEDWRITE); |
bf07bb7d N |
1143 | |
1144 | if (bitmap->need_sync && | |
1145 | mddev->bitmap_info.external == 0) { | |
1146 | /* Arrange for superblock update as well as | |
1147 | * other changes */ | |
1148 | bitmap_super_t *sb; | |
1149 | bitmap->need_sync = 0; | |
1ec885cd N |
1150 | if (bitmap->storage.filemap) { |
1151 | sb = kmap_atomic(bitmap->storage.sb_page); | |
ef99bf48 N |
1152 | sb->events_cleared = |
1153 | cpu_to_le64(bitmap->events_cleared); | |
1154 | kunmap_atomic(sb); | |
d189122d | 1155 | set_page_attr(bitmap, 0, |
ef99bf48 N |
1156 | BITMAP_PAGE_NEEDWRITE); |
1157 | } | |
bf07bb7d N |
1158 | } |
1159 | /* Now look at the bitmap counters and if any are '2' or '1', | |
1160 | * decrement and handle accordingly. | |
1161 | */ | |
40cffcc0 N |
1162 | counts = &bitmap->counts; |
1163 | spin_lock_irq(&counts->lock); | |
bf07bb7d | 1164 | nextpage = 0; |
40cffcc0 | 1165 | for (j = 0; j < counts->chunks; j++) { |
32a7627c | 1166 | bitmap_counter_t *bmc; |
40cffcc0 | 1167 | sector_t block = (sector_t)j << counts->chunkshift; |
3520fa4d | 1168 | |
bf07bb7d N |
1169 | if (j == nextpage) { |
1170 | nextpage += PAGE_COUNTER_RATIO; | |
40cffcc0 | 1171 | if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) { |
bf07bb7d | 1172 | j |= PAGE_COUNTER_MASK; |
aa3163f8 N |
1173 | continue; |
1174 | } | |
40cffcc0 | 1175 | counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; |
32a7627c | 1176 | } |
40cffcc0 | 1177 | bmc = bitmap_get_counter(counts, |
ef99bf48 | 1178 | block, |
db305e50 | 1179 | &blocks, 0); |
bf07bb7d N |
1180 | |
1181 | if (!bmc) { | |
5a537df4 | 1182 | j |= PAGE_COUNTER_MASK; |
bf07bb7d N |
1183 | continue; |
1184 | } | |
1185 | if (*bmc == 1 && !bitmap->need_sync) { | |
1186 | /* We can clear the bit */ | |
bf07bb7d | 1187 | *bmc = 0; |
40cffcc0 | 1188 | bitmap_count_page(counts, block, -1); |
ef99bf48 | 1189 | bitmap_file_clear_bit(bitmap, block); |
bf07bb7d N |
1190 | } else if (*bmc && *bmc <= 2) { |
1191 | *bmc = 1; | |
40cffcc0 | 1192 | bitmap_set_pending(counts, block); |
bf07bb7d | 1193 | bitmap->allclean = 0; |
5a537df4 | 1194 | } |
32a7627c | 1195 | } |
40cffcc0 | 1196 | spin_unlock_irq(&counts->lock); |
32a7627c | 1197 | |
bf07bb7d N |
1198 | /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. |
1199 | * DIRTY pages need to be written by bitmap_unplug so it can wait | |
1200 | * for them. | |
1201 | * If we find any DIRTY page we stop there and let bitmap_unplug | |
1202 | * handle all the rest. This is important in the case where | |
1203 | * the first blocking holds the superblock and it has been updated. | |
1204 | * We mustn't write any other blocks before the superblock. | |
1205 | */ | |
62f82faa N |
1206 | for (j = 0; |
1207 | j < bitmap->storage.file_pages | |
1208 | && !test_bit(BITMAP_STALE, &bitmap->flags); | |
1209 | j++) { | |
bf07bb7d | 1210 | |
d189122d | 1211 | if (test_page_attr(bitmap, j, |
bf07bb7d N |
1212 | BITMAP_PAGE_DIRTY)) |
1213 | /* bitmap_unplug will handle the rest */ | |
1214 | break; | |
bdfd1140 N |
1215 | if (test_and_clear_page_attr(bitmap, j, |
1216 | BITMAP_PAGE_NEEDWRITE)) { | |
1ec885cd | 1217 | write_page(bitmap, bitmap->storage.filemap[j], 0); |
32a7627c | 1218 | } |
32a7627c N |
1219 | } |
1220 | ||
7be3dfec | 1221 | done: |
8311c29d | 1222 | if (bitmap->allclean == 0) |
2e61ebbc N |
1223 | mddev->thread->timeout = |
1224 | mddev->bitmap_info.daemon_sleep; | |
c3d9714e | 1225 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1226 | } |
1227 | ||
40cffcc0 | 1228 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1229 | sector_t offset, sector_t *blocks, |
32a7627c | 1230 | int create) |
ee305ace N |
1231 | __releases(bitmap->lock) |
1232 | __acquires(bitmap->lock) | |
32a7627c N |
1233 | { |
1234 | /* If 'create', we might release the lock and reclaim it. | |
1235 | * The lock must have been taken with interrupts enabled. | |
1236 | * If !create, we don't release the lock. | |
1237 | */ | |
61a0d80c | 1238 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1239 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1240 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1241 | sector_t csize; | |
ef425673 | 1242 | int err; |
32a7627c | 1243 | |
ef425673 N |
1244 | err = bitmap_checkpage(bitmap, page, create); |
1245 | ||
1246 | if (bitmap->bp[page].hijacked || | |
1247 | bitmap->bp[page].map == NULL) | |
61a0d80c | 1248 | csize = ((sector_t)1) << (bitmap->chunkshift + |
ef425673 N |
1249 | PAGE_COUNTER_SHIFT - 1); |
1250 | else | |
61a0d80c | 1251 | csize = ((sector_t)1) << bitmap->chunkshift; |
ef425673 N |
1252 | *blocks = csize - (offset & (csize - 1)); |
1253 | ||
1254 | if (err < 0) | |
32a7627c | 1255 | return NULL; |
ef425673 | 1256 | |
32a7627c N |
1257 | /* now locked ... */ |
1258 | ||
1259 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1260 | /* should we use the first or second counter field | |
1261 | * of the hijacked pointer? */ | |
1262 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1263 | return &((bitmap_counter_t *) |
1264 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1265 | } else /* page is allocated */ |
32a7627c N |
1266 | return (bitmap_counter_t *) |
1267 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1268 | } |
1269 | ||
4b6d287f | 1270 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1271 | { |
ac2f40be N |
1272 | if (!bitmap) |
1273 | return 0; | |
4b6d287f N |
1274 | |
1275 | if (behind) { | |
696fcd53 | 1276 | int bw; |
4b6d287f | 1277 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1278 | bw = atomic_read(&bitmap->behind_writes); |
1279 | if (bw > bitmap->behind_writes_used) | |
1280 | bitmap->behind_writes_used = bw; | |
1281 | ||
36a4e1fe N |
1282 | pr_debug("inc write-behind count %d/%lu\n", |
1283 | bw, bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1284 | } |
1285 | ||
32a7627c | 1286 | while (sectors) { |
57dab0bd | 1287 | sector_t blocks; |
32a7627c N |
1288 | bitmap_counter_t *bmc; |
1289 | ||
40cffcc0 N |
1290 | spin_lock_irq(&bitmap->counts.lock); |
1291 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1); | |
32a7627c | 1292 | if (!bmc) { |
40cffcc0 | 1293 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1294 | return 0; |
1295 | } | |
1296 | ||
27d5ea04 | 1297 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
da6e1a32 NB |
1298 | DEFINE_WAIT(__wait); |
1299 | /* note that it is safe to do the prepare_to_wait | |
1300 | * after the test as long as we do it before dropping | |
1301 | * the spinlock. | |
1302 | */ | |
1303 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1304 | TASK_UNINTERRUPTIBLE); | |
40cffcc0 | 1305 | spin_unlock_irq(&bitmap->counts.lock); |
f54a9d0e | 1306 | schedule(); |
da6e1a32 NB |
1307 | finish_wait(&bitmap->overflow_wait, &__wait); |
1308 | continue; | |
1309 | } | |
1310 | ||
ac2f40be | 1311 | switch (*bmc) { |
32a7627c N |
1312 | case 0: |
1313 | bitmap_file_set_bit(bitmap, offset); | |
40cffcc0 | 1314 | bitmap_count_page(&bitmap->counts, offset, 1); |
32a7627c N |
1315 | /* fall through */ |
1316 | case 1: | |
1317 | *bmc = 2; | |
1318 | } | |
da6e1a32 | 1319 | |
32a7627c N |
1320 | (*bmc)++; |
1321 | ||
40cffcc0 | 1322 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1323 | |
1324 | offset += blocks; | |
1325 | if (sectors > blocks) | |
1326 | sectors -= blocks; | |
ac2f40be N |
1327 | else |
1328 | sectors = 0; | |
32a7627c N |
1329 | } |
1330 | return 0; | |
1331 | } | |
ac2f40be | 1332 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1333 | |
1334 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1335 | int success, int behind) |
32a7627c | 1336 | { |
ac2f40be N |
1337 | if (!bitmap) |
1338 | return; | |
4b6d287f | 1339 | if (behind) { |
e555190d N |
1340 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1341 | wake_up(&bitmap->behind_wait); | |
36a4e1fe N |
1342 | pr_debug("dec write-behind count %d/%lu\n", |
1343 | atomic_read(&bitmap->behind_writes), | |
1344 | bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1345 | } |
1346 | ||
32a7627c | 1347 | while (sectors) { |
57dab0bd | 1348 | sector_t blocks; |
32a7627c N |
1349 | unsigned long flags; |
1350 | bitmap_counter_t *bmc; | |
1351 | ||
40cffcc0 N |
1352 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1353 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0); | |
32a7627c | 1354 | if (!bmc) { |
40cffcc0 | 1355 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1356 | return; |
1357 | } | |
1358 | ||
961902c0 | 1359 | if (success && !bitmap->mddev->degraded && |
a0da84f3 NB |
1360 | bitmap->events_cleared < bitmap->mddev->events) { |
1361 | bitmap->events_cleared = bitmap->mddev->events; | |
1362 | bitmap->need_sync = 1; | |
5ff5afff | 1363 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1364 | } |
1365 | ||
27d5ea04 | 1366 | if (!success && !NEEDED(*bmc)) |
32a7627c N |
1367 | *bmc |= NEEDED_MASK; |
1368 | ||
27d5ea04 | 1369 | if (COUNTER(*bmc) == COUNTER_MAX) |
da6e1a32 NB |
1370 | wake_up(&bitmap->overflow_wait); |
1371 | ||
32a7627c | 1372 | (*bmc)--; |
2585f3ef | 1373 | if (*bmc <= 2) { |
40cffcc0 | 1374 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1375 | bitmap->allclean = 0; |
1376 | } | |
40cffcc0 | 1377 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1378 | offset += blocks; |
1379 | if (sectors > blocks) | |
1380 | sectors -= blocks; | |
ac2f40be N |
1381 | else |
1382 | sectors = 0; | |
32a7627c N |
1383 | } |
1384 | } | |
ac2f40be | 1385 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1386 | |
57dab0bd | 1387 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1388 | int degraded) |
32a7627c N |
1389 | { |
1390 | bitmap_counter_t *bmc; | |
1391 | int rv; | |
1392 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1393 | *blocks = 1024; | |
1394 | return 1; /* always resync if no bitmap */ | |
1395 | } | |
40cffcc0 N |
1396 | spin_lock_irq(&bitmap->counts.lock); |
1397 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1398 | rv = 0; |
1399 | if (bmc) { | |
1400 | /* locked */ | |
1401 | if (RESYNC(*bmc)) | |
1402 | rv = 1; | |
1403 | else if (NEEDED(*bmc)) { | |
1404 | rv = 1; | |
6a806c51 N |
1405 | if (!degraded) { /* don't set/clear bits if degraded */ |
1406 | *bmc |= RESYNC_MASK; | |
1407 | *bmc &= ~NEEDED_MASK; | |
1408 | } | |
32a7627c N |
1409 | } |
1410 | } | |
40cffcc0 | 1411 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1412 | return rv; |
1413 | } | |
1414 | ||
57dab0bd | 1415 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a N |
1416 | int degraded) |
1417 | { | |
1418 | /* bitmap_start_sync must always report on multiples of whole | |
1419 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1420 | * get confused. | |
1421 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1422 | * At least PAGE_SIZE>>9 blocks are covered. | |
1423 | * Return the 'or' of the result. | |
1424 | */ | |
1425 | int rv = 0; | |
57dab0bd | 1426 | sector_t blocks1; |
1187cf0a N |
1427 | |
1428 | *blocks = 0; | |
1429 | while (*blocks < (PAGE_SIZE>>9)) { | |
1430 | rv |= __bitmap_start_sync(bitmap, offset, | |
1431 | &blocks1, degraded); | |
1432 | offset += blocks1; | |
1433 | *blocks += blocks1; | |
1434 | } | |
1435 | return rv; | |
1436 | } | |
ac2f40be | 1437 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1438 | |
57dab0bd | 1439 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1440 | { |
1441 | bitmap_counter_t *bmc; | |
1442 | unsigned long flags; | |
ac2f40be N |
1443 | |
1444 | if (bitmap == NULL) { | |
32a7627c N |
1445 | *blocks = 1024; |
1446 | return; | |
1447 | } | |
40cffcc0 N |
1448 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1449 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1450 | if (bmc == NULL) |
1451 | goto unlock; | |
1452 | /* locked */ | |
32a7627c N |
1453 | if (RESYNC(*bmc)) { |
1454 | *bmc &= ~RESYNC_MASK; | |
1455 | ||
1456 | if (!NEEDED(*bmc) && aborted) | |
1457 | *bmc |= NEEDED_MASK; | |
1458 | else { | |
2585f3ef | 1459 | if (*bmc <= 2) { |
40cffcc0 | 1460 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1461 | bitmap->allclean = 0; |
1462 | } | |
32a7627c N |
1463 | } |
1464 | } | |
1465 | unlock: | |
40cffcc0 | 1466 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c | 1467 | } |
ac2f40be | 1468 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1469 | |
1470 | void bitmap_close_sync(struct bitmap *bitmap) | |
1471 | { | |
1472 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1473 | * properly have been aborted. It remains to us to clear the | |
1474 | * RESYNC bit wherever it is still on | |
1475 | */ | |
1476 | sector_t sector = 0; | |
57dab0bd | 1477 | sector_t blocks; |
b47490c9 N |
1478 | if (!bitmap) |
1479 | return; | |
32a7627c N |
1480 | while (sector < bitmap->mddev->resync_max_sectors) { |
1481 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1482 | sector += blocks; |
1483 | } | |
1484 | } | |
ac2f40be | 1485 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 N |
1486 | |
1487 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1488 | { | |
1489 | sector_t s = 0; | |
57dab0bd | 1490 | sector_t blocks; |
b47490c9 N |
1491 | |
1492 | if (!bitmap) | |
1493 | return; | |
1494 | if (sector == 0) { | |
1495 | bitmap->last_end_sync = jiffies; | |
1496 | return; | |
1497 | } | |
1498 | if (time_before(jiffies, (bitmap->last_end_sync | |
1b04be96 | 1499 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1500 | return; |
1501 | wait_event(bitmap->mddev->recovery_wait, | |
1502 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1503 | ||
75d3da43 | 1504 | bitmap->mddev->curr_resync_completed = sector; |
070dc6dd | 1505 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
40cffcc0 | 1506 | sector &= ~((1ULL << bitmap->counts.chunkshift) - 1); |
b47490c9 N |
1507 | s = 0; |
1508 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1509 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1510 | s += blocks; | |
32a7627c | 1511 | } |
b47490c9 | 1512 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1513 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1514 | } |
ac2f40be | 1515 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1516 | |
6a07997f | 1517 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1518 | { |
1519 | /* For each chunk covered by any of these sectors, set the | |
ef99bf48 | 1520 | * counter to 2 and possibly set resync_needed. They should all |
32a7627c N |
1521 | * be 0 at this point |
1522 | */ | |
193f1c93 | 1523 | |
57dab0bd | 1524 | sector_t secs; |
193f1c93 | 1525 | bitmap_counter_t *bmc; |
40cffcc0 N |
1526 | spin_lock_irq(&bitmap->counts.lock); |
1527 | bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1); | |
193f1c93 | 1528 | if (!bmc) { |
40cffcc0 | 1529 | spin_unlock_irq(&bitmap->counts.lock); |
193f1c93 | 1530 | return; |
32a7627c | 1531 | } |
ac2f40be | 1532 | if (!*bmc) { |
915c420d | 1533 | *bmc = 2 | (needed ? NEEDED_MASK : 0); |
40cffcc0 N |
1534 | bitmap_count_page(&bitmap->counts, offset, 1); |
1535 | bitmap_set_pending(&bitmap->counts, offset); | |
2585f3ef | 1536 | bitmap->allclean = 0; |
193f1c93 | 1537 | } |
40cffcc0 | 1538 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1539 | } |
1540 | ||
9b1d1dac PC |
1541 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1542 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1543 | { | |
1544 | unsigned long chunk; | |
1545 | ||
1546 | for (chunk = s; chunk <= e; chunk++) { | |
40cffcc0 | 1547 | sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift; |
9b1d1dac PC |
1548 | bitmap_set_memory_bits(bitmap, sec, 1); |
1549 | bitmap_file_set_bit(bitmap, sec); | |
ffa23322 N |
1550 | if (sec < bitmap->mddev->recovery_cp) |
1551 | /* We are asserting that the array is dirty, | |
1552 | * so move the recovery_cp address back so | |
1553 | * that it is obvious that it is dirty | |
1554 | */ | |
1555 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1556 | } |
1557 | } | |
1558 | ||
6b8b3e8a N |
1559 | /* |
1560 | * flush out any pending updates | |
1561 | */ | |
fd01b88c | 1562 | void bitmap_flush(struct mddev *mddev) |
6b8b3e8a N |
1563 | { |
1564 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1565 | long sleep; |
6b8b3e8a N |
1566 | |
1567 | if (!bitmap) /* there was no bitmap */ | |
1568 | return; | |
1569 | ||
1570 | /* run the daemon_work three time to ensure everything is flushed | |
1571 | * that can be | |
1572 | */ | |
1b04be96 | 1573 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1574 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1575 | bitmap_daemon_work(mddev); |
42a04b50 | 1576 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1577 | bitmap_daemon_work(mddev); |
42a04b50 | 1578 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1579 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1580 | bitmap_update_sb(bitmap); |
1581 | } | |
1582 | ||
32a7627c N |
1583 | /* |
1584 | * free memory that was allocated | |
1585 | */ | |
3178b0db | 1586 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1587 | { |
1588 | unsigned long k, pages; | |
1589 | struct bitmap_page *bp; | |
32a7627c N |
1590 | |
1591 | if (!bitmap) /* there was no bitmap */ | |
1592 | return; | |
1593 | ||
fae7d326 N |
1594 | /* Shouldn't be needed - but just in case.... */ |
1595 | wait_event(bitmap->write_wait, | |
1596 | atomic_read(&bitmap->pending_writes) == 0); | |
1597 | ||
1598 | /* release the bitmap file */ | |
1599 | bitmap_file_unmap(&bitmap->storage); | |
32a7627c | 1600 | |
40cffcc0 N |
1601 | bp = bitmap->counts.bp; |
1602 | pages = bitmap->counts.pages; | |
32a7627c N |
1603 | |
1604 | /* free all allocated memory */ | |
1605 | ||
32a7627c N |
1606 | if (bp) /* deallocate the page memory */ |
1607 | for (k = 0; k < pages; k++) | |
1608 | if (bp[k].map && !bp[k].hijacked) | |
1609 | kfree(bp[k].map); | |
1610 | kfree(bp); | |
1611 | kfree(bitmap); | |
1612 | } | |
aa5cbd10 | 1613 | |
fd01b88c | 1614 | void bitmap_destroy(struct mddev *mddev) |
3178b0db N |
1615 | { |
1616 | struct bitmap *bitmap = mddev->bitmap; | |
1617 | ||
1618 | if (!bitmap) /* there was no bitmap */ | |
1619 | return; | |
1620 | ||
c3d9714e | 1621 | mutex_lock(&mddev->bitmap_info.mutex); |
978a7a47 | 1622 | spin_lock(&mddev->lock); |
3178b0db | 1623 | mddev->bitmap = NULL; /* disconnect from the md device */ |
978a7a47 | 1624 | spin_unlock(&mddev->lock); |
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; |
324a56e1 | 1645 | struct kernfs_node *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 | 1663 | if (mddev->kobj.sd) |
388975cc | 1664 | bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap"); |
ece5cff0 | 1665 | if (bm) { |
388975cc | 1666 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "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; | |
43a70507 N |
1998 | fput(f); |
1999 | } | |
2000 | } else { | |
2001 | /* No bitmap, OK to set a location */ | |
2002 | long long offset; | |
2003 | if (strncmp(buf, "none", 4) == 0) | |
2004 | /* nothing to be done */; | |
2005 | else if (strncmp(buf, "file:", 5) == 0) { | |
2006 | /* Not supported yet */ | |
2007 | return -EINVAL; | |
2008 | } else { | |
2009 | int rv; | |
2010 | if (buf[0] == '+') | |
b29bebd6 | 2011 | rv = kstrtoll(buf+1, 10, &offset); |
43a70507 | 2012 | else |
b29bebd6 | 2013 | rv = kstrtoll(buf, 10, &offset); |
43a70507 N |
2014 | if (rv) |
2015 | return rv; | |
2016 | if (offset == 0) | |
2017 | return -EINVAL; | |
ece5cff0 N |
2018 | if (mddev->bitmap_info.external == 0 && |
2019 | mddev->major_version == 0 && | |
43a70507 N |
2020 | offset != mddev->bitmap_info.default_offset) |
2021 | return -EINVAL; | |
2022 | mddev->bitmap_info.offset = offset; | |
2023 | if (mddev->pers) { | |
2024 | mddev->pers->quiesce(mddev, 1); | |
2025 | rv = bitmap_create(mddev); | |
4474ca42 N |
2026 | if (!rv) |
2027 | rv = bitmap_load(mddev); | |
43a70507 N |
2028 | if (rv) { |
2029 | bitmap_destroy(mddev); | |
2030 | mddev->bitmap_info.offset = 0; | |
2031 | } | |
2032 | mddev->pers->quiesce(mddev, 0); | |
2033 | if (rv) | |
2034 | return rv; | |
2035 | } | |
2036 | } | |
2037 | } | |
2038 | if (!mddev->external) { | |
2039 | /* Ensure new bitmap info is stored in | |
2040 | * metadata promptly. | |
2041 | */ | |
2042 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2043 | md_wakeup_thread(mddev->thread); | |
2044 | } | |
2045 | return len; | |
2046 | } | |
2047 | ||
2048 | static struct md_sysfs_entry bitmap_location = | |
2049 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
2050 | ||
6409bb05 N |
2051 | /* 'bitmap/space' is the space available at 'location' for the |
2052 | * bitmap. This allows the kernel to know when it is safe to | |
2053 | * resize the bitmap to match a resized array. | |
2054 | */ | |
2055 | static ssize_t | |
2056 | space_show(struct mddev *mddev, char *page) | |
2057 | { | |
2058 | return sprintf(page, "%lu\n", mddev->bitmap_info.space); | |
2059 | } | |
2060 | ||
2061 | static ssize_t | |
2062 | space_store(struct mddev *mddev, const char *buf, size_t len) | |
2063 | { | |
2064 | unsigned long sectors; | |
2065 | int rv; | |
2066 | ||
2067 | rv = kstrtoul(buf, 10, §ors); | |
2068 | if (rv) | |
2069 | return rv; | |
2070 | ||
2071 | if (sectors == 0) | |
2072 | return -EINVAL; | |
2073 | ||
2074 | if (mddev->bitmap && | |
9b1215c1 | 2075 | sectors < (mddev->bitmap->storage.bytes + 511) >> 9) |
6409bb05 N |
2076 | return -EFBIG; /* Bitmap is too big for this small space */ |
2077 | ||
2078 | /* could make sure it isn't too big, but that isn't really | |
2079 | * needed - user-space should be careful. | |
2080 | */ | |
2081 | mddev->bitmap_info.space = sectors; | |
2082 | return len; | |
2083 | } | |
2084 | ||
2085 | static struct md_sysfs_entry bitmap_space = | |
2086 | __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); | |
2087 | ||
43a70507 | 2088 | static ssize_t |
fd01b88c | 2089 | timeout_show(struct mddev *mddev, char *page) |
43a70507 N |
2090 | { |
2091 | ssize_t len; | |
2092 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
2093 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 2094 | |
43a70507 N |
2095 | len = sprintf(page, "%lu", secs); |
2096 | if (jifs) | |
2097 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
2098 | len += sprintf(page+len, "\n"); | |
2099 | return len; | |
2100 | } | |
2101 | ||
2102 | static ssize_t | |
fd01b88c | 2103 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2104 | { |
2105 | /* timeout can be set at any time */ | |
2106 | unsigned long timeout; | |
2107 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
2108 | if (rv) | |
2109 | return rv; | |
2110 | ||
2111 | /* just to make sure we don't overflow... */ | |
2112 | if (timeout >= LONG_MAX / HZ) | |
2113 | return -EINVAL; | |
2114 | ||
2115 | timeout = timeout * HZ / 10000; | |
2116 | ||
2117 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
2118 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
2119 | if (timeout < 1) | |
2120 | timeout = 1; | |
2121 | mddev->bitmap_info.daemon_sleep = timeout; | |
2122 | if (mddev->thread) { | |
2123 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
2124 | * the bitmap is all clean and we don't need to | |
2125 | * adjust the timeout right now | |
2126 | */ | |
2127 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2128 | mddev->thread->timeout = timeout; | |
2129 | md_wakeup_thread(mddev->thread); | |
2130 | } | |
2131 | } | |
2132 | return len; | |
2133 | } | |
2134 | ||
2135 | static struct md_sysfs_entry bitmap_timeout = | |
2136 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2137 | ||
2138 | static ssize_t | |
fd01b88c | 2139 | backlog_show(struct mddev *mddev, char *page) |
43a70507 N |
2140 | { |
2141 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2142 | } | |
2143 | ||
2144 | static ssize_t | |
fd01b88c | 2145 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2146 | { |
2147 | unsigned long backlog; | |
b29bebd6 | 2148 | int rv = kstrtoul(buf, 10, &backlog); |
43a70507 N |
2149 | if (rv) |
2150 | return rv; | |
2151 | if (backlog > COUNTER_MAX) | |
2152 | return -EINVAL; | |
2153 | mddev->bitmap_info.max_write_behind = backlog; | |
2154 | return len; | |
2155 | } | |
2156 | ||
2157 | static struct md_sysfs_entry bitmap_backlog = | |
2158 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2159 | ||
2160 | static ssize_t | |
fd01b88c | 2161 | chunksize_show(struct mddev *mddev, char *page) |
43a70507 N |
2162 | { |
2163 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2164 | } | |
2165 | ||
2166 | static ssize_t | |
fd01b88c | 2167 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2168 | { |
2169 | /* Can only be changed when no bitmap is active */ | |
2170 | int rv; | |
2171 | unsigned long csize; | |
2172 | if (mddev->bitmap) | |
2173 | return -EBUSY; | |
b29bebd6 | 2174 | rv = kstrtoul(buf, 10, &csize); |
43a70507 N |
2175 | if (rv) |
2176 | return rv; | |
2177 | if (csize < 512 || | |
2178 | !is_power_of_2(csize)) | |
2179 | return -EINVAL; | |
2180 | mddev->bitmap_info.chunksize = csize; | |
2181 | return len; | |
2182 | } | |
2183 | ||
2184 | static struct md_sysfs_entry bitmap_chunksize = | |
2185 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2186 | ||
fd01b88c | 2187 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2188 | { |
2189 | return sprintf(page, "%s\n", (mddev->bitmap_info.external | |
2190 | ? "external" : "internal")); | |
2191 | } | |
2192 | ||
fd01b88c | 2193 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2194 | { |
2195 | if (mddev->bitmap || | |
2196 | mddev->bitmap_info.file || | |
2197 | mddev->bitmap_info.offset) | |
2198 | return -EBUSY; | |
2199 | if (strncmp(buf, "external", 8) == 0) | |
2200 | mddev->bitmap_info.external = 1; | |
2201 | else if (strncmp(buf, "internal", 8) == 0) | |
2202 | mddev->bitmap_info.external = 0; | |
2203 | else | |
2204 | return -EINVAL; | |
2205 | return len; | |
2206 | } | |
2207 | ||
2208 | static struct md_sysfs_entry bitmap_metadata = | |
2209 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2210 | ||
fd01b88c | 2211 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2212 | { |
2213 | int len; | |
b7b17c9b | 2214 | spin_lock(&mddev->lock); |
ece5cff0 N |
2215 | if (mddev->bitmap) |
2216 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2217 | "false" : "true")); | |
2218 | else | |
2219 | len = sprintf(page, "\n"); | |
b7b17c9b | 2220 | spin_unlock(&mddev->lock); |
ece5cff0 N |
2221 | return len; |
2222 | } | |
2223 | ||
fd01b88c | 2224 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2225 | { |
2226 | if (mddev->bitmap == NULL) | |
2227 | return -ENOENT; | |
2228 | if (strncmp(buf, "false", 5) == 0) | |
2229 | mddev->bitmap->need_sync = 1; | |
2230 | else if (strncmp(buf, "true", 4) == 0) { | |
2231 | if (mddev->degraded) | |
2232 | return -EBUSY; | |
2233 | mddev->bitmap->need_sync = 0; | |
2234 | } else | |
2235 | return -EINVAL; | |
2236 | return len; | |
2237 | } | |
2238 | ||
2239 | static struct md_sysfs_entry bitmap_can_clear = | |
2240 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2241 | ||
696fcd53 | 2242 | static ssize_t |
fd01b88c | 2243 | behind_writes_used_show(struct mddev *mddev, char *page) |
696fcd53 | 2244 | { |
b7b17c9b N |
2245 | ssize_t ret; |
2246 | spin_lock(&mddev->lock); | |
696fcd53 | 2247 | if (mddev->bitmap == NULL) |
b7b17c9b N |
2248 | ret = sprintf(page, "0\n"); |
2249 | else | |
2250 | ret = sprintf(page, "%lu\n", | |
2251 | mddev->bitmap->behind_writes_used); | |
2252 | spin_unlock(&mddev->lock); | |
2253 | return ret; | |
696fcd53 PC |
2254 | } |
2255 | ||
2256 | static ssize_t | |
fd01b88c | 2257 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
696fcd53 PC |
2258 | { |
2259 | if (mddev->bitmap) | |
2260 | mddev->bitmap->behind_writes_used = 0; | |
2261 | return len; | |
2262 | } | |
2263 | ||
2264 | static struct md_sysfs_entry max_backlog_used = | |
2265 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2266 | behind_writes_used_show, behind_writes_used_reset); | |
2267 | ||
43a70507 N |
2268 | static struct attribute *md_bitmap_attrs[] = { |
2269 | &bitmap_location.attr, | |
6409bb05 | 2270 | &bitmap_space.attr, |
43a70507 N |
2271 | &bitmap_timeout.attr, |
2272 | &bitmap_backlog.attr, | |
2273 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2274 | &bitmap_metadata.attr, |
2275 | &bitmap_can_clear.attr, | |
696fcd53 | 2276 | &max_backlog_used.attr, |
43a70507 N |
2277 | NULL |
2278 | }; | |
2279 | struct attribute_group md_bitmap_group = { | |
2280 | .name = "bitmap", | |
2281 | .attrs = md_bitmap_attrs, | |
2282 | }; | |
2283 |