]>
Commit | Line | Data |
---|---|---|
32a7627c N |
1 | /* |
2 | * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003 | |
3 | * | |
4 | * bitmap_create - sets up the bitmap structure | |
5 | * bitmap_destroy - destroys the bitmap structure | |
6 | * | |
7 | * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.: | |
8 | * - added disk storage for bitmap | |
9 | * - changes to allow various bitmap chunk sizes | |
32a7627c N |
10 | */ |
11 | ||
12 | /* | |
13 | * Still to do: | |
14 | * | |
15 | * flush after percent set rather than just time based. (maybe both). | |
32a7627c N |
16 | */ |
17 | ||
bff61975 | 18 | #include <linux/blkdev.h> |
32a7627c | 19 | #include <linux/module.h> |
32a7627c N |
20 | #include <linux/errno.h> |
21 | #include <linux/slab.h> | |
22 | #include <linux/init.h> | |
32a7627c N |
23 | #include <linux/timer.h> |
24 | #include <linux/sched.h> | |
25 | #include <linux/list.h> | |
26 | #include <linux/file.h> | |
27 | #include <linux/mount.h> | |
28 | #include <linux/buffer_head.h> | |
57148964 | 29 | #include <linux/seq_file.h> |
43b2e5d8 | 30 | #include "md.h" |
ef740c37 | 31 | #include "bitmap.h" |
32a7627c | 32 | |
ac2f40be | 33 | static inline char *bmname(struct bitmap *bitmap) |
32a7627c N |
34 | { |
35 | return bitmap->mddev ? mdname(bitmap->mddev) : "mdX"; | |
36 | } | |
37 | ||
32a7627c N |
38 | /* |
39 | * check a page and, if necessary, allocate it (or hijack it if the alloc fails) | |
40 | * | |
41 | * 1) check to see if this page is allocated, if it's not then try to alloc | |
42 | * 2) if the alloc fails, set the page's hijacked flag so we'll use the | |
43 | * page pointer directly as a counter | |
44 | * | |
45 | * if we find our page, we increment the page's refcount so that it stays | |
46 | * allocated while we're using it | |
47 | */ | |
40cffcc0 | 48 | static int bitmap_checkpage(struct bitmap_counts *bitmap, |
ac2f40be | 49 | unsigned long page, int create) |
ee305ace N |
50 | __releases(bitmap->lock) |
51 | __acquires(bitmap->lock) | |
32a7627c N |
52 | { |
53 | unsigned char *mappage; | |
54 | ||
55 | if (page >= bitmap->pages) { | |
1187cf0a N |
56 | /* This can happen if bitmap_start_sync goes beyond |
57 | * End-of-device while looking for a whole page. | |
58 | * It is harmless. | |
59 | */ | |
32a7627c N |
60 | return -EINVAL; |
61 | } | |
62 | ||
32a7627c N |
63 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ |
64 | return 0; | |
65 | ||
66 | if (bitmap->bp[page].map) /* page is already allocated, just return */ | |
67 | return 0; | |
68 | ||
69 | if (!create) | |
70 | return -ENOENT; | |
71 | ||
32a7627c N |
72 | /* this page has not been allocated yet */ |
73 | ||
ac2f40be | 74 | spin_unlock_irq(&bitmap->lock); |
d9590143 N |
75 | /* It is possible that this is being called inside a |
76 | * prepare_to_wait/finish_wait loop from raid5c:make_request(). | |
77 | * In general it is not permitted to sleep in that context as it | |
78 | * can cause the loop to spin freely. | |
79 | * That doesn't apply here as we can only reach this point | |
80 | * once with any loop. | |
81 | * When this function completes, either bp[page].map or | |
82 | * bp[page].hijacked. In either case, this function will | |
83 | * abort before getting to this point again. So there is | |
84 | * no risk of a free-spin, and so it is safe to assert | |
85 | * that sleeping here is allowed. | |
86 | */ | |
87 | sched_annotate_sleep(); | |
792a1d4b | 88 | mappage = kzalloc(PAGE_SIZE, GFP_NOIO); |
ac2f40be N |
89 | spin_lock_irq(&bitmap->lock); |
90 | ||
91 | if (mappage == NULL) { | |
40cffcc0 | 92 | pr_debug("md/bitmap: map page allocation failed, hijacking\n"); |
32a7627c N |
93 | /* failed - set the hijacked flag so that we can use the |
94 | * pointer as a counter */ | |
32a7627c N |
95 | if (!bitmap->bp[page].map) |
96 | bitmap->bp[page].hijacked = 1; | |
ac2f40be N |
97 | } else if (bitmap->bp[page].map || |
98 | bitmap->bp[page].hijacked) { | |
32a7627c | 99 | /* somebody beat us to getting the page */ |
792a1d4b | 100 | kfree(mappage); |
ac2f40be | 101 | } else { |
32a7627c | 102 | |
ac2f40be | 103 | /* no page was in place and we have one, so install it */ |
32a7627c | 104 | |
ac2f40be N |
105 | bitmap->bp[page].map = mappage; |
106 | bitmap->missing_pages--; | |
107 | } | |
32a7627c N |
108 | return 0; |
109 | } | |
110 | ||
32a7627c N |
111 | /* if page is completely empty, put it back on the free list, or dealloc it */ |
112 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
113 | /* Note: lock should be held when calling this */ | |
40cffcc0 | 114 | static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page) |
32a7627c N |
115 | { |
116 | char *ptr; | |
117 | ||
118 | if (bitmap->bp[page].count) /* page is still busy */ | |
119 | return; | |
120 | ||
121 | /* page is no longer in use, it can be released */ | |
122 | ||
123 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
124 | bitmap->bp[page].hijacked = 0; | |
125 | bitmap->bp[page].map = NULL; | |
ac2f40be N |
126 | } else { |
127 | /* normal case, free the page */ | |
128 | ptr = bitmap->bp[page].map; | |
129 | bitmap->bp[page].map = NULL; | |
130 | bitmap->missing_pages++; | |
792a1d4b | 131 | kfree(ptr); |
32a7627c | 132 | } |
32a7627c N |
133 | } |
134 | ||
32a7627c N |
135 | /* |
136 | * bitmap file handling - read and write the bitmap file and its superblock | |
137 | */ | |
138 | ||
32a7627c N |
139 | /* |
140 | * basic page I/O operations | |
141 | */ | |
142 | ||
a654b9d8 | 143 | /* IO operations when bitmap is stored near all superblocks */ |
27581e5a N |
144 | static int read_sb_page(struct mddev *mddev, loff_t offset, |
145 | struct page *page, | |
146 | unsigned long index, int size) | |
a654b9d8 N |
147 | { |
148 | /* choose a good rdev and read the page from there */ | |
149 | ||
3cb03002 | 150 | struct md_rdev *rdev; |
a654b9d8 | 151 | sector_t target; |
a654b9d8 | 152 | |
dafb20fa | 153 | rdev_for_each(rdev, mddev) { |
b2d444d7 N |
154 | if (! test_bit(In_sync, &rdev->flags) |
155 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
156 | continue; |
157 | ||
ccebd4c4 | 158 | target = offset + index * (PAGE_SIZE/512); |
a654b9d8 | 159 | |
2b193363 | 160 | if (sync_page_io(rdev, target, |
e1defc4f | 161 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
ccebd4c4 | 162 | page, READ, true)) { |
ab904d63 | 163 | page->index = index; |
27581e5a | 164 | return 0; |
ab904d63 N |
165 | } |
166 | } | |
27581e5a | 167 | return -EIO; |
a654b9d8 N |
168 | } |
169 | ||
fd01b88c | 170 | static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) |
b2d2c4ce N |
171 | { |
172 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
173 | * linked list, and raising the refcount of devices we return to ensure | |
174 | * they don't disappear while in use. | |
175 | * As devices are only added or removed when raid_disk is < 0 and | |
176 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
177 | * still be in the same position on the list when we re-enter | |
fd177481 | 178 | * list_for_each_entry_continue_rcu. |
8532e343 N |
179 | * |
180 | * Note that if entered with 'rdev == NULL' to start at the | |
181 | * beginning, we temporarily assign 'rdev' to an address which | |
182 | * isn't really an rdev, but which can be used by | |
183 | * list_for_each_entry_continue_rcu() to find the first entry. | |
b2d2c4ce | 184 | */ |
b2d2c4ce N |
185 | rcu_read_lock(); |
186 | if (rdev == NULL) | |
187 | /* start at the beginning */ | |
8532e343 | 188 | rdev = list_entry(&mddev->disks, struct md_rdev, same_set); |
b2d2c4ce N |
189 | else { |
190 | /* release the previous rdev and start from there. */ | |
191 | rdev_dec_pending(rdev, mddev); | |
b2d2c4ce | 192 | } |
fd177481 | 193 | list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) { |
b2d2c4ce | 194 | if (rdev->raid_disk >= 0 && |
b2d2c4ce N |
195 | !test_bit(Faulty, &rdev->flags)) { |
196 | /* this is a usable devices */ | |
197 | atomic_inc(&rdev->nr_pending); | |
198 | rcu_read_unlock(); | |
199 | return rdev; | |
200 | } | |
201 | } | |
202 | rcu_read_unlock(); | |
203 | return NULL; | |
204 | } | |
205 | ||
ab6085c7 | 206 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 207 | { |
3cb03002 | 208 | struct md_rdev *rdev = NULL; |
a6ff7e08 | 209 | struct block_device *bdev; |
fd01b88c | 210 | struct mddev *mddev = bitmap->mddev; |
1ec885cd | 211 | struct bitmap_storage *store = &bitmap->storage; |
a654b9d8 | 212 | |
b2d2c4ce | 213 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ac2f40be N |
214 | int size = PAGE_SIZE; |
215 | loff_t offset = mddev->bitmap_info.offset; | |
a6ff7e08 JB |
216 | |
217 | bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; | |
218 | ||
9b1215c1 N |
219 | if (page->index == store->file_pages-1) { |
220 | int last_page_size = store->bytes & (PAGE_SIZE-1); | |
221 | if (last_page_size == 0) | |
222 | last_page_size = PAGE_SIZE; | |
223 | size = roundup(last_page_size, | |
a6ff7e08 | 224 | bdev_logical_block_size(bdev)); |
9b1215c1 | 225 | } |
ac2f40be N |
226 | /* Just make sure we aren't corrupting data or |
227 | * metadata | |
228 | */ | |
229 | if (mddev->external) { | |
230 | /* Bitmap could be anywhere. */ | |
231 | if (rdev->sb_start + offset + (page->index | |
232 | * (PAGE_SIZE/512)) | |
233 | > rdev->data_offset | |
234 | && | |
235 | rdev->sb_start + offset | |
236 | < (rdev->data_offset + mddev->dev_sectors | |
237 | + (PAGE_SIZE/512))) | |
238 | goto bad_alignment; | |
239 | } else if (offset < 0) { | |
240 | /* DATA BITMAP METADATA */ | |
241 | if (offset | |
242 | + (long)(page->index * (PAGE_SIZE/512)) | |
243 | + size/512 > 0) | |
244 | /* bitmap runs in to metadata */ | |
245 | goto bad_alignment; | |
246 | if (rdev->data_offset + mddev->dev_sectors | |
247 | > rdev->sb_start + offset) | |
248 | /* data runs in to bitmap */ | |
249 | goto bad_alignment; | |
250 | } else if (rdev->sb_start < rdev->data_offset) { | |
251 | /* METADATA BITMAP DATA */ | |
252 | if (rdev->sb_start | |
253 | + offset | |
254 | + page->index*(PAGE_SIZE/512) + size/512 | |
255 | > rdev->data_offset) | |
256 | /* bitmap runs in to data */ | |
257 | goto bad_alignment; | |
258 | } else { | |
259 | /* DATA METADATA BITMAP - no problems */ | |
260 | } | |
261 | md_super_write(mddev, rdev, | |
262 | rdev->sb_start + offset | |
263 | + page->index * (PAGE_SIZE/512), | |
264 | size, | |
265 | page); | |
b2d2c4ce | 266 | } |
a654b9d8 N |
267 | |
268 | if (wait) | |
a9701a30 | 269 | md_super_wait(mddev); |
a654b9d8 | 270 | return 0; |
4b80991c N |
271 | |
272 | bad_alignment: | |
4b80991c | 273 | return -EINVAL; |
a654b9d8 N |
274 | } |
275 | ||
4ad13663 | 276 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 277 | /* |
a654b9d8 | 278 | * write out a page to a file |
32a7627c | 279 | */ |
4ad13663 | 280 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 281 | { |
d785a06a | 282 | struct buffer_head *bh; |
32a7627c | 283 | |
1ec885cd | 284 | if (bitmap->storage.file == NULL) { |
f0d76d70 N |
285 | switch (write_sb_page(bitmap, page, wait)) { |
286 | case -EINVAL: | |
b405fe91 | 287 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); |
f0d76d70 | 288 | } |
4ad13663 | 289 | } else { |
a654b9d8 | 290 | |
4ad13663 | 291 | bh = page_buffers(page); |
c708443c | 292 | |
4ad13663 N |
293 | while (bh && bh->b_blocknr) { |
294 | atomic_inc(&bitmap->pending_writes); | |
295 | set_buffer_locked(bh); | |
296 | set_buffer_mapped(bh); | |
721a9602 | 297 | submit_bh(WRITE | REQ_SYNC, bh); |
4ad13663 N |
298 | bh = bh->b_this_page; |
299 | } | |
d785a06a | 300 | |
ac2f40be | 301 | if (wait) |
4ad13663 N |
302 | wait_event(bitmap->write_wait, |
303 | atomic_read(&bitmap->pending_writes)==0); | |
8a5e9cf1 | 304 | } |
b405fe91 | 305 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
4ad13663 | 306 | bitmap_file_kick(bitmap); |
d785a06a N |
307 | } |
308 | ||
309 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
310 | { | |
311 | struct bitmap *bitmap = bh->b_private; | |
32a7627c | 312 | |
b405fe91 N |
313 | if (!uptodate) |
314 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); | |
d785a06a N |
315 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
316 | wake_up(&bitmap->write_wait); | |
317 | } | |
32a7627c | 318 | |
d785a06a N |
319 | /* copied from buffer.c */ |
320 | static void | |
321 | __clear_page_buffers(struct page *page) | |
322 | { | |
323 | ClearPagePrivate(page); | |
324 | set_page_private(page, 0); | |
09cbfeaf | 325 | put_page(page); |
d785a06a N |
326 | } |
327 | static void free_buffers(struct page *page) | |
328 | { | |
27581e5a | 329 | struct buffer_head *bh; |
77ad4bc7 | 330 | |
27581e5a N |
331 | if (!PagePrivate(page)) |
332 | return; | |
333 | ||
334 | bh = page_buffers(page); | |
d785a06a N |
335 | while (bh) { |
336 | struct buffer_head *next = bh->b_this_page; | |
337 | free_buffer_head(bh); | |
338 | bh = next; | |
77ad4bc7 | 339 | } |
d785a06a N |
340 | __clear_page_buffers(page); |
341 | put_page(page); | |
32a7627c N |
342 | } |
343 | ||
d785a06a N |
344 | /* read a page from a file. |
345 | * We both read the page, and attach buffers to the page to record the | |
346 | * address of each block (using bmap). These addresses will be used | |
347 | * to write the block later, completely bypassing the filesystem. | |
348 | * This usage is similar to how swap files are handled, and allows us | |
349 | * to write to a file with no concerns of memory allocation failing. | |
350 | */ | |
27581e5a N |
351 | static int read_page(struct file *file, unsigned long index, |
352 | struct bitmap *bitmap, | |
353 | unsigned long count, | |
354 | struct page *page) | |
32a7627c | 355 | { |
27581e5a | 356 | int ret = 0; |
496ad9aa | 357 | struct inode *inode = file_inode(file); |
d785a06a N |
358 | struct buffer_head *bh; |
359 | sector_t block; | |
32a7627c | 360 | |
36a4e1fe N |
361 | pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
362 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 363 | |
d785a06a N |
364 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
365 | if (!bh) { | |
27581e5a | 366 | ret = -ENOMEM; |
32a7627c N |
367 | goto out; |
368 | } | |
d785a06a N |
369 | attach_page_buffers(page, bh); |
370 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
371 | while (bh) { | |
372 | if (count == 0) | |
373 | bh->b_blocknr = 0; | |
374 | else { | |
375 | bh->b_blocknr = bmap(inode, block); | |
376 | if (bh->b_blocknr == 0) { | |
377 | /* Cannot use this file! */ | |
27581e5a | 378 | ret = -EINVAL; |
d785a06a N |
379 | goto out; |
380 | } | |
381 | bh->b_bdev = inode->i_sb->s_bdev; | |
382 | if (count < (1<<inode->i_blkbits)) | |
383 | count = 0; | |
384 | else | |
385 | count -= (1<<inode->i_blkbits); | |
386 | ||
387 | bh->b_end_io = end_bitmap_write; | |
388 | bh->b_private = bitmap; | |
ce25c31b N |
389 | atomic_inc(&bitmap->pending_writes); |
390 | set_buffer_locked(bh); | |
391 | set_buffer_mapped(bh); | |
392 | submit_bh(READ, bh); | |
d785a06a N |
393 | } |
394 | block++; | |
395 | bh = bh->b_this_page; | |
396 | } | |
d785a06a | 397 | page->index = index; |
ce25c31b N |
398 | |
399 | wait_event(bitmap->write_wait, | |
400 | atomic_read(&bitmap->pending_writes)==0); | |
b405fe91 | 401 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
27581e5a | 402 | ret = -EIO; |
32a7627c | 403 | out: |
27581e5a N |
404 | if (ret) |
405 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n", | |
2d1f3b5d N |
406 | (int)PAGE_SIZE, |
407 | (unsigned long long)index << PAGE_SHIFT, | |
27581e5a N |
408 | ret); |
409 | return ret; | |
32a7627c N |
410 | } |
411 | ||
412 | /* | |
413 | * bitmap file superblock operations | |
414 | */ | |
415 | ||
416 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 417 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
418 | { |
419 | bitmap_super_t *sb; | |
32a7627c N |
420 | |
421 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 422 | return; |
ece5cff0 N |
423 | if (bitmap->mddev->bitmap_info.external) |
424 | return; | |
1ec885cd | 425 | if (!bitmap->storage.sb_page) /* no superblock */ |
4ad13663 | 426 | return; |
1ec885cd | 427 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 428 | sb->events = cpu_to_le64(bitmap->mddev->events); |
8258c532 | 429 | if (bitmap->mddev->events < bitmap->events_cleared) |
a0da84f3 NB |
430 | /* rocking back to read-only */ |
431 | bitmap->events_cleared = bitmap->mddev->events; | |
8258c532 N |
432 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); |
433 | sb->state = cpu_to_le32(bitmap->flags); | |
43a70507 N |
434 | /* Just in case these have been changed via sysfs: */ |
435 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
436 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
b81a0404 N |
437 | /* This might have been changed by a reshape */ |
438 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
439 | sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize); | |
c4ce867f | 440 | sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes); |
1dff2b87 N |
441 | sb->sectors_reserved = cpu_to_le32(bitmap->mddev-> |
442 | bitmap_info.space); | |
b2f46e68 | 443 | kunmap_atomic(sb); |
1ec885cd | 444 | write_page(bitmap, bitmap->storage.sb_page, 1); |
32a7627c N |
445 | } |
446 | ||
447 | /* print out the bitmap file superblock */ | |
448 | void bitmap_print_sb(struct bitmap *bitmap) | |
449 | { | |
450 | bitmap_super_t *sb; | |
451 | ||
1ec885cd | 452 | if (!bitmap || !bitmap->storage.sb_page) |
32a7627c | 453 | return; |
1ec885cd | 454 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 455 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
456 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
457 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
458 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
459 | *(__u32 *)(sb->uuid+0), |
460 | *(__u32 *)(sb->uuid+4), | |
461 | *(__u32 *)(sb->uuid+8), | |
462 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 463 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 464 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 465 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 466 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
467 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
468 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
469 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
470 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
471 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 472 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
b2f46e68 | 473 | kunmap_atomic(sb); |
32a7627c N |
474 | } |
475 | ||
9c81075f JB |
476 | /* |
477 | * bitmap_new_disk_sb | |
478 | * @bitmap | |
479 | * | |
480 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb | |
481 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. | |
482 | * This function verifies 'bitmap_info' and populates the on-disk bitmap | |
483 | * structure, which is to be written to disk. | |
484 | * | |
485 | * Returns: 0 on success, -Exxx on error | |
486 | */ | |
487 | static int bitmap_new_disk_sb(struct bitmap *bitmap) | |
488 | { | |
489 | bitmap_super_t *sb; | |
490 | unsigned long chunksize, daemon_sleep, write_behind; | |
9c81075f | 491 | |
d3b178ad | 492 | bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
582e2e05 JM |
493 | if (bitmap->storage.sb_page == NULL) |
494 | return -ENOMEM; | |
1ec885cd | 495 | bitmap->storage.sb_page->index = 0; |
9c81075f | 496 | |
1ec885cd | 497 | sb = kmap_atomic(bitmap->storage.sb_page); |
9c81075f JB |
498 | |
499 | sb->magic = cpu_to_le32(BITMAP_MAGIC); | |
500 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); | |
501 | ||
502 | chunksize = bitmap->mddev->bitmap_info.chunksize; | |
503 | BUG_ON(!chunksize); | |
504 | if (!is_power_of_2(chunksize)) { | |
b2f46e68 | 505 | kunmap_atomic(sb); |
9c81075f JB |
506 | printk(KERN_ERR "bitmap chunksize not a power of 2\n"); |
507 | return -EINVAL; | |
508 | } | |
509 | sb->chunksize = cpu_to_le32(chunksize); | |
510 | ||
511 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; | |
c97e0602 | 512 | if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { |
9c81075f JB |
513 | printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n"); |
514 | daemon_sleep = 5 * HZ; | |
515 | } | |
516 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); | |
517 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
518 | ||
519 | /* | |
520 | * FIXME: write_behind for RAID1. If not specified, what | |
521 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. | |
522 | */ | |
523 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; | |
524 | if (write_behind > COUNTER_MAX) | |
525 | write_behind = COUNTER_MAX / 2; | |
526 | sb->write_behind = cpu_to_le32(write_behind); | |
527 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
528 | ||
529 | /* keep the array size field of the bitmap superblock up to date */ | |
530 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
531 | ||
532 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); | |
533 | ||
b405fe91 | 534 | set_bit(BITMAP_STALE, &bitmap->flags); |
84e92345 | 535 | sb->state = cpu_to_le32(bitmap->flags); |
9c81075f JB |
536 | bitmap->events_cleared = bitmap->mddev->events; |
537 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
d3b178ad | 538 | bitmap->mddev->bitmap_info.nodes = 0; |
9c81075f | 539 | |
b2f46e68 | 540 | kunmap_atomic(sb); |
9c81075f JB |
541 | |
542 | return 0; | |
543 | } | |
544 | ||
32a7627c N |
545 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
546 | static int bitmap_read_sb(struct bitmap *bitmap) | |
547 | { | |
548 | char *reason = NULL; | |
549 | bitmap_super_t *sb; | |
4b6d287f | 550 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c | 551 | unsigned long long events; |
c4ce867f | 552 | int nodes = 0; |
1dff2b87 | 553 | unsigned long sectors_reserved = 0; |
32a7627c | 554 | int err = -EINVAL; |
27581e5a | 555 | struct page *sb_page; |
33e38ac6 | 556 | loff_t offset = bitmap->mddev->bitmap_info.offset; |
32a7627c | 557 | |
1ec885cd | 558 | if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { |
ef99bf48 N |
559 | chunksize = 128 * 1024 * 1024; |
560 | daemon_sleep = 5 * HZ; | |
561 | write_behind = 0; | |
b405fe91 | 562 | set_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
563 | err = 0; |
564 | goto out_no_sb; | |
565 | } | |
32a7627c | 566 | /* page 0 is the superblock, read it... */ |
27581e5a N |
567 | sb_page = alloc_page(GFP_KERNEL); |
568 | if (!sb_page) | |
569 | return -ENOMEM; | |
1ec885cd | 570 | bitmap->storage.sb_page = sb_page; |
27581e5a | 571 | |
b97e9257 | 572 | re_read: |
f9209a32 GR |
573 | /* If cluster_slot is set, the cluster is setup */ |
574 | if (bitmap->cluster_slot >= 0) { | |
3b0e6aac | 575 | sector_t bm_blocks = bitmap->mddev->resync_max_sectors; |
f9209a32 | 576 | |
3b0e6aac SR |
577 | sector_div(bm_blocks, |
578 | bitmap->mddev->bitmap_info.chunksize >> 9); | |
124eb761 GR |
579 | /* bits to bytes */ |
580 | bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t); | |
581 | /* to 4k blocks */ | |
935f3d4f | 582 | bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096); |
33e38ac6 | 583 | offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3)); |
f9209a32 | 584 | pr_info("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__, |
33e38ac6 | 585 | bitmap->cluster_slot, offset); |
f9209a32 GR |
586 | } |
587 | ||
1ec885cd N |
588 | if (bitmap->storage.file) { |
589 | loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); | |
f49d5e62 N |
590 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; |
591 | ||
1ec885cd | 592 | err = read_page(bitmap->storage.file, 0, |
27581e5a | 593 | bitmap, bytes, sb_page); |
f49d5e62 | 594 | } else { |
27581e5a | 595 | err = read_sb_page(bitmap->mddev, |
33e38ac6 | 596 | offset, |
27581e5a N |
597 | sb_page, |
598 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 599 | } |
27581e5a | 600 | if (err) |
32a7627c | 601 | return err; |
32a7627c | 602 | |
b97e9257 | 603 | err = -EINVAL; |
27581e5a | 604 | sb = kmap_atomic(sb_page); |
32a7627c | 605 | |
32a7627c | 606 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 607 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 608 | write_behind = le32_to_cpu(sb->write_behind); |
1dff2b87 | 609 | sectors_reserved = le32_to_cpu(sb->sectors_reserved); |
3c462c88 GR |
610 | /* Setup nodes/clustername only if bitmap version is |
611 | * cluster-compatible | |
d3b178ad | 612 | */ |
3c462c88 | 613 | if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) { |
d3b178ad GR |
614 | nodes = le32_to_cpu(sb->nodes); |
615 | strlcpy(bitmap->mddev->bitmap_info.cluster_name, | |
616 | sb->cluster_name, 64); | |
617 | } | |
32a7627c N |
618 | |
619 | /* verify that the bitmap-specific fields are valid */ | |
620 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
621 | reason = "bad magic"; | |
bd926c63 | 622 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
3c462c88 | 623 | le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED) |
32a7627c | 624 | reason = "unrecognized superblock version"; |
1187cf0a | 625 | else if (chunksize < 512) |
7dd5d34c | 626 | reason = "bitmap chunksize too small"; |
d744540c | 627 | else if (!is_power_of_2(chunksize)) |
32a7627c | 628 | reason = "bitmap chunksize not a power of 2"; |
1b04be96 | 629 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 630 | reason = "daemon sleep period out of range"; |
4b6d287f N |
631 | else if (write_behind > COUNTER_MAX) |
632 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
633 | if (reason) { |
634 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
635 | bmname(bitmap), reason); | |
636 | goto out; | |
637 | } | |
638 | ||
639 | /* keep the array size field of the bitmap superblock up to date */ | |
640 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
641 | ||
278c1ca2 N |
642 | if (bitmap->mddev->persistent) { |
643 | /* | |
644 | * We have a persistent array superblock, so compare the | |
645 | * bitmap's UUID and event counter to the mddev's | |
646 | */ | |
647 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
648 | printk(KERN_INFO | |
649 | "%s: bitmap superblock UUID mismatch\n", | |
650 | bmname(bitmap)); | |
651 | goto out; | |
652 | } | |
653 | events = le64_to_cpu(sb->events); | |
b97e9257 | 654 | if (!nodes && (events < bitmap->mddev->events)) { |
278c1ca2 N |
655 | printk(KERN_INFO |
656 | "%s: bitmap file is out of date (%llu < %llu) " | |
657 | "-- forcing full recovery\n", | |
658 | bmname(bitmap), events, | |
659 | (unsigned long long) bitmap->mddev->events); | |
b405fe91 | 660 | set_bit(BITMAP_STALE, &bitmap->flags); |
278c1ca2 | 661 | } |
32a7627c | 662 | } |
278c1ca2 | 663 | |
32a7627c | 664 | /* assign fields using values from superblock */ |
4f2e639a | 665 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 | 666 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
b405fe91 | 667 | set_bit(BITMAP_HOSTENDIAN, &bitmap->flags); |
32a7627c | 668 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
cf921cc1 | 669 | strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64); |
32a7627c | 670 | err = 0; |
b97e9257 | 671 | |
32a7627c | 672 | out: |
b2f46e68 | 673 | kunmap_atomic(sb); |
f9209a32 GR |
674 | /* Assiging chunksize is required for "re_read" */ |
675 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
f7357273 | 676 | if (err == 0 && nodes && (bitmap->cluster_slot < 0)) { |
b97e9257 GR |
677 | err = md_setup_cluster(bitmap->mddev, nodes); |
678 | if (err) { | |
679 | pr_err("%s: Could not setup cluster service (%d)\n", | |
680 | bmname(bitmap), err); | |
681 | goto out_no_sb; | |
682 | } | |
683 | bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev); | |
b97e9257 GR |
684 | goto re_read; |
685 | } | |
686 | ||
687 | ||
ef99bf48 | 688 | out_no_sb: |
b405fe91 | 689 | if (test_bit(BITMAP_STALE, &bitmap->flags)) |
ef99bf48 N |
690 | bitmap->events_cleared = bitmap->mddev->events; |
691 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
692 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
693 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
c4ce867f | 694 | bitmap->mddev->bitmap_info.nodes = nodes; |
1dff2b87 N |
695 | if (bitmap->mddev->bitmap_info.space == 0 || |
696 | bitmap->mddev->bitmap_info.space > sectors_reserved) | |
697 | bitmap->mddev->bitmap_info.space = sectors_reserved; | |
b97e9257 | 698 | if (err) { |
32a7627c | 699 | bitmap_print_sb(bitmap); |
f9209a32 | 700 | if (bitmap->cluster_slot < 0) |
b97e9257 GR |
701 | md_cluster_stop(bitmap->mddev); |
702 | } | |
32a7627c N |
703 | return err; |
704 | } | |
705 | ||
32a7627c N |
706 | /* |
707 | * general bitmap file operations | |
708 | */ | |
709 | ||
ece5cff0 N |
710 | /* |
711 | * on-disk bitmap: | |
712 | * | |
713 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
714 | * file a page at a time. There's a superblock at the start of the file. | |
715 | */ | |
32a7627c | 716 | /* calculate the index of the page that contains this bit */ |
1ec885cd N |
717 | static inline unsigned long file_page_index(struct bitmap_storage *store, |
718 | unsigned long chunk) | |
32a7627c | 719 | { |
1ec885cd | 720 | if (store->sb_page) |
ece5cff0 N |
721 | chunk += sizeof(bitmap_super_t) << 3; |
722 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
723 | } |
724 | ||
725 | /* calculate the (bit) offset of this bit within a page */ | |
1ec885cd N |
726 | static inline unsigned long file_page_offset(struct bitmap_storage *store, |
727 | unsigned long chunk) | |
32a7627c | 728 | { |
1ec885cd | 729 | if (store->sb_page) |
ece5cff0 N |
730 | chunk += sizeof(bitmap_super_t) << 3; |
731 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
732 | } |
733 | ||
734 | /* | |
735 | * return a pointer to the page in the filemap that contains the given bit | |
736 | * | |
32a7627c | 737 | */ |
1ec885cd | 738 | static inline struct page *filemap_get_page(struct bitmap_storage *store, |
3520fa4d | 739 | unsigned long chunk) |
32a7627c | 740 | { |
1ec885cd | 741 | if (file_page_index(store, chunk) >= store->file_pages) |
ac2f40be | 742 | return NULL; |
f2e06c58 | 743 | return store->filemap[file_page_index(store, chunk)]; |
32a7627c N |
744 | } |
745 | ||
d1244cb0 | 746 | static int bitmap_storage_alloc(struct bitmap_storage *store, |
b97e9257 GR |
747 | unsigned long chunks, int with_super, |
748 | int slot_number) | |
d1244cb0 | 749 | { |
b97e9257 | 750 | int pnum, offset = 0; |
d1244cb0 N |
751 | unsigned long num_pages; |
752 | unsigned long bytes; | |
753 | ||
754 | bytes = DIV_ROUND_UP(chunks, 8); | |
755 | if (with_super) | |
756 | bytes += sizeof(bitmap_super_t); | |
757 | ||
758 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); | |
b97e9257 | 759 | offset = slot_number * (num_pages - 1); |
d1244cb0 N |
760 | |
761 | store->filemap = kmalloc(sizeof(struct page *) | |
762 | * num_pages, GFP_KERNEL); | |
763 | if (!store->filemap) | |
764 | return -ENOMEM; | |
765 | ||
766 | if (with_super && !store->sb_page) { | |
d60b479d | 767 | store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
768 | if (store->sb_page == NULL) |
769 | return -ENOMEM; | |
d1244cb0 | 770 | } |
b97e9257 | 771 | |
d1244cb0 N |
772 | pnum = 0; |
773 | if (store->sb_page) { | |
774 | store->filemap[0] = store->sb_page; | |
775 | pnum = 1; | |
b97e9257 | 776 | store->sb_page->index = offset; |
d1244cb0 | 777 | } |
b97e9257 | 778 | |
d1244cb0 | 779 | for ( ; pnum < num_pages; pnum++) { |
d60b479d | 780 | store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
781 | if (!store->filemap[pnum]) { |
782 | store->file_pages = pnum; | |
783 | return -ENOMEM; | |
784 | } | |
b97e9257 | 785 | store->filemap[pnum]->index = pnum + offset; |
d1244cb0 N |
786 | } |
787 | store->file_pages = pnum; | |
788 | ||
789 | /* We need 4 bits per page, rounded up to a multiple | |
790 | * of sizeof(unsigned long) */ | |
791 | store->filemap_attr = kzalloc( | |
792 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), | |
793 | GFP_KERNEL); | |
794 | if (!store->filemap_attr) | |
795 | return -ENOMEM; | |
796 | ||
797 | store->bytes = bytes; | |
798 | ||
799 | return 0; | |
800 | } | |
801 | ||
fae7d326 | 802 | static void bitmap_file_unmap(struct bitmap_storage *store) |
32a7627c N |
803 | { |
804 | struct page **map, *sb_page; | |
32a7627c | 805 | int pages; |
fae7d326 | 806 | struct file *file; |
32a7627c | 807 | |
fae7d326 | 808 | file = store->file; |
1ec885cd | 809 | map = store->filemap; |
1ec885cd | 810 | pages = store->file_pages; |
1ec885cd | 811 | sb_page = store->sb_page; |
32a7627c N |
812 | |
813 | while (pages--) | |
ece5cff0 | 814 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 815 | free_buffers(map[pages]); |
32a7627c | 816 | kfree(map); |
fae7d326 | 817 | kfree(store->filemap_attr); |
32a7627c | 818 | |
d785a06a N |
819 | if (sb_page) |
820 | free_buffers(sb_page); | |
32a7627c | 821 | |
d785a06a | 822 | if (file) { |
496ad9aa | 823 | struct inode *inode = file_inode(file); |
fc0ecff6 | 824 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 825 | fput(file); |
d785a06a | 826 | } |
32a7627c N |
827 | } |
828 | ||
32a7627c N |
829 | /* |
830 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
831 | * then it is no longer reliable, so we stop using it and we mark the file | |
832 | * as failed in the superblock | |
833 | */ | |
834 | static void bitmap_file_kick(struct bitmap *bitmap) | |
835 | { | |
836 | char *path, *ptr = NULL; | |
837 | ||
b405fe91 | 838 | if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) { |
4ad13663 | 839 | bitmap_update_sb(bitmap); |
32a7627c | 840 | |
1ec885cd | 841 | if (bitmap->storage.file) { |
4ad13663 N |
842 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); |
843 | if (path) | |
9bf39ab2 | 844 | ptr = file_path(bitmap->storage.file, |
1ec885cd | 845 | path, PAGE_SIZE); |
6bcfd601 | 846 | |
4ad13663 N |
847 | printk(KERN_ALERT |
848 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 849 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 850 | |
4ad13663 N |
851 | kfree(path); |
852 | } else | |
853 | printk(KERN_ALERT | |
854 | "%s: disabling internal bitmap due to errors\n", | |
855 | bmname(bitmap)); | |
a654b9d8 | 856 | } |
32a7627c N |
857 | } |
858 | ||
859 | enum bitmap_page_attr { | |
ac2f40be | 860 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
5a537df4 N |
861 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
862 | * i.e. counter is 1 or 2. */ | |
ac2f40be | 863 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
32a7627c N |
864 | }; |
865 | ||
d189122d N |
866 | static inline void set_page_attr(struct bitmap *bitmap, int pnum, |
867 | enum bitmap_page_attr attr) | |
32a7627c | 868 | { |
bdfd1140 | 869 | set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
870 | } |
871 | ||
d189122d N |
872 | static inline void clear_page_attr(struct bitmap *bitmap, int pnum, |
873 | enum bitmap_page_attr attr) | |
32a7627c | 874 | { |
bdfd1140 | 875 | clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
876 | } |
877 | ||
bdfd1140 N |
878 | static inline int test_page_attr(struct bitmap *bitmap, int pnum, |
879 | enum bitmap_page_attr attr) | |
32a7627c | 880 | { |
1ec885cd | 881 | return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
882 | } |
883 | ||
bdfd1140 N |
884 | static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum, |
885 | enum bitmap_page_attr attr) | |
886 | { | |
887 | return test_and_clear_bit((pnum<<2) + attr, | |
888 | bitmap->storage.filemap_attr); | |
889 | } | |
32a7627c N |
890 | /* |
891 | * bitmap_file_set_bit -- called before performing a write to the md device | |
892 | * to set (and eventually sync) a particular bit in the bitmap file | |
893 | * | |
894 | * we set the bit immediately, then we record the page number so that | |
895 | * when an unplug occurs, we can flush the dirty pages out to disk | |
896 | */ | |
897 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
898 | { | |
899 | unsigned long bit; | |
3520fa4d | 900 | struct page *page; |
32a7627c | 901 | void *kaddr; |
40cffcc0 | 902 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
32a7627c | 903 | |
1ec885cd | 904 | page = filemap_get_page(&bitmap->storage, chunk); |
3520fa4d JB |
905 | if (!page) |
906 | return; | |
1ec885cd | 907 | bit = file_page_offset(&bitmap->storage, chunk); |
32a7627c | 908 | |
3520fa4d | 909 | /* set the bit */ |
b2f46e68 | 910 | kaddr = kmap_atomic(page); |
b405fe91 | 911 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
3520fa4d JB |
912 | set_bit(bit, kaddr); |
913 | else | |
3f810b6c | 914 | set_bit_le(bit, kaddr); |
b2f46e68 | 915 | kunmap_atomic(kaddr); |
36a4e1fe | 916 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
32a7627c | 917 | /* record page number so it gets flushed to disk when unplug occurs */ |
d189122d | 918 | set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY); |
32a7627c N |
919 | } |
920 | ||
ef99bf48 N |
921 | static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) |
922 | { | |
923 | unsigned long bit; | |
924 | struct page *page; | |
925 | void *paddr; | |
40cffcc0 | 926 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
ef99bf48 | 927 | |
1ec885cd | 928 | page = filemap_get_page(&bitmap->storage, chunk); |
ef99bf48 N |
929 | if (!page) |
930 | return; | |
1ec885cd | 931 | bit = file_page_offset(&bitmap->storage, chunk); |
ef99bf48 | 932 | paddr = kmap_atomic(page); |
b405fe91 | 933 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ef99bf48 N |
934 | clear_bit(bit, paddr); |
935 | else | |
3f810b6c | 936 | clear_bit_le(bit, paddr); |
ef99bf48 | 937 | kunmap_atomic(paddr); |
d189122d N |
938 | if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) { |
939 | set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING); | |
ef99bf48 N |
940 | bitmap->allclean = 0; |
941 | } | |
942 | } | |
943 | ||
11dd35da GR |
944 | static int bitmap_file_test_bit(struct bitmap *bitmap, sector_t block) |
945 | { | |
946 | unsigned long bit; | |
947 | struct page *page; | |
948 | void *paddr; | |
949 | unsigned long chunk = block >> bitmap->counts.chunkshift; | |
950 | int set = 0; | |
951 | ||
952 | page = filemap_get_page(&bitmap->storage, chunk); | |
953 | if (!page) | |
954 | return -EINVAL; | |
955 | bit = file_page_offset(&bitmap->storage, chunk); | |
956 | paddr = kmap_atomic(page); | |
957 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) | |
958 | set = test_bit(bit, paddr); | |
959 | else | |
960 | set = test_bit_le(bit, paddr); | |
961 | kunmap_atomic(paddr); | |
962 | return set; | |
963 | } | |
964 | ||
965 | ||
32a7627c N |
966 | /* this gets called when the md device is ready to unplug its underlying |
967 | * (slave) device queues -- before we let any writes go down, we need to | |
968 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 969 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 970 | { |
74667123 | 971 | unsigned long i; |
ec7a3197 | 972 | int dirty, need_write; |
32a7627c | 973 | |
62f82faa N |
974 | if (!bitmap || !bitmap->storage.filemap || |
975 | test_bit(BITMAP_STALE, &bitmap->flags)) | |
4ad13663 | 976 | return; |
32a7627c N |
977 | |
978 | /* look at each page to see if there are any set bits that need to be | |
979 | * flushed out to disk */ | |
1ec885cd | 980 | for (i = 0; i < bitmap->storage.file_pages; i++) { |
bdfd1140 | 981 | if (!bitmap->storage.filemap) |
4ad13663 | 982 | return; |
bdfd1140 N |
983 | dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
984 | need_write = test_and_clear_page_attr(bitmap, i, | |
985 | BITMAP_PAGE_NEEDWRITE); | |
986 | if (dirty || need_write) { | |
d189122d | 987 | clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); |
bdfd1140 N |
988 | write_page(bitmap, bitmap->storage.filemap[i], 0); |
989 | } | |
32a7627c | 990 | } |
4b5060dd N |
991 | if (bitmap->storage.file) |
992 | wait_event(bitmap->write_wait, | |
993 | atomic_read(&bitmap->pending_writes)==0); | |
994 | else | |
995 | md_super_wait(bitmap->mddev); | |
996 | ||
b405fe91 | 997 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
d785a06a | 998 | bitmap_file_kick(bitmap); |
32a7627c | 999 | } |
ac2f40be | 1000 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 1001 | |
6a07997f | 1002 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
1003 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
1004 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
1005 | * memory mapping of the bitmap file | |
1006 | * Special cases: | |
1007 | * if there's no bitmap file, or if the bitmap file had been | |
1008 | * previously kicked from the array, we mark all the bits as | |
1009 | * 1's in order to cause a full resync. | |
6a07997f N |
1010 | * |
1011 | * We ignore all bits for sectors that end earlier than 'start'. | |
1012 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 1013 | */ |
6a07997f | 1014 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c | 1015 | { |
b97e9257 | 1016 | unsigned long i, chunks, index, oldindex, bit, node_offset = 0; |
27581e5a | 1017 | struct page *page = NULL; |
d1244cb0 | 1018 | unsigned long bit_cnt = 0; |
32a7627c | 1019 | struct file *file; |
d1244cb0 | 1020 | unsigned long offset; |
32a7627c N |
1021 | int outofdate; |
1022 | int ret = -ENOSPC; | |
ea03aff9 | 1023 | void *paddr; |
1ec885cd | 1024 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c | 1025 | |
40cffcc0 | 1026 | chunks = bitmap->counts.chunks; |
1ec885cd | 1027 | file = store->file; |
32a7627c | 1028 | |
ef99bf48 N |
1029 | if (!file && !bitmap->mddev->bitmap_info.offset) { |
1030 | /* No permanent bitmap - fill with '1s'. */ | |
1ec885cd N |
1031 | store->filemap = NULL; |
1032 | store->file_pages = 0; | |
ef99bf48 N |
1033 | for (i = 0; i < chunks ; i++) { |
1034 | /* if the disk bit is set, set the memory bit */ | |
40cffcc0 | 1035 | int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift) |
ef99bf48 N |
1036 | >= start); |
1037 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 1038 | (sector_t)i << bitmap->counts.chunkshift, |
ef99bf48 N |
1039 | needed); |
1040 | } | |
1041 | return 0; | |
1042 | } | |
32a7627c | 1043 | |
b405fe91 | 1044 | outofdate = test_bit(BITMAP_STALE, &bitmap->flags); |
32a7627c N |
1045 | if (outofdate) |
1046 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
1047 | "recovery\n", bmname(bitmap)); | |
1048 | ||
d1244cb0 | 1049 | if (file && i_size_read(file->f_mapping->host) < store->bytes) { |
32a7627c | 1050 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
d1244cb0 N |
1051 | bmname(bitmap), |
1052 | (unsigned long) i_size_read(file->f_mapping->host), | |
1053 | store->bytes); | |
4ad13663 | 1054 | goto err; |
32a7627c | 1055 | } |
bc7f77de | 1056 | |
d1244cb0 | 1057 | oldindex = ~0L; |
27581e5a | 1058 | offset = 0; |
d1244cb0 | 1059 | if (!bitmap->mddev->bitmap_info.external) |
27581e5a | 1060 | offset = sizeof(bitmap_super_t); |
32a7627c | 1061 | |
b97e9257 GR |
1062 | if (mddev_is_clustered(bitmap->mddev)) |
1063 | node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE)); | |
1064 | ||
32a7627c | 1065 | for (i = 0; i < chunks; i++) { |
bd926c63 | 1066 | int b; |
1ec885cd N |
1067 | index = file_page_index(&bitmap->storage, i); |
1068 | bit = file_page_offset(&bitmap->storage, i); | |
32a7627c | 1069 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 1070 | int count; |
32a7627c | 1071 | /* unmap the old page, we're done with it */ |
d1244cb0 N |
1072 | if (index == store->file_pages-1) |
1073 | count = store->bytes - index * PAGE_SIZE; | |
d785a06a N |
1074 | else |
1075 | count = PAGE_SIZE; | |
1ec885cd | 1076 | page = store->filemap[index]; |
27581e5a N |
1077 | if (file) |
1078 | ret = read_page(file, index, bitmap, | |
1079 | count, page); | |
1080 | else | |
1081 | ret = read_sb_page( | |
1082 | bitmap->mddev, | |
1083 | bitmap->mddev->bitmap_info.offset, | |
1084 | page, | |
b97e9257 | 1085 | index + node_offset, count); |
27581e5a N |
1086 | |
1087 | if (ret) | |
4ad13663 | 1088 | goto err; |
a654b9d8 | 1089 | |
32a7627c | 1090 | oldindex = index; |
32a7627c N |
1091 | |
1092 | if (outofdate) { | |
1093 | /* | |
1094 | * if bitmap is out of date, dirty the | |
ac2f40be | 1095 | * whole page and write it out |
32a7627c | 1096 | */ |
b2f46e68 | 1097 | paddr = kmap_atomic(page); |
ea03aff9 | 1098 | memset(paddr + offset, 0xff, |
6a07997f | 1099 | PAGE_SIZE - offset); |
b2f46e68 | 1100 | kunmap_atomic(paddr); |
4ad13663 N |
1101 | write_page(bitmap, page, 1); |
1102 | ||
1103 | ret = -EIO; | |
b405fe91 N |
1104 | if (test_bit(BITMAP_WRITE_ERROR, |
1105 | &bitmap->flags)) | |
4ad13663 | 1106 | goto err; |
32a7627c | 1107 | } |
32a7627c | 1108 | } |
b2f46e68 | 1109 | paddr = kmap_atomic(page); |
b405fe91 | 1110 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ea03aff9 | 1111 | b = test_bit(bit, paddr); |
bd926c63 | 1112 | else |
6b33aff3 | 1113 | b = test_bit_le(bit, paddr); |
b2f46e68 | 1114 | kunmap_atomic(paddr); |
bd926c63 | 1115 | if (b) { |
32a7627c | 1116 | /* if the disk bit is set, set the memory bit */ |
40cffcc0 | 1117 | int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift |
db305e50 N |
1118 | >= start); |
1119 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 1120 | (sector_t)i << bitmap->counts.chunkshift, |
db305e50 | 1121 | needed); |
32a7627c N |
1122 | bit_cnt++; |
1123 | } | |
27581e5a | 1124 | offset = 0; |
32a7627c N |
1125 | } |
1126 | ||
32a7627c | 1127 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
d1244cb0 | 1128 | "read %lu pages, set %lu of %lu bits\n", |
1ec885cd | 1129 | bmname(bitmap), store->file_pages, |
d1244cb0 | 1130 | bit_cnt, chunks); |
4ad13663 N |
1131 | |
1132 | return 0; | |
32a7627c | 1133 | |
4ad13663 N |
1134 | err: |
1135 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1136 | bmname(bitmap), ret); | |
32a7627c N |
1137 | return ret; |
1138 | } | |
1139 | ||
a654b9d8 N |
1140 | void bitmap_write_all(struct bitmap *bitmap) |
1141 | { | |
1142 | /* We don't actually write all bitmap blocks here, | |
1143 | * just flag them as needing to be written | |
1144 | */ | |
ec7a3197 | 1145 | int i; |
a654b9d8 | 1146 | |
1ec885cd | 1147 | if (!bitmap || !bitmap->storage.filemap) |
ef99bf48 | 1148 | return; |
1ec885cd | 1149 | if (bitmap->storage.file) |
ef99bf48 N |
1150 | /* Only one copy, so nothing needed */ |
1151 | return; | |
1152 | ||
1ec885cd | 1153 | for (i = 0; i < bitmap->storage.file_pages; i++) |
d189122d | 1154 | set_page_attr(bitmap, i, |
ec7a3197 | 1155 | BITMAP_PAGE_NEEDWRITE); |
2585f3ef | 1156 | bitmap->allclean = 0; |
a654b9d8 N |
1157 | } |
1158 | ||
40cffcc0 N |
1159 | static void bitmap_count_page(struct bitmap_counts *bitmap, |
1160 | sector_t offset, int inc) | |
32a7627c | 1161 | { |
61a0d80c | 1162 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1163 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1164 | bitmap->bp[page].count += inc; | |
32a7627c N |
1165 | bitmap_checkfree(bitmap, page); |
1166 | } | |
bf07bb7d | 1167 | |
40cffcc0 | 1168 | static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset) |
bf07bb7d N |
1169 | { |
1170 | sector_t chunk = offset >> bitmap->chunkshift; | |
1171 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1172 | struct bitmap_page *bp = &bitmap->bp[page]; | |
1173 | ||
1174 | if (!bp->pending) | |
1175 | bp->pending = 1; | |
1176 | } | |
1177 | ||
40cffcc0 | 1178 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1179 | sector_t offset, sector_t *blocks, |
32a7627c N |
1180 | int create); |
1181 | ||
1182 | /* | |
1183 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1184 | * out to disk | |
1185 | */ | |
1186 | ||
fd01b88c | 1187 | void bitmap_daemon_work(struct mddev *mddev) |
32a7627c | 1188 | { |
aa5cbd10 | 1189 | struct bitmap *bitmap; |
aa3163f8 | 1190 | unsigned long j; |
bf07bb7d | 1191 | unsigned long nextpage; |
57dab0bd | 1192 | sector_t blocks; |
40cffcc0 | 1193 | struct bitmap_counts *counts; |
32a7627c | 1194 | |
aa5cbd10 N |
1195 | /* Use a mutex to guard daemon_work against |
1196 | * bitmap_destroy. | |
1197 | */ | |
c3d9714e | 1198 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1199 | bitmap = mddev->bitmap; |
1200 | if (bitmap == NULL) { | |
c3d9714e | 1201 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1202 | return; |
aa5cbd10 | 1203 | } |
42a04b50 | 1204 | if (time_before(jiffies, bitmap->daemon_lastrun |
2e61ebbc | 1205 | + mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1206 | goto done; |
1207 | ||
32a7627c | 1208 | bitmap->daemon_lastrun = jiffies; |
8311c29d | 1209 | if (bitmap->allclean) { |
2e61ebbc | 1210 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
aa5cbd10 | 1211 | goto done; |
8311c29d N |
1212 | } |
1213 | bitmap->allclean = 1; | |
32a7627c | 1214 | |
bf07bb7d N |
1215 | /* Any file-page which is PENDING now needs to be written. |
1216 | * So set NEEDWRITE now, then after we make any last-minute changes | |
1217 | * we will write it. | |
1218 | */ | |
1ec885cd | 1219 | for (j = 0; j < bitmap->storage.file_pages; j++) |
bdfd1140 N |
1220 | if (test_and_clear_page_attr(bitmap, j, |
1221 | BITMAP_PAGE_PENDING)) | |
d189122d | 1222 | set_page_attr(bitmap, j, |
bf07bb7d | 1223 | BITMAP_PAGE_NEEDWRITE); |
bf07bb7d N |
1224 | |
1225 | if (bitmap->need_sync && | |
1226 | mddev->bitmap_info.external == 0) { | |
1227 | /* Arrange for superblock update as well as | |
1228 | * other changes */ | |
1229 | bitmap_super_t *sb; | |
1230 | bitmap->need_sync = 0; | |
1ec885cd N |
1231 | if (bitmap->storage.filemap) { |
1232 | sb = kmap_atomic(bitmap->storage.sb_page); | |
ef99bf48 N |
1233 | sb->events_cleared = |
1234 | cpu_to_le64(bitmap->events_cleared); | |
1235 | kunmap_atomic(sb); | |
d189122d | 1236 | set_page_attr(bitmap, 0, |
ef99bf48 N |
1237 | BITMAP_PAGE_NEEDWRITE); |
1238 | } | |
bf07bb7d N |
1239 | } |
1240 | /* Now look at the bitmap counters and if any are '2' or '1', | |
1241 | * decrement and handle accordingly. | |
1242 | */ | |
40cffcc0 N |
1243 | counts = &bitmap->counts; |
1244 | spin_lock_irq(&counts->lock); | |
bf07bb7d | 1245 | nextpage = 0; |
40cffcc0 | 1246 | for (j = 0; j < counts->chunks; j++) { |
32a7627c | 1247 | bitmap_counter_t *bmc; |
40cffcc0 | 1248 | sector_t block = (sector_t)j << counts->chunkshift; |
3520fa4d | 1249 | |
bf07bb7d N |
1250 | if (j == nextpage) { |
1251 | nextpage += PAGE_COUNTER_RATIO; | |
40cffcc0 | 1252 | if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) { |
bf07bb7d | 1253 | j |= PAGE_COUNTER_MASK; |
aa3163f8 N |
1254 | continue; |
1255 | } | |
40cffcc0 | 1256 | counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; |
32a7627c | 1257 | } |
40cffcc0 | 1258 | bmc = bitmap_get_counter(counts, |
ef99bf48 | 1259 | block, |
db305e50 | 1260 | &blocks, 0); |
bf07bb7d N |
1261 | |
1262 | if (!bmc) { | |
5a537df4 | 1263 | j |= PAGE_COUNTER_MASK; |
bf07bb7d N |
1264 | continue; |
1265 | } | |
1266 | if (*bmc == 1 && !bitmap->need_sync) { | |
1267 | /* We can clear the bit */ | |
bf07bb7d | 1268 | *bmc = 0; |
40cffcc0 | 1269 | bitmap_count_page(counts, block, -1); |
ef99bf48 | 1270 | bitmap_file_clear_bit(bitmap, block); |
bf07bb7d N |
1271 | } else if (*bmc && *bmc <= 2) { |
1272 | *bmc = 1; | |
40cffcc0 | 1273 | bitmap_set_pending(counts, block); |
bf07bb7d | 1274 | bitmap->allclean = 0; |
5a537df4 | 1275 | } |
32a7627c | 1276 | } |
40cffcc0 | 1277 | spin_unlock_irq(&counts->lock); |
32a7627c | 1278 | |
bf07bb7d N |
1279 | /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. |
1280 | * DIRTY pages need to be written by bitmap_unplug so it can wait | |
1281 | * for them. | |
1282 | * If we find any DIRTY page we stop there and let bitmap_unplug | |
1283 | * handle all the rest. This is important in the case where | |
1284 | * the first blocking holds the superblock and it has been updated. | |
1285 | * We mustn't write any other blocks before the superblock. | |
1286 | */ | |
62f82faa N |
1287 | for (j = 0; |
1288 | j < bitmap->storage.file_pages | |
1289 | && !test_bit(BITMAP_STALE, &bitmap->flags); | |
1290 | j++) { | |
d189122d | 1291 | if (test_page_attr(bitmap, j, |
bf07bb7d N |
1292 | BITMAP_PAGE_DIRTY)) |
1293 | /* bitmap_unplug will handle the rest */ | |
1294 | break; | |
bdfd1140 N |
1295 | if (test_and_clear_page_attr(bitmap, j, |
1296 | BITMAP_PAGE_NEEDWRITE)) { | |
1ec885cd | 1297 | write_page(bitmap, bitmap->storage.filemap[j], 0); |
32a7627c | 1298 | } |
32a7627c N |
1299 | } |
1300 | ||
7be3dfec | 1301 | done: |
8311c29d | 1302 | if (bitmap->allclean == 0) |
2e61ebbc N |
1303 | mddev->thread->timeout = |
1304 | mddev->bitmap_info.daemon_sleep; | |
c3d9714e | 1305 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1306 | } |
1307 | ||
40cffcc0 | 1308 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1309 | sector_t offset, sector_t *blocks, |
32a7627c | 1310 | int create) |
ee305ace N |
1311 | __releases(bitmap->lock) |
1312 | __acquires(bitmap->lock) | |
32a7627c N |
1313 | { |
1314 | /* If 'create', we might release the lock and reclaim it. | |
1315 | * The lock must have been taken with interrupts enabled. | |
1316 | * If !create, we don't release the lock. | |
1317 | */ | |
61a0d80c | 1318 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1319 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1320 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1321 | sector_t csize; | |
ef425673 | 1322 | int err; |
32a7627c | 1323 | |
ef425673 N |
1324 | err = bitmap_checkpage(bitmap, page, create); |
1325 | ||
1326 | if (bitmap->bp[page].hijacked || | |
1327 | bitmap->bp[page].map == NULL) | |
61a0d80c | 1328 | csize = ((sector_t)1) << (bitmap->chunkshift + |
ef425673 N |
1329 | PAGE_COUNTER_SHIFT - 1); |
1330 | else | |
61a0d80c | 1331 | csize = ((sector_t)1) << bitmap->chunkshift; |
ef425673 N |
1332 | *blocks = csize - (offset & (csize - 1)); |
1333 | ||
1334 | if (err < 0) | |
32a7627c | 1335 | return NULL; |
ef425673 | 1336 | |
32a7627c N |
1337 | /* now locked ... */ |
1338 | ||
1339 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1340 | /* should we use the first or second counter field | |
1341 | * of the hijacked pointer? */ | |
1342 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1343 | return &((bitmap_counter_t *) |
1344 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1345 | } else /* page is allocated */ |
32a7627c N |
1346 | return (bitmap_counter_t *) |
1347 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1348 | } |
1349 | ||
4b6d287f | 1350 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1351 | { |
ac2f40be N |
1352 | if (!bitmap) |
1353 | return 0; | |
4b6d287f N |
1354 | |
1355 | if (behind) { | |
696fcd53 | 1356 | int bw; |
4b6d287f | 1357 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1358 | bw = atomic_read(&bitmap->behind_writes); |
1359 | if (bw > bitmap->behind_writes_used) | |
1360 | bitmap->behind_writes_used = bw; | |
1361 | ||
36a4e1fe N |
1362 | pr_debug("inc write-behind count %d/%lu\n", |
1363 | bw, bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1364 | } |
1365 | ||
32a7627c | 1366 | while (sectors) { |
57dab0bd | 1367 | sector_t blocks; |
32a7627c N |
1368 | bitmap_counter_t *bmc; |
1369 | ||
40cffcc0 N |
1370 | spin_lock_irq(&bitmap->counts.lock); |
1371 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1); | |
32a7627c | 1372 | if (!bmc) { |
40cffcc0 | 1373 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1374 | return 0; |
1375 | } | |
1376 | ||
27d5ea04 | 1377 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
da6e1a32 NB |
1378 | DEFINE_WAIT(__wait); |
1379 | /* note that it is safe to do the prepare_to_wait | |
1380 | * after the test as long as we do it before dropping | |
1381 | * the spinlock. | |
1382 | */ | |
1383 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1384 | TASK_UNINTERRUPTIBLE); | |
40cffcc0 | 1385 | spin_unlock_irq(&bitmap->counts.lock); |
f54a9d0e | 1386 | schedule(); |
da6e1a32 NB |
1387 | finish_wait(&bitmap->overflow_wait, &__wait); |
1388 | continue; | |
1389 | } | |
1390 | ||
ac2f40be | 1391 | switch (*bmc) { |
32a7627c N |
1392 | case 0: |
1393 | bitmap_file_set_bit(bitmap, offset); | |
40cffcc0 | 1394 | bitmap_count_page(&bitmap->counts, offset, 1); |
32a7627c N |
1395 | /* fall through */ |
1396 | case 1: | |
1397 | *bmc = 2; | |
1398 | } | |
da6e1a32 | 1399 | |
32a7627c N |
1400 | (*bmc)++; |
1401 | ||
40cffcc0 | 1402 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1403 | |
1404 | offset += blocks; | |
1405 | if (sectors > blocks) | |
1406 | sectors -= blocks; | |
ac2f40be N |
1407 | else |
1408 | sectors = 0; | |
32a7627c N |
1409 | } |
1410 | return 0; | |
1411 | } | |
ac2f40be | 1412 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1413 | |
1414 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1415 | int success, int behind) |
32a7627c | 1416 | { |
ac2f40be N |
1417 | if (!bitmap) |
1418 | return; | |
4b6d287f | 1419 | if (behind) { |
e555190d N |
1420 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1421 | wake_up(&bitmap->behind_wait); | |
36a4e1fe N |
1422 | pr_debug("dec write-behind count %d/%lu\n", |
1423 | atomic_read(&bitmap->behind_writes), | |
1424 | bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1425 | } |
1426 | ||
32a7627c | 1427 | while (sectors) { |
57dab0bd | 1428 | sector_t blocks; |
32a7627c N |
1429 | unsigned long flags; |
1430 | bitmap_counter_t *bmc; | |
1431 | ||
40cffcc0 N |
1432 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1433 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0); | |
32a7627c | 1434 | if (!bmc) { |
40cffcc0 | 1435 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1436 | return; |
1437 | } | |
1438 | ||
961902c0 | 1439 | if (success && !bitmap->mddev->degraded && |
a0da84f3 NB |
1440 | bitmap->events_cleared < bitmap->mddev->events) { |
1441 | bitmap->events_cleared = bitmap->mddev->events; | |
1442 | bitmap->need_sync = 1; | |
5ff5afff | 1443 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1444 | } |
1445 | ||
27d5ea04 | 1446 | if (!success && !NEEDED(*bmc)) |
32a7627c N |
1447 | *bmc |= NEEDED_MASK; |
1448 | ||
27d5ea04 | 1449 | if (COUNTER(*bmc) == COUNTER_MAX) |
da6e1a32 NB |
1450 | wake_up(&bitmap->overflow_wait); |
1451 | ||
32a7627c | 1452 | (*bmc)--; |
2585f3ef | 1453 | if (*bmc <= 2) { |
40cffcc0 | 1454 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1455 | bitmap->allclean = 0; |
1456 | } | |
40cffcc0 | 1457 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1458 | offset += blocks; |
1459 | if (sectors > blocks) | |
1460 | sectors -= blocks; | |
ac2f40be N |
1461 | else |
1462 | sectors = 0; | |
32a7627c N |
1463 | } |
1464 | } | |
ac2f40be | 1465 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1466 | |
57dab0bd | 1467 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1468 | int degraded) |
32a7627c N |
1469 | { |
1470 | bitmap_counter_t *bmc; | |
1471 | int rv; | |
1472 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1473 | *blocks = 1024; | |
1474 | return 1; /* always resync if no bitmap */ | |
1475 | } | |
40cffcc0 N |
1476 | spin_lock_irq(&bitmap->counts.lock); |
1477 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1478 | rv = 0; |
1479 | if (bmc) { | |
1480 | /* locked */ | |
1481 | if (RESYNC(*bmc)) | |
1482 | rv = 1; | |
1483 | else if (NEEDED(*bmc)) { | |
1484 | rv = 1; | |
6a806c51 N |
1485 | if (!degraded) { /* don't set/clear bits if degraded */ |
1486 | *bmc |= RESYNC_MASK; | |
1487 | *bmc &= ~NEEDED_MASK; | |
1488 | } | |
32a7627c N |
1489 | } |
1490 | } | |
40cffcc0 | 1491 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1492 | return rv; |
1493 | } | |
1494 | ||
57dab0bd | 1495 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a N |
1496 | int degraded) |
1497 | { | |
1498 | /* bitmap_start_sync must always report on multiples of whole | |
1499 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1500 | * get confused. | |
1501 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1502 | * At least PAGE_SIZE>>9 blocks are covered. | |
1503 | * Return the 'or' of the result. | |
1504 | */ | |
1505 | int rv = 0; | |
57dab0bd | 1506 | sector_t blocks1; |
1187cf0a N |
1507 | |
1508 | *blocks = 0; | |
1509 | while (*blocks < (PAGE_SIZE>>9)) { | |
1510 | rv |= __bitmap_start_sync(bitmap, offset, | |
1511 | &blocks1, degraded); | |
1512 | offset += blocks1; | |
1513 | *blocks += blocks1; | |
1514 | } | |
1515 | return rv; | |
1516 | } | |
ac2f40be | 1517 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1518 | |
57dab0bd | 1519 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1520 | { |
1521 | bitmap_counter_t *bmc; | |
1522 | unsigned long flags; | |
ac2f40be N |
1523 | |
1524 | if (bitmap == NULL) { | |
32a7627c N |
1525 | *blocks = 1024; |
1526 | return; | |
1527 | } | |
40cffcc0 N |
1528 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1529 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1530 | if (bmc == NULL) |
1531 | goto unlock; | |
1532 | /* locked */ | |
32a7627c N |
1533 | if (RESYNC(*bmc)) { |
1534 | *bmc &= ~RESYNC_MASK; | |
1535 | ||
1536 | if (!NEEDED(*bmc) && aborted) | |
1537 | *bmc |= NEEDED_MASK; | |
1538 | else { | |
2585f3ef | 1539 | if (*bmc <= 2) { |
40cffcc0 | 1540 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1541 | bitmap->allclean = 0; |
1542 | } | |
32a7627c N |
1543 | } |
1544 | } | |
1545 | unlock: | |
40cffcc0 | 1546 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c | 1547 | } |
ac2f40be | 1548 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1549 | |
1550 | void bitmap_close_sync(struct bitmap *bitmap) | |
1551 | { | |
1552 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1553 | * properly have been aborted. It remains to us to clear the | |
1554 | * RESYNC bit wherever it is still on | |
1555 | */ | |
1556 | sector_t sector = 0; | |
57dab0bd | 1557 | sector_t blocks; |
b47490c9 N |
1558 | if (!bitmap) |
1559 | return; | |
32a7627c N |
1560 | while (sector < bitmap->mddev->resync_max_sectors) { |
1561 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1562 | sector += blocks; |
1563 | } | |
1564 | } | |
ac2f40be | 1565 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 | 1566 | |
c40f341f | 1567 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force) |
b47490c9 N |
1568 | { |
1569 | sector_t s = 0; | |
57dab0bd | 1570 | sector_t blocks; |
b47490c9 N |
1571 | |
1572 | if (!bitmap) | |
1573 | return; | |
1574 | if (sector == 0) { | |
1575 | bitmap->last_end_sync = jiffies; | |
1576 | return; | |
1577 | } | |
c40f341f | 1578 | if (!force && time_before(jiffies, (bitmap->last_end_sync |
1b04be96 | 1579 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1580 | return; |
1581 | wait_event(bitmap->mddev->recovery_wait, | |
1582 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1583 | ||
75d3da43 | 1584 | bitmap->mddev->curr_resync_completed = sector; |
070dc6dd | 1585 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
40cffcc0 | 1586 | sector &= ~((1ULL << bitmap->counts.chunkshift) - 1); |
b47490c9 N |
1587 | s = 0; |
1588 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1589 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1590 | s += blocks; | |
32a7627c | 1591 | } |
b47490c9 | 1592 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1593 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1594 | } |
ac2f40be | 1595 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1596 | |
6a07997f | 1597 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1598 | { |
1599 | /* For each chunk covered by any of these sectors, set the | |
ef99bf48 | 1600 | * counter to 2 and possibly set resync_needed. They should all |
32a7627c N |
1601 | * be 0 at this point |
1602 | */ | |
193f1c93 | 1603 | |
57dab0bd | 1604 | sector_t secs; |
193f1c93 | 1605 | bitmap_counter_t *bmc; |
40cffcc0 N |
1606 | spin_lock_irq(&bitmap->counts.lock); |
1607 | bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1); | |
193f1c93 | 1608 | if (!bmc) { |
40cffcc0 | 1609 | spin_unlock_irq(&bitmap->counts.lock); |
193f1c93 | 1610 | return; |
32a7627c | 1611 | } |
ac2f40be | 1612 | if (!*bmc) { |
11dd35da | 1613 | *bmc = 2; |
40cffcc0 N |
1614 | bitmap_count_page(&bitmap->counts, offset, 1); |
1615 | bitmap_set_pending(&bitmap->counts, offset); | |
2585f3ef | 1616 | bitmap->allclean = 0; |
193f1c93 | 1617 | } |
11dd35da GR |
1618 | if (needed) |
1619 | *bmc |= NEEDED_MASK; | |
40cffcc0 | 1620 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1621 | } |
1622 | ||
9b1d1dac PC |
1623 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1624 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1625 | { | |
1626 | unsigned long chunk; | |
1627 | ||
1628 | for (chunk = s; chunk <= e; chunk++) { | |
40cffcc0 | 1629 | sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift; |
9b1d1dac PC |
1630 | bitmap_set_memory_bits(bitmap, sec, 1); |
1631 | bitmap_file_set_bit(bitmap, sec); | |
ffa23322 N |
1632 | if (sec < bitmap->mddev->recovery_cp) |
1633 | /* We are asserting that the array is dirty, | |
1634 | * so move the recovery_cp address back so | |
1635 | * that it is obvious that it is dirty | |
1636 | */ | |
1637 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1638 | } |
1639 | } | |
1640 | ||
6b8b3e8a N |
1641 | /* |
1642 | * flush out any pending updates | |
1643 | */ | |
fd01b88c | 1644 | void bitmap_flush(struct mddev *mddev) |
6b8b3e8a N |
1645 | { |
1646 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1647 | long sleep; |
6b8b3e8a N |
1648 | |
1649 | if (!bitmap) /* there was no bitmap */ | |
1650 | return; | |
1651 | ||
1652 | /* run the daemon_work three time to ensure everything is flushed | |
1653 | * that can be | |
1654 | */ | |
1b04be96 | 1655 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1656 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1657 | bitmap_daemon_work(mddev); |
42a04b50 | 1658 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1659 | bitmap_daemon_work(mddev); |
42a04b50 | 1660 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1661 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1662 | bitmap_update_sb(bitmap); |
1663 | } | |
1664 | ||
32a7627c N |
1665 | /* |
1666 | * free memory that was allocated | |
1667 | */ | |
3178b0db | 1668 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1669 | { |
1670 | unsigned long k, pages; | |
1671 | struct bitmap_page *bp; | |
32a7627c N |
1672 | |
1673 | if (!bitmap) /* there was no bitmap */ | |
1674 | return; | |
1675 | ||
f9a67b11 GJ |
1676 | if (bitmap->sysfs_can_clear) |
1677 | sysfs_put(bitmap->sysfs_can_clear); | |
1678 | ||
f9209a32 GR |
1679 | if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info && |
1680 | bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev)) | |
b97e9257 GR |
1681 | md_cluster_stop(bitmap->mddev); |
1682 | ||
fae7d326 N |
1683 | /* Shouldn't be needed - but just in case.... */ |
1684 | wait_event(bitmap->write_wait, | |
1685 | atomic_read(&bitmap->pending_writes) == 0); | |
1686 | ||
1687 | /* release the bitmap file */ | |
1688 | bitmap_file_unmap(&bitmap->storage); | |
32a7627c | 1689 | |
40cffcc0 N |
1690 | bp = bitmap->counts.bp; |
1691 | pages = bitmap->counts.pages; | |
32a7627c N |
1692 | |
1693 | /* free all allocated memory */ | |
1694 | ||
32a7627c N |
1695 | if (bp) /* deallocate the page memory */ |
1696 | for (k = 0; k < pages; k++) | |
1697 | if (bp[k].map && !bp[k].hijacked) | |
1698 | kfree(bp[k].map); | |
1699 | kfree(bp); | |
1700 | kfree(bitmap); | |
1701 | } | |
aa5cbd10 | 1702 | |
fd01b88c | 1703 | void bitmap_destroy(struct mddev *mddev) |
3178b0db N |
1704 | { |
1705 | struct bitmap *bitmap = mddev->bitmap; | |
1706 | ||
1707 | if (!bitmap) /* there was no bitmap */ | |
1708 | return; | |
1709 | ||
c3d9714e | 1710 | mutex_lock(&mddev->bitmap_info.mutex); |
978a7a47 | 1711 | spin_lock(&mddev->lock); |
3178b0db | 1712 | mddev->bitmap = NULL; /* disconnect from the md device */ |
978a7a47 | 1713 | spin_unlock(&mddev->lock); |
c3d9714e | 1714 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1715 | if (mddev->thread) |
1716 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db N |
1717 | |
1718 | bitmap_free(bitmap); | |
1719 | } | |
32a7627c N |
1720 | |
1721 | /* | |
1722 | * initialize the bitmap structure | |
1723 | * if this returns an error, bitmap_destroy must be called to do clean up | |
f9a67b11 | 1724 | * once mddev->bitmap is set |
32a7627c | 1725 | */ |
f9209a32 | 1726 | struct bitmap *bitmap_create(struct mddev *mddev, int slot) |
32a7627c N |
1727 | { |
1728 | struct bitmap *bitmap; | |
1f593903 | 1729 | sector_t blocks = mddev->resync_max_sectors; |
c3d9714e | 1730 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1731 | int err; |
324a56e1 | 1732 | struct kernfs_node *bm = NULL; |
32a7627c | 1733 | |
5f6e3c83 | 1734 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1735 | |
c3d9714e | 1736 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1737 | |
9ffae0cf | 1738 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c | 1739 | if (!bitmap) |
f9209a32 | 1740 | return ERR_PTR(-ENOMEM); |
32a7627c | 1741 | |
40cffcc0 | 1742 | spin_lock_init(&bitmap->counts.lock); |
ce25c31b N |
1743 | atomic_set(&bitmap->pending_writes, 0); |
1744 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1745 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1746 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1747 | |
32a7627c | 1748 | bitmap->mddev = mddev; |
f9209a32 | 1749 | bitmap->cluster_slot = slot; |
32a7627c | 1750 | |
5ff5afff | 1751 | if (mddev->kobj.sd) |
388975cc | 1752 | bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap"); |
ece5cff0 | 1753 | if (bm) { |
388975cc | 1754 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear"); |
ece5cff0 N |
1755 | sysfs_put(bm); |
1756 | } else | |
1757 | bitmap->sysfs_can_clear = NULL; | |
1758 | ||
1ec885cd | 1759 | bitmap->storage.file = file; |
ce25c31b N |
1760 | if (file) { |
1761 | get_file(file); | |
ae8fa283 N |
1762 | /* As future accesses to this file will use bmap, |
1763 | * and bypass the page cache, we must sync the file | |
1764 | * first. | |
1765 | */ | |
8018ab05 | 1766 | vfs_fsync(file, 1); |
ce25c31b | 1767 | } |
42a04b50 | 1768 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
9c81075f JB |
1769 | if (!mddev->bitmap_info.external) { |
1770 | /* | |
1771 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is | |
1772 | * instructing us to create a new on-disk bitmap instance. | |
1773 | */ | |
1774 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) | |
1775 | err = bitmap_new_disk_sb(bitmap); | |
1776 | else | |
1777 | err = bitmap_read_sb(bitmap); | |
1778 | } else { | |
ece5cff0 N |
1779 | err = 0; |
1780 | if (mddev->bitmap_info.chunksize == 0 || | |
1781 | mddev->bitmap_info.daemon_sleep == 0) | |
1782 | /* chunksize and time_base need to be | |
1783 | * set first. */ | |
1784 | err = -EINVAL; | |
1785 | } | |
32a7627c | 1786 | if (err) |
3178b0db | 1787 | goto error; |
32a7627c | 1788 | |
624ce4f5 | 1789 | bitmap->daemon_lastrun = jiffies; |
d60b479d N |
1790 | err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1); |
1791 | if (err) | |
3178b0db | 1792 | goto error; |
32a7627c | 1793 | |
69e51b44 | 1794 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", |
d60b479d | 1795 | bitmap->counts.pages, bmname(bitmap)); |
69e51b44 | 1796 | |
f9209a32 GR |
1797 | err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0; |
1798 | if (err) | |
1799 | goto error; | |
69e51b44 | 1800 | |
f9209a32 | 1801 | return bitmap; |
69e51b44 N |
1802 | error: |
1803 | bitmap_free(bitmap); | |
f9209a32 | 1804 | return ERR_PTR(err); |
69e51b44 N |
1805 | } |
1806 | ||
fd01b88c | 1807 | int bitmap_load(struct mddev *mddev) |
69e51b44 N |
1808 | { |
1809 | int err = 0; | |
3520fa4d | 1810 | sector_t start = 0; |
69e51b44 N |
1811 | sector_t sector = 0; |
1812 | struct bitmap *bitmap = mddev->bitmap; | |
1813 | ||
1814 | if (!bitmap) | |
1815 | goto out; | |
1816 | ||
1817 | /* Clear out old bitmap info first: Either there is none, or we | |
1818 | * are resuming after someone else has possibly changed things, | |
1819 | * so we should forget old cached info. | |
1820 | * All chunks should be clean, but some might need_sync. | |
1821 | */ | |
1822 | while (sector < mddev->resync_max_sectors) { | |
57dab0bd | 1823 | sector_t blocks; |
69e51b44 N |
1824 | bitmap_start_sync(bitmap, sector, &blocks, 0); |
1825 | sector += blocks; | |
1826 | } | |
1827 | bitmap_close_sync(bitmap); | |
1828 | ||
3520fa4d JB |
1829 | if (mddev->degraded == 0 |
1830 | || bitmap->events_cleared == mddev->events) | |
1831 | /* no need to keep dirty bits to optimise a | |
1832 | * re-add of a missing device */ | |
1833 | start = mddev->recovery_cp; | |
1834 | ||
afbaa90b | 1835 | mutex_lock(&mddev->bitmap_info.mutex); |
3520fa4d | 1836 | err = bitmap_init_from_disk(bitmap, start); |
afbaa90b | 1837 | mutex_unlock(&mddev->bitmap_info.mutex); |
3520fa4d | 1838 | |
32a7627c | 1839 | if (err) |
69e51b44 | 1840 | goto out; |
b405fe91 | 1841 | clear_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
1842 | |
1843 | /* Kick recovery in case any bits were set */ | |
1844 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
3178b0db | 1845 | |
1b04be96 | 1846 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1847 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1848 | |
4ad13663 N |
1849 | bitmap_update_sb(bitmap); |
1850 | ||
b405fe91 | 1851 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
69e51b44 N |
1852 | err = -EIO; |
1853 | out: | |
3178b0db | 1854 | return err; |
32a7627c | 1855 | } |
69e51b44 | 1856 | EXPORT_SYMBOL_GPL(bitmap_load); |
32a7627c | 1857 | |
11dd35da GR |
1858 | /* Loads the bitmap associated with slot and copies the resync information |
1859 | * to our bitmap | |
1860 | */ | |
1861 | int bitmap_copy_from_slot(struct mddev *mddev, int slot, | |
97f6cd39 | 1862 | sector_t *low, sector_t *high, bool clear_bits) |
11dd35da GR |
1863 | { |
1864 | int rv = 0, i, j; | |
1865 | sector_t block, lo = 0, hi = 0; | |
1866 | struct bitmap_counts *counts; | |
1867 | struct bitmap *bitmap = bitmap_create(mddev, slot); | |
1868 | ||
f9a67b11 GJ |
1869 | if (IS_ERR(bitmap)) { |
1870 | bitmap_free(bitmap); | |
11dd35da | 1871 | return PTR_ERR(bitmap); |
f9a67b11 | 1872 | } |
11dd35da | 1873 | |
11dd35da GR |
1874 | rv = bitmap_init_from_disk(bitmap, 0); |
1875 | if (rv) | |
1876 | goto err; | |
1877 | ||
1878 | counts = &bitmap->counts; | |
1879 | for (j = 0; j < counts->chunks; j++) { | |
1880 | block = (sector_t)j << counts->chunkshift; | |
1881 | if (bitmap_file_test_bit(bitmap, block)) { | |
1882 | if (!lo) | |
1883 | lo = block; | |
1884 | hi = block; | |
1885 | bitmap_file_clear_bit(bitmap, block); | |
1886 | bitmap_set_memory_bits(mddev->bitmap, block, 1); | |
1887 | bitmap_file_set_bit(mddev->bitmap, block); | |
1888 | } | |
1889 | } | |
1890 | ||
97f6cd39 GR |
1891 | if (clear_bits) { |
1892 | bitmap_update_sb(bitmap); | |
1893 | /* Setting this for the ev_page should be enough. | |
1894 | * And we do not require both write_all and PAGE_DIRT either | |
1895 | */ | |
1896 | for (i = 0; i < bitmap->storage.file_pages; i++) | |
1897 | set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); | |
1898 | bitmap_write_all(bitmap); | |
1899 | bitmap_unplug(bitmap); | |
1900 | } | |
11dd35da GR |
1901 | *low = lo; |
1902 | *high = hi; | |
1903 | err: | |
1904 | bitmap_free(bitmap); | |
1905 | return rv; | |
1906 | } | |
1907 | EXPORT_SYMBOL_GPL(bitmap_copy_from_slot); | |
1908 | ||
1909 | ||
57148964 N |
1910 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap) |
1911 | { | |
1912 | unsigned long chunk_kb; | |
40cffcc0 | 1913 | struct bitmap_counts *counts; |
57148964 N |
1914 | |
1915 | if (!bitmap) | |
1916 | return; | |
1917 | ||
40cffcc0 N |
1918 | counts = &bitmap->counts; |
1919 | ||
57148964 N |
1920 | chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; |
1921 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
1922 | "%lu%s chunk", | |
40cffcc0 N |
1923 | counts->pages - counts->missing_pages, |
1924 | counts->pages, | |
1925 | (counts->pages - counts->missing_pages) | |
57148964 N |
1926 | << (PAGE_SHIFT - 10), |
1927 | chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, | |
1928 | chunk_kb ? "KB" : "B"); | |
1ec885cd | 1929 | if (bitmap->storage.file) { |
57148964 | 1930 | seq_printf(seq, ", file: "); |
2726d566 | 1931 | seq_file_path(seq, bitmap->storage.file, " \t\n"); |
57148964 N |
1932 | } |
1933 | ||
1934 | seq_printf(seq, "\n"); | |
57148964 N |
1935 | } |
1936 | ||
d60b479d N |
1937 | int bitmap_resize(struct bitmap *bitmap, sector_t blocks, |
1938 | int chunksize, int init) | |
1939 | { | |
1940 | /* If chunk_size is 0, choose an appropriate chunk size. | |
1941 | * Then possibly allocate new storage space. | |
1942 | * Then quiesce, copy bits, replace bitmap, and re-start | |
1943 | * | |
1944 | * This function is called both to set up the initial bitmap | |
1945 | * and to resize the bitmap while the array is active. | |
1946 | * If this happens as a result of the array being resized, | |
1947 | * chunksize will be zero, and we need to choose a suitable | |
1948 | * chunksize, otherwise we use what we are given. | |
1949 | */ | |
1950 | struct bitmap_storage store; | |
1951 | struct bitmap_counts old_counts; | |
1952 | unsigned long chunks; | |
1953 | sector_t block; | |
1954 | sector_t old_blocks, new_blocks; | |
1955 | int chunkshift; | |
1956 | int ret = 0; | |
1957 | long pages; | |
1958 | struct bitmap_page *new_bp; | |
1959 | ||
1960 | if (chunksize == 0) { | |
1961 | /* If there is enough space, leave the chunk size unchanged, | |
1962 | * else increase by factor of two until there is enough space. | |
1963 | */ | |
1964 | long bytes; | |
1965 | long space = bitmap->mddev->bitmap_info.space; | |
1966 | ||
1967 | if (space == 0) { | |
1968 | /* We don't know how much space there is, so limit | |
1969 | * to current size - in sectors. | |
1970 | */ | |
1971 | bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8); | |
1972 | if (!bitmap->mddev->bitmap_info.external) | |
1973 | bytes += sizeof(bitmap_super_t); | |
1974 | space = DIV_ROUND_UP(bytes, 512); | |
1975 | bitmap->mddev->bitmap_info.space = space; | |
1976 | } | |
1977 | chunkshift = bitmap->counts.chunkshift; | |
1978 | chunkshift--; | |
1979 | do { | |
1980 | /* 'chunkshift' is shift from block size to chunk size */ | |
1981 | chunkshift++; | |
1982 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
1983 | bytes = DIV_ROUND_UP(chunks, 8); | |
1984 | if (!bitmap->mddev->bitmap_info.external) | |
1985 | bytes += sizeof(bitmap_super_t); | |
1986 | } while (bytes > (space << 9)); | |
1987 | } else | |
1988 | chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT; | |
1989 | ||
1990 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
1991 | memset(&store, 0, sizeof(store)); | |
1992 | if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file) | |
1993 | ret = bitmap_storage_alloc(&store, chunks, | |
b97e9257 | 1994 | !bitmap->mddev->bitmap_info.external, |
da6fb7a9 N |
1995 | mddev_is_clustered(bitmap->mddev) |
1996 | ? bitmap->cluster_slot : 0); | |
d60b479d N |
1997 | if (ret) |
1998 | goto err; | |
1999 | ||
2000 | pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO); | |
2001 | ||
2002 | new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL); | |
2003 | ret = -ENOMEM; | |
2004 | if (!new_bp) { | |
2005 | bitmap_file_unmap(&store); | |
2006 | goto err; | |
2007 | } | |
2008 | ||
2009 | if (!init) | |
2010 | bitmap->mddev->pers->quiesce(bitmap->mddev, 1); | |
2011 | ||
2012 | store.file = bitmap->storage.file; | |
2013 | bitmap->storage.file = NULL; | |
2014 | ||
2015 | if (store.sb_page && bitmap->storage.sb_page) | |
2016 | memcpy(page_address(store.sb_page), | |
2017 | page_address(bitmap->storage.sb_page), | |
2018 | sizeof(bitmap_super_t)); | |
2019 | bitmap_file_unmap(&bitmap->storage); | |
2020 | bitmap->storage = store; | |
2021 | ||
2022 | old_counts = bitmap->counts; | |
2023 | bitmap->counts.bp = new_bp; | |
2024 | bitmap->counts.pages = pages; | |
2025 | bitmap->counts.missing_pages = pages; | |
2026 | bitmap->counts.chunkshift = chunkshift; | |
2027 | bitmap->counts.chunks = chunks; | |
2028 | bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift + | |
2029 | BITMAP_BLOCK_SHIFT); | |
2030 | ||
2031 | blocks = min(old_counts.chunks << old_counts.chunkshift, | |
2032 | chunks << chunkshift); | |
2033 | ||
2034 | spin_lock_irq(&bitmap->counts.lock); | |
2035 | for (block = 0; block < blocks; ) { | |
2036 | bitmap_counter_t *bmc_old, *bmc_new; | |
2037 | int set; | |
2038 | ||
2039 | bmc_old = bitmap_get_counter(&old_counts, block, | |
2040 | &old_blocks, 0); | |
2041 | set = bmc_old && NEEDED(*bmc_old); | |
2042 | ||
2043 | if (set) { | |
2044 | bmc_new = bitmap_get_counter(&bitmap->counts, block, | |
2045 | &new_blocks, 1); | |
2046 | if (*bmc_new == 0) { | |
2047 | /* need to set on-disk bits too. */ | |
2048 | sector_t end = block + new_blocks; | |
2049 | sector_t start = block >> chunkshift; | |
2050 | start <<= chunkshift; | |
2051 | while (start < end) { | |
2052 | bitmap_file_set_bit(bitmap, block); | |
2053 | start += 1 << chunkshift; | |
2054 | } | |
2055 | *bmc_new = 2; | |
2056 | bitmap_count_page(&bitmap->counts, | |
2057 | block, 1); | |
2058 | bitmap_set_pending(&bitmap->counts, | |
2059 | block); | |
2060 | } | |
2061 | *bmc_new |= NEEDED_MASK; | |
2062 | if (new_blocks < old_blocks) | |
2063 | old_blocks = new_blocks; | |
2064 | } | |
2065 | block += old_blocks; | |
2066 | } | |
2067 | ||
2068 | if (!init) { | |
2069 | int i; | |
2070 | while (block < (chunks << chunkshift)) { | |
2071 | bitmap_counter_t *bmc; | |
2072 | bmc = bitmap_get_counter(&bitmap->counts, block, | |
2073 | &new_blocks, 1); | |
2074 | if (bmc) { | |
2075 | /* new space. It needs to be resynced, so | |
2076 | * we set NEEDED_MASK. | |
2077 | */ | |
2078 | if (*bmc == 0) { | |
2079 | *bmc = NEEDED_MASK | 2; | |
2080 | bitmap_count_page(&bitmap->counts, | |
2081 | block, 1); | |
2082 | bitmap_set_pending(&bitmap->counts, | |
2083 | block); | |
2084 | } | |
2085 | } | |
2086 | block += new_blocks; | |
2087 | } | |
2088 | for (i = 0; i < bitmap->storage.file_pages; i++) | |
2089 | set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); | |
2090 | } | |
2091 | spin_unlock_irq(&bitmap->counts.lock); | |
2092 | ||
2093 | if (!init) { | |
2094 | bitmap_unplug(bitmap); | |
2095 | bitmap->mddev->pers->quiesce(bitmap->mddev, 0); | |
2096 | } | |
2097 | ret = 0; | |
2098 | err: | |
2099 | return ret; | |
2100 | } | |
2101 | EXPORT_SYMBOL_GPL(bitmap_resize); | |
2102 | ||
43a70507 | 2103 | static ssize_t |
fd01b88c | 2104 | location_show(struct mddev *mddev, char *page) |
43a70507 N |
2105 | { |
2106 | ssize_t len; | |
ac2f40be | 2107 | if (mddev->bitmap_info.file) |
43a70507 | 2108 | len = sprintf(page, "file"); |
ac2f40be | 2109 | else if (mddev->bitmap_info.offset) |
43a70507 | 2110 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 2111 | else |
43a70507 N |
2112 | len = sprintf(page, "none"); |
2113 | len += sprintf(page+len, "\n"); | |
2114 | return len; | |
2115 | } | |
2116 | ||
2117 | static ssize_t | |
fd01b88c | 2118 | location_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2119 | { |
2120 | ||
2121 | if (mddev->pers) { | |
2122 | if (!mddev->pers->quiesce) | |
2123 | return -EBUSY; | |
2124 | if (mddev->recovery || mddev->sync_thread) | |
2125 | return -EBUSY; | |
2126 | } | |
2127 | ||
2128 | if (mddev->bitmap || mddev->bitmap_info.file || | |
2129 | mddev->bitmap_info.offset) { | |
2130 | /* bitmap already configured. Only option is to clear it */ | |
2131 | if (strncmp(buf, "none", 4) != 0) | |
2132 | return -EBUSY; | |
2133 | if (mddev->pers) { | |
2134 | mddev->pers->quiesce(mddev, 1); | |
2135 | bitmap_destroy(mddev); | |
2136 | mddev->pers->quiesce(mddev, 0); | |
2137 | } | |
2138 | mddev->bitmap_info.offset = 0; | |
2139 | if (mddev->bitmap_info.file) { | |
2140 | struct file *f = mddev->bitmap_info.file; | |
2141 | mddev->bitmap_info.file = NULL; | |
43a70507 N |
2142 | fput(f); |
2143 | } | |
2144 | } else { | |
2145 | /* No bitmap, OK to set a location */ | |
2146 | long long offset; | |
2147 | if (strncmp(buf, "none", 4) == 0) | |
2148 | /* nothing to be done */; | |
2149 | else if (strncmp(buf, "file:", 5) == 0) { | |
2150 | /* Not supported yet */ | |
2151 | return -EINVAL; | |
2152 | } else { | |
2153 | int rv; | |
2154 | if (buf[0] == '+') | |
b29bebd6 | 2155 | rv = kstrtoll(buf+1, 10, &offset); |
43a70507 | 2156 | else |
b29bebd6 | 2157 | rv = kstrtoll(buf, 10, &offset); |
43a70507 N |
2158 | if (rv) |
2159 | return rv; | |
2160 | if (offset == 0) | |
2161 | return -EINVAL; | |
ece5cff0 N |
2162 | if (mddev->bitmap_info.external == 0 && |
2163 | mddev->major_version == 0 && | |
43a70507 N |
2164 | offset != mddev->bitmap_info.default_offset) |
2165 | return -EINVAL; | |
2166 | mddev->bitmap_info.offset = offset; | |
2167 | if (mddev->pers) { | |
f9209a32 | 2168 | struct bitmap *bitmap; |
43a70507 | 2169 | mddev->pers->quiesce(mddev, 1); |
f9209a32 GR |
2170 | bitmap = bitmap_create(mddev, -1); |
2171 | if (IS_ERR(bitmap)) | |
2172 | rv = PTR_ERR(bitmap); | |
2173 | else { | |
2174 | mddev->bitmap = bitmap; | |
4474ca42 | 2175 | rv = bitmap_load(mddev); |
f9a67b11 | 2176 | if (rv) |
f9209a32 | 2177 | mddev->bitmap_info.offset = 0; |
43a70507 N |
2178 | } |
2179 | mddev->pers->quiesce(mddev, 0); | |
f9a67b11 GJ |
2180 | if (rv) { |
2181 | bitmap_destroy(mddev); | |
43a70507 | 2182 | return rv; |
f9a67b11 | 2183 | } |
43a70507 N |
2184 | } |
2185 | } | |
2186 | } | |
2187 | if (!mddev->external) { | |
2188 | /* Ensure new bitmap info is stored in | |
2189 | * metadata promptly. | |
2190 | */ | |
2191 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2192 | md_wakeup_thread(mddev->thread); | |
2193 | } | |
2194 | return len; | |
2195 | } | |
2196 | ||
2197 | static struct md_sysfs_entry bitmap_location = | |
2198 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
2199 | ||
6409bb05 N |
2200 | /* 'bitmap/space' is the space available at 'location' for the |
2201 | * bitmap. This allows the kernel to know when it is safe to | |
2202 | * resize the bitmap to match a resized array. | |
2203 | */ | |
2204 | static ssize_t | |
2205 | space_show(struct mddev *mddev, char *page) | |
2206 | { | |
2207 | return sprintf(page, "%lu\n", mddev->bitmap_info.space); | |
2208 | } | |
2209 | ||
2210 | static ssize_t | |
2211 | space_store(struct mddev *mddev, const char *buf, size_t len) | |
2212 | { | |
2213 | unsigned long sectors; | |
2214 | int rv; | |
2215 | ||
2216 | rv = kstrtoul(buf, 10, §ors); | |
2217 | if (rv) | |
2218 | return rv; | |
2219 | ||
2220 | if (sectors == 0) | |
2221 | return -EINVAL; | |
2222 | ||
2223 | if (mddev->bitmap && | |
9b1215c1 | 2224 | sectors < (mddev->bitmap->storage.bytes + 511) >> 9) |
6409bb05 N |
2225 | return -EFBIG; /* Bitmap is too big for this small space */ |
2226 | ||
2227 | /* could make sure it isn't too big, but that isn't really | |
2228 | * needed - user-space should be careful. | |
2229 | */ | |
2230 | mddev->bitmap_info.space = sectors; | |
2231 | return len; | |
2232 | } | |
2233 | ||
2234 | static struct md_sysfs_entry bitmap_space = | |
2235 | __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); | |
2236 | ||
43a70507 | 2237 | static ssize_t |
fd01b88c | 2238 | timeout_show(struct mddev *mddev, char *page) |
43a70507 N |
2239 | { |
2240 | ssize_t len; | |
2241 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
2242 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 2243 | |
43a70507 N |
2244 | len = sprintf(page, "%lu", secs); |
2245 | if (jifs) | |
2246 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
2247 | len += sprintf(page+len, "\n"); | |
2248 | return len; | |
2249 | } | |
2250 | ||
2251 | static ssize_t | |
fd01b88c | 2252 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2253 | { |
2254 | /* timeout can be set at any time */ | |
2255 | unsigned long timeout; | |
2256 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
2257 | if (rv) | |
2258 | return rv; | |
2259 | ||
2260 | /* just to make sure we don't overflow... */ | |
2261 | if (timeout >= LONG_MAX / HZ) | |
2262 | return -EINVAL; | |
2263 | ||
2264 | timeout = timeout * HZ / 10000; | |
2265 | ||
2266 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
2267 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
2268 | if (timeout < 1) | |
2269 | timeout = 1; | |
2270 | mddev->bitmap_info.daemon_sleep = timeout; | |
2271 | if (mddev->thread) { | |
2272 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
2273 | * the bitmap is all clean and we don't need to | |
2274 | * adjust the timeout right now | |
2275 | */ | |
2276 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2277 | mddev->thread->timeout = timeout; | |
2278 | md_wakeup_thread(mddev->thread); | |
2279 | } | |
2280 | } | |
2281 | return len; | |
2282 | } | |
2283 | ||
2284 | static struct md_sysfs_entry bitmap_timeout = | |
2285 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2286 | ||
2287 | static ssize_t | |
fd01b88c | 2288 | backlog_show(struct mddev *mddev, char *page) |
43a70507 N |
2289 | { |
2290 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2291 | } | |
2292 | ||
2293 | static ssize_t | |
fd01b88c | 2294 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2295 | { |
2296 | unsigned long backlog; | |
b29bebd6 | 2297 | int rv = kstrtoul(buf, 10, &backlog); |
43a70507 N |
2298 | if (rv) |
2299 | return rv; | |
2300 | if (backlog > COUNTER_MAX) | |
2301 | return -EINVAL; | |
2302 | mddev->bitmap_info.max_write_behind = backlog; | |
2303 | return len; | |
2304 | } | |
2305 | ||
2306 | static struct md_sysfs_entry bitmap_backlog = | |
2307 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2308 | ||
2309 | static ssize_t | |
fd01b88c | 2310 | chunksize_show(struct mddev *mddev, char *page) |
43a70507 N |
2311 | { |
2312 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2313 | } | |
2314 | ||
2315 | static ssize_t | |
fd01b88c | 2316 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2317 | { |
2318 | /* Can only be changed when no bitmap is active */ | |
2319 | int rv; | |
2320 | unsigned long csize; | |
2321 | if (mddev->bitmap) | |
2322 | return -EBUSY; | |
b29bebd6 | 2323 | rv = kstrtoul(buf, 10, &csize); |
43a70507 N |
2324 | if (rv) |
2325 | return rv; | |
2326 | if (csize < 512 || | |
2327 | !is_power_of_2(csize)) | |
2328 | return -EINVAL; | |
2329 | mddev->bitmap_info.chunksize = csize; | |
2330 | return len; | |
2331 | } | |
2332 | ||
2333 | static struct md_sysfs_entry bitmap_chunksize = | |
2334 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2335 | ||
fd01b88c | 2336 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
ece5cff0 | 2337 | { |
c4ce867f GR |
2338 | if (mddev_is_clustered(mddev)) |
2339 | return sprintf(page, "clustered\n"); | |
ece5cff0 N |
2340 | return sprintf(page, "%s\n", (mddev->bitmap_info.external |
2341 | ? "external" : "internal")); | |
2342 | } | |
2343 | ||
fd01b88c | 2344 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2345 | { |
2346 | if (mddev->bitmap || | |
2347 | mddev->bitmap_info.file || | |
2348 | mddev->bitmap_info.offset) | |
2349 | return -EBUSY; | |
2350 | if (strncmp(buf, "external", 8) == 0) | |
2351 | mddev->bitmap_info.external = 1; | |
c4ce867f GR |
2352 | else if ((strncmp(buf, "internal", 8) == 0) || |
2353 | (strncmp(buf, "clustered", 9) == 0)) | |
ece5cff0 N |
2354 | mddev->bitmap_info.external = 0; |
2355 | else | |
2356 | return -EINVAL; | |
2357 | return len; | |
2358 | } | |
2359 | ||
2360 | static struct md_sysfs_entry bitmap_metadata = | |
2361 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2362 | ||
fd01b88c | 2363 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2364 | { |
2365 | int len; | |
b7b17c9b | 2366 | spin_lock(&mddev->lock); |
ece5cff0 N |
2367 | if (mddev->bitmap) |
2368 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2369 | "false" : "true")); | |
2370 | else | |
2371 | len = sprintf(page, "\n"); | |
b7b17c9b | 2372 | spin_unlock(&mddev->lock); |
ece5cff0 N |
2373 | return len; |
2374 | } | |
2375 | ||
fd01b88c | 2376 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2377 | { |
2378 | if (mddev->bitmap == NULL) | |
2379 | return -ENOENT; | |
2380 | if (strncmp(buf, "false", 5) == 0) | |
2381 | mddev->bitmap->need_sync = 1; | |
2382 | else if (strncmp(buf, "true", 4) == 0) { | |
2383 | if (mddev->degraded) | |
2384 | return -EBUSY; | |
2385 | mddev->bitmap->need_sync = 0; | |
2386 | } else | |
2387 | return -EINVAL; | |
2388 | return len; | |
2389 | } | |
2390 | ||
2391 | static struct md_sysfs_entry bitmap_can_clear = | |
2392 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2393 | ||
696fcd53 | 2394 | static ssize_t |
fd01b88c | 2395 | behind_writes_used_show(struct mddev *mddev, char *page) |
696fcd53 | 2396 | { |
b7b17c9b N |
2397 | ssize_t ret; |
2398 | spin_lock(&mddev->lock); | |
696fcd53 | 2399 | if (mddev->bitmap == NULL) |
b7b17c9b N |
2400 | ret = sprintf(page, "0\n"); |
2401 | else | |
2402 | ret = sprintf(page, "%lu\n", | |
2403 | mddev->bitmap->behind_writes_used); | |
2404 | spin_unlock(&mddev->lock); | |
2405 | return ret; | |
696fcd53 PC |
2406 | } |
2407 | ||
2408 | static ssize_t | |
fd01b88c | 2409 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
696fcd53 PC |
2410 | { |
2411 | if (mddev->bitmap) | |
2412 | mddev->bitmap->behind_writes_used = 0; | |
2413 | return len; | |
2414 | } | |
2415 | ||
2416 | static struct md_sysfs_entry max_backlog_used = | |
2417 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2418 | behind_writes_used_show, behind_writes_used_reset); | |
2419 | ||
43a70507 N |
2420 | static struct attribute *md_bitmap_attrs[] = { |
2421 | &bitmap_location.attr, | |
6409bb05 | 2422 | &bitmap_space.attr, |
43a70507 N |
2423 | &bitmap_timeout.attr, |
2424 | &bitmap_backlog.attr, | |
2425 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2426 | &bitmap_metadata.attr, |
2427 | &bitmap_can_clear.attr, | |
696fcd53 | 2428 | &max_backlog_used.attr, |
43a70507 N |
2429 | NULL |
2430 | }; | |
2431 | struct attribute_group md_bitmap_group = { | |
2432 | .name = "bitmap", | |
2433 | .attrs = md_bitmap_attrs, | |
2434 | }; | |
2435 |