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, |
c9d65032 | 49 | unsigned long page, int create, int no_hijack) |
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"); |
c9d65032 GJ |
93 | /* We don't support hijack for cluster raid */ |
94 | if (no_hijack) | |
95 | return -ENOMEM; | |
32a7627c N |
96 | /* failed - set the hijacked flag so that we can use the |
97 | * pointer as a counter */ | |
32a7627c N |
98 | if (!bitmap->bp[page].map) |
99 | bitmap->bp[page].hijacked = 1; | |
ac2f40be N |
100 | } else if (bitmap->bp[page].map || |
101 | bitmap->bp[page].hijacked) { | |
32a7627c | 102 | /* somebody beat us to getting the page */ |
792a1d4b | 103 | kfree(mappage); |
ac2f40be | 104 | } else { |
32a7627c | 105 | |
ac2f40be | 106 | /* no page was in place and we have one, so install it */ |
32a7627c | 107 | |
ac2f40be N |
108 | bitmap->bp[page].map = mappage; |
109 | bitmap->missing_pages--; | |
110 | } | |
32a7627c N |
111 | return 0; |
112 | } | |
113 | ||
32a7627c N |
114 | /* if page is completely empty, put it back on the free list, or dealloc it */ |
115 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
116 | /* Note: lock should be held when calling this */ | |
40cffcc0 | 117 | static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page) |
32a7627c N |
118 | { |
119 | char *ptr; | |
120 | ||
121 | if (bitmap->bp[page].count) /* page is still busy */ | |
122 | return; | |
123 | ||
124 | /* page is no longer in use, it can be released */ | |
125 | ||
126 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
127 | bitmap->bp[page].hijacked = 0; | |
128 | bitmap->bp[page].map = NULL; | |
ac2f40be N |
129 | } else { |
130 | /* normal case, free the page */ | |
131 | ptr = bitmap->bp[page].map; | |
132 | bitmap->bp[page].map = NULL; | |
133 | bitmap->missing_pages++; | |
792a1d4b | 134 | kfree(ptr); |
32a7627c | 135 | } |
32a7627c N |
136 | } |
137 | ||
32a7627c N |
138 | /* |
139 | * bitmap file handling - read and write the bitmap file and its superblock | |
140 | */ | |
141 | ||
32a7627c N |
142 | /* |
143 | * basic page I/O operations | |
144 | */ | |
145 | ||
a654b9d8 | 146 | /* IO operations when bitmap is stored near all superblocks */ |
27581e5a N |
147 | static int read_sb_page(struct mddev *mddev, loff_t offset, |
148 | struct page *page, | |
149 | unsigned long index, int size) | |
a654b9d8 N |
150 | { |
151 | /* choose a good rdev and read the page from there */ | |
152 | ||
3cb03002 | 153 | struct md_rdev *rdev; |
a654b9d8 | 154 | sector_t target; |
a654b9d8 | 155 | |
dafb20fa | 156 | rdev_for_each(rdev, mddev) { |
b2d444d7 N |
157 | if (! test_bit(In_sync, &rdev->flags) |
158 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
159 | continue; |
160 | ||
ccebd4c4 | 161 | target = offset + index * (PAGE_SIZE/512); |
a654b9d8 | 162 | |
2b193363 | 163 | if (sync_page_io(rdev, target, |
e1defc4f | 164 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
ccebd4c4 | 165 | page, READ, true)) { |
ab904d63 | 166 | page->index = index; |
27581e5a | 167 | return 0; |
ab904d63 N |
168 | } |
169 | } | |
27581e5a | 170 | return -EIO; |
a654b9d8 N |
171 | } |
172 | ||
fd01b88c | 173 | static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) |
b2d2c4ce N |
174 | { |
175 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
176 | * linked list, and raising the refcount of devices we return to ensure | |
177 | * they don't disappear while in use. | |
178 | * As devices are only added or removed when raid_disk is < 0 and | |
179 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
180 | * still be in the same position on the list when we re-enter | |
fd177481 | 181 | * list_for_each_entry_continue_rcu. |
8532e343 N |
182 | * |
183 | * Note that if entered with 'rdev == NULL' to start at the | |
184 | * beginning, we temporarily assign 'rdev' to an address which | |
185 | * isn't really an rdev, but which can be used by | |
186 | * list_for_each_entry_continue_rcu() to find the first entry. | |
b2d2c4ce | 187 | */ |
b2d2c4ce N |
188 | rcu_read_lock(); |
189 | if (rdev == NULL) | |
190 | /* start at the beginning */ | |
8532e343 | 191 | rdev = list_entry(&mddev->disks, struct md_rdev, same_set); |
b2d2c4ce N |
192 | else { |
193 | /* release the previous rdev and start from there. */ | |
194 | rdev_dec_pending(rdev, mddev); | |
b2d2c4ce | 195 | } |
fd177481 | 196 | list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) { |
b2d2c4ce | 197 | if (rdev->raid_disk >= 0 && |
b2d2c4ce N |
198 | !test_bit(Faulty, &rdev->flags)) { |
199 | /* this is a usable devices */ | |
200 | atomic_inc(&rdev->nr_pending); | |
201 | rcu_read_unlock(); | |
202 | return rdev; | |
203 | } | |
204 | } | |
205 | rcu_read_unlock(); | |
206 | return NULL; | |
207 | } | |
208 | ||
ab6085c7 | 209 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 210 | { |
3cb03002 | 211 | struct md_rdev *rdev = NULL; |
a6ff7e08 | 212 | struct block_device *bdev; |
fd01b88c | 213 | struct mddev *mddev = bitmap->mddev; |
1ec885cd | 214 | struct bitmap_storage *store = &bitmap->storage; |
a654b9d8 | 215 | |
b2d2c4ce | 216 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ac2f40be N |
217 | int size = PAGE_SIZE; |
218 | loff_t offset = mddev->bitmap_info.offset; | |
a6ff7e08 JB |
219 | |
220 | bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; | |
221 | ||
9b1215c1 N |
222 | if (page->index == store->file_pages-1) { |
223 | int last_page_size = store->bytes & (PAGE_SIZE-1); | |
224 | if (last_page_size == 0) | |
225 | last_page_size = PAGE_SIZE; | |
226 | size = roundup(last_page_size, | |
a6ff7e08 | 227 | bdev_logical_block_size(bdev)); |
9b1215c1 | 228 | } |
ac2f40be N |
229 | /* Just make sure we aren't corrupting data or |
230 | * metadata | |
231 | */ | |
232 | if (mddev->external) { | |
233 | /* Bitmap could be anywhere. */ | |
234 | if (rdev->sb_start + offset + (page->index | |
235 | * (PAGE_SIZE/512)) | |
236 | > rdev->data_offset | |
237 | && | |
238 | rdev->sb_start + offset | |
239 | < (rdev->data_offset + mddev->dev_sectors | |
240 | + (PAGE_SIZE/512))) | |
241 | goto bad_alignment; | |
242 | } else if (offset < 0) { | |
243 | /* DATA BITMAP METADATA */ | |
244 | if (offset | |
245 | + (long)(page->index * (PAGE_SIZE/512)) | |
246 | + size/512 > 0) | |
247 | /* bitmap runs in to metadata */ | |
248 | goto bad_alignment; | |
249 | if (rdev->data_offset + mddev->dev_sectors | |
250 | > rdev->sb_start + offset) | |
251 | /* data runs in to bitmap */ | |
252 | goto bad_alignment; | |
253 | } else if (rdev->sb_start < rdev->data_offset) { | |
254 | /* METADATA BITMAP DATA */ | |
255 | if (rdev->sb_start | |
256 | + offset | |
257 | + page->index*(PAGE_SIZE/512) + size/512 | |
258 | > rdev->data_offset) | |
259 | /* bitmap runs in to data */ | |
260 | goto bad_alignment; | |
261 | } else { | |
262 | /* DATA METADATA BITMAP - no problems */ | |
263 | } | |
264 | md_super_write(mddev, rdev, | |
265 | rdev->sb_start + offset | |
266 | + page->index * (PAGE_SIZE/512), | |
267 | size, | |
268 | page); | |
b2d2c4ce | 269 | } |
a654b9d8 N |
270 | |
271 | if (wait) | |
a9701a30 | 272 | md_super_wait(mddev); |
a654b9d8 | 273 | return 0; |
4b80991c N |
274 | |
275 | bad_alignment: | |
4b80991c | 276 | return -EINVAL; |
a654b9d8 N |
277 | } |
278 | ||
4ad13663 | 279 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 280 | /* |
a654b9d8 | 281 | * write out a page to a file |
32a7627c | 282 | */ |
4ad13663 | 283 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 284 | { |
d785a06a | 285 | struct buffer_head *bh; |
32a7627c | 286 | |
1ec885cd | 287 | if (bitmap->storage.file == NULL) { |
f0d76d70 N |
288 | switch (write_sb_page(bitmap, page, wait)) { |
289 | case -EINVAL: | |
b405fe91 | 290 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); |
f0d76d70 | 291 | } |
4ad13663 | 292 | } else { |
a654b9d8 | 293 | |
4ad13663 | 294 | bh = page_buffers(page); |
c708443c | 295 | |
4ad13663 N |
296 | while (bh && bh->b_blocknr) { |
297 | atomic_inc(&bitmap->pending_writes); | |
298 | set_buffer_locked(bh); | |
299 | set_buffer_mapped(bh); | |
721a9602 | 300 | submit_bh(WRITE | REQ_SYNC, bh); |
4ad13663 N |
301 | bh = bh->b_this_page; |
302 | } | |
d785a06a | 303 | |
ac2f40be | 304 | if (wait) |
4ad13663 N |
305 | wait_event(bitmap->write_wait, |
306 | atomic_read(&bitmap->pending_writes)==0); | |
8a5e9cf1 | 307 | } |
b405fe91 | 308 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
4ad13663 | 309 | bitmap_file_kick(bitmap); |
d785a06a N |
310 | } |
311 | ||
312 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
313 | { | |
314 | struct bitmap *bitmap = bh->b_private; | |
32a7627c | 315 | |
b405fe91 N |
316 | if (!uptodate) |
317 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); | |
d785a06a N |
318 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
319 | wake_up(&bitmap->write_wait); | |
320 | } | |
32a7627c | 321 | |
d785a06a N |
322 | /* copied from buffer.c */ |
323 | static void | |
324 | __clear_page_buffers(struct page *page) | |
325 | { | |
326 | ClearPagePrivate(page); | |
327 | set_page_private(page, 0); | |
09cbfeaf | 328 | put_page(page); |
d785a06a N |
329 | } |
330 | static void free_buffers(struct page *page) | |
331 | { | |
27581e5a | 332 | struct buffer_head *bh; |
77ad4bc7 | 333 | |
27581e5a N |
334 | if (!PagePrivate(page)) |
335 | return; | |
336 | ||
337 | bh = page_buffers(page); | |
d785a06a N |
338 | while (bh) { |
339 | struct buffer_head *next = bh->b_this_page; | |
340 | free_buffer_head(bh); | |
341 | bh = next; | |
77ad4bc7 | 342 | } |
d785a06a N |
343 | __clear_page_buffers(page); |
344 | put_page(page); | |
32a7627c N |
345 | } |
346 | ||
d785a06a N |
347 | /* read a page from a file. |
348 | * We both read the page, and attach buffers to the page to record the | |
349 | * address of each block (using bmap). These addresses will be used | |
350 | * to write the block later, completely bypassing the filesystem. | |
351 | * This usage is similar to how swap files are handled, and allows us | |
352 | * to write to a file with no concerns of memory allocation failing. | |
353 | */ | |
27581e5a N |
354 | static int read_page(struct file *file, unsigned long index, |
355 | struct bitmap *bitmap, | |
356 | unsigned long count, | |
357 | struct page *page) | |
32a7627c | 358 | { |
27581e5a | 359 | int ret = 0; |
496ad9aa | 360 | struct inode *inode = file_inode(file); |
d785a06a N |
361 | struct buffer_head *bh; |
362 | sector_t block; | |
32a7627c | 363 | |
36a4e1fe N |
364 | pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
365 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 366 | |
d785a06a N |
367 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
368 | if (!bh) { | |
27581e5a | 369 | ret = -ENOMEM; |
32a7627c N |
370 | goto out; |
371 | } | |
d785a06a N |
372 | attach_page_buffers(page, bh); |
373 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
374 | while (bh) { | |
375 | if (count == 0) | |
376 | bh->b_blocknr = 0; | |
377 | else { | |
378 | bh->b_blocknr = bmap(inode, block); | |
379 | if (bh->b_blocknr == 0) { | |
380 | /* Cannot use this file! */ | |
27581e5a | 381 | ret = -EINVAL; |
d785a06a N |
382 | goto out; |
383 | } | |
384 | bh->b_bdev = inode->i_sb->s_bdev; | |
385 | if (count < (1<<inode->i_blkbits)) | |
386 | count = 0; | |
387 | else | |
388 | count -= (1<<inode->i_blkbits); | |
389 | ||
390 | bh->b_end_io = end_bitmap_write; | |
391 | bh->b_private = bitmap; | |
ce25c31b N |
392 | atomic_inc(&bitmap->pending_writes); |
393 | set_buffer_locked(bh); | |
394 | set_buffer_mapped(bh); | |
395 | submit_bh(READ, bh); | |
d785a06a N |
396 | } |
397 | block++; | |
398 | bh = bh->b_this_page; | |
399 | } | |
d785a06a | 400 | page->index = index; |
ce25c31b N |
401 | |
402 | wait_event(bitmap->write_wait, | |
403 | atomic_read(&bitmap->pending_writes)==0); | |
b405fe91 | 404 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
27581e5a | 405 | ret = -EIO; |
32a7627c | 406 | out: |
27581e5a N |
407 | if (ret) |
408 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n", | |
2d1f3b5d N |
409 | (int)PAGE_SIZE, |
410 | (unsigned long long)index << PAGE_SHIFT, | |
27581e5a N |
411 | ret); |
412 | return ret; | |
32a7627c N |
413 | } |
414 | ||
415 | /* | |
416 | * bitmap file superblock operations | |
417 | */ | |
418 | ||
419 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 420 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
421 | { |
422 | bitmap_super_t *sb; | |
32a7627c N |
423 | |
424 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 425 | return; |
ece5cff0 N |
426 | if (bitmap->mddev->bitmap_info.external) |
427 | return; | |
1ec885cd | 428 | if (!bitmap->storage.sb_page) /* no superblock */ |
4ad13663 | 429 | return; |
1ec885cd | 430 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 431 | sb->events = cpu_to_le64(bitmap->mddev->events); |
8258c532 | 432 | if (bitmap->mddev->events < bitmap->events_cleared) |
a0da84f3 NB |
433 | /* rocking back to read-only */ |
434 | bitmap->events_cleared = bitmap->mddev->events; | |
8258c532 N |
435 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); |
436 | sb->state = cpu_to_le32(bitmap->flags); | |
43a70507 N |
437 | /* Just in case these have been changed via sysfs: */ |
438 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
439 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
b81a0404 N |
440 | /* This might have been changed by a reshape */ |
441 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
442 | sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize); | |
c4ce867f | 443 | sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes); |
1dff2b87 N |
444 | sb->sectors_reserved = cpu_to_le32(bitmap->mddev-> |
445 | bitmap_info.space); | |
b2f46e68 | 446 | kunmap_atomic(sb); |
1ec885cd | 447 | write_page(bitmap, bitmap->storage.sb_page, 1); |
32a7627c N |
448 | } |
449 | ||
450 | /* print out the bitmap file superblock */ | |
451 | void bitmap_print_sb(struct bitmap *bitmap) | |
452 | { | |
453 | bitmap_super_t *sb; | |
454 | ||
1ec885cd | 455 | if (!bitmap || !bitmap->storage.sb_page) |
32a7627c | 456 | return; |
1ec885cd | 457 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 458 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
459 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
460 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
461 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
462 | *(__u32 *)(sb->uuid+0), |
463 | *(__u32 *)(sb->uuid+4), | |
464 | *(__u32 *)(sb->uuid+8), | |
465 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 466 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 467 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 468 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 469 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
470 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
471 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
472 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
473 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
474 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 475 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
b2f46e68 | 476 | kunmap_atomic(sb); |
32a7627c N |
477 | } |
478 | ||
9c81075f JB |
479 | /* |
480 | * bitmap_new_disk_sb | |
481 | * @bitmap | |
482 | * | |
483 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb | |
484 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. | |
485 | * This function verifies 'bitmap_info' and populates the on-disk bitmap | |
486 | * structure, which is to be written to disk. | |
487 | * | |
488 | * Returns: 0 on success, -Exxx on error | |
489 | */ | |
490 | static int bitmap_new_disk_sb(struct bitmap *bitmap) | |
491 | { | |
492 | bitmap_super_t *sb; | |
493 | unsigned long chunksize, daemon_sleep, write_behind; | |
9c81075f | 494 | |
d3b178ad | 495 | bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
582e2e05 JM |
496 | if (bitmap->storage.sb_page == NULL) |
497 | return -ENOMEM; | |
1ec885cd | 498 | bitmap->storage.sb_page->index = 0; |
9c81075f | 499 | |
1ec885cd | 500 | sb = kmap_atomic(bitmap->storage.sb_page); |
9c81075f JB |
501 | |
502 | sb->magic = cpu_to_le32(BITMAP_MAGIC); | |
503 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); | |
504 | ||
505 | chunksize = bitmap->mddev->bitmap_info.chunksize; | |
506 | BUG_ON(!chunksize); | |
507 | if (!is_power_of_2(chunksize)) { | |
b2f46e68 | 508 | kunmap_atomic(sb); |
9c81075f JB |
509 | printk(KERN_ERR "bitmap chunksize not a power of 2\n"); |
510 | return -EINVAL; | |
511 | } | |
512 | sb->chunksize = cpu_to_le32(chunksize); | |
513 | ||
514 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; | |
c97e0602 | 515 | if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { |
9c81075f JB |
516 | printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n"); |
517 | daemon_sleep = 5 * HZ; | |
518 | } | |
519 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); | |
520 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
521 | ||
522 | /* | |
523 | * FIXME: write_behind for RAID1. If not specified, what | |
524 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. | |
525 | */ | |
526 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; | |
527 | if (write_behind > COUNTER_MAX) | |
528 | write_behind = COUNTER_MAX / 2; | |
529 | sb->write_behind = cpu_to_le32(write_behind); | |
530 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
531 | ||
532 | /* keep the array size field of the bitmap superblock up to date */ | |
533 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
534 | ||
535 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); | |
536 | ||
b405fe91 | 537 | set_bit(BITMAP_STALE, &bitmap->flags); |
84e92345 | 538 | sb->state = cpu_to_le32(bitmap->flags); |
9c81075f JB |
539 | bitmap->events_cleared = bitmap->mddev->events; |
540 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
d3b178ad | 541 | bitmap->mddev->bitmap_info.nodes = 0; |
9c81075f | 542 | |
b2f46e68 | 543 | kunmap_atomic(sb); |
9c81075f JB |
544 | |
545 | return 0; | |
546 | } | |
547 | ||
32a7627c N |
548 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
549 | static int bitmap_read_sb(struct bitmap *bitmap) | |
550 | { | |
551 | char *reason = NULL; | |
552 | bitmap_super_t *sb; | |
4b6d287f | 553 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c | 554 | unsigned long long events; |
c4ce867f | 555 | int nodes = 0; |
1dff2b87 | 556 | unsigned long sectors_reserved = 0; |
32a7627c | 557 | int err = -EINVAL; |
27581e5a | 558 | struct page *sb_page; |
33e38ac6 | 559 | loff_t offset = bitmap->mddev->bitmap_info.offset; |
32a7627c | 560 | |
1ec885cd | 561 | if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { |
ef99bf48 N |
562 | chunksize = 128 * 1024 * 1024; |
563 | daemon_sleep = 5 * HZ; | |
564 | write_behind = 0; | |
b405fe91 | 565 | set_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
566 | err = 0; |
567 | goto out_no_sb; | |
568 | } | |
32a7627c | 569 | /* page 0 is the superblock, read it... */ |
27581e5a N |
570 | sb_page = alloc_page(GFP_KERNEL); |
571 | if (!sb_page) | |
572 | return -ENOMEM; | |
1ec885cd | 573 | bitmap->storage.sb_page = sb_page; |
27581e5a | 574 | |
b97e9257 | 575 | re_read: |
f9209a32 GR |
576 | /* If cluster_slot is set, the cluster is setup */ |
577 | if (bitmap->cluster_slot >= 0) { | |
3b0e6aac | 578 | sector_t bm_blocks = bitmap->mddev->resync_max_sectors; |
f9209a32 | 579 | |
3b0e6aac SR |
580 | sector_div(bm_blocks, |
581 | bitmap->mddev->bitmap_info.chunksize >> 9); | |
124eb761 GR |
582 | /* bits to bytes */ |
583 | bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t); | |
584 | /* to 4k blocks */ | |
935f3d4f | 585 | bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096); |
33e38ac6 | 586 | offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3)); |
f9209a32 | 587 | pr_info("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__, |
33e38ac6 | 588 | bitmap->cluster_slot, offset); |
f9209a32 GR |
589 | } |
590 | ||
1ec885cd N |
591 | if (bitmap->storage.file) { |
592 | loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); | |
f49d5e62 N |
593 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; |
594 | ||
1ec885cd | 595 | err = read_page(bitmap->storage.file, 0, |
27581e5a | 596 | bitmap, bytes, sb_page); |
f49d5e62 | 597 | } else { |
27581e5a | 598 | err = read_sb_page(bitmap->mddev, |
33e38ac6 | 599 | offset, |
27581e5a N |
600 | sb_page, |
601 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 602 | } |
27581e5a | 603 | if (err) |
32a7627c | 604 | return err; |
32a7627c | 605 | |
b97e9257 | 606 | err = -EINVAL; |
27581e5a | 607 | sb = kmap_atomic(sb_page); |
32a7627c | 608 | |
32a7627c | 609 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 610 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 611 | write_behind = le32_to_cpu(sb->write_behind); |
1dff2b87 | 612 | sectors_reserved = le32_to_cpu(sb->sectors_reserved); |
3c462c88 GR |
613 | /* Setup nodes/clustername only if bitmap version is |
614 | * cluster-compatible | |
d3b178ad | 615 | */ |
3c462c88 | 616 | if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) { |
d3b178ad GR |
617 | nodes = le32_to_cpu(sb->nodes); |
618 | strlcpy(bitmap->mddev->bitmap_info.cluster_name, | |
619 | sb->cluster_name, 64); | |
620 | } | |
32a7627c N |
621 | |
622 | /* verify that the bitmap-specific fields are valid */ | |
623 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
624 | reason = "bad magic"; | |
bd926c63 | 625 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
3c462c88 | 626 | le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED) |
32a7627c | 627 | reason = "unrecognized superblock version"; |
1187cf0a | 628 | else if (chunksize < 512) |
7dd5d34c | 629 | reason = "bitmap chunksize too small"; |
d744540c | 630 | else if (!is_power_of_2(chunksize)) |
32a7627c | 631 | reason = "bitmap chunksize not a power of 2"; |
1b04be96 | 632 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 633 | reason = "daemon sleep period out of range"; |
4b6d287f N |
634 | else if (write_behind > COUNTER_MAX) |
635 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
636 | if (reason) { |
637 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
638 | bmname(bitmap), reason); | |
639 | goto out; | |
640 | } | |
641 | ||
642 | /* keep the array size field of the bitmap superblock up to date */ | |
643 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
644 | ||
278c1ca2 N |
645 | if (bitmap->mddev->persistent) { |
646 | /* | |
647 | * We have a persistent array superblock, so compare the | |
648 | * bitmap's UUID and event counter to the mddev's | |
649 | */ | |
650 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
651 | printk(KERN_INFO | |
652 | "%s: bitmap superblock UUID mismatch\n", | |
653 | bmname(bitmap)); | |
654 | goto out; | |
655 | } | |
656 | events = le64_to_cpu(sb->events); | |
b97e9257 | 657 | if (!nodes && (events < bitmap->mddev->events)) { |
278c1ca2 N |
658 | printk(KERN_INFO |
659 | "%s: bitmap file is out of date (%llu < %llu) " | |
660 | "-- forcing full recovery\n", | |
661 | bmname(bitmap), events, | |
662 | (unsigned long long) bitmap->mddev->events); | |
b405fe91 | 663 | set_bit(BITMAP_STALE, &bitmap->flags); |
278c1ca2 | 664 | } |
32a7627c | 665 | } |
278c1ca2 | 666 | |
32a7627c | 667 | /* assign fields using values from superblock */ |
4f2e639a | 668 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 | 669 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
b405fe91 | 670 | set_bit(BITMAP_HOSTENDIAN, &bitmap->flags); |
32a7627c | 671 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
cf921cc1 | 672 | strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64); |
32a7627c | 673 | err = 0; |
b97e9257 | 674 | |
32a7627c | 675 | out: |
b2f46e68 | 676 | kunmap_atomic(sb); |
f9209a32 GR |
677 | /* Assiging chunksize is required for "re_read" */ |
678 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
f7357273 | 679 | if (err == 0 && nodes && (bitmap->cluster_slot < 0)) { |
b97e9257 GR |
680 | err = md_setup_cluster(bitmap->mddev, nodes); |
681 | if (err) { | |
682 | pr_err("%s: Could not setup cluster service (%d)\n", | |
683 | bmname(bitmap), err); | |
684 | goto out_no_sb; | |
685 | } | |
686 | bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev); | |
b97e9257 GR |
687 | goto re_read; |
688 | } | |
689 | ||
690 | ||
ef99bf48 | 691 | out_no_sb: |
b405fe91 | 692 | if (test_bit(BITMAP_STALE, &bitmap->flags)) |
ef99bf48 N |
693 | bitmap->events_cleared = bitmap->mddev->events; |
694 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
695 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
696 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
c4ce867f | 697 | bitmap->mddev->bitmap_info.nodes = nodes; |
1dff2b87 N |
698 | if (bitmap->mddev->bitmap_info.space == 0 || |
699 | bitmap->mddev->bitmap_info.space > sectors_reserved) | |
700 | bitmap->mddev->bitmap_info.space = sectors_reserved; | |
b97e9257 | 701 | if (err) { |
32a7627c | 702 | bitmap_print_sb(bitmap); |
f9209a32 | 703 | if (bitmap->cluster_slot < 0) |
b97e9257 GR |
704 | md_cluster_stop(bitmap->mddev); |
705 | } | |
32a7627c N |
706 | return err; |
707 | } | |
708 | ||
32a7627c N |
709 | /* |
710 | * general bitmap file operations | |
711 | */ | |
712 | ||
ece5cff0 N |
713 | /* |
714 | * on-disk bitmap: | |
715 | * | |
716 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
717 | * file a page at a time. There's a superblock at the start of the file. | |
718 | */ | |
32a7627c | 719 | /* calculate the index of the page that contains this bit */ |
1ec885cd N |
720 | static inline unsigned long file_page_index(struct bitmap_storage *store, |
721 | unsigned long chunk) | |
32a7627c | 722 | { |
1ec885cd | 723 | if (store->sb_page) |
ece5cff0 N |
724 | chunk += sizeof(bitmap_super_t) << 3; |
725 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
726 | } |
727 | ||
728 | /* calculate the (bit) offset of this bit within a page */ | |
1ec885cd N |
729 | static inline unsigned long file_page_offset(struct bitmap_storage *store, |
730 | unsigned long chunk) | |
32a7627c | 731 | { |
1ec885cd | 732 | if (store->sb_page) |
ece5cff0 N |
733 | chunk += sizeof(bitmap_super_t) << 3; |
734 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
735 | } |
736 | ||
737 | /* | |
738 | * return a pointer to the page in the filemap that contains the given bit | |
739 | * | |
32a7627c | 740 | */ |
1ec885cd | 741 | static inline struct page *filemap_get_page(struct bitmap_storage *store, |
3520fa4d | 742 | unsigned long chunk) |
32a7627c | 743 | { |
1ec885cd | 744 | if (file_page_index(store, chunk) >= store->file_pages) |
ac2f40be | 745 | return NULL; |
f2e06c58 | 746 | return store->filemap[file_page_index(store, chunk)]; |
32a7627c N |
747 | } |
748 | ||
d1244cb0 | 749 | static int bitmap_storage_alloc(struct bitmap_storage *store, |
b97e9257 GR |
750 | unsigned long chunks, int with_super, |
751 | int slot_number) | |
d1244cb0 | 752 | { |
b97e9257 | 753 | int pnum, offset = 0; |
d1244cb0 N |
754 | unsigned long num_pages; |
755 | unsigned long bytes; | |
756 | ||
757 | bytes = DIV_ROUND_UP(chunks, 8); | |
758 | if (with_super) | |
759 | bytes += sizeof(bitmap_super_t); | |
760 | ||
761 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); | |
7f86ffed | 762 | offset = slot_number * num_pages; |
d1244cb0 N |
763 | |
764 | store->filemap = kmalloc(sizeof(struct page *) | |
765 | * num_pages, GFP_KERNEL); | |
766 | if (!store->filemap) | |
767 | return -ENOMEM; | |
768 | ||
769 | if (with_super && !store->sb_page) { | |
d60b479d | 770 | store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
771 | if (store->sb_page == NULL) |
772 | return -ENOMEM; | |
d1244cb0 | 773 | } |
b97e9257 | 774 | |
d1244cb0 N |
775 | pnum = 0; |
776 | if (store->sb_page) { | |
777 | store->filemap[0] = store->sb_page; | |
778 | pnum = 1; | |
b97e9257 | 779 | store->sb_page->index = offset; |
d1244cb0 | 780 | } |
b97e9257 | 781 | |
d1244cb0 | 782 | for ( ; pnum < num_pages; pnum++) { |
d60b479d | 783 | store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
784 | if (!store->filemap[pnum]) { |
785 | store->file_pages = pnum; | |
786 | return -ENOMEM; | |
787 | } | |
b97e9257 | 788 | store->filemap[pnum]->index = pnum + offset; |
d1244cb0 N |
789 | } |
790 | store->file_pages = pnum; | |
791 | ||
792 | /* We need 4 bits per page, rounded up to a multiple | |
793 | * of sizeof(unsigned long) */ | |
794 | store->filemap_attr = kzalloc( | |
795 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), | |
796 | GFP_KERNEL); | |
797 | if (!store->filemap_attr) | |
798 | return -ENOMEM; | |
799 | ||
800 | store->bytes = bytes; | |
801 | ||
802 | return 0; | |
803 | } | |
804 | ||
fae7d326 | 805 | static void bitmap_file_unmap(struct bitmap_storage *store) |
32a7627c N |
806 | { |
807 | struct page **map, *sb_page; | |
32a7627c | 808 | int pages; |
fae7d326 | 809 | struct file *file; |
32a7627c | 810 | |
fae7d326 | 811 | file = store->file; |
1ec885cd | 812 | map = store->filemap; |
1ec885cd | 813 | pages = store->file_pages; |
1ec885cd | 814 | sb_page = store->sb_page; |
32a7627c N |
815 | |
816 | while (pages--) | |
ece5cff0 | 817 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 818 | free_buffers(map[pages]); |
32a7627c | 819 | kfree(map); |
fae7d326 | 820 | kfree(store->filemap_attr); |
32a7627c | 821 | |
d785a06a N |
822 | if (sb_page) |
823 | free_buffers(sb_page); | |
32a7627c | 824 | |
d785a06a | 825 | if (file) { |
496ad9aa | 826 | struct inode *inode = file_inode(file); |
fc0ecff6 | 827 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 828 | fput(file); |
d785a06a | 829 | } |
32a7627c N |
830 | } |
831 | ||
32a7627c N |
832 | /* |
833 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
834 | * then it is no longer reliable, so we stop using it and we mark the file | |
835 | * as failed in the superblock | |
836 | */ | |
837 | static void bitmap_file_kick(struct bitmap *bitmap) | |
838 | { | |
839 | char *path, *ptr = NULL; | |
840 | ||
b405fe91 | 841 | if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) { |
4ad13663 | 842 | bitmap_update_sb(bitmap); |
32a7627c | 843 | |
1ec885cd | 844 | if (bitmap->storage.file) { |
4ad13663 N |
845 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); |
846 | if (path) | |
9bf39ab2 | 847 | ptr = file_path(bitmap->storage.file, |
1ec885cd | 848 | path, PAGE_SIZE); |
6bcfd601 | 849 | |
4ad13663 N |
850 | printk(KERN_ALERT |
851 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 852 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 853 | |
4ad13663 N |
854 | kfree(path); |
855 | } else | |
856 | printk(KERN_ALERT | |
857 | "%s: disabling internal bitmap due to errors\n", | |
858 | bmname(bitmap)); | |
a654b9d8 | 859 | } |
32a7627c N |
860 | } |
861 | ||
862 | enum bitmap_page_attr { | |
ac2f40be | 863 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
5a537df4 N |
864 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
865 | * i.e. counter is 1 or 2. */ | |
ac2f40be | 866 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
32a7627c N |
867 | }; |
868 | ||
d189122d N |
869 | static inline void set_page_attr(struct bitmap *bitmap, int pnum, |
870 | enum bitmap_page_attr attr) | |
32a7627c | 871 | { |
bdfd1140 | 872 | set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
873 | } |
874 | ||
d189122d N |
875 | static inline void clear_page_attr(struct bitmap *bitmap, int pnum, |
876 | enum bitmap_page_attr attr) | |
32a7627c | 877 | { |
bdfd1140 | 878 | clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
879 | } |
880 | ||
bdfd1140 N |
881 | static inline int test_page_attr(struct bitmap *bitmap, int pnum, |
882 | enum bitmap_page_attr attr) | |
32a7627c | 883 | { |
1ec885cd | 884 | return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
885 | } |
886 | ||
bdfd1140 N |
887 | static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum, |
888 | enum bitmap_page_attr attr) | |
889 | { | |
890 | return test_and_clear_bit((pnum<<2) + attr, | |
891 | bitmap->storage.filemap_attr); | |
892 | } | |
32a7627c N |
893 | /* |
894 | * bitmap_file_set_bit -- called before performing a write to the md device | |
895 | * to set (and eventually sync) a particular bit in the bitmap file | |
896 | * | |
897 | * we set the bit immediately, then we record the page number so that | |
898 | * when an unplug occurs, we can flush the dirty pages out to disk | |
899 | */ | |
900 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
901 | { | |
902 | unsigned long bit; | |
3520fa4d | 903 | struct page *page; |
32a7627c | 904 | void *kaddr; |
40cffcc0 | 905 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
32a7627c | 906 | |
1ec885cd | 907 | page = filemap_get_page(&bitmap->storage, chunk); |
3520fa4d JB |
908 | if (!page) |
909 | return; | |
1ec885cd | 910 | bit = file_page_offset(&bitmap->storage, chunk); |
32a7627c | 911 | |
3520fa4d | 912 | /* set the bit */ |
b2f46e68 | 913 | kaddr = kmap_atomic(page); |
b405fe91 | 914 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
3520fa4d JB |
915 | set_bit(bit, kaddr); |
916 | else | |
3f810b6c | 917 | set_bit_le(bit, kaddr); |
b2f46e68 | 918 | kunmap_atomic(kaddr); |
36a4e1fe | 919 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
32a7627c | 920 | /* record page number so it gets flushed to disk when unplug occurs */ |
d189122d | 921 | set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY); |
32a7627c N |
922 | } |
923 | ||
ef99bf48 N |
924 | static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) |
925 | { | |
926 | unsigned long bit; | |
927 | struct page *page; | |
928 | void *paddr; | |
40cffcc0 | 929 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
ef99bf48 | 930 | |
1ec885cd | 931 | page = filemap_get_page(&bitmap->storage, chunk); |
ef99bf48 N |
932 | if (!page) |
933 | return; | |
1ec885cd | 934 | bit = file_page_offset(&bitmap->storage, chunk); |
ef99bf48 | 935 | paddr = kmap_atomic(page); |
b405fe91 | 936 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ef99bf48 N |
937 | clear_bit(bit, paddr); |
938 | else | |
3f810b6c | 939 | clear_bit_le(bit, paddr); |
ef99bf48 | 940 | kunmap_atomic(paddr); |
d189122d N |
941 | if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) { |
942 | set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING); | |
ef99bf48 N |
943 | bitmap->allclean = 0; |
944 | } | |
945 | } | |
946 | ||
11dd35da GR |
947 | static int bitmap_file_test_bit(struct bitmap *bitmap, sector_t block) |
948 | { | |
949 | unsigned long bit; | |
950 | struct page *page; | |
951 | void *paddr; | |
952 | unsigned long chunk = block >> bitmap->counts.chunkshift; | |
953 | int set = 0; | |
954 | ||
955 | page = filemap_get_page(&bitmap->storage, chunk); | |
956 | if (!page) | |
957 | return -EINVAL; | |
958 | bit = file_page_offset(&bitmap->storage, chunk); | |
959 | paddr = kmap_atomic(page); | |
960 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) | |
961 | set = test_bit(bit, paddr); | |
962 | else | |
963 | set = test_bit_le(bit, paddr); | |
964 | kunmap_atomic(paddr); | |
965 | return set; | |
966 | } | |
967 | ||
968 | ||
32a7627c N |
969 | /* this gets called when the md device is ready to unplug its underlying |
970 | * (slave) device queues -- before we let any writes go down, we need to | |
971 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 972 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 973 | { |
74667123 | 974 | unsigned long i; |
ec7a3197 | 975 | int dirty, need_write; |
32a7627c | 976 | |
62f82faa N |
977 | if (!bitmap || !bitmap->storage.filemap || |
978 | test_bit(BITMAP_STALE, &bitmap->flags)) | |
4ad13663 | 979 | return; |
32a7627c N |
980 | |
981 | /* look at each page to see if there are any set bits that need to be | |
982 | * flushed out to disk */ | |
1ec885cd | 983 | for (i = 0; i < bitmap->storage.file_pages; i++) { |
bdfd1140 | 984 | if (!bitmap->storage.filemap) |
4ad13663 | 985 | return; |
bdfd1140 N |
986 | dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
987 | need_write = test_and_clear_page_attr(bitmap, i, | |
988 | BITMAP_PAGE_NEEDWRITE); | |
989 | if (dirty || need_write) { | |
d189122d | 990 | clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); |
bdfd1140 N |
991 | write_page(bitmap, bitmap->storage.filemap[i], 0); |
992 | } | |
32a7627c | 993 | } |
4b5060dd N |
994 | if (bitmap->storage.file) |
995 | wait_event(bitmap->write_wait, | |
996 | atomic_read(&bitmap->pending_writes)==0); | |
997 | else | |
998 | md_super_wait(bitmap->mddev); | |
999 | ||
b405fe91 | 1000 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
d785a06a | 1001 | bitmap_file_kick(bitmap); |
32a7627c | 1002 | } |
ac2f40be | 1003 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 1004 | |
6a07997f | 1005 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
1006 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
1007 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
1008 | * memory mapping of the bitmap file | |
1009 | * Special cases: | |
1010 | * if there's no bitmap file, or if the bitmap file had been | |
1011 | * previously kicked from the array, we mark all the bits as | |
1012 | * 1's in order to cause a full resync. | |
6a07997f N |
1013 | * |
1014 | * We ignore all bits for sectors that end earlier than 'start'. | |
1015 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 1016 | */ |
6a07997f | 1017 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c | 1018 | { |
b97e9257 | 1019 | unsigned long i, chunks, index, oldindex, bit, node_offset = 0; |
27581e5a | 1020 | struct page *page = NULL; |
d1244cb0 | 1021 | unsigned long bit_cnt = 0; |
32a7627c | 1022 | struct file *file; |
d1244cb0 | 1023 | unsigned long offset; |
32a7627c N |
1024 | int outofdate; |
1025 | int ret = -ENOSPC; | |
ea03aff9 | 1026 | void *paddr; |
1ec885cd | 1027 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c | 1028 | |
40cffcc0 | 1029 | chunks = bitmap->counts.chunks; |
1ec885cd | 1030 | file = store->file; |
32a7627c | 1031 | |
ef99bf48 N |
1032 | if (!file && !bitmap->mddev->bitmap_info.offset) { |
1033 | /* No permanent bitmap - fill with '1s'. */ | |
1ec885cd N |
1034 | store->filemap = NULL; |
1035 | store->file_pages = 0; | |
ef99bf48 N |
1036 | for (i = 0; i < chunks ; i++) { |
1037 | /* if the disk bit is set, set the memory bit */ | |
40cffcc0 | 1038 | int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift) |
ef99bf48 N |
1039 | >= start); |
1040 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 1041 | (sector_t)i << bitmap->counts.chunkshift, |
ef99bf48 N |
1042 | needed); |
1043 | } | |
1044 | return 0; | |
1045 | } | |
32a7627c | 1046 | |
b405fe91 | 1047 | outofdate = test_bit(BITMAP_STALE, &bitmap->flags); |
32a7627c N |
1048 | if (outofdate) |
1049 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
1050 | "recovery\n", bmname(bitmap)); | |
1051 | ||
d1244cb0 | 1052 | if (file && i_size_read(file->f_mapping->host) < store->bytes) { |
32a7627c | 1053 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
d1244cb0 N |
1054 | bmname(bitmap), |
1055 | (unsigned long) i_size_read(file->f_mapping->host), | |
1056 | store->bytes); | |
4ad13663 | 1057 | goto err; |
32a7627c | 1058 | } |
bc7f77de | 1059 | |
d1244cb0 | 1060 | oldindex = ~0L; |
27581e5a | 1061 | offset = 0; |
d1244cb0 | 1062 | if (!bitmap->mddev->bitmap_info.external) |
27581e5a | 1063 | offset = sizeof(bitmap_super_t); |
32a7627c | 1064 | |
b97e9257 GR |
1065 | if (mddev_is_clustered(bitmap->mddev)) |
1066 | node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE)); | |
1067 | ||
32a7627c | 1068 | for (i = 0; i < chunks; i++) { |
bd926c63 | 1069 | int b; |
1ec885cd N |
1070 | index = file_page_index(&bitmap->storage, i); |
1071 | bit = file_page_offset(&bitmap->storage, i); | |
32a7627c | 1072 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 1073 | int count; |
32a7627c | 1074 | /* unmap the old page, we're done with it */ |
d1244cb0 N |
1075 | if (index == store->file_pages-1) |
1076 | count = store->bytes - index * PAGE_SIZE; | |
d785a06a N |
1077 | else |
1078 | count = PAGE_SIZE; | |
1ec885cd | 1079 | page = store->filemap[index]; |
27581e5a N |
1080 | if (file) |
1081 | ret = read_page(file, index, bitmap, | |
1082 | count, page); | |
1083 | else | |
1084 | ret = read_sb_page( | |
1085 | bitmap->mddev, | |
1086 | bitmap->mddev->bitmap_info.offset, | |
1087 | page, | |
b97e9257 | 1088 | index + node_offset, count); |
27581e5a N |
1089 | |
1090 | if (ret) | |
4ad13663 | 1091 | goto err; |
a654b9d8 | 1092 | |
32a7627c | 1093 | oldindex = index; |
32a7627c N |
1094 | |
1095 | if (outofdate) { | |
1096 | /* | |
1097 | * if bitmap is out of date, dirty the | |
ac2f40be | 1098 | * whole page and write it out |
32a7627c | 1099 | */ |
b2f46e68 | 1100 | paddr = kmap_atomic(page); |
ea03aff9 | 1101 | memset(paddr + offset, 0xff, |
6a07997f | 1102 | PAGE_SIZE - offset); |
b2f46e68 | 1103 | kunmap_atomic(paddr); |
4ad13663 N |
1104 | write_page(bitmap, page, 1); |
1105 | ||
1106 | ret = -EIO; | |
b405fe91 N |
1107 | if (test_bit(BITMAP_WRITE_ERROR, |
1108 | &bitmap->flags)) | |
4ad13663 | 1109 | goto err; |
32a7627c | 1110 | } |
32a7627c | 1111 | } |
b2f46e68 | 1112 | paddr = kmap_atomic(page); |
b405fe91 | 1113 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ea03aff9 | 1114 | b = test_bit(bit, paddr); |
bd926c63 | 1115 | else |
6b33aff3 | 1116 | b = test_bit_le(bit, paddr); |
b2f46e68 | 1117 | kunmap_atomic(paddr); |
bd926c63 | 1118 | if (b) { |
32a7627c | 1119 | /* if the disk bit is set, set the memory bit */ |
40cffcc0 | 1120 | int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift |
db305e50 N |
1121 | >= start); |
1122 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 1123 | (sector_t)i << bitmap->counts.chunkshift, |
db305e50 | 1124 | needed); |
32a7627c N |
1125 | bit_cnt++; |
1126 | } | |
27581e5a | 1127 | offset = 0; |
32a7627c N |
1128 | } |
1129 | ||
32a7627c | 1130 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
d1244cb0 | 1131 | "read %lu pages, set %lu of %lu bits\n", |
1ec885cd | 1132 | bmname(bitmap), store->file_pages, |
d1244cb0 | 1133 | bit_cnt, chunks); |
4ad13663 N |
1134 | |
1135 | return 0; | |
32a7627c | 1136 | |
4ad13663 N |
1137 | err: |
1138 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1139 | bmname(bitmap), ret); | |
32a7627c N |
1140 | return ret; |
1141 | } | |
1142 | ||
a654b9d8 N |
1143 | void bitmap_write_all(struct bitmap *bitmap) |
1144 | { | |
1145 | /* We don't actually write all bitmap blocks here, | |
1146 | * just flag them as needing to be written | |
1147 | */ | |
ec7a3197 | 1148 | int i; |
a654b9d8 | 1149 | |
1ec885cd | 1150 | if (!bitmap || !bitmap->storage.filemap) |
ef99bf48 | 1151 | return; |
1ec885cd | 1152 | if (bitmap->storage.file) |
ef99bf48 N |
1153 | /* Only one copy, so nothing needed */ |
1154 | return; | |
1155 | ||
1ec885cd | 1156 | for (i = 0; i < bitmap->storage.file_pages; i++) |
d189122d | 1157 | set_page_attr(bitmap, i, |
ec7a3197 | 1158 | BITMAP_PAGE_NEEDWRITE); |
2585f3ef | 1159 | bitmap->allclean = 0; |
a654b9d8 N |
1160 | } |
1161 | ||
40cffcc0 N |
1162 | static void bitmap_count_page(struct bitmap_counts *bitmap, |
1163 | sector_t offset, int inc) | |
32a7627c | 1164 | { |
61a0d80c | 1165 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1166 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1167 | bitmap->bp[page].count += inc; | |
32a7627c N |
1168 | bitmap_checkfree(bitmap, page); |
1169 | } | |
bf07bb7d | 1170 | |
40cffcc0 | 1171 | static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset) |
bf07bb7d N |
1172 | { |
1173 | sector_t chunk = offset >> bitmap->chunkshift; | |
1174 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1175 | struct bitmap_page *bp = &bitmap->bp[page]; | |
1176 | ||
1177 | if (!bp->pending) | |
1178 | bp->pending = 1; | |
1179 | } | |
1180 | ||
40cffcc0 | 1181 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1182 | sector_t offset, sector_t *blocks, |
32a7627c N |
1183 | int create); |
1184 | ||
1185 | /* | |
1186 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1187 | * out to disk | |
1188 | */ | |
1189 | ||
fd01b88c | 1190 | void bitmap_daemon_work(struct mddev *mddev) |
32a7627c | 1191 | { |
aa5cbd10 | 1192 | struct bitmap *bitmap; |
aa3163f8 | 1193 | unsigned long j; |
bf07bb7d | 1194 | unsigned long nextpage; |
57dab0bd | 1195 | sector_t blocks; |
40cffcc0 | 1196 | struct bitmap_counts *counts; |
32a7627c | 1197 | |
aa5cbd10 N |
1198 | /* Use a mutex to guard daemon_work against |
1199 | * bitmap_destroy. | |
1200 | */ | |
c3d9714e | 1201 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1202 | bitmap = mddev->bitmap; |
1203 | if (bitmap == NULL) { | |
c3d9714e | 1204 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1205 | return; |
aa5cbd10 | 1206 | } |
42a04b50 | 1207 | if (time_before(jiffies, bitmap->daemon_lastrun |
2e61ebbc | 1208 | + mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1209 | goto done; |
1210 | ||
32a7627c | 1211 | bitmap->daemon_lastrun = jiffies; |
8311c29d | 1212 | if (bitmap->allclean) { |
2e61ebbc | 1213 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
aa5cbd10 | 1214 | goto done; |
8311c29d N |
1215 | } |
1216 | bitmap->allclean = 1; | |
32a7627c | 1217 | |
bf07bb7d N |
1218 | /* Any file-page which is PENDING now needs to be written. |
1219 | * So set NEEDWRITE now, then after we make any last-minute changes | |
1220 | * we will write it. | |
1221 | */ | |
1ec885cd | 1222 | for (j = 0; j < bitmap->storage.file_pages; j++) |
bdfd1140 N |
1223 | if (test_and_clear_page_attr(bitmap, j, |
1224 | BITMAP_PAGE_PENDING)) | |
d189122d | 1225 | set_page_attr(bitmap, j, |
bf07bb7d | 1226 | BITMAP_PAGE_NEEDWRITE); |
bf07bb7d N |
1227 | |
1228 | if (bitmap->need_sync && | |
1229 | mddev->bitmap_info.external == 0) { | |
1230 | /* Arrange for superblock update as well as | |
1231 | * other changes */ | |
1232 | bitmap_super_t *sb; | |
1233 | bitmap->need_sync = 0; | |
1ec885cd N |
1234 | if (bitmap->storage.filemap) { |
1235 | sb = kmap_atomic(bitmap->storage.sb_page); | |
ef99bf48 N |
1236 | sb->events_cleared = |
1237 | cpu_to_le64(bitmap->events_cleared); | |
1238 | kunmap_atomic(sb); | |
d189122d | 1239 | set_page_attr(bitmap, 0, |
ef99bf48 N |
1240 | BITMAP_PAGE_NEEDWRITE); |
1241 | } | |
bf07bb7d N |
1242 | } |
1243 | /* Now look at the bitmap counters and if any are '2' or '1', | |
1244 | * decrement and handle accordingly. | |
1245 | */ | |
40cffcc0 N |
1246 | counts = &bitmap->counts; |
1247 | spin_lock_irq(&counts->lock); | |
bf07bb7d | 1248 | nextpage = 0; |
40cffcc0 | 1249 | for (j = 0; j < counts->chunks; j++) { |
32a7627c | 1250 | bitmap_counter_t *bmc; |
40cffcc0 | 1251 | sector_t block = (sector_t)j << counts->chunkshift; |
3520fa4d | 1252 | |
bf07bb7d N |
1253 | if (j == nextpage) { |
1254 | nextpage += PAGE_COUNTER_RATIO; | |
40cffcc0 | 1255 | if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) { |
bf07bb7d | 1256 | j |= PAGE_COUNTER_MASK; |
aa3163f8 N |
1257 | continue; |
1258 | } | |
40cffcc0 | 1259 | counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; |
32a7627c | 1260 | } |
40cffcc0 | 1261 | bmc = bitmap_get_counter(counts, |
ef99bf48 | 1262 | block, |
db305e50 | 1263 | &blocks, 0); |
bf07bb7d N |
1264 | |
1265 | if (!bmc) { | |
5a537df4 | 1266 | j |= PAGE_COUNTER_MASK; |
bf07bb7d N |
1267 | continue; |
1268 | } | |
1269 | if (*bmc == 1 && !bitmap->need_sync) { | |
1270 | /* We can clear the bit */ | |
bf07bb7d | 1271 | *bmc = 0; |
40cffcc0 | 1272 | bitmap_count_page(counts, block, -1); |
ef99bf48 | 1273 | bitmap_file_clear_bit(bitmap, block); |
bf07bb7d N |
1274 | } else if (*bmc && *bmc <= 2) { |
1275 | *bmc = 1; | |
40cffcc0 | 1276 | bitmap_set_pending(counts, block); |
bf07bb7d | 1277 | bitmap->allclean = 0; |
5a537df4 | 1278 | } |
32a7627c | 1279 | } |
40cffcc0 | 1280 | spin_unlock_irq(&counts->lock); |
32a7627c | 1281 | |
bf07bb7d N |
1282 | /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. |
1283 | * DIRTY pages need to be written by bitmap_unplug so it can wait | |
1284 | * for them. | |
1285 | * If we find any DIRTY page we stop there and let bitmap_unplug | |
1286 | * handle all the rest. This is important in the case where | |
1287 | * the first blocking holds the superblock and it has been updated. | |
1288 | * We mustn't write any other blocks before the superblock. | |
1289 | */ | |
62f82faa N |
1290 | for (j = 0; |
1291 | j < bitmap->storage.file_pages | |
1292 | && !test_bit(BITMAP_STALE, &bitmap->flags); | |
1293 | j++) { | |
d189122d | 1294 | if (test_page_attr(bitmap, j, |
bf07bb7d N |
1295 | BITMAP_PAGE_DIRTY)) |
1296 | /* bitmap_unplug will handle the rest */ | |
1297 | break; | |
bdfd1140 N |
1298 | if (test_and_clear_page_attr(bitmap, j, |
1299 | BITMAP_PAGE_NEEDWRITE)) { | |
1ec885cd | 1300 | write_page(bitmap, bitmap->storage.filemap[j], 0); |
32a7627c | 1301 | } |
32a7627c N |
1302 | } |
1303 | ||
7be3dfec | 1304 | done: |
8311c29d | 1305 | if (bitmap->allclean == 0) |
2e61ebbc N |
1306 | mddev->thread->timeout = |
1307 | mddev->bitmap_info.daemon_sleep; | |
c3d9714e | 1308 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1309 | } |
1310 | ||
40cffcc0 | 1311 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1312 | sector_t offset, sector_t *blocks, |
32a7627c | 1313 | int create) |
ee305ace N |
1314 | __releases(bitmap->lock) |
1315 | __acquires(bitmap->lock) | |
32a7627c N |
1316 | { |
1317 | /* If 'create', we might release the lock and reclaim it. | |
1318 | * The lock must have been taken with interrupts enabled. | |
1319 | * If !create, we don't release the lock. | |
1320 | */ | |
61a0d80c | 1321 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1322 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1323 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1324 | sector_t csize; | |
ef425673 | 1325 | int err; |
32a7627c | 1326 | |
c9d65032 | 1327 | err = bitmap_checkpage(bitmap, page, create, 0); |
ef425673 N |
1328 | |
1329 | if (bitmap->bp[page].hijacked || | |
1330 | bitmap->bp[page].map == NULL) | |
61a0d80c | 1331 | csize = ((sector_t)1) << (bitmap->chunkshift + |
ef425673 N |
1332 | PAGE_COUNTER_SHIFT - 1); |
1333 | else | |
61a0d80c | 1334 | csize = ((sector_t)1) << bitmap->chunkshift; |
ef425673 N |
1335 | *blocks = csize - (offset & (csize - 1)); |
1336 | ||
1337 | if (err < 0) | |
32a7627c | 1338 | return NULL; |
ef425673 | 1339 | |
32a7627c N |
1340 | /* now locked ... */ |
1341 | ||
1342 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1343 | /* should we use the first or second counter field | |
1344 | * of the hijacked pointer? */ | |
1345 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1346 | return &((bitmap_counter_t *) |
1347 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1348 | } else /* page is allocated */ |
32a7627c N |
1349 | return (bitmap_counter_t *) |
1350 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1351 | } |
1352 | ||
4b6d287f | 1353 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1354 | { |
ac2f40be N |
1355 | if (!bitmap) |
1356 | return 0; | |
4b6d287f N |
1357 | |
1358 | if (behind) { | |
696fcd53 | 1359 | int bw; |
4b6d287f | 1360 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1361 | bw = atomic_read(&bitmap->behind_writes); |
1362 | if (bw > bitmap->behind_writes_used) | |
1363 | bitmap->behind_writes_used = bw; | |
1364 | ||
36a4e1fe N |
1365 | pr_debug("inc write-behind count %d/%lu\n", |
1366 | bw, bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1367 | } |
1368 | ||
32a7627c | 1369 | while (sectors) { |
57dab0bd | 1370 | sector_t blocks; |
32a7627c N |
1371 | bitmap_counter_t *bmc; |
1372 | ||
40cffcc0 N |
1373 | spin_lock_irq(&bitmap->counts.lock); |
1374 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1); | |
32a7627c | 1375 | if (!bmc) { |
40cffcc0 | 1376 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1377 | return 0; |
1378 | } | |
1379 | ||
27d5ea04 | 1380 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
da6e1a32 NB |
1381 | DEFINE_WAIT(__wait); |
1382 | /* note that it is safe to do the prepare_to_wait | |
1383 | * after the test as long as we do it before dropping | |
1384 | * the spinlock. | |
1385 | */ | |
1386 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1387 | TASK_UNINTERRUPTIBLE); | |
40cffcc0 | 1388 | spin_unlock_irq(&bitmap->counts.lock); |
f54a9d0e | 1389 | schedule(); |
da6e1a32 NB |
1390 | finish_wait(&bitmap->overflow_wait, &__wait); |
1391 | continue; | |
1392 | } | |
1393 | ||
ac2f40be | 1394 | switch (*bmc) { |
32a7627c N |
1395 | case 0: |
1396 | bitmap_file_set_bit(bitmap, offset); | |
40cffcc0 | 1397 | bitmap_count_page(&bitmap->counts, offset, 1); |
32a7627c N |
1398 | /* fall through */ |
1399 | case 1: | |
1400 | *bmc = 2; | |
1401 | } | |
da6e1a32 | 1402 | |
32a7627c N |
1403 | (*bmc)++; |
1404 | ||
40cffcc0 | 1405 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1406 | |
1407 | offset += blocks; | |
1408 | if (sectors > blocks) | |
1409 | sectors -= blocks; | |
ac2f40be N |
1410 | else |
1411 | sectors = 0; | |
32a7627c N |
1412 | } |
1413 | return 0; | |
1414 | } | |
ac2f40be | 1415 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1416 | |
1417 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1418 | int success, int behind) |
32a7627c | 1419 | { |
ac2f40be N |
1420 | if (!bitmap) |
1421 | return; | |
4b6d287f | 1422 | if (behind) { |
e555190d N |
1423 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1424 | wake_up(&bitmap->behind_wait); | |
36a4e1fe N |
1425 | pr_debug("dec write-behind count %d/%lu\n", |
1426 | atomic_read(&bitmap->behind_writes), | |
1427 | bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1428 | } |
1429 | ||
32a7627c | 1430 | while (sectors) { |
57dab0bd | 1431 | sector_t blocks; |
32a7627c N |
1432 | unsigned long flags; |
1433 | bitmap_counter_t *bmc; | |
1434 | ||
40cffcc0 N |
1435 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1436 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0); | |
32a7627c | 1437 | if (!bmc) { |
40cffcc0 | 1438 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1439 | return; |
1440 | } | |
1441 | ||
961902c0 | 1442 | if (success && !bitmap->mddev->degraded && |
a0da84f3 NB |
1443 | bitmap->events_cleared < bitmap->mddev->events) { |
1444 | bitmap->events_cleared = bitmap->mddev->events; | |
1445 | bitmap->need_sync = 1; | |
5ff5afff | 1446 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1447 | } |
1448 | ||
27d5ea04 | 1449 | if (!success && !NEEDED(*bmc)) |
32a7627c N |
1450 | *bmc |= NEEDED_MASK; |
1451 | ||
27d5ea04 | 1452 | if (COUNTER(*bmc) == COUNTER_MAX) |
da6e1a32 NB |
1453 | wake_up(&bitmap->overflow_wait); |
1454 | ||
32a7627c | 1455 | (*bmc)--; |
2585f3ef | 1456 | if (*bmc <= 2) { |
40cffcc0 | 1457 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1458 | bitmap->allclean = 0; |
1459 | } | |
40cffcc0 | 1460 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1461 | offset += blocks; |
1462 | if (sectors > blocks) | |
1463 | sectors -= blocks; | |
ac2f40be N |
1464 | else |
1465 | sectors = 0; | |
32a7627c N |
1466 | } |
1467 | } | |
ac2f40be | 1468 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1469 | |
57dab0bd | 1470 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1471 | int degraded) |
32a7627c N |
1472 | { |
1473 | bitmap_counter_t *bmc; | |
1474 | int rv; | |
1475 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1476 | *blocks = 1024; | |
1477 | return 1; /* always resync if no bitmap */ | |
1478 | } | |
40cffcc0 N |
1479 | spin_lock_irq(&bitmap->counts.lock); |
1480 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1481 | rv = 0; |
1482 | if (bmc) { | |
1483 | /* locked */ | |
1484 | if (RESYNC(*bmc)) | |
1485 | rv = 1; | |
1486 | else if (NEEDED(*bmc)) { | |
1487 | rv = 1; | |
6a806c51 N |
1488 | if (!degraded) { /* don't set/clear bits if degraded */ |
1489 | *bmc |= RESYNC_MASK; | |
1490 | *bmc &= ~NEEDED_MASK; | |
1491 | } | |
32a7627c N |
1492 | } |
1493 | } | |
40cffcc0 | 1494 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1495 | return rv; |
1496 | } | |
1497 | ||
57dab0bd | 1498 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a N |
1499 | int degraded) |
1500 | { | |
1501 | /* bitmap_start_sync must always report on multiples of whole | |
1502 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1503 | * get confused. | |
1504 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1505 | * At least PAGE_SIZE>>9 blocks are covered. | |
1506 | * Return the 'or' of the result. | |
1507 | */ | |
1508 | int rv = 0; | |
57dab0bd | 1509 | sector_t blocks1; |
1187cf0a N |
1510 | |
1511 | *blocks = 0; | |
1512 | while (*blocks < (PAGE_SIZE>>9)) { | |
1513 | rv |= __bitmap_start_sync(bitmap, offset, | |
1514 | &blocks1, degraded); | |
1515 | offset += blocks1; | |
1516 | *blocks += blocks1; | |
1517 | } | |
1518 | return rv; | |
1519 | } | |
ac2f40be | 1520 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1521 | |
57dab0bd | 1522 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1523 | { |
1524 | bitmap_counter_t *bmc; | |
1525 | unsigned long flags; | |
ac2f40be N |
1526 | |
1527 | if (bitmap == NULL) { | |
32a7627c N |
1528 | *blocks = 1024; |
1529 | return; | |
1530 | } | |
40cffcc0 N |
1531 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1532 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1533 | if (bmc == NULL) |
1534 | goto unlock; | |
1535 | /* locked */ | |
32a7627c N |
1536 | if (RESYNC(*bmc)) { |
1537 | *bmc &= ~RESYNC_MASK; | |
1538 | ||
1539 | if (!NEEDED(*bmc) && aborted) | |
1540 | *bmc |= NEEDED_MASK; | |
1541 | else { | |
2585f3ef | 1542 | if (*bmc <= 2) { |
40cffcc0 | 1543 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1544 | bitmap->allclean = 0; |
1545 | } | |
32a7627c N |
1546 | } |
1547 | } | |
1548 | unlock: | |
40cffcc0 | 1549 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c | 1550 | } |
ac2f40be | 1551 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1552 | |
1553 | void bitmap_close_sync(struct bitmap *bitmap) | |
1554 | { | |
1555 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1556 | * properly have been aborted. It remains to us to clear the | |
1557 | * RESYNC bit wherever it is still on | |
1558 | */ | |
1559 | sector_t sector = 0; | |
57dab0bd | 1560 | sector_t blocks; |
b47490c9 N |
1561 | if (!bitmap) |
1562 | return; | |
32a7627c N |
1563 | while (sector < bitmap->mddev->resync_max_sectors) { |
1564 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1565 | sector += blocks; |
1566 | } | |
1567 | } | |
ac2f40be | 1568 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 | 1569 | |
c40f341f | 1570 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force) |
b47490c9 N |
1571 | { |
1572 | sector_t s = 0; | |
57dab0bd | 1573 | sector_t blocks; |
b47490c9 N |
1574 | |
1575 | if (!bitmap) | |
1576 | return; | |
1577 | if (sector == 0) { | |
1578 | bitmap->last_end_sync = jiffies; | |
1579 | return; | |
1580 | } | |
c40f341f | 1581 | if (!force && time_before(jiffies, (bitmap->last_end_sync |
1b04be96 | 1582 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1583 | return; |
1584 | wait_event(bitmap->mddev->recovery_wait, | |
1585 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1586 | ||
75d3da43 | 1587 | bitmap->mddev->curr_resync_completed = sector; |
070dc6dd | 1588 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
40cffcc0 | 1589 | sector &= ~((1ULL << bitmap->counts.chunkshift) - 1); |
b47490c9 N |
1590 | s = 0; |
1591 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1592 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1593 | s += blocks; | |
32a7627c | 1594 | } |
b47490c9 | 1595 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1596 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1597 | } |
ac2f40be | 1598 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1599 | |
18c9ff7f GJ |
1600 | void bitmap_sync_with_cluster(struct mddev *mddev, |
1601 | sector_t old_lo, sector_t old_hi, | |
1602 | sector_t new_lo, sector_t new_hi) | |
1603 | { | |
1604 | struct bitmap *bitmap = mddev->bitmap; | |
1605 | sector_t sector, blocks = 0; | |
1606 | ||
1607 | for (sector = old_lo; sector < new_lo; ) { | |
1608 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
1609 | sector += blocks; | |
1610 | } | |
1611 | WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n"); | |
1612 | ||
1613 | for (sector = old_hi; sector < new_hi; ) { | |
1614 | bitmap_start_sync(bitmap, sector, &blocks, 0); | |
1615 | sector += blocks; | |
1616 | } | |
1617 | WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n"); | |
1618 | } | |
1619 | EXPORT_SYMBOL(bitmap_sync_with_cluster); | |
1620 | ||
6a07997f | 1621 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1622 | { |
1623 | /* For each chunk covered by any of these sectors, set the | |
ef99bf48 | 1624 | * counter to 2 and possibly set resync_needed. They should all |
32a7627c N |
1625 | * be 0 at this point |
1626 | */ | |
193f1c93 | 1627 | |
57dab0bd | 1628 | sector_t secs; |
193f1c93 | 1629 | bitmap_counter_t *bmc; |
40cffcc0 N |
1630 | spin_lock_irq(&bitmap->counts.lock); |
1631 | bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1); | |
193f1c93 | 1632 | if (!bmc) { |
40cffcc0 | 1633 | spin_unlock_irq(&bitmap->counts.lock); |
193f1c93 | 1634 | return; |
32a7627c | 1635 | } |
ac2f40be | 1636 | if (!*bmc) { |
11dd35da | 1637 | *bmc = 2; |
40cffcc0 N |
1638 | bitmap_count_page(&bitmap->counts, offset, 1); |
1639 | bitmap_set_pending(&bitmap->counts, offset); | |
2585f3ef | 1640 | bitmap->allclean = 0; |
193f1c93 | 1641 | } |
11dd35da GR |
1642 | if (needed) |
1643 | *bmc |= NEEDED_MASK; | |
40cffcc0 | 1644 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1645 | } |
1646 | ||
9b1d1dac PC |
1647 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1648 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1649 | { | |
1650 | unsigned long chunk; | |
1651 | ||
1652 | for (chunk = s; chunk <= e; chunk++) { | |
40cffcc0 | 1653 | sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift; |
9b1d1dac PC |
1654 | bitmap_set_memory_bits(bitmap, sec, 1); |
1655 | bitmap_file_set_bit(bitmap, sec); | |
ffa23322 N |
1656 | if (sec < bitmap->mddev->recovery_cp) |
1657 | /* We are asserting that the array is dirty, | |
1658 | * so move the recovery_cp address back so | |
1659 | * that it is obvious that it is dirty | |
1660 | */ | |
1661 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1662 | } |
1663 | } | |
1664 | ||
6b8b3e8a N |
1665 | /* |
1666 | * flush out any pending updates | |
1667 | */ | |
fd01b88c | 1668 | void bitmap_flush(struct mddev *mddev) |
6b8b3e8a N |
1669 | { |
1670 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1671 | long sleep; |
6b8b3e8a N |
1672 | |
1673 | if (!bitmap) /* there was no bitmap */ | |
1674 | return; | |
1675 | ||
1676 | /* run the daemon_work three time to ensure everything is flushed | |
1677 | * that can be | |
1678 | */ | |
1b04be96 | 1679 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1680 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1681 | bitmap_daemon_work(mddev); |
42a04b50 | 1682 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1683 | bitmap_daemon_work(mddev); |
42a04b50 | 1684 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1685 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1686 | bitmap_update_sb(bitmap); |
1687 | } | |
1688 | ||
32a7627c N |
1689 | /* |
1690 | * free memory that was allocated | |
1691 | */ | |
3178b0db | 1692 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1693 | { |
1694 | unsigned long k, pages; | |
1695 | struct bitmap_page *bp; | |
32a7627c N |
1696 | |
1697 | if (!bitmap) /* there was no bitmap */ | |
1698 | return; | |
1699 | ||
f9a67b11 GJ |
1700 | if (bitmap->sysfs_can_clear) |
1701 | sysfs_put(bitmap->sysfs_can_clear); | |
1702 | ||
f9209a32 GR |
1703 | if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info && |
1704 | bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev)) | |
b97e9257 GR |
1705 | md_cluster_stop(bitmap->mddev); |
1706 | ||
fae7d326 N |
1707 | /* Shouldn't be needed - but just in case.... */ |
1708 | wait_event(bitmap->write_wait, | |
1709 | atomic_read(&bitmap->pending_writes) == 0); | |
1710 | ||
1711 | /* release the bitmap file */ | |
1712 | bitmap_file_unmap(&bitmap->storage); | |
32a7627c | 1713 | |
40cffcc0 N |
1714 | bp = bitmap->counts.bp; |
1715 | pages = bitmap->counts.pages; | |
32a7627c N |
1716 | |
1717 | /* free all allocated memory */ | |
1718 | ||
32a7627c N |
1719 | if (bp) /* deallocate the page memory */ |
1720 | for (k = 0; k < pages; k++) | |
1721 | if (bp[k].map && !bp[k].hijacked) | |
1722 | kfree(bp[k].map); | |
1723 | kfree(bp); | |
1724 | kfree(bitmap); | |
1725 | } | |
aa5cbd10 | 1726 | |
fd01b88c | 1727 | void bitmap_destroy(struct mddev *mddev) |
3178b0db N |
1728 | { |
1729 | struct bitmap *bitmap = mddev->bitmap; | |
1730 | ||
1731 | if (!bitmap) /* there was no bitmap */ | |
1732 | return; | |
1733 | ||
c3d9714e | 1734 | mutex_lock(&mddev->bitmap_info.mutex); |
978a7a47 | 1735 | spin_lock(&mddev->lock); |
3178b0db | 1736 | mddev->bitmap = NULL; /* disconnect from the md device */ |
978a7a47 | 1737 | spin_unlock(&mddev->lock); |
c3d9714e | 1738 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1739 | if (mddev->thread) |
1740 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db N |
1741 | |
1742 | bitmap_free(bitmap); | |
1743 | } | |
32a7627c N |
1744 | |
1745 | /* | |
1746 | * initialize the bitmap structure | |
1747 | * if this returns an error, bitmap_destroy must be called to do clean up | |
f9a67b11 | 1748 | * once mddev->bitmap is set |
32a7627c | 1749 | */ |
f9209a32 | 1750 | struct bitmap *bitmap_create(struct mddev *mddev, int slot) |
32a7627c N |
1751 | { |
1752 | struct bitmap *bitmap; | |
1f593903 | 1753 | sector_t blocks = mddev->resync_max_sectors; |
c3d9714e | 1754 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1755 | int err; |
324a56e1 | 1756 | struct kernfs_node *bm = NULL; |
32a7627c | 1757 | |
5f6e3c83 | 1758 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1759 | |
c3d9714e | 1760 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1761 | |
9ffae0cf | 1762 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c | 1763 | if (!bitmap) |
f9209a32 | 1764 | return ERR_PTR(-ENOMEM); |
32a7627c | 1765 | |
40cffcc0 | 1766 | spin_lock_init(&bitmap->counts.lock); |
ce25c31b N |
1767 | atomic_set(&bitmap->pending_writes, 0); |
1768 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1769 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1770 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1771 | |
32a7627c | 1772 | bitmap->mddev = mddev; |
f9209a32 | 1773 | bitmap->cluster_slot = slot; |
32a7627c | 1774 | |
5ff5afff | 1775 | if (mddev->kobj.sd) |
388975cc | 1776 | bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap"); |
ece5cff0 | 1777 | if (bm) { |
388975cc | 1778 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear"); |
ece5cff0 N |
1779 | sysfs_put(bm); |
1780 | } else | |
1781 | bitmap->sysfs_can_clear = NULL; | |
1782 | ||
1ec885cd | 1783 | bitmap->storage.file = file; |
ce25c31b N |
1784 | if (file) { |
1785 | get_file(file); | |
ae8fa283 N |
1786 | /* As future accesses to this file will use bmap, |
1787 | * and bypass the page cache, we must sync the file | |
1788 | * first. | |
1789 | */ | |
8018ab05 | 1790 | vfs_fsync(file, 1); |
ce25c31b | 1791 | } |
42a04b50 | 1792 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
9c81075f JB |
1793 | if (!mddev->bitmap_info.external) { |
1794 | /* | |
1795 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is | |
1796 | * instructing us to create a new on-disk bitmap instance. | |
1797 | */ | |
1798 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) | |
1799 | err = bitmap_new_disk_sb(bitmap); | |
1800 | else | |
1801 | err = bitmap_read_sb(bitmap); | |
1802 | } else { | |
ece5cff0 N |
1803 | err = 0; |
1804 | if (mddev->bitmap_info.chunksize == 0 || | |
1805 | mddev->bitmap_info.daemon_sleep == 0) | |
1806 | /* chunksize and time_base need to be | |
1807 | * set first. */ | |
1808 | err = -EINVAL; | |
1809 | } | |
32a7627c | 1810 | if (err) |
3178b0db | 1811 | goto error; |
32a7627c | 1812 | |
624ce4f5 | 1813 | bitmap->daemon_lastrun = jiffies; |
d60b479d N |
1814 | err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1); |
1815 | if (err) | |
3178b0db | 1816 | goto error; |
32a7627c | 1817 | |
69e51b44 | 1818 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", |
d60b479d | 1819 | bitmap->counts.pages, bmname(bitmap)); |
69e51b44 | 1820 | |
f9209a32 GR |
1821 | err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0; |
1822 | if (err) | |
1823 | goto error; | |
69e51b44 | 1824 | |
f9209a32 | 1825 | return bitmap; |
69e51b44 N |
1826 | error: |
1827 | bitmap_free(bitmap); | |
f9209a32 | 1828 | return ERR_PTR(err); |
69e51b44 N |
1829 | } |
1830 | ||
fd01b88c | 1831 | int bitmap_load(struct mddev *mddev) |
69e51b44 N |
1832 | { |
1833 | int err = 0; | |
3520fa4d | 1834 | sector_t start = 0; |
69e51b44 N |
1835 | sector_t sector = 0; |
1836 | struct bitmap *bitmap = mddev->bitmap; | |
1837 | ||
1838 | if (!bitmap) | |
1839 | goto out; | |
1840 | ||
1841 | /* Clear out old bitmap info first: Either there is none, or we | |
1842 | * are resuming after someone else has possibly changed things, | |
1843 | * so we should forget old cached info. | |
1844 | * All chunks should be clean, but some might need_sync. | |
1845 | */ | |
1846 | while (sector < mddev->resync_max_sectors) { | |
57dab0bd | 1847 | sector_t blocks; |
69e51b44 N |
1848 | bitmap_start_sync(bitmap, sector, &blocks, 0); |
1849 | sector += blocks; | |
1850 | } | |
1851 | bitmap_close_sync(bitmap); | |
1852 | ||
3520fa4d JB |
1853 | if (mddev->degraded == 0 |
1854 | || bitmap->events_cleared == mddev->events) | |
1855 | /* no need to keep dirty bits to optimise a | |
1856 | * re-add of a missing device */ | |
1857 | start = mddev->recovery_cp; | |
1858 | ||
afbaa90b | 1859 | mutex_lock(&mddev->bitmap_info.mutex); |
3520fa4d | 1860 | err = bitmap_init_from_disk(bitmap, start); |
afbaa90b | 1861 | mutex_unlock(&mddev->bitmap_info.mutex); |
3520fa4d | 1862 | |
32a7627c | 1863 | if (err) |
69e51b44 | 1864 | goto out; |
b405fe91 | 1865 | clear_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
1866 | |
1867 | /* Kick recovery in case any bits were set */ | |
1868 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
3178b0db | 1869 | |
1b04be96 | 1870 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1871 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1872 | |
4ad13663 N |
1873 | bitmap_update_sb(bitmap); |
1874 | ||
b405fe91 | 1875 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
69e51b44 N |
1876 | err = -EIO; |
1877 | out: | |
3178b0db | 1878 | return err; |
32a7627c | 1879 | } |
69e51b44 | 1880 | EXPORT_SYMBOL_GPL(bitmap_load); |
32a7627c | 1881 | |
11dd35da GR |
1882 | /* Loads the bitmap associated with slot and copies the resync information |
1883 | * to our bitmap | |
1884 | */ | |
1885 | int bitmap_copy_from_slot(struct mddev *mddev, int slot, | |
97f6cd39 | 1886 | sector_t *low, sector_t *high, bool clear_bits) |
11dd35da GR |
1887 | { |
1888 | int rv = 0, i, j; | |
1889 | sector_t block, lo = 0, hi = 0; | |
1890 | struct bitmap_counts *counts; | |
1891 | struct bitmap *bitmap = bitmap_create(mddev, slot); | |
1892 | ||
f9a67b11 GJ |
1893 | if (IS_ERR(bitmap)) { |
1894 | bitmap_free(bitmap); | |
11dd35da | 1895 | return PTR_ERR(bitmap); |
f9a67b11 | 1896 | } |
11dd35da | 1897 | |
11dd35da GR |
1898 | rv = bitmap_init_from_disk(bitmap, 0); |
1899 | if (rv) | |
1900 | goto err; | |
1901 | ||
1902 | counts = &bitmap->counts; | |
1903 | for (j = 0; j < counts->chunks; j++) { | |
1904 | block = (sector_t)j << counts->chunkshift; | |
1905 | if (bitmap_file_test_bit(bitmap, block)) { | |
1906 | if (!lo) | |
1907 | lo = block; | |
1908 | hi = block; | |
1909 | bitmap_file_clear_bit(bitmap, block); | |
1910 | bitmap_set_memory_bits(mddev->bitmap, block, 1); | |
1911 | bitmap_file_set_bit(mddev->bitmap, block); | |
1912 | } | |
1913 | } | |
1914 | ||
97f6cd39 GR |
1915 | if (clear_bits) { |
1916 | bitmap_update_sb(bitmap); | |
1917 | /* Setting this for the ev_page should be enough. | |
1918 | * And we do not require both write_all and PAGE_DIRT either | |
1919 | */ | |
1920 | for (i = 0; i < bitmap->storage.file_pages; i++) | |
1921 | set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); | |
1922 | bitmap_write_all(bitmap); | |
1923 | bitmap_unplug(bitmap); | |
1924 | } | |
11dd35da GR |
1925 | *low = lo; |
1926 | *high = hi; | |
1927 | err: | |
1928 | bitmap_free(bitmap); | |
1929 | return rv; | |
1930 | } | |
1931 | EXPORT_SYMBOL_GPL(bitmap_copy_from_slot); | |
1932 | ||
1933 | ||
57148964 N |
1934 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap) |
1935 | { | |
1936 | unsigned long chunk_kb; | |
40cffcc0 | 1937 | struct bitmap_counts *counts; |
57148964 N |
1938 | |
1939 | if (!bitmap) | |
1940 | return; | |
1941 | ||
40cffcc0 N |
1942 | counts = &bitmap->counts; |
1943 | ||
57148964 N |
1944 | chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; |
1945 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
1946 | "%lu%s chunk", | |
40cffcc0 N |
1947 | counts->pages - counts->missing_pages, |
1948 | counts->pages, | |
1949 | (counts->pages - counts->missing_pages) | |
57148964 N |
1950 | << (PAGE_SHIFT - 10), |
1951 | chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, | |
1952 | chunk_kb ? "KB" : "B"); | |
1ec885cd | 1953 | if (bitmap->storage.file) { |
57148964 | 1954 | seq_printf(seq, ", file: "); |
2726d566 | 1955 | seq_file_path(seq, bitmap->storage.file, " \t\n"); |
57148964 N |
1956 | } |
1957 | ||
1958 | seq_printf(seq, "\n"); | |
57148964 N |
1959 | } |
1960 | ||
d60b479d N |
1961 | int bitmap_resize(struct bitmap *bitmap, sector_t blocks, |
1962 | int chunksize, int init) | |
1963 | { | |
1964 | /* If chunk_size is 0, choose an appropriate chunk size. | |
1965 | * Then possibly allocate new storage space. | |
1966 | * Then quiesce, copy bits, replace bitmap, and re-start | |
1967 | * | |
1968 | * This function is called both to set up the initial bitmap | |
1969 | * and to resize the bitmap while the array is active. | |
1970 | * If this happens as a result of the array being resized, | |
1971 | * chunksize will be zero, and we need to choose a suitable | |
1972 | * chunksize, otherwise we use what we are given. | |
1973 | */ | |
1974 | struct bitmap_storage store; | |
1975 | struct bitmap_counts old_counts; | |
1976 | unsigned long chunks; | |
1977 | sector_t block; | |
1978 | sector_t old_blocks, new_blocks; | |
1979 | int chunkshift; | |
1980 | int ret = 0; | |
1981 | long pages; | |
1982 | struct bitmap_page *new_bp; | |
1983 | ||
1984 | if (chunksize == 0) { | |
1985 | /* If there is enough space, leave the chunk size unchanged, | |
1986 | * else increase by factor of two until there is enough space. | |
1987 | */ | |
1988 | long bytes; | |
1989 | long space = bitmap->mddev->bitmap_info.space; | |
1990 | ||
1991 | if (space == 0) { | |
1992 | /* We don't know how much space there is, so limit | |
1993 | * to current size - in sectors. | |
1994 | */ | |
1995 | bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8); | |
1996 | if (!bitmap->mddev->bitmap_info.external) | |
1997 | bytes += sizeof(bitmap_super_t); | |
1998 | space = DIV_ROUND_UP(bytes, 512); | |
1999 | bitmap->mddev->bitmap_info.space = space; | |
2000 | } | |
2001 | chunkshift = bitmap->counts.chunkshift; | |
2002 | chunkshift--; | |
2003 | do { | |
2004 | /* 'chunkshift' is shift from block size to chunk size */ | |
2005 | chunkshift++; | |
2006 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
2007 | bytes = DIV_ROUND_UP(chunks, 8); | |
2008 | if (!bitmap->mddev->bitmap_info.external) | |
2009 | bytes += sizeof(bitmap_super_t); | |
2010 | } while (bytes > (space << 9)); | |
2011 | } else | |
2012 | chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT; | |
2013 | ||
2014 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
2015 | memset(&store, 0, sizeof(store)); | |
2016 | if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file) | |
2017 | ret = bitmap_storage_alloc(&store, chunks, | |
b97e9257 | 2018 | !bitmap->mddev->bitmap_info.external, |
da6fb7a9 N |
2019 | mddev_is_clustered(bitmap->mddev) |
2020 | ? bitmap->cluster_slot : 0); | |
d60b479d N |
2021 | if (ret) |
2022 | goto err; | |
2023 | ||
2024 | pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO); | |
2025 | ||
2026 | new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL); | |
2027 | ret = -ENOMEM; | |
2028 | if (!new_bp) { | |
2029 | bitmap_file_unmap(&store); | |
2030 | goto err; | |
2031 | } | |
2032 | ||
2033 | if (!init) | |
2034 | bitmap->mddev->pers->quiesce(bitmap->mddev, 1); | |
2035 | ||
2036 | store.file = bitmap->storage.file; | |
2037 | bitmap->storage.file = NULL; | |
2038 | ||
2039 | if (store.sb_page && bitmap->storage.sb_page) | |
2040 | memcpy(page_address(store.sb_page), | |
2041 | page_address(bitmap->storage.sb_page), | |
2042 | sizeof(bitmap_super_t)); | |
2043 | bitmap_file_unmap(&bitmap->storage); | |
2044 | bitmap->storage = store; | |
2045 | ||
2046 | old_counts = bitmap->counts; | |
2047 | bitmap->counts.bp = new_bp; | |
2048 | bitmap->counts.pages = pages; | |
2049 | bitmap->counts.missing_pages = pages; | |
2050 | bitmap->counts.chunkshift = chunkshift; | |
2051 | bitmap->counts.chunks = chunks; | |
2052 | bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift + | |
2053 | BITMAP_BLOCK_SHIFT); | |
2054 | ||
2055 | blocks = min(old_counts.chunks << old_counts.chunkshift, | |
2056 | chunks << chunkshift); | |
2057 | ||
2058 | spin_lock_irq(&bitmap->counts.lock); | |
c9d65032 GJ |
2059 | /* For cluster raid, need to pre-allocate bitmap */ |
2060 | if (mddev_is_clustered(bitmap->mddev)) { | |
2061 | unsigned long page; | |
2062 | for (page = 0; page < pages; page++) { | |
2063 | ret = bitmap_checkpage(&bitmap->counts, page, 1, 1); | |
2064 | if (ret) { | |
2065 | unsigned long k; | |
2066 | ||
2067 | /* deallocate the page memory */ | |
2068 | for (k = 0; k < page; k++) { | |
2069 | if (new_bp[k].map) | |
2070 | kfree(new_bp[k].map); | |
2071 | } | |
2072 | ||
2073 | /* restore some fields from old_counts */ | |
2074 | bitmap->counts.bp = old_counts.bp; | |
2075 | bitmap->counts.pages = old_counts.pages; | |
2076 | bitmap->counts.missing_pages = old_counts.pages; | |
2077 | bitmap->counts.chunkshift = old_counts.chunkshift; | |
2078 | bitmap->counts.chunks = old_counts.chunks; | |
2079 | bitmap->mddev->bitmap_info.chunksize = 1 << (old_counts.chunkshift + | |
2080 | BITMAP_BLOCK_SHIFT); | |
2081 | blocks = old_counts.chunks << old_counts.chunkshift; | |
2082 | pr_err("Could not pre-allocate in-memory bitmap for cluster raid\n"); | |
2083 | break; | |
2084 | } else | |
2085 | bitmap->counts.bp[page].count += 1; | |
2086 | } | |
2087 | } | |
2088 | ||
d60b479d N |
2089 | for (block = 0; block < blocks; ) { |
2090 | bitmap_counter_t *bmc_old, *bmc_new; | |
2091 | int set; | |
2092 | ||
2093 | bmc_old = bitmap_get_counter(&old_counts, block, | |
2094 | &old_blocks, 0); | |
2095 | set = bmc_old && NEEDED(*bmc_old); | |
2096 | ||
2097 | if (set) { | |
2098 | bmc_new = bitmap_get_counter(&bitmap->counts, block, | |
2099 | &new_blocks, 1); | |
2100 | if (*bmc_new == 0) { | |
2101 | /* need to set on-disk bits too. */ | |
2102 | sector_t end = block + new_blocks; | |
2103 | sector_t start = block >> chunkshift; | |
2104 | start <<= chunkshift; | |
2105 | while (start < end) { | |
2106 | bitmap_file_set_bit(bitmap, block); | |
2107 | start += 1 << chunkshift; | |
2108 | } | |
2109 | *bmc_new = 2; | |
2110 | bitmap_count_page(&bitmap->counts, | |
2111 | block, 1); | |
2112 | bitmap_set_pending(&bitmap->counts, | |
2113 | block); | |
2114 | } | |
2115 | *bmc_new |= NEEDED_MASK; | |
2116 | if (new_blocks < old_blocks) | |
2117 | old_blocks = new_blocks; | |
2118 | } | |
2119 | block += old_blocks; | |
2120 | } | |
2121 | ||
2122 | if (!init) { | |
2123 | int i; | |
2124 | while (block < (chunks << chunkshift)) { | |
2125 | bitmap_counter_t *bmc; | |
2126 | bmc = bitmap_get_counter(&bitmap->counts, block, | |
2127 | &new_blocks, 1); | |
2128 | if (bmc) { | |
2129 | /* new space. It needs to be resynced, so | |
2130 | * we set NEEDED_MASK. | |
2131 | */ | |
2132 | if (*bmc == 0) { | |
2133 | *bmc = NEEDED_MASK | 2; | |
2134 | bitmap_count_page(&bitmap->counts, | |
2135 | block, 1); | |
2136 | bitmap_set_pending(&bitmap->counts, | |
2137 | block); | |
2138 | } | |
2139 | } | |
2140 | block += new_blocks; | |
2141 | } | |
2142 | for (i = 0; i < bitmap->storage.file_pages; i++) | |
2143 | set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); | |
2144 | } | |
2145 | spin_unlock_irq(&bitmap->counts.lock); | |
2146 | ||
2147 | if (!init) { | |
2148 | bitmap_unplug(bitmap); | |
2149 | bitmap->mddev->pers->quiesce(bitmap->mddev, 0); | |
2150 | } | |
2151 | ret = 0; | |
2152 | err: | |
2153 | return ret; | |
2154 | } | |
2155 | EXPORT_SYMBOL_GPL(bitmap_resize); | |
2156 | ||
43a70507 | 2157 | static ssize_t |
fd01b88c | 2158 | location_show(struct mddev *mddev, char *page) |
43a70507 N |
2159 | { |
2160 | ssize_t len; | |
ac2f40be | 2161 | if (mddev->bitmap_info.file) |
43a70507 | 2162 | len = sprintf(page, "file"); |
ac2f40be | 2163 | else if (mddev->bitmap_info.offset) |
43a70507 | 2164 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 2165 | else |
43a70507 N |
2166 | len = sprintf(page, "none"); |
2167 | len += sprintf(page+len, "\n"); | |
2168 | return len; | |
2169 | } | |
2170 | ||
2171 | static ssize_t | |
fd01b88c | 2172 | location_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2173 | { |
2174 | ||
2175 | if (mddev->pers) { | |
2176 | if (!mddev->pers->quiesce) | |
2177 | return -EBUSY; | |
2178 | if (mddev->recovery || mddev->sync_thread) | |
2179 | return -EBUSY; | |
2180 | } | |
2181 | ||
2182 | if (mddev->bitmap || mddev->bitmap_info.file || | |
2183 | mddev->bitmap_info.offset) { | |
2184 | /* bitmap already configured. Only option is to clear it */ | |
2185 | if (strncmp(buf, "none", 4) != 0) | |
2186 | return -EBUSY; | |
2187 | if (mddev->pers) { | |
2188 | mddev->pers->quiesce(mddev, 1); | |
2189 | bitmap_destroy(mddev); | |
2190 | mddev->pers->quiesce(mddev, 0); | |
2191 | } | |
2192 | mddev->bitmap_info.offset = 0; | |
2193 | if (mddev->bitmap_info.file) { | |
2194 | struct file *f = mddev->bitmap_info.file; | |
2195 | mddev->bitmap_info.file = NULL; | |
43a70507 N |
2196 | fput(f); |
2197 | } | |
2198 | } else { | |
2199 | /* No bitmap, OK to set a location */ | |
2200 | long long offset; | |
2201 | if (strncmp(buf, "none", 4) == 0) | |
2202 | /* nothing to be done */; | |
2203 | else if (strncmp(buf, "file:", 5) == 0) { | |
2204 | /* Not supported yet */ | |
2205 | return -EINVAL; | |
2206 | } else { | |
2207 | int rv; | |
2208 | if (buf[0] == '+') | |
b29bebd6 | 2209 | rv = kstrtoll(buf+1, 10, &offset); |
43a70507 | 2210 | else |
b29bebd6 | 2211 | rv = kstrtoll(buf, 10, &offset); |
43a70507 N |
2212 | if (rv) |
2213 | return rv; | |
2214 | if (offset == 0) | |
2215 | return -EINVAL; | |
ece5cff0 N |
2216 | if (mddev->bitmap_info.external == 0 && |
2217 | mddev->major_version == 0 && | |
43a70507 N |
2218 | offset != mddev->bitmap_info.default_offset) |
2219 | return -EINVAL; | |
2220 | mddev->bitmap_info.offset = offset; | |
2221 | if (mddev->pers) { | |
f9209a32 | 2222 | struct bitmap *bitmap; |
43a70507 | 2223 | mddev->pers->quiesce(mddev, 1); |
f9209a32 GR |
2224 | bitmap = bitmap_create(mddev, -1); |
2225 | if (IS_ERR(bitmap)) | |
2226 | rv = PTR_ERR(bitmap); | |
2227 | else { | |
2228 | mddev->bitmap = bitmap; | |
4474ca42 | 2229 | rv = bitmap_load(mddev); |
f9a67b11 | 2230 | if (rv) |
f9209a32 | 2231 | mddev->bitmap_info.offset = 0; |
43a70507 N |
2232 | } |
2233 | mddev->pers->quiesce(mddev, 0); | |
f9a67b11 GJ |
2234 | if (rv) { |
2235 | bitmap_destroy(mddev); | |
43a70507 | 2236 | return rv; |
f9a67b11 | 2237 | } |
43a70507 N |
2238 | } |
2239 | } | |
2240 | } | |
2241 | if (!mddev->external) { | |
2242 | /* Ensure new bitmap info is stored in | |
2243 | * metadata promptly. | |
2244 | */ | |
2245 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2246 | md_wakeup_thread(mddev->thread); | |
2247 | } | |
2248 | return len; | |
2249 | } | |
2250 | ||
2251 | static struct md_sysfs_entry bitmap_location = | |
2252 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
2253 | ||
6409bb05 N |
2254 | /* 'bitmap/space' is the space available at 'location' for the |
2255 | * bitmap. This allows the kernel to know when it is safe to | |
2256 | * resize the bitmap to match a resized array. | |
2257 | */ | |
2258 | static ssize_t | |
2259 | space_show(struct mddev *mddev, char *page) | |
2260 | { | |
2261 | return sprintf(page, "%lu\n", mddev->bitmap_info.space); | |
2262 | } | |
2263 | ||
2264 | static ssize_t | |
2265 | space_store(struct mddev *mddev, const char *buf, size_t len) | |
2266 | { | |
2267 | unsigned long sectors; | |
2268 | int rv; | |
2269 | ||
2270 | rv = kstrtoul(buf, 10, §ors); | |
2271 | if (rv) | |
2272 | return rv; | |
2273 | ||
2274 | if (sectors == 0) | |
2275 | return -EINVAL; | |
2276 | ||
2277 | if (mddev->bitmap && | |
9b1215c1 | 2278 | sectors < (mddev->bitmap->storage.bytes + 511) >> 9) |
6409bb05 N |
2279 | return -EFBIG; /* Bitmap is too big for this small space */ |
2280 | ||
2281 | /* could make sure it isn't too big, but that isn't really | |
2282 | * needed - user-space should be careful. | |
2283 | */ | |
2284 | mddev->bitmap_info.space = sectors; | |
2285 | return len; | |
2286 | } | |
2287 | ||
2288 | static struct md_sysfs_entry bitmap_space = | |
2289 | __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); | |
2290 | ||
43a70507 | 2291 | static ssize_t |
fd01b88c | 2292 | timeout_show(struct mddev *mddev, char *page) |
43a70507 N |
2293 | { |
2294 | ssize_t len; | |
2295 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
2296 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 2297 | |
43a70507 N |
2298 | len = sprintf(page, "%lu", secs); |
2299 | if (jifs) | |
2300 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
2301 | len += sprintf(page+len, "\n"); | |
2302 | return len; | |
2303 | } | |
2304 | ||
2305 | static ssize_t | |
fd01b88c | 2306 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2307 | { |
2308 | /* timeout can be set at any time */ | |
2309 | unsigned long timeout; | |
2310 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
2311 | if (rv) | |
2312 | return rv; | |
2313 | ||
2314 | /* just to make sure we don't overflow... */ | |
2315 | if (timeout >= LONG_MAX / HZ) | |
2316 | return -EINVAL; | |
2317 | ||
2318 | timeout = timeout * HZ / 10000; | |
2319 | ||
2320 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
2321 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
2322 | if (timeout < 1) | |
2323 | timeout = 1; | |
2324 | mddev->bitmap_info.daemon_sleep = timeout; | |
2325 | if (mddev->thread) { | |
2326 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
2327 | * the bitmap is all clean and we don't need to | |
2328 | * adjust the timeout right now | |
2329 | */ | |
2330 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2331 | mddev->thread->timeout = timeout; | |
2332 | md_wakeup_thread(mddev->thread); | |
2333 | } | |
2334 | } | |
2335 | return len; | |
2336 | } | |
2337 | ||
2338 | static struct md_sysfs_entry bitmap_timeout = | |
2339 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2340 | ||
2341 | static ssize_t | |
fd01b88c | 2342 | backlog_show(struct mddev *mddev, char *page) |
43a70507 N |
2343 | { |
2344 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2345 | } | |
2346 | ||
2347 | static ssize_t | |
fd01b88c | 2348 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2349 | { |
2350 | unsigned long backlog; | |
b29bebd6 | 2351 | int rv = kstrtoul(buf, 10, &backlog); |
43a70507 N |
2352 | if (rv) |
2353 | return rv; | |
2354 | if (backlog > COUNTER_MAX) | |
2355 | return -EINVAL; | |
2356 | mddev->bitmap_info.max_write_behind = backlog; | |
2357 | return len; | |
2358 | } | |
2359 | ||
2360 | static struct md_sysfs_entry bitmap_backlog = | |
2361 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2362 | ||
2363 | static ssize_t | |
fd01b88c | 2364 | chunksize_show(struct mddev *mddev, char *page) |
43a70507 N |
2365 | { |
2366 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2367 | } | |
2368 | ||
2369 | static ssize_t | |
fd01b88c | 2370 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2371 | { |
2372 | /* Can only be changed when no bitmap is active */ | |
2373 | int rv; | |
2374 | unsigned long csize; | |
2375 | if (mddev->bitmap) | |
2376 | return -EBUSY; | |
b29bebd6 | 2377 | rv = kstrtoul(buf, 10, &csize); |
43a70507 N |
2378 | if (rv) |
2379 | return rv; | |
2380 | if (csize < 512 || | |
2381 | !is_power_of_2(csize)) | |
2382 | return -EINVAL; | |
2383 | mddev->bitmap_info.chunksize = csize; | |
2384 | return len; | |
2385 | } | |
2386 | ||
2387 | static struct md_sysfs_entry bitmap_chunksize = | |
2388 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2389 | ||
fd01b88c | 2390 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
ece5cff0 | 2391 | { |
c4ce867f GR |
2392 | if (mddev_is_clustered(mddev)) |
2393 | return sprintf(page, "clustered\n"); | |
ece5cff0 N |
2394 | return sprintf(page, "%s\n", (mddev->bitmap_info.external |
2395 | ? "external" : "internal")); | |
2396 | } | |
2397 | ||
fd01b88c | 2398 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2399 | { |
2400 | if (mddev->bitmap || | |
2401 | mddev->bitmap_info.file || | |
2402 | mddev->bitmap_info.offset) | |
2403 | return -EBUSY; | |
2404 | if (strncmp(buf, "external", 8) == 0) | |
2405 | mddev->bitmap_info.external = 1; | |
c4ce867f GR |
2406 | else if ((strncmp(buf, "internal", 8) == 0) || |
2407 | (strncmp(buf, "clustered", 9) == 0)) | |
ece5cff0 N |
2408 | mddev->bitmap_info.external = 0; |
2409 | else | |
2410 | return -EINVAL; | |
2411 | return len; | |
2412 | } | |
2413 | ||
2414 | static struct md_sysfs_entry bitmap_metadata = | |
2415 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2416 | ||
fd01b88c | 2417 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2418 | { |
2419 | int len; | |
b7b17c9b | 2420 | spin_lock(&mddev->lock); |
ece5cff0 N |
2421 | if (mddev->bitmap) |
2422 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2423 | "false" : "true")); | |
2424 | else | |
2425 | len = sprintf(page, "\n"); | |
b7b17c9b | 2426 | spin_unlock(&mddev->lock); |
ece5cff0 N |
2427 | return len; |
2428 | } | |
2429 | ||
fd01b88c | 2430 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2431 | { |
2432 | if (mddev->bitmap == NULL) | |
2433 | return -ENOENT; | |
2434 | if (strncmp(buf, "false", 5) == 0) | |
2435 | mddev->bitmap->need_sync = 1; | |
2436 | else if (strncmp(buf, "true", 4) == 0) { | |
2437 | if (mddev->degraded) | |
2438 | return -EBUSY; | |
2439 | mddev->bitmap->need_sync = 0; | |
2440 | } else | |
2441 | return -EINVAL; | |
2442 | return len; | |
2443 | } | |
2444 | ||
2445 | static struct md_sysfs_entry bitmap_can_clear = | |
2446 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2447 | ||
696fcd53 | 2448 | static ssize_t |
fd01b88c | 2449 | behind_writes_used_show(struct mddev *mddev, char *page) |
696fcd53 | 2450 | { |
b7b17c9b N |
2451 | ssize_t ret; |
2452 | spin_lock(&mddev->lock); | |
696fcd53 | 2453 | if (mddev->bitmap == NULL) |
b7b17c9b N |
2454 | ret = sprintf(page, "0\n"); |
2455 | else | |
2456 | ret = sprintf(page, "%lu\n", | |
2457 | mddev->bitmap->behind_writes_used); | |
2458 | spin_unlock(&mddev->lock); | |
2459 | return ret; | |
696fcd53 PC |
2460 | } |
2461 | ||
2462 | static ssize_t | |
fd01b88c | 2463 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
696fcd53 PC |
2464 | { |
2465 | if (mddev->bitmap) | |
2466 | mddev->bitmap->behind_writes_used = 0; | |
2467 | return len; | |
2468 | } | |
2469 | ||
2470 | static struct md_sysfs_entry max_backlog_used = | |
2471 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2472 | behind_writes_used_show, behind_writes_used_reset); | |
2473 | ||
43a70507 N |
2474 | static struct attribute *md_bitmap_attrs[] = { |
2475 | &bitmap_location.attr, | |
6409bb05 | 2476 | &bitmap_space.attr, |
43a70507 N |
2477 | &bitmap_timeout.attr, |
2478 | &bitmap_backlog.attr, | |
2479 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2480 | &bitmap_metadata.attr, |
2481 | &bitmap_can_clear.attr, | |
696fcd53 | 2482 | &max_backlog_used.attr, |
43a70507 N |
2483 | NULL |
2484 | }; | |
2485 | struct attribute_group md_bitmap_group = { | |
2486 | .name = "bitmap", | |
2487 | .attrs = md_bitmap_attrs, | |
2488 | }; | |
2489 |