]>
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 | */ | |
ac2f40be N |
48 | static int bitmap_checkpage(struct bitmap *bitmap, |
49 | unsigned long page, int create) | |
ee305ace N |
50 | __releases(bitmap->lock) |
51 | __acquires(bitmap->lock) | |
32a7627c N |
52 | { |
53 | unsigned char *mappage; | |
54 | ||
55 | if (page >= bitmap->pages) { | |
1187cf0a N |
56 | /* This can happen if bitmap_start_sync goes beyond |
57 | * End-of-device while looking for a whole page. | |
58 | * It is harmless. | |
59 | */ | |
32a7627c N |
60 | return -EINVAL; |
61 | } | |
62 | ||
32a7627c N |
63 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ |
64 | return 0; | |
65 | ||
66 | if (bitmap->bp[page].map) /* page is already allocated, just return */ | |
67 | return 0; | |
68 | ||
69 | if (!create) | |
70 | return -ENOENT; | |
71 | ||
32a7627c N |
72 | /* this page has not been allocated yet */ |
73 | ||
ac2f40be | 74 | spin_unlock_irq(&bitmap->lock); |
792a1d4b | 75 | mappage = kzalloc(PAGE_SIZE, GFP_NOIO); |
ac2f40be N |
76 | spin_lock_irq(&bitmap->lock); |
77 | ||
78 | if (mappage == NULL) { | |
36a4e1fe N |
79 | pr_debug("%s: bitmap map page allocation failed, hijacking\n", |
80 | bmname(bitmap)); | |
32a7627c N |
81 | /* failed - set the hijacked flag so that we can use the |
82 | * pointer as a counter */ | |
32a7627c N |
83 | if (!bitmap->bp[page].map) |
84 | bitmap->bp[page].hijacked = 1; | |
ac2f40be N |
85 | } else if (bitmap->bp[page].map || |
86 | bitmap->bp[page].hijacked) { | |
32a7627c | 87 | /* somebody beat us to getting the page */ |
792a1d4b | 88 | kfree(mappage); |
32a7627c | 89 | return 0; |
ac2f40be | 90 | } else { |
32a7627c | 91 | |
ac2f40be | 92 | /* no page was in place and we have one, so install it */ |
32a7627c | 93 | |
ac2f40be N |
94 | bitmap->bp[page].map = mappage; |
95 | bitmap->missing_pages--; | |
96 | } | |
32a7627c N |
97 | return 0; |
98 | } | |
99 | ||
32a7627c N |
100 | /* if page is completely empty, put it back on the free list, or dealloc it */ |
101 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
102 | /* Note: lock should be held when calling this */ | |
858119e1 | 103 | static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page) |
32a7627c N |
104 | { |
105 | char *ptr; | |
106 | ||
107 | if (bitmap->bp[page].count) /* page is still busy */ | |
108 | return; | |
109 | ||
110 | /* page is no longer in use, it can be released */ | |
111 | ||
112 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
113 | bitmap->bp[page].hijacked = 0; | |
114 | bitmap->bp[page].map = NULL; | |
ac2f40be N |
115 | } else { |
116 | /* normal case, free the page */ | |
117 | ptr = bitmap->bp[page].map; | |
118 | bitmap->bp[page].map = NULL; | |
119 | bitmap->missing_pages++; | |
792a1d4b | 120 | kfree(ptr); |
32a7627c | 121 | } |
32a7627c N |
122 | } |
123 | ||
32a7627c N |
124 | /* |
125 | * bitmap file handling - read and write the bitmap file and its superblock | |
126 | */ | |
127 | ||
32a7627c N |
128 | /* |
129 | * basic page I/O operations | |
130 | */ | |
131 | ||
a654b9d8 | 132 | /* IO operations when bitmap is stored near all superblocks */ |
27581e5a N |
133 | static int read_sb_page(struct mddev *mddev, loff_t offset, |
134 | struct page *page, | |
135 | unsigned long index, int size) | |
a654b9d8 N |
136 | { |
137 | /* choose a good rdev and read the page from there */ | |
138 | ||
3cb03002 | 139 | struct md_rdev *rdev; |
a654b9d8 | 140 | sector_t target; |
a654b9d8 | 141 | |
dafb20fa | 142 | rdev_for_each(rdev, mddev) { |
b2d444d7 N |
143 | if (! test_bit(In_sync, &rdev->flags) |
144 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
145 | continue; |
146 | ||
ccebd4c4 | 147 | target = offset + index * (PAGE_SIZE/512); |
a654b9d8 | 148 | |
2b193363 | 149 | if (sync_page_io(rdev, target, |
e1defc4f | 150 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
ccebd4c4 | 151 | page, READ, true)) { |
ab904d63 | 152 | page->index = index; |
27581e5a | 153 | return 0; |
ab904d63 N |
154 | } |
155 | } | |
27581e5a | 156 | return -EIO; |
a654b9d8 N |
157 | } |
158 | ||
fd01b88c | 159 | static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) |
b2d2c4ce N |
160 | { |
161 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
162 | * linked list, and raising the refcount of devices we return to ensure | |
163 | * they don't disappear while in use. | |
164 | * As devices are only added or removed when raid_disk is < 0 and | |
165 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
166 | * still be in the same position on the list when we re-enter | |
167 | * list_for_each_continue_rcu. | |
168 | */ | |
169 | struct list_head *pos; | |
170 | rcu_read_lock(); | |
171 | if (rdev == NULL) | |
172 | /* start at the beginning */ | |
173 | pos = &mddev->disks; | |
174 | else { | |
175 | /* release the previous rdev and start from there. */ | |
176 | rdev_dec_pending(rdev, mddev); | |
177 | pos = &rdev->same_set; | |
178 | } | |
179 | list_for_each_continue_rcu(pos, &mddev->disks) { | |
3cb03002 | 180 | rdev = list_entry(pos, struct md_rdev, same_set); |
b2d2c4ce | 181 | if (rdev->raid_disk >= 0 && |
b2d2c4ce N |
182 | !test_bit(Faulty, &rdev->flags)) { |
183 | /* this is a usable devices */ | |
184 | atomic_inc(&rdev->nr_pending); | |
185 | rcu_read_unlock(); | |
186 | return rdev; | |
187 | } | |
188 | } | |
189 | rcu_read_unlock(); | |
190 | return NULL; | |
191 | } | |
192 | ||
ab6085c7 | 193 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 194 | { |
3cb03002 | 195 | struct md_rdev *rdev = NULL; |
a6ff7e08 | 196 | struct block_device *bdev; |
fd01b88c | 197 | struct mddev *mddev = bitmap->mddev; |
1ec885cd | 198 | struct bitmap_storage *store = &bitmap->storage; |
a654b9d8 | 199 | |
b2d2c4ce | 200 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ac2f40be N |
201 | int size = PAGE_SIZE; |
202 | loff_t offset = mddev->bitmap_info.offset; | |
a6ff7e08 JB |
203 | |
204 | bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; | |
205 | ||
9b1215c1 N |
206 | if (page->index == store->file_pages-1) { |
207 | int last_page_size = store->bytes & (PAGE_SIZE-1); | |
208 | if (last_page_size == 0) | |
209 | last_page_size = PAGE_SIZE; | |
210 | size = roundup(last_page_size, | |
a6ff7e08 | 211 | bdev_logical_block_size(bdev)); |
9b1215c1 | 212 | } |
ac2f40be N |
213 | /* Just make sure we aren't corrupting data or |
214 | * metadata | |
215 | */ | |
216 | if (mddev->external) { | |
217 | /* Bitmap could be anywhere. */ | |
218 | if (rdev->sb_start + offset + (page->index | |
219 | * (PAGE_SIZE/512)) | |
220 | > rdev->data_offset | |
221 | && | |
222 | rdev->sb_start + offset | |
223 | < (rdev->data_offset + mddev->dev_sectors | |
224 | + (PAGE_SIZE/512))) | |
225 | goto bad_alignment; | |
226 | } else if (offset < 0) { | |
227 | /* DATA BITMAP METADATA */ | |
228 | if (offset | |
229 | + (long)(page->index * (PAGE_SIZE/512)) | |
230 | + size/512 > 0) | |
231 | /* bitmap runs in to metadata */ | |
232 | goto bad_alignment; | |
233 | if (rdev->data_offset + mddev->dev_sectors | |
234 | > rdev->sb_start + offset) | |
235 | /* data runs in to bitmap */ | |
236 | goto bad_alignment; | |
237 | } else if (rdev->sb_start < rdev->data_offset) { | |
238 | /* METADATA BITMAP DATA */ | |
239 | if (rdev->sb_start | |
240 | + offset | |
241 | + page->index*(PAGE_SIZE/512) + size/512 | |
242 | > rdev->data_offset) | |
243 | /* bitmap runs in to data */ | |
244 | goto bad_alignment; | |
245 | } else { | |
246 | /* DATA METADATA BITMAP - no problems */ | |
247 | } | |
248 | md_super_write(mddev, rdev, | |
249 | rdev->sb_start + offset | |
250 | + page->index * (PAGE_SIZE/512), | |
251 | size, | |
252 | page); | |
b2d2c4ce | 253 | } |
a654b9d8 N |
254 | |
255 | if (wait) | |
a9701a30 | 256 | md_super_wait(mddev); |
a654b9d8 | 257 | return 0; |
4b80991c N |
258 | |
259 | bad_alignment: | |
4b80991c | 260 | return -EINVAL; |
a654b9d8 N |
261 | } |
262 | ||
4ad13663 | 263 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 264 | /* |
a654b9d8 | 265 | * write out a page to a file |
32a7627c | 266 | */ |
4ad13663 | 267 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 268 | { |
d785a06a | 269 | struct buffer_head *bh; |
32a7627c | 270 | |
1ec885cd | 271 | if (bitmap->storage.file == NULL) { |
f0d76d70 N |
272 | switch (write_sb_page(bitmap, page, wait)) { |
273 | case -EINVAL: | |
274 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
f0d76d70 | 275 | } |
4ad13663 | 276 | } else { |
a654b9d8 | 277 | |
4ad13663 | 278 | bh = page_buffers(page); |
c708443c | 279 | |
4ad13663 N |
280 | while (bh && bh->b_blocknr) { |
281 | atomic_inc(&bitmap->pending_writes); | |
282 | set_buffer_locked(bh); | |
283 | set_buffer_mapped(bh); | |
721a9602 | 284 | submit_bh(WRITE | REQ_SYNC, bh); |
4ad13663 N |
285 | bh = bh->b_this_page; |
286 | } | |
d785a06a | 287 | |
ac2f40be | 288 | if (wait) |
4ad13663 N |
289 | wait_event(bitmap->write_wait, |
290 | atomic_read(&bitmap->pending_writes)==0); | |
8a5e9cf1 | 291 | } |
4ad13663 N |
292 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
293 | bitmap_file_kick(bitmap); | |
d785a06a N |
294 | } |
295 | ||
296 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
297 | { | |
298 | struct bitmap *bitmap = bh->b_private; | |
299 | unsigned long flags; | |
32a7627c | 300 | |
d785a06a N |
301 | if (!uptodate) { |
302 | spin_lock_irqsave(&bitmap->lock, flags); | |
303 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
304 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
32a7627c | 305 | } |
d785a06a N |
306 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
307 | wake_up(&bitmap->write_wait); | |
308 | } | |
32a7627c | 309 | |
d785a06a N |
310 | /* copied from buffer.c */ |
311 | static void | |
312 | __clear_page_buffers(struct page *page) | |
313 | { | |
314 | ClearPagePrivate(page); | |
315 | set_page_private(page, 0); | |
316 | page_cache_release(page); | |
317 | } | |
318 | static void free_buffers(struct page *page) | |
319 | { | |
27581e5a | 320 | struct buffer_head *bh; |
77ad4bc7 | 321 | |
27581e5a N |
322 | if (!PagePrivate(page)) |
323 | return; | |
324 | ||
325 | bh = page_buffers(page); | |
d785a06a N |
326 | while (bh) { |
327 | struct buffer_head *next = bh->b_this_page; | |
328 | free_buffer_head(bh); | |
329 | bh = next; | |
77ad4bc7 | 330 | } |
d785a06a N |
331 | __clear_page_buffers(page); |
332 | put_page(page); | |
32a7627c N |
333 | } |
334 | ||
d785a06a N |
335 | /* read a page from a file. |
336 | * We both read the page, and attach buffers to the page to record the | |
337 | * address of each block (using bmap). These addresses will be used | |
338 | * to write the block later, completely bypassing the filesystem. | |
339 | * This usage is similar to how swap files are handled, and allows us | |
340 | * to write to a file with no concerns of memory allocation failing. | |
341 | */ | |
27581e5a N |
342 | static int read_page(struct file *file, unsigned long index, |
343 | struct bitmap *bitmap, | |
344 | unsigned long count, | |
345 | struct page *page) | |
32a7627c | 346 | { |
27581e5a | 347 | int ret = 0; |
c649bb9c | 348 | struct inode *inode = file->f_path.dentry->d_inode; |
d785a06a N |
349 | struct buffer_head *bh; |
350 | sector_t block; | |
32a7627c | 351 | |
36a4e1fe N |
352 | pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
353 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 354 | |
d785a06a N |
355 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
356 | if (!bh) { | |
27581e5a | 357 | ret = -ENOMEM; |
32a7627c N |
358 | goto out; |
359 | } | |
d785a06a N |
360 | attach_page_buffers(page, bh); |
361 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
362 | while (bh) { | |
363 | if (count == 0) | |
364 | bh->b_blocknr = 0; | |
365 | else { | |
366 | bh->b_blocknr = bmap(inode, block); | |
367 | if (bh->b_blocknr == 0) { | |
368 | /* Cannot use this file! */ | |
27581e5a | 369 | ret = -EINVAL; |
d785a06a N |
370 | goto out; |
371 | } | |
372 | bh->b_bdev = inode->i_sb->s_bdev; | |
373 | if (count < (1<<inode->i_blkbits)) | |
374 | count = 0; | |
375 | else | |
376 | count -= (1<<inode->i_blkbits); | |
377 | ||
378 | bh->b_end_io = end_bitmap_write; | |
379 | bh->b_private = bitmap; | |
ce25c31b N |
380 | atomic_inc(&bitmap->pending_writes); |
381 | set_buffer_locked(bh); | |
382 | set_buffer_mapped(bh); | |
383 | submit_bh(READ, bh); | |
d785a06a N |
384 | } |
385 | block++; | |
386 | bh = bh->b_this_page; | |
387 | } | |
d785a06a | 388 | page->index = index; |
ce25c31b N |
389 | |
390 | wait_event(bitmap->write_wait, | |
391 | atomic_read(&bitmap->pending_writes)==0); | |
27581e5a N |
392 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
393 | ret = -EIO; | |
32a7627c | 394 | out: |
27581e5a N |
395 | if (ret) |
396 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n", | |
2d1f3b5d N |
397 | (int)PAGE_SIZE, |
398 | (unsigned long long)index << PAGE_SHIFT, | |
27581e5a N |
399 | ret); |
400 | return ret; | |
32a7627c N |
401 | } |
402 | ||
403 | /* | |
404 | * bitmap file superblock operations | |
405 | */ | |
406 | ||
407 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 408 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
409 | { |
410 | bitmap_super_t *sb; | |
32a7627c N |
411 | |
412 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 413 | return; |
ece5cff0 N |
414 | if (bitmap->mddev->bitmap_info.external) |
415 | return; | |
1ec885cd | 416 | if (!bitmap->storage.sb_page) /* no superblock */ |
4ad13663 | 417 | return; |
1ec885cd | 418 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 419 | sb->events = cpu_to_le64(bitmap->mddev->events); |
8258c532 | 420 | if (bitmap->mddev->events < bitmap->events_cleared) |
a0da84f3 NB |
421 | /* rocking back to read-only */ |
422 | bitmap->events_cleared = bitmap->mddev->events; | |
8258c532 N |
423 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); |
424 | sb->state = cpu_to_le32(bitmap->flags); | |
43a70507 N |
425 | /* Just in case these have been changed via sysfs: */ |
426 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
427 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
b2f46e68 | 428 | kunmap_atomic(sb); |
1ec885cd | 429 | write_page(bitmap, bitmap->storage.sb_page, 1); |
32a7627c N |
430 | } |
431 | ||
432 | /* print out the bitmap file superblock */ | |
433 | void bitmap_print_sb(struct bitmap *bitmap) | |
434 | { | |
435 | bitmap_super_t *sb; | |
436 | ||
1ec885cd | 437 | if (!bitmap || !bitmap->storage.sb_page) |
32a7627c | 438 | return; |
1ec885cd | 439 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 440 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
441 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
442 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
443 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
444 | *(__u32 *)(sb->uuid+0), |
445 | *(__u32 *)(sb->uuid+4), | |
446 | *(__u32 *)(sb->uuid+8), | |
447 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 448 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 449 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 450 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 451 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
452 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
453 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
454 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
455 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
456 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 457 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
b2f46e68 | 458 | kunmap_atomic(sb); |
32a7627c N |
459 | } |
460 | ||
9c81075f JB |
461 | /* |
462 | * bitmap_new_disk_sb | |
463 | * @bitmap | |
464 | * | |
465 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb | |
466 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. | |
467 | * This function verifies 'bitmap_info' and populates the on-disk bitmap | |
468 | * structure, which is to be written to disk. | |
469 | * | |
470 | * Returns: 0 on success, -Exxx on error | |
471 | */ | |
472 | static int bitmap_new_disk_sb(struct bitmap *bitmap) | |
473 | { | |
474 | bitmap_super_t *sb; | |
475 | unsigned long chunksize, daemon_sleep, write_behind; | |
476 | int err = -EINVAL; | |
477 | ||
1ec885cd N |
478 | bitmap->storage.sb_page = alloc_page(GFP_KERNEL); |
479 | if (IS_ERR(bitmap->storage.sb_page)) { | |
480 | err = PTR_ERR(bitmap->storage.sb_page); | |
481 | bitmap->storage.sb_page = NULL; | |
9c81075f JB |
482 | return err; |
483 | } | |
1ec885cd | 484 | bitmap->storage.sb_page->index = 0; |
9c81075f | 485 | |
1ec885cd | 486 | sb = kmap_atomic(bitmap->storage.sb_page); |
9c81075f JB |
487 | |
488 | sb->magic = cpu_to_le32(BITMAP_MAGIC); | |
489 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); | |
490 | ||
491 | chunksize = bitmap->mddev->bitmap_info.chunksize; | |
492 | BUG_ON(!chunksize); | |
493 | if (!is_power_of_2(chunksize)) { | |
b2f46e68 | 494 | kunmap_atomic(sb); |
9c81075f JB |
495 | printk(KERN_ERR "bitmap chunksize not a power of 2\n"); |
496 | return -EINVAL; | |
497 | } | |
498 | sb->chunksize = cpu_to_le32(chunksize); | |
499 | ||
500 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; | |
501 | if (!daemon_sleep || | |
502 | (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { | |
503 | printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n"); | |
504 | daemon_sleep = 5 * HZ; | |
505 | } | |
506 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); | |
507 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
508 | ||
509 | /* | |
510 | * FIXME: write_behind for RAID1. If not specified, what | |
511 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. | |
512 | */ | |
513 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; | |
514 | if (write_behind > COUNTER_MAX) | |
515 | write_behind = COUNTER_MAX / 2; | |
516 | sb->write_behind = cpu_to_le32(write_behind); | |
517 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
518 | ||
519 | /* keep the array size field of the bitmap superblock up to date */ | |
520 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
521 | ||
522 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); | |
523 | ||
524 | bitmap->flags |= BITMAP_STALE; | |
525 | sb->state |= cpu_to_le32(BITMAP_STALE); | |
526 | bitmap->events_cleared = bitmap->mddev->events; | |
527 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
528 | ||
b2f46e68 | 529 | kunmap_atomic(sb); |
9c81075f JB |
530 | |
531 | return 0; | |
532 | } | |
533 | ||
32a7627c N |
534 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
535 | static int bitmap_read_sb(struct bitmap *bitmap) | |
536 | { | |
537 | char *reason = NULL; | |
538 | bitmap_super_t *sb; | |
4b6d287f | 539 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c N |
540 | unsigned long long events; |
541 | int err = -EINVAL; | |
27581e5a | 542 | struct page *sb_page; |
32a7627c | 543 | |
1ec885cd | 544 | if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { |
ef99bf48 N |
545 | chunksize = 128 * 1024 * 1024; |
546 | daemon_sleep = 5 * HZ; | |
547 | write_behind = 0; | |
548 | bitmap->flags = BITMAP_STALE; | |
549 | err = 0; | |
550 | goto out_no_sb; | |
551 | } | |
32a7627c | 552 | /* page 0 is the superblock, read it... */ |
27581e5a N |
553 | sb_page = alloc_page(GFP_KERNEL); |
554 | if (!sb_page) | |
555 | return -ENOMEM; | |
1ec885cd | 556 | bitmap->storage.sb_page = sb_page; |
27581e5a | 557 | |
1ec885cd N |
558 | if (bitmap->storage.file) { |
559 | loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); | |
f49d5e62 N |
560 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; |
561 | ||
1ec885cd | 562 | err = read_page(bitmap->storage.file, 0, |
27581e5a | 563 | bitmap, bytes, sb_page); |
f49d5e62 | 564 | } else { |
27581e5a N |
565 | err = read_sb_page(bitmap->mddev, |
566 | bitmap->mddev->bitmap_info.offset, | |
567 | sb_page, | |
568 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 569 | } |
27581e5a | 570 | if (err) |
32a7627c | 571 | return err; |
32a7627c | 572 | |
27581e5a | 573 | sb = kmap_atomic(sb_page); |
32a7627c | 574 | |
32a7627c | 575 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 576 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 577 | write_behind = le32_to_cpu(sb->write_behind); |
32a7627c N |
578 | |
579 | /* verify that the bitmap-specific fields are valid */ | |
580 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
581 | reason = "bad magic"; | |
bd926c63 N |
582 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
583 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 584 | reason = "unrecognized superblock version"; |
1187cf0a | 585 | else if (chunksize < 512) |
7dd5d34c | 586 | reason = "bitmap chunksize too small"; |
d744540c | 587 | else if (!is_power_of_2(chunksize)) |
32a7627c | 588 | reason = "bitmap chunksize not a power of 2"; |
1b04be96 | 589 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 590 | reason = "daemon sleep period out of range"; |
4b6d287f N |
591 | else if (write_behind > COUNTER_MAX) |
592 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
593 | if (reason) { |
594 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
595 | bmname(bitmap), reason); | |
596 | goto out; | |
597 | } | |
598 | ||
599 | /* keep the array size field of the bitmap superblock up to date */ | |
600 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
601 | ||
278c1ca2 N |
602 | if (bitmap->mddev->persistent) { |
603 | /* | |
604 | * We have a persistent array superblock, so compare the | |
605 | * bitmap's UUID and event counter to the mddev's | |
606 | */ | |
607 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
608 | printk(KERN_INFO | |
609 | "%s: bitmap superblock UUID mismatch\n", | |
610 | bmname(bitmap)); | |
611 | goto out; | |
612 | } | |
613 | events = le64_to_cpu(sb->events); | |
614 | if (events < bitmap->mddev->events) { | |
615 | printk(KERN_INFO | |
616 | "%s: bitmap file is out of date (%llu < %llu) " | |
617 | "-- forcing full recovery\n", | |
618 | bmname(bitmap), events, | |
619 | (unsigned long long) bitmap->mddev->events); | |
620 | sb->state |= cpu_to_le32(BITMAP_STALE); | |
621 | } | |
32a7627c | 622 | } |
278c1ca2 | 623 | |
32a7627c | 624 | /* assign fields using values from superblock */ |
4f2e639a | 625 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 N |
626 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
627 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
32a7627c N |
628 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
629 | err = 0; | |
630 | out: | |
b2f46e68 | 631 | kunmap_atomic(sb); |
ef99bf48 N |
632 | out_no_sb: |
633 | if (bitmap->flags & BITMAP_STALE) | |
634 | bitmap->events_cleared = bitmap->mddev->events; | |
635 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
636 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
637 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
32a7627c N |
638 | if (err) |
639 | bitmap_print_sb(bitmap); | |
640 | return err; | |
641 | } | |
642 | ||
32a7627c N |
643 | /* |
644 | * general bitmap file operations | |
645 | */ | |
646 | ||
ece5cff0 N |
647 | /* |
648 | * on-disk bitmap: | |
649 | * | |
650 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
651 | * file a page at a time. There's a superblock at the start of the file. | |
652 | */ | |
32a7627c | 653 | /* calculate the index of the page that contains this bit */ |
1ec885cd N |
654 | static inline unsigned long file_page_index(struct bitmap_storage *store, |
655 | unsigned long chunk) | |
32a7627c | 656 | { |
1ec885cd | 657 | if (store->sb_page) |
ece5cff0 N |
658 | chunk += sizeof(bitmap_super_t) << 3; |
659 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
660 | } |
661 | ||
662 | /* calculate the (bit) offset of this bit within a page */ | |
1ec885cd N |
663 | static inline unsigned long file_page_offset(struct bitmap_storage *store, |
664 | unsigned long chunk) | |
32a7627c | 665 | { |
1ec885cd | 666 | if (store->sb_page) |
ece5cff0 N |
667 | chunk += sizeof(bitmap_super_t) << 3; |
668 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
669 | } |
670 | ||
671 | /* | |
672 | * return a pointer to the page in the filemap that contains the given bit | |
673 | * | |
674 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
675 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
676 | * 0 or page 1 | |
677 | */ | |
1ec885cd | 678 | static inline struct page *filemap_get_page(struct bitmap_storage *store, |
3520fa4d | 679 | unsigned long chunk) |
32a7627c | 680 | { |
1ec885cd | 681 | if (file_page_index(store, chunk) >= store->file_pages) |
ac2f40be | 682 | return NULL; |
1ec885cd N |
683 | return store->filemap[file_page_index(store, chunk) |
684 | - file_page_index(store, 0)]; | |
32a7627c N |
685 | } |
686 | ||
d1244cb0 N |
687 | static int bitmap_storage_alloc(struct bitmap_storage *store, |
688 | unsigned long chunks, int with_super) | |
689 | { | |
690 | int pnum; | |
691 | unsigned long num_pages; | |
692 | unsigned long bytes; | |
693 | ||
694 | bytes = DIV_ROUND_UP(chunks, 8); | |
695 | if (with_super) | |
696 | bytes += sizeof(bitmap_super_t); | |
697 | ||
698 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); | |
699 | ||
700 | store->filemap = kmalloc(sizeof(struct page *) | |
701 | * num_pages, GFP_KERNEL); | |
702 | if (!store->filemap) | |
703 | return -ENOMEM; | |
704 | ||
705 | if (with_super && !store->sb_page) { | |
706 | store->sb_page = alloc_page(GFP_KERNEL); | |
707 | if (store->sb_page == NULL) | |
708 | return -ENOMEM; | |
709 | store->sb_page->index = 0; | |
710 | } | |
711 | pnum = 0; | |
712 | if (store->sb_page) { | |
713 | store->filemap[0] = store->sb_page; | |
714 | pnum = 1; | |
715 | } | |
716 | for ( ; pnum < num_pages; pnum++) { | |
717 | store->filemap[pnum] = alloc_page(GFP_KERNEL); | |
718 | if (!store->filemap[pnum]) { | |
719 | store->file_pages = pnum; | |
720 | return -ENOMEM; | |
721 | } | |
722 | store->filemap[pnum]->index = pnum; | |
723 | } | |
724 | store->file_pages = pnum; | |
725 | ||
726 | /* We need 4 bits per page, rounded up to a multiple | |
727 | * of sizeof(unsigned long) */ | |
728 | store->filemap_attr = kzalloc( | |
729 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), | |
730 | GFP_KERNEL); | |
731 | if (!store->filemap_attr) | |
732 | return -ENOMEM; | |
733 | ||
734 | store->bytes = bytes; | |
735 | ||
736 | return 0; | |
737 | } | |
738 | ||
32a7627c N |
739 | static void bitmap_file_unmap(struct bitmap *bitmap) |
740 | { | |
741 | struct page **map, *sb_page; | |
742 | unsigned long *attr; | |
743 | int pages; | |
744 | unsigned long flags; | |
1ec885cd | 745 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c N |
746 | |
747 | spin_lock_irqsave(&bitmap->lock, flags); | |
1ec885cd N |
748 | map = store->filemap; |
749 | store->filemap = NULL; | |
750 | attr = store->filemap_attr; | |
751 | store->filemap_attr = NULL; | |
752 | pages = store->file_pages; | |
753 | store->file_pages = 0; | |
754 | sb_page = store->sb_page; | |
755 | store->sb_page = NULL; | |
32a7627c N |
756 | spin_unlock_irqrestore(&bitmap->lock, flags); |
757 | ||
758 | while (pages--) | |
ece5cff0 | 759 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 760 | free_buffers(map[pages]); |
32a7627c N |
761 | kfree(map); |
762 | kfree(attr); | |
763 | ||
d785a06a N |
764 | if (sb_page) |
765 | free_buffers(sb_page); | |
32a7627c N |
766 | } |
767 | ||
768 | static void bitmap_file_put(struct bitmap *bitmap) | |
769 | { | |
770 | struct file *file; | |
32a7627c N |
771 | unsigned long flags; |
772 | ||
773 | spin_lock_irqsave(&bitmap->lock, flags); | |
1ec885cd N |
774 | file = bitmap->storage.file; |
775 | bitmap->storage.file = NULL; | |
32a7627c N |
776 | spin_unlock_irqrestore(&bitmap->lock, flags); |
777 | ||
d785a06a N |
778 | if (file) |
779 | wait_event(bitmap->write_wait, | |
780 | atomic_read(&bitmap->pending_writes)==0); | |
32a7627c N |
781 | bitmap_file_unmap(bitmap); |
782 | ||
d785a06a | 783 | if (file) { |
c649bb9c | 784 | struct inode *inode = file->f_path.dentry->d_inode; |
fc0ecff6 | 785 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 786 | fput(file); |
d785a06a | 787 | } |
32a7627c N |
788 | } |
789 | ||
32a7627c N |
790 | /* |
791 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
792 | * then it is no longer reliable, so we stop using it and we mark the file | |
793 | * as failed in the superblock | |
794 | */ | |
795 | static void bitmap_file_kick(struct bitmap *bitmap) | |
796 | { | |
797 | char *path, *ptr = NULL; | |
798 | ||
edbb79df N |
799 | if (!(bitmap->flags & BITMAP_STALE)) { |
800 | bitmap->flags |= BITMAP_STALE; | |
4ad13663 | 801 | bitmap_update_sb(bitmap); |
32a7627c | 802 | |
1ec885cd | 803 | if (bitmap->storage.file) { |
4ad13663 N |
804 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); |
805 | if (path) | |
1ec885cd N |
806 | ptr = d_path(&bitmap->storage.file->f_path, |
807 | path, PAGE_SIZE); | |
6bcfd601 | 808 | |
4ad13663 N |
809 | printk(KERN_ALERT |
810 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 811 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 812 | |
4ad13663 N |
813 | kfree(path); |
814 | } else | |
815 | printk(KERN_ALERT | |
816 | "%s: disabling internal bitmap due to errors\n", | |
817 | bmname(bitmap)); | |
a654b9d8 | 818 | } |
32a7627c N |
819 | |
820 | bitmap_file_put(bitmap); | |
821 | ||
822 | return; | |
823 | } | |
824 | ||
825 | enum bitmap_page_attr { | |
ac2f40be | 826 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
5a537df4 N |
827 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
828 | * i.e. counter is 1 or 2. */ | |
ac2f40be | 829 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
32a7627c N |
830 | }; |
831 | ||
d189122d N |
832 | static inline void set_page_attr(struct bitmap *bitmap, int pnum, |
833 | enum bitmap_page_attr attr) | |
32a7627c | 834 | { |
1ec885cd | 835 | __set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
836 | } |
837 | ||
d189122d N |
838 | static inline void clear_page_attr(struct bitmap *bitmap, int pnum, |
839 | enum bitmap_page_attr attr) | |
32a7627c | 840 | { |
1ec885cd | 841 | __clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
842 | } |
843 | ||
d189122d | 844 | static inline unsigned long test_page_attr(struct bitmap *bitmap, int pnum, |
ec7a3197 | 845 | enum bitmap_page_attr attr) |
32a7627c | 846 | { |
1ec885cd | 847 | return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
848 | } |
849 | ||
850 | /* | |
851 | * bitmap_file_set_bit -- called before performing a write to the md device | |
852 | * to set (and eventually sync) a particular bit in the bitmap file | |
853 | * | |
854 | * we set the bit immediately, then we record the page number so that | |
855 | * when an unplug occurs, we can flush the dirty pages out to disk | |
856 | */ | |
857 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
858 | { | |
859 | unsigned long bit; | |
3520fa4d | 860 | struct page *page; |
32a7627c | 861 | void *kaddr; |
61a0d80c | 862 | unsigned long chunk = block >> bitmap->chunkshift; |
32a7627c | 863 | |
1ec885cd | 864 | page = filemap_get_page(&bitmap->storage, chunk); |
3520fa4d JB |
865 | if (!page) |
866 | return; | |
1ec885cd | 867 | bit = file_page_offset(&bitmap->storage, chunk); |
32a7627c | 868 | |
3520fa4d | 869 | /* set the bit */ |
b2f46e68 | 870 | kaddr = kmap_atomic(page); |
3520fa4d JB |
871 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
872 | set_bit(bit, kaddr); | |
873 | else | |
874 | __set_bit_le(bit, kaddr); | |
b2f46e68 | 875 | kunmap_atomic(kaddr); |
36a4e1fe | 876 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
32a7627c | 877 | /* record page number so it gets flushed to disk when unplug occurs */ |
d189122d | 878 | set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY); |
32a7627c N |
879 | } |
880 | ||
ef99bf48 N |
881 | static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) |
882 | { | |
883 | unsigned long bit; | |
884 | struct page *page; | |
885 | void *paddr; | |
886 | unsigned long chunk = block >> bitmap->chunkshift; | |
887 | ||
1ec885cd | 888 | page = filemap_get_page(&bitmap->storage, chunk); |
ef99bf48 N |
889 | if (!page) |
890 | return; | |
1ec885cd | 891 | bit = file_page_offset(&bitmap->storage, chunk); |
ef99bf48 N |
892 | paddr = kmap_atomic(page); |
893 | if (bitmap->flags & BITMAP_HOSTENDIAN) | |
894 | clear_bit(bit, paddr); | |
895 | else | |
896 | __clear_bit_le(bit, paddr); | |
897 | kunmap_atomic(paddr); | |
d189122d N |
898 | if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) { |
899 | set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING); | |
ef99bf48 N |
900 | bitmap->allclean = 0; |
901 | } | |
902 | } | |
903 | ||
32a7627c N |
904 | /* this gets called when the md device is ready to unplug its underlying |
905 | * (slave) device queues -- before we let any writes go down, we need to | |
906 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 907 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 908 | { |
ec7a3197 N |
909 | unsigned long i, flags; |
910 | int dirty, need_write; | |
32a7627c N |
911 | int wait = 0; |
912 | ||
1ec885cd | 913 | if (!bitmap || !bitmap->storage.filemap) |
4ad13663 | 914 | return; |
32a7627c N |
915 | |
916 | /* look at each page to see if there are any set bits that need to be | |
917 | * flushed out to disk */ | |
1ec885cd | 918 | for (i = 0; i < bitmap->storage.file_pages; i++) { |
32a7627c | 919 | spin_lock_irqsave(&bitmap->lock, flags); |
1ec885cd | 920 | if (!bitmap->storage.filemap) { |
32a7627c | 921 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 922 | return; |
32a7627c | 923 | } |
d189122d N |
924 | dirty = test_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
925 | need_write = test_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE); | |
926 | clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); | |
927 | clear_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE); | |
bf07bb7d | 928 | if (dirty || need_write) |
d189122d | 929 | clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); |
ec7a3197 | 930 | if (dirty) |
32a7627c N |
931 | wait = 1; |
932 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
933 | ||
ac2f40be | 934 | if (dirty || need_write) |
1ec885cd | 935 | write_page(bitmap, bitmap->storage.filemap[i], 0); |
32a7627c N |
936 | } |
937 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
1ec885cd | 938 | if (bitmap->storage.file) |
d785a06a N |
939 | wait_event(bitmap->write_wait, |
940 | atomic_read(&bitmap->pending_writes)==0); | |
0b79ccf0 | 941 | else |
a9701a30 | 942 | md_super_wait(bitmap->mddev); |
32a7627c | 943 | } |
d785a06a N |
944 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
945 | bitmap_file_kick(bitmap); | |
32a7627c | 946 | } |
ac2f40be | 947 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 948 | |
6a07997f | 949 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
950 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
951 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
952 | * memory mapping of the bitmap file | |
953 | * Special cases: | |
954 | * if there's no bitmap file, or if the bitmap file had been | |
955 | * previously kicked from the array, we mark all the bits as | |
956 | * 1's in order to cause a full resync. | |
6a07997f N |
957 | * |
958 | * We ignore all bits for sectors that end earlier than 'start'. | |
959 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 960 | */ |
6a07997f | 961 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
962 | { |
963 | unsigned long i, chunks, index, oldindex, bit; | |
27581e5a | 964 | struct page *page = NULL; |
d1244cb0 | 965 | unsigned long bit_cnt = 0; |
32a7627c | 966 | struct file *file; |
d1244cb0 | 967 | unsigned long offset; |
32a7627c N |
968 | int outofdate; |
969 | int ret = -ENOSPC; | |
ea03aff9 | 970 | void *paddr; |
1ec885cd | 971 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c N |
972 | |
973 | chunks = bitmap->chunks; | |
1ec885cd | 974 | file = store->file; |
32a7627c | 975 | |
ef99bf48 N |
976 | if (!file && !bitmap->mddev->bitmap_info.offset) { |
977 | /* No permanent bitmap - fill with '1s'. */ | |
1ec885cd N |
978 | store->filemap = NULL; |
979 | store->file_pages = 0; | |
ef99bf48 N |
980 | for (i = 0; i < chunks ; i++) { |
981 | /* if the disk bit is set, set the memory bit */ | |
982 | int needed = ((sector_t)(i+1) << (bitmap->chunkshift) | |
983 | >= start); | |
984 | bitmap_set_memory_bits(bitmap, | |
985 | (sector_t)i << bitmap->chunkshift, | |
986 | needed); | |
987 | } | |
988 | return 0; | |
989 | } | |
32a7627c | 990 | |
32a7627c | 991 | outofdate = bitmap->flags & BITMAP_STALE; |
32a7627c N |
992 | if (outofdate) |
993 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
994 | "recovery\n", bmname(bitmap)); | |
995 | ||
d1244cb0 | 996 | if (file && i_size_read(file->f_mapping->host) < store->bytes) { |
32a7627c | 997 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
d1244cb0 N |
998 | bmname(bitmap), |
999 | (unsigned long) i_size_read(file->f_mapping->host), | |
1000 | store->bytes); | |
4ad13663 | 1001 | goto err; |
32a7627c | 1002 | } |
bc7f77de | 1003 | |
d1244cb0 | 1004 | oldindex = ~0L; |
27581e5a | 1005 | offset = 0; |
d1244cb0 | 1006 | if (!bitmap->mddev->bitmap_info.external) |
27581e5a | 1007 | offset = sizeof(bitmap_super_t); |
32a7627c N |
1008 | |
1009 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 1010 | int b; |
1ec885cd N |
1011 | index = file_page_index(&bitmap->storage, i); |
1012 | bit = file_page_offset(&bitmap->storage, i); | |
32a7627c | 1013 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 1014 | int count; |
32a7627c | 1015 | /* unmap the old page, we're done with it */ |
d1244cb0 N |
1016 | if (index == store->file_pages-1) |
1017 | count = store->bytes - index * PAGE_SIZE; | |
d785a06a N |
1018 | else |
1019 | count = PAGE_SIZE; | |
1ec885cd | 1020 | page = store->filemap[index]; |
27581e5a N |
1021 | if (file) |
1022 | ret = read_page(file, index, bitmap, | |
1023 | count, page); | |
1024 | else | |
1025 | ret = read_sb_page( | |
1026 | bitmap->mddev, | |
1027 | bitmap->mddev->bitmap_info.offset, | |
1028 | page, | |
1029 | index, count); | |
1030 | ||
1031 | if (ret) | |
4ad13663 | 1032 | goto err; |
a654b9d8 | 1033 | |
32a7627c | 1034 | oldindex = index; |
32a7627c N |
1035 | |
1036 | if (outofdate) { | |
1037 | /* | |
1038 | * if bitmap is out of date, dirty the | |
ac2f40be | 1039 | * whole page and write it out |
32a7627c | 1040 | */ |
b2f46e68 | 1041 | paddr = kmap_atomic(page); |
ea03aff9 | 1042 | memset(paddr + offset, 0xff, |
6a07997f | 1043 | PAGE_SIZE - offset); |
b2f46e68 | 1044 | kunmap_atomic(paddr); |
4ad13663 N |
1045 | write_page(bitmap, page, 1); |
1046 | ||
1047 | ret = -EIO; | |
b74fd282 | 1048 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
4ad13663 | 1049 | goto err; |
32a7627c | 1050 | } |
32a7627c | 1051 | } |
b2f46e68 | 1052 | paddr = kmap_atomic(page); |
bd926c63 | 1053 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1054 | b = test_bit(bit, paddr); |
bd926c63 | 1055 | else |
6b33aff3 | 1056 | b = test_bit_le(bit, paddr); |
b2f46e68 | 1057 | kunmap_atomic(paddr); |
bd926c63 | 1058 | if (b) { |
32a7627c | 1059 | /* if the disk bit is set, set the memory bit */ |
61a0d80c | 1060 | int needed = ((sector_t)(i+1) << bitmap->chunkshift |
db305e50 N |
1061 | >= start); |
1062 | bitmap_set_memory_bits(bitmap, | |
61a0d80c | 1063 | (sector_t)i << bitmap->chunkshift, |
db305e50 | 1064 | needed); |
32a7627c N |
1065 | bit_cnt++; |
1066 | } | |
27581e5a | 1067 | offset = 0; |
32a7627c N |
1068 | } |
1069 | ||
32a7627c | 1070 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
d1244cb0 | 1071 | "read %lu pages, set %lu of %lu bits\n", |
1ec885cd | 1072 | bmname(bitmap), store->file_pages, |
d1244cb0 | 1073 | bit_cnt, chunks); |
4ad13663 N |
1074 | |
1075 | return 0; | |
32a7627c | 1076 | |
4ad13663 N |
1077 | err: |
1078 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1079 | bmname(bitmap), ret); | |
32a7627c N |
1080 | return ret; |
1081 | } | |
1082 | ||
a654b9d8 N |
1083 | void bitmap_write_all(struct bitmap *bitmap) |
1084 | { | |
1085 | /* We don't actually write all bitmap blocks here, | |
1086 | * just flag them as needing to be written | |
1087 | */ | |
ec7a3197 | 1088 | int i; |
a654b9d8 | 1089 | |
1ec885cd | 1090 | if (!bitmap || !bitmap->storage.filemap) |
ef99bf48 | 1091 | return; |
1ec885cd | 1092 | if (bitmap->storage.file) |
ef99bf48 N |
1093 | /* Only one copy, so nothing needed */ |
1094 | return; | |
1095 | ||
7c8f4247 | 1096 | spin_lock_irq(&bitmap->lock); |
1ec885cd | 1097 | for (i = 0; i < bitmap->storage.file_pages; i++) |
d189122d | 1098 | set_page_attr(bitmap, i, |
ec7a3197 | 1099 | BITMAP_PAGE_NEEDWRITE); |
2585f3ef | 1100 | bitmap->allclean = 0; |
7c8f4247 | 1101 | spin_unlock_irq(&bitmap->lock); |
a654b9d8 N |
1102 | } |
1103 | ||
32a7627c N |
1104 | static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) |
1105 | { | |
61a0d80c | 1106 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1107 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1108 | bitmap->bp[page].count += inc; | |
32a7627c N |
1109 | bitmap_checkfree(bitmap, page); |
1110 | } | |
bf07bb7d N |
1111 | |
1112 | static void bitmap_set_pending(struct bitmap *bitmap, sector_t offset) | |
1113 | { | |
1114 | sector_t chunk = offset >> bitmap->chunkshift; | |
1115 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1116 | struct bitmap_page *bp = &bitmap->bp[page]; | |
1117 | ||
1118 | if (!bp->pending) | |
1119 | bp->pending = 1; | |
1120 | } | |
1121 | ||
32a7627c | 1122 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
57dab0bd | 1123 | sector_t offset, sector_t *blocks, |
32a7627c N |
1124 | int create); |
1125 | ||
1126 | /* | |
1127 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1128 | * out to disk | |
1129 | */ | |
1130 | ||
fd01b88c | 1131 | void bitmap_daemon_work(struct mddev *mddev) |
32a7627c | 1132 | { |
aa5cbd10 | 1133 | struct bitmap *bitmap; |
aa3163f8 | 1134 | unsigned long j; |
bf07bb7d | 1135 | unsigned long nextpage; |
32a7627c | 1136 | unsigned long flags; |
57dab0bd | 1137 | sector_t blocks; |
32a7627c | 1138 | |
aa5cbd10 N |
1139 | /* Use a mutex to guard daemon_work against |
1140 | * bitmap_destroy. | |
1141 | */ | |
c3d9714e | 1142 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1143 | bitmap = mddev->bitmap; |
1144 | if (bitmap == NULL) { | |
c3d9714e | 1145 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1146 | return; |
aa5cbd10 | 1147 | } |
42a04b50 | 1148 | if (time_before(jiffies, bitmap->daemon_lastrun |
2e61ebbc | 1149 | + mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1150 | goto done; |
1151 | ||
32a7627c | 1152 | bitmap->daemon_lastrun = jiffies; |
8311c29d | 1153 | if (bitmap->allclean) { |
2e61ebbc | 1154 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
aa5cbd10 | 1155 | goto done; |
8311c29d N |
1156 | } |
1157 | bitmap->allclean = 1; | |
32a7627c | 1158 | |
bf07bb7d N |
1159 | /* Any file-page which is PENDING now needs to be written. |
1160 | * So set NEEDWRITE now, then after we make any last-minute changes | |
1161 | * we will write it. | |
1162 | */ | |
be512691 | 1163 | spin_lock_irqsave(&bitmap->lock, flags); |
1ec885cd | 1164 | for (j = 0; j < bitmap->storage.file_pages; j++) |
d189122d | 1165 | if (test_page_attr(bitmap, j, |
bf07bb7d | 1166 | BITMAP_PAGE_PENDING)) { |
d189122d | 1167 | set_page_attr(bitmap, j, |
bf07bb7d | 1168 | BITMAP_PAGE_NEEDWRITE); |
d189122d | 1169 | clear_page_attr(bitmap, j, |
bf07bb7d N |
1170 | BITMAP_PAGE_PENDING); |
1171 | } | |
1172 | ||
1173 | if (bitmap->need_sync && | |
1174 | mddev->bitmap_info.external == 0) { | |
1175 | /* Arrange for superblock update as well as | |
1176 | * other changes */ | |
1177 | bitmap_super_t *sb; | |
1178 | bitmap->need_sync = 0; | |
1ec885cd N |
1179 | if (bitmap->storage.filemap) { |
1180 | sb = kmap_atomic(bitmap->storage.sb_page); | |
ef99bf48 N |
1181 | sb->events_cleared = |
1182 | cpu_to_le64(bitmap->events_cleared); | |
1183 | kunmap_atomic(sb); | |
d189122d | 1184 | set_page_attr(bitmap, 0, |
ef99bf48 N |
1185 | BITMAP_PAGE_NEEDWRITE); |
1186 | } | |
bf07bb7d N |
1187 | } |
1188 | /* Now look at the bitmap counters and if any are '2' or '1', | |
1189 | * decrement and handle accordingly. | |
1190 | */ | |
1191 | nextpage = 0; | |
32a7627c N |
1192 | for (j = 0; j < bitmap->chunks; j++) { |
1193 | bitmap_counter_t *bmc; | |
ef99bf48 | 1194 | sector_t block = (sector_t)j << bitmap->chunkshift; |
3520fa4d | 1195 | |
bf07bb7d N |
1196 | if (j == nextpage) { |
1197 | nextpage += PAGE_COUNTER_RATIO; | |
1198 | if (!bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending) { | |
1199 | j |= PAGE_COUNTER_MASK; | |
aa3163f8 N |
1200 | continue; |
1201 | } | |
bf07bb7d | 1202 | bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; |
32a7627c | 1203 | } |
db305e50 | 1204 | bmc = bitmap_get_counter(bitmap, |
ef99bf48 | 1205 | block, |
db305e50 | 1206 | &blocks, 0); |
bf07bb7d N |
1207 | |
1208 | if (!bmc) { | |
5a537df4 | 1209 | j |= PAGE_COUNTER_MASK; |
bf07bb7d N |
1210 | continue; |
1211 | } | |
1212 | if (*bmc == 1 && !bitmap->need_sync) { | |
1213 | /* We can clear the bit */ | |
bf07bb7d | 1214 | *bmc = 0; |
ef99bf48 N |
1215 | bitmap_count_page(bitmap, block, -1); |
1216 | bitmap_file_clear_bit(bitmap, block); | |
bf07bb7d N |
1217 | } else if (*bmc && *bmc <= 2) { |
1218 | *bmc = 1; | |
ef99bf48 | 1219 | bitmap_set_pending(bitmap, block); |
bf07bb7d | 1220 | bitmap->allclean = 0; |
5a537df4 | 1221 | } |
32a7627c N |
1222 | } |
1223 | ||
bf07bb7d N |
1224 | /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. |
1225 | * DIRTY pages need to be written by bitmap_unplug so it can wait | |
1226 | * for them. | |
1227 | * If we find any DIRTY page we stop there and let bitmap_unplug | |
1228 | * handle all the rest. This is important in the case where | |
1229 | * the first blocking holds the superblock and it has been updated. | |
1230 | * We mustn't write any other blocks before the superblock. | |
1231 | */ | |
1ec885cd | 1232 | for (j = 0; j < bitmap->storage.file_pages; j++) { |
bf07bb7d | 1233 | |
d189122d | 1234 | if (test_page_attr(bitmap, j, |
bf07bb7d N |
1235 | BITMAP_PAGE_DIRTY)) |
1236 | /* bitmap_unplug will handle the rest */ | |
1237 | break; | |
d189122d | 1238 | if (test_page_attr(bitmap, j, |
bf07bb7d | 1239 | BITMAP_PAGE_NEEDWRITE)) { |
d189122d | 1240 | clear_page_attr(bitmap, j, |
bf07bb7d | 1241 | BITMAP_PAGE_NEEDWRITE); |
32a7627c | 1242 | spin_unlock_irqrestore(&bitmap->lock, flags); |
1ec885cd | 1243 | write_page(bitmap, bitmap->storage.filemap[j], 0); |
bf07bb7d | 1244 | spin_lock_irqsave(&bitmap->lock, flags); |
1ec885cd | 1245 | if (!bitmap->storage.filemap) |
bf07bb7d | 1246 | break; |
32a7627c | 1247 | } |
32a7627c | 1248 | } |
bf07bb7d | 1249 | spin_unlock_irqrestore(&bitmap->lock, flags); |
32a7627c | 1250 | |
7be3dfec | 1251 | done: |
8311c29d | 1252 | if (bitmap->allclean == 0) |
2e61ebbc N |
1253 | mddev->thread->timeout = |
1254 | mddev->bitmap_info.daemon_sleep; | |
c3d9714e | 1255 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1256 | } |
1257 | ||
32a7627c | 1258 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
57dab0bd | 1259 | sector_t offset, sector_t *blocks, |
32a7627c | 1260 | int create) |
ee305ace N |
1261 | __releases(bitmap->lock) |
1262 | __acquires(bitmap->lock) | |
32a7627c N |
1263 | { |
1264 | /* If 'create', we might release the lock and reclaim it. | |
1265 | * The lock must have been taken with interrupts enabled. | |
1266 | * If !create, we don't release the lock. | |
1267 | */ | |
61a0d80c | 1268 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1269 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1270 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1271 | sector_t csize; | |
ef425673 | 1272 | int err; |
32a7627c | 1273 | |
ef425673 N |
1274 | err = bitmap_checkpage(bitmap, page, create); |
1275 | ||
1276 | if (bitmap->bp[page].hijacked || | |
1277 | bitmap->bp[page].map == NULL) | |
61a0d80c | 1278 | csize = ((sector_t)1) << (bitmap->chunkshift + |
ef425673 N |
1279 | PAGE_COUNTER_SHIFT - 1); |
1280 | else | |
61a0d80c | 1281 | csize = ((sector_t)1) << bitmap->chunkshift; |
ef425673 N |
1282 | *blocks = csize - (offset & (csize - 1)); |
1283 | ||
1284 | if (err < 0) | |
32a7627c | 1285 | return NULL; |
ef425673 | 1286 | |
32a7627c N |
1287 | /* now locked ... */ |
1288 | ||
1289 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1290 | /* should we use the first or second counter field | |
1291 | * of the hijacked pointer? */ | |
1292 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1293 | return &((bitmap_counter_t *) |
1294 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1295 | } else /* page is allocated */ |
32a7627c N |
1296 | return (bitmap_counter_t *) |
1297 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1298 | } |
1299 | ||
4b6d287f | 1300 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1301 | { |
ac2f40be N |
1302 | if (!bitmap) |
1303 | return 0; | |
4b6d287f N |
1304 | |
1305 | if (behind) { | |
696fcd53 | 1306 | int bw; |
4b6d287f | 1307 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1308 | bw = atomic_read(&bitmap->behind_writes); |
1309 | if (bw > bitmap->behind_writes_used) | |
1310 | bitmap->behind_writes_used = bw; | |
1311 | ||
36a4e1fe N |
1312 | pr_debug("inc write-behind count %d/%lu\n", |
1313 | bw, bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1314 | } |
1315 | ||
32a7627c | 1316 | while (sectors) { |
57dab0bd | 1317 | sector_t blocks; |
32a7627c N |
1318 | bitmap_counter_t *bmc; |
1319 | ||
1320 | spin_lock_irq(&bitmap->lock); | |
1321 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); | |
1322 | if (!bmc) { | |
1323 | spin_unlock_irq(&bitmap->lock); | |
1324 | return 0; | |
1325 | } | |
1326 | ||
27d5ea04 | 1327 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
da6e1a32 NB |
1328 | DEFINE_WAIT(__wait); |
1329 | /* note that it is safe to do the prepare_to_wait | |
1330 | * after the test as long as we do it before dropping | |
1331 | * the spinlock. | |
1332 | */ | |
1333 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1334 | TASK_UNINTERRUPTIBLE); | |
1335 | spin_unlock_irq(&bitmap->lock); | |
7eaceacc | 1336 | io_schedule(); |
da6e1a32 NB |
1337 | finish_wait(&bitmap->overflow_wait, &__wait); |
1338 | continue; | |
1339 | } | |
1340 | ||
ac2f40be | 1341 | switch (*bmc) { |
32a7627c N |
1342 | case 0: |
1343 | bitmap_file_set_bit(bitmap, offset); | |
ac2f40be | 1344 | bitmap_count_page(bitmap, offset, 1); |
32a7627c N |
1345 | /* fall through */ |
1346 | case 1: | |
1347 | *bmc = 2; | |
1348 | } | |
da6e1a32 | 1349 | |
32a7627c N |
1350 | (*bmc)++; |
1351 | ||
1352 | spin_unlock_irq(&bitmap->lock); | |
1353 | ||
1354 | offset += blocks; | |
1355 | if (sectors > blocks) | |
1356 | sectors -= blocks; | |
ac2f40be N |
1357 | else |
1358 | sectors = 0; | |
32a7627c N |
1359 | } |
1360 | return 0; | |
1361 | } | |
ac2f40be | 1362 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1363 | |
1364 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1365 | int success, int behind) |
32a7627c | 1366 | { |
ac2f40be N |
1367 | if (!bitmap) |
1368 | return; | |
4b6d287f | 1369 | if (behind) { |
e555190d N |
1370 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1371 | wake_up(&bitmap->behind_wait); | |
36a4e1fe N |
1372 | pr_debug("dec write-behind count %d/%lu\n", |
1373 | atomic_read(&bitmap->behind_writes), | |
1374 | bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1375 | } |
1376 | ||
32a7627c | 1377 | while (sectors) { |
57dab0bd | 1378 | sector_t blocks; |
32a7627c N |
1379 | unsigned long flags; |
1380 | bitmap_counter_t *bmc; | |
1381 | ||
1382 | spin_lock_irqsave(&bitmap->lock, flags); | |
1383 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); | |
1384 | if (!bmc) { | |
1385 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1386 | return; | |
1387 | } | |
1388 | ||
961902c0 | 1389 | if (success && !bitmap->mddev->degraded && |
a0da84f3 NB |
1390 | bitmap->events_cleared < bitmap->mddev->events) { |
1391 | bitmap->events_cleared = bitmap->mddev->events; | |
1392 | bitmap->need_sync = 1; | |
5ff5afff | 1393 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1394 | } |
1395 | ||
27d5ea04 | 1396 | if (!success && !NEEDED(*bmc)) |
32a7627c N |
1397 | *bmc |= NEEDED_MASK; |
1398 | ||
27d5ea04 | 1399 | if (COUNTER(*bmc) == COUNTER_MAX) |
da6e1a32 NB |
1400 | wake_up(&bitmap->overflow_wait); |
1401 | ||
32a7627c | 1402 | (*bmc)--; |
2585f3ef | 1403 | if (*bmc <= 2) { |
bf07bb7d | 1404 | bitmap_set_pending(bitmap, offset); |
2585f3ef N |
1405 | bitmap->allclean = 0; |
1406 | } | |
32a7627c N |
1407 | spin_unlock_irqrestore(&bitmap->lock, flags); |
1408 | offset += blocks; | |
1409 | if (sectors > blocks) | |
1410 | sectors -= blocks; | |
ac2f40be N |
1411 | else |
1412 | sectors = 0; | |
32a7627c N |
1413 | } |
1414 | } | |
ac2f40be | 1415 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1416 | |
57dab0bd | 1417 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1418 | int degraded) |
32a7627c N |
1419 | { |
1420 | bitmap_counter_t *bmc; | |
1421 | int rv; | |
1422 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1423 | *blocks = 1024; | |
1424 | return 1; /* always resync if no bitmap */ | |
1425 | } | |
1426 | spin_lock_irq(&bitmap->lock); | |
1427 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1428 | rv = 0; | |
1429 | if (bmc) { | |
1430 | /* locked */ | |
1431 | if (RESYNC(*bmc)) | |
1432 | rv = 1; | |
1433 | else if (NEEDED(*bmc)) { | |
1434 | rv = 1; | |
6a806c51 N |
1435 | if (!degraded) { /* don't set/clear bits if degraded */ |
1436 | *bmc |= RESYNC_MASK; | |
1437 | *bmc &= ~NEEDED_MASK; | |
1438 | } | |
32a7627c N |
1439 | } |
1440 | } | |
1441 | spin_unlock_irq(&bitmap->lock); | |
1442 | return rv; | |
1443 | } | |
1444 | ||
57dab0bd | 1445 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a N |
1446 | int degraded) |
1447 | { | |
1448 | /* bitmap_start_sync must always report on multiples of whole | |
1449 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1450 | * get confused. | |
1451 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1452 | * At least PAGE_SIZE>>9 blocks are covered. | |
1453 | * Return the 'or' of the result. | |
1454 | */ | |
1455 | int rv = 0; | |
57dab0bd | 1456 | sector_t blocks1; |
1187cf0a N |
1457 | |
1458 | *blocks = 0; | |
1459 | while (*blocks < (PAGE_SIZE>>9)) { | |
1460 | rv |= __bitmap_start_sync(bitmap, offset, | |
1461 | &blocks1, degraded); | |
1462 | offset += blocks1; | |
1463 | *blocks += blocks1; | |
1464 | } | |
1465 | return rv; | |
1466 | } | |
ac2f40be | 1467 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1468 | |
57dab0bd | 1469 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1470 | { |
1471 | bitmap_counter_t *bmc; | |
1472 | unsigned long flags; | |
ac2f40be N |
1473 | |
1474 | if (bitmap == NULL) { | |
32a7627c N |
1475 | *blocks = 1024; |
1476 | return; | |
1477 | } | |
1478 | spin_lock_irqsave(&bitmap->lock, flags); | |
1479 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1480 | if (bmc == NULL) | |
1481 | goto unlock; | |
1482 | /* locked */ | |
32a7627c N |
1483 | if (RESYNC(*bmc)) { |
1484 | *bmc &= ~RESYNC_MASK; | |
1485 | ||
1486 | if (!NEEDED(*bmc) && aborted) | |
1487 | *bmc |= NEEDED_MASK; | |
1488 | else { | |
2585f3ef | 1489 | if (*bmc <= 2) { |
bf07bb7d | 1490 | bitmap_set_pending(bitmap, offset); |
2585f3ef N |
1491 | bitmap->allclean = 0; |
1492 | } | |
32a7627c N |
1493 | } |
1494 | } | |
1495 | unlock: | |
1496 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1497 | } | |
ac2f40be | 1498 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1499 | |
1500 | void bitmap_close_sync(struct bitmap *bitmap) | |
1501 | { | |
1502 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1503 | * properly have been aborted. It remains to us to clear the | |
1504 | * RESYNC bit wherever it is still on | |
1505 | */ | |
1506 | sector_t sector = 0; | |
57dab0bd | 1507 | sector_t blocks; |
b47490c9 N |
1508 | if (!bitmap) |
1509 | return; | |
32a7627c N |
1510 | while (sector < bitmap->mddev->resync_max_sectors) { |
1511 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1512 | sector += blocks; |
1513 | } | |
1514 | } | |
ac2f40be | 1515 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 N |
1516 | |
1517 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1518 | { | |
1519 | sector_t s = 0; | |
57dab0bd | 1520 | sector_t blocks; |
b47490c9 N |
1521 | |
1522 | if (!bitmap) | |
1523 | return; | |
1524 | if (sector == 0) { | |
1525 | bitmap->last_end_sync = jiffies; | |
1526 | return; | |
1527 | } | |
1528 | if (time_before(jiffies, (bitmap->last_end_sync | |
1b04be96 | 1529 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1530 | return; |
1531 | wait_event(bitmap->mddev->recovery_wait, | |
1532 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1533 | ||
75d3da43 | 1534 | bitmap->mddev->curr_resync_completed = sector; |
070dc6dd | 1535 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
61a0d80c | 1536 | sector &= ~((1ULL << bitmap->chunkshift) - 1); |
b47490c9 N |
1537 | s = 0; |
1538 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1539 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1540 | s += blocks; | |
32a7627c | 1541 | } |
b47490c9 | 1542 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1543 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1544 | } |
ac2f40be | 1545 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1546 | |
6a07997f | 1547 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1548 | { |
1549 | /* For each chunk covered by any of these sectors, set the | |
ef99bf48 | 1550 | * counter to 2 and possibly set resync_needed. They should all |
32a7627c N |
1551 | * be 0 at this point |
1552 | */ | |
193f1c93 | 1553 | |
57dab0bd | 1554 | sector_t secs; |
193f1c93 N |
1555 | bitmap_counter_t *bmc; |
1556 | spin_lock_irq(&bitmap->lock); | |
1557 | bmc = bitmap_get_counter(bitmap, offset, &secs, 1); | |
1558 | if (!bmc) { | |
32a7627c | 1559 | spin_unlock_irq(&bitmap->lock); |
193f1c93 | 1560 | return; |
32a7627c | 1561 | } |
ac2f40be | 1562 | if (!*bmc) { |
915c420d | 1563 | *bmc = 2 | (needed ? NEEDED_MASK : 0); |
193f1c93 | 1564 | bitmap_count_page(bitmap, offset, 1); |
bf07bb7d | 1565 | bitmap_set_pending(bitmap, offset); |
2585f3ef | 1566 | bitmap->allclean = 0; |
193f1c93 N |
1567 | } |
1568 | spin_unlock_irq(&bitmap->lock); | |
32a7627c N |
1569 | } |
1570 | ||
9b1d1dac PC |
1571 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1572 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1573 | { | |
1574 | unsigned long chunk; | |
1575 | ||
1576 | for (chunk = s; chunk <= e; chunk++) { | |
61a0d80c | 1577 | sector_t sec = (sector_t)chunk << bitmap->chunkshift; |
9b1d1dac | 1578 | bitmap_set_memory_bits(bitmap, sec, 1); |
7c8f4247 | 1579 | spin_lock_irq(&bitmap->lock); |
9b1d1dac | 1580 | bitmap_file_set_bit(bitmap, sec); |
7c8f4247 | 1581 | spin_unlock_irq(&bitmap->lock); |
ffa23322 N |
1582 | if (sec < bitmap->mddev->recovery_cp) |
1583 | /* We are asserting that the array is dirty, | |
1584 | * so move the recovery_cp address back so | |
1585 | * that it is obvious that it is dirty | |
1586 | */ | |
1587 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1588 | } |
1589 | } | |
1590 | ||
6b8b3e8a N |
1591 | /* |
1592 | * flush out any pending updates | |
1593 | */ | |
fd01b88c | 1594 | void bitmap_flush(struct mddev *mddev) |
6b8b3e8a N |
1595 | { |
1596 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1597 | long sleep; |
6b8b3e8a N |
1598 | |
1599 | if (!bitmap) /* there was no bitmap */ | |
1600 | return; | |
1601 | ||
1602 | /* run the daemon_work three time to ensure everything is flushed | |
1603 | * that can be | |
1604 | */ | |
1b04be96 | 1605 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1606 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1607 | bitmap_daemon_work(mddev); |
42a04b50 | 1608 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1609 | bitmap_daemon_work(mddev); |
42a04b50 | 1610 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1611 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1612 | bitmap_update_sb(bitmap); |
1613 | } | |
1614 | ||
32a7627c N |
1615 | /* |
1616 | * free memory that was allocated | |
1617 | */ | |
3178b0db | 1618 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1619 | { |
1620 | unsigned long k, pages; | |
1621 | struct bitmap_page *bp; | |
32a7627c N |
1622 | |
1623 | if (!bitmap) /* there was no bitmap */ | |
1624 | return; | |
1625 | ||
32a7627c N |
1626 | /* release the bitmap file and kill the daemon */ |
1627 | bitmap_file_put(bitmap); | |
1628 | ||
1629 | bp = bitmap->bp; | |
1630 | pages = bitmap->pages; | |
1631 | ||
1632 | /* free all allocated memory */ | |
1633 | ||
32a7627c N |
1634 | if (bp) /* deallocate the page memory */ |
1635 | for (k = 0; k < pages; k++) | |
1636 | if (bp[k].map && !bp[k].hijacked) | |
1637 | kfree(bp[k].map); | |
1638 | kfree(bp); | |
1639 | kfree(bitmap); | |
1640 | } | |
aa5cbd10 | 1641 | |
fd01b88c | 1642 | void bitmap_destroy(struct mddev *mddev) |
3178b0db N |
1643 | { |
1644 | struct bitmap *bitmap = mddev->bitmap; | |
1645 | ||
1646 | if (!bitmap) /* there was no bitmap */ | |
1647 | return; | |
1648 | ||
c3d9714e | 1649 | mutex_lock(&mddev->bitmap_info.mutex); |
3178b0db | 1650 | mddev->bitmap = NULL; /* disconnect from the md device */ |
c3d9714e | 1651 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1652 | if (mddev->thread) |
1653 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db | 1654 | |
ece5cff0 N |
1655 | if (bitmap->sysfs_can_clear) |
1656 | sysfs_put(bitmap->sysfs_can_clear); | |
1657 | ||
3178b0db N |
1658 | bitmap_free(bitmap); |
1659 | } | |
32a7627c N |
1660 | |
1661 | /* | |
1662 | * initialize the bitmap structure | |
1663 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1664 | */ | |
fd01b88c | 1665 | int bitmap_create(struct mddev *mddev) |
32a7627c N |
1666 | { |
1667 | struct bitmap *bitmap; | |
1f593903 | 1668 | sector_t blocks = mddev->resync_max_sectors; |
32a7627c N |
1669 | unsigned long chunks; |
1670 | unsigned long pages; | |
c3d9714e | 1671 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1672 | int err; |
5ff5afff | 1673 | struct sysfs_dirent *bm = NULL; |
32a7627c | 1674 | |
5f6e3c83 | 1675 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1676 | |
c3d9714e | 1677 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1678 | |
9ffae0cf | 1679 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1680 | if (!bitmap) |
1681 | return -ENOMEM; | |
1682 | ||
32a7627c | 1683 | spin_lock_init(&bitmap->lock); |
ce25c31b N |
1684 | atomic_set(&bitmap->pending_writes, 0); |
1685 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1686 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1687 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1688 | |
32a7627c | 1689 | bitmap->mddev = mddev; |
32a7627c | 1690 | |
5ff5afff N |
1691 | if (mddev->kobj.sd) |
1692 | bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap"); | |
ece5cff0 | 1693 | if (bm) { |
3ff195b0 | 1694 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear"); |
ece5cff0 N |
1695 | sysfs_put(bm); |
1696 | } else | |
1697 | bitmap->sysfs_can_clear = NULL; | |
1698 | ||
1ec885cd | 1699 | bitmap->storage.file = file; |
ce25c31b N |
1700 | if (file) { |
1701 | get_file(file); | |
ae8fa283 N |
1702 | /* As future accesses to this file will use bmap, |
1703 | * and bypass the page cache, we must sync the file | |
1704 | * first. | |
1705 | */ | |
8018ab05 | 1706 | vfs_fsync(file, 1); |
ce25c31b | 1707 | } |
42a04b50 | 1708 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
9c81075f JB |
1709 | if (!mddev->bitmap_info.external) { |
1710 | /* | |
1711 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is | |
1712 | * instructing us to create a new on-disk bitmap instance. | |
1713 | */ | |
1714 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) | |
1715 | err = bitmap_new_disk_sb(bitmap); | |
1716 | else | |
1717 | err = bitmap_read_sb(bitmap); | |
1718 | } else { | |
ece5cff0 N |
1719 | err = 0; |
1720 | if (mddev->bitmap_info.chunksize == 0 || | |
1721 | mddev->bitmap_info.daemon_sleep == 0) | |
1722 | /* chunksize and time_base need to be | |
1723 | * set first. */ | |
1724 | err = -EINVAL; | |
1725 | } | |
32a7627c | 1726 | if (err) |
3178b0db | 1727 | goto error; |
32a7627c | 1728 | |
624ce4f5 | 1729 | bitmap->daemon_lastrun = jiffies; |
61a0d80c N |
1730 | bitmap->chunkshift = (ffz(~mddev->bitmap_info.chunksize) |
1731 | - BITMAP_BLOCK_SHIFT); | |
32a7627c | 1732 | |
b16b1b6c | 1733 | chunks = (blocks + (1 << bitmap->chunkshift) - 1) >> |
61a0d80c | 1734 | bitmap->chunkshift; |
ac2f40be | 1735 | pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; |
32a7627c N |
1736 | |
1737 | BUG_ON(!pages); | |
1738 | ||
1739 | bitmap->chunks = chunks; | |
1740 | bitmap->pages = pages; | |
1741 | bitmap->missing_pages = pages; | |
32a7627c | 1742 | |
9ffae0cf | 1743 | bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); |
29d3247e | 1744 | |
3178b0db | 1745 | err = -ENOMEM; |
32a7627c | 1746 | if (!bitmap->bp) |
3178b0db | 1747 | goto error; |
32a7627c | 1748 | |
bc9891a8 N |
1749 | if (file || mddev->bitmap_info.offset) { |
1750 | err = bitmap_storage_alloc(&bitmap->storage, bitmap->chunks, | |
1751 | !mddev->bitmap_info.external); | |
1752 | if (err) | |
1753 | goto error; | |
1754 | } | |
69e51b44 N |
1755 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", |
1756 | pages, bmname(bitmap)); | |
1757 | ||
1758 | mddev->bitmap = bitmap; | |
1759 | ||
1760 | ||
1761 | return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0; | |
1762 | ||
1763 | error: | |
1764 | bitmap_free(bitmap); | |
1765 | return err; | |
1766 | } | |
1767 | ||
fd01b88c | 1768 | int bitmap_load(struct mddev *mddev) |
69e51b44 N |
1769 | { |
1770 | int err = 0; | |
3520fa4d | 1771 | sector_t start = 0; |
69e51b44 N |
1772 | sector_t sector = 0; |
1773 | struct bitmap *bitmap = mddev->bitmap; | |
1774 | ||
1775 | if (!bitmap) | |
1776 | goto out; | |
1777 | ||
1778 | /* Clear out old bitmap info first: Either there is none, or we | |
1779 | * are resuming after someone else has possibly changed things, | |
1780 | * so we should forget old cached info. | |
1781 | * All chunks should be clean, but some might need_sync. | |
1782 | */ | |
1783 | while (sector < mddev->resync_max_sectors) { | |
57dab0bd | 1784 | sector_t blocks; |
69e51b44 N |
1785 | bitmap_start_sync(bitmap, sector, &blocks, 0); |
1786 | sector += blocks; | |
1787 | } | |
1788 | bitmap_close_sync(bitmap); | |
1789 | ||
3520fa4d JB |
1790 | if (mddev->degraded == 0 |
1791 | || bitmap->events_cleared == mddev->events) | |
1792 | /* no need to keep dirty bits to optimise a | |
1793 | * re-add of a missing device */ | |
1794 | start = mddev->recovery_cp; | |
1795 | ||
afbaa90b | 1796 | mutex_lock(&mddev->bitmap_info.mutex); |
3520fa4d | 1797 | err = bitmap_init_from_disk(bitmap, start); |
afbaa90b | 1798 | mutex_unlock(&mddev->bitmap_info.mutex); |
3520fa4d | 1799 | |
32a7627c | 1800 | if (err) |
69e51b44 | 1801 | goto out; |
edbb79df | 1802 | bitmap->flags &= ~BITMAP_STALE; |
ef99bf48 N |
1803 | |
1804 | /* Kick recovery in case any bits were set */ | |
1805 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
3178b0db | 1806 | |
1b04be96 | 1807 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1808 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1809 | |
4ad13663 N |
1810 | bitmap_update_sb(bitmap); |
1811 | ||
69e51b44 N |
1812 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
1813 | err = -EIO; | |
1814 | out: | |
3178b0db | 1815 | return err; |
32a7627c | 1816 | } |
69e51b44 | 1817 | EXPORT_SYMBOL_GPL(bitmap_load); |
32a7627c | 1818 | |
57148964 N |
1819 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap) |
1820 | { | |
1821 | unsigned long chunk_kb; | |
1822 | unsigned long flags; | |
1823 | ||
1824 | if (!bitmap) | |
1825 | return; | |
1826 | ||
1827 | spin_lock_irqsave(&bitmap->lock, flags); | |
1828 | chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; | |
1829 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
1830 | "%lu%s chunk", | |
1831 | bitmap->pages - bitmap->missing_pages, | |
1832 | bitmap->pages, | |
1833 | (bitmap->pages - bitmap->missing_pages) | |
1834 | << (PAGE_SHIFT - 10), | |
1835 | chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, | |
1836 | chunk_kb ? "KB" : "B"); | |
1ec885cd | 1837 | if (bitmap->storage.file) { |
57148964 | 1838 | seq_printf(seq, ", file: "); |
1ec885cd | 1839 | seq_path(seq, &bitmap->storage.file->f_path, " \t\n"); |
57148964 N |
1840 | } |
1841 | ||
1842 | seq_printf(seq, "\n"); | |
1843 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1844 | } | |
1845 | ||
43a70507 | 1846 | static ssize_t |
fd01b88c | 1847 | location_show(struct mddev *mddev, char *page) |
43a70507 N |
1848 | { |
1849 | ssize_t len; | |
ac2f40be | 1850 | if (mddev->bitmap_info.file) |
43a70507 | 1851 | len = sprintf(page, "file"); |
ac2f40be | 1852 | else if (mddev->bitmap_info.offset) |
43a70507 | 1853 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 1854 | else |
43a70507 N |
1855 | len = sprintf(page, "none"); |
1856 | len += sprintf(page+len, "\n"); | |
1857 | return len; | |
1858 | } | |
1859 | ||
1860 | static ssize_t | |
fd01b88c | 1861 | location_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
1862 | { |
1863 | ||
1864 | if (mddev->pers) { | |
1865 | if (!mddev->pers->quiesce) | |
1866 | return -EBUSY; | |
1867 | if (mddev->recovery || mddev->sync_thread) | |
1868 | return -EBUSY; | |
1869 | } | |
1870 | ||
1871 | if (mddev->bitmap || mddev->bitmap_info.file || | |
1872 | mddev->bitmap_info.offset) { | |
1873 | /* bitmap already configured. Only option is to clear it */ | |
1874 | if (strncmp(buf, "none", 4) != 0) | |
1875 | return -EBUSY; | |
1876 | if (mddev->pers) { | |
1877 | mddev->pers->quiesce(mddev, 1); | |
1878 | bitmap_destroy(mddev); | |
1879 | mddev->pers->quiesce(mddev, 0); | |
1880 | } | |
1881 | mddev->bitmap_info.offset = 0; | |
1882 | if (mddev->bitmap_info.file) { | |
1883 | struct file *f = mddev->bitmap_info.file; | |
1884 | mddev->bitmap_info.file = NULL; | |
1885 | restore_bitmap_write_access(f); | |
1886 | fput(f); | |
1887 | } | |
1888 | } else { | |
1889 | /* No bitmap, OK to set a location */ | |
1890 | long long offset; | |
1891 | if (strncmp(buf, "none", 4) == 0) | |
1892 | /* nothing to be done */; | |
1893 | else if (strncmp(buf, "file:", 5) == 0) { | |
1894 | /* Not supported yet */ | |
1895 | return -EINVAL; | |
1896 | } else { | |
1897 | int rv; | |
1898 | if (buf[0] == '+') | |
1899 | rv = strict_strtoll(buf+1, 10, &offset); | |
1900 | else | |
1901 | rv = strict_strtoll(buf, 10, &offset); | |
1902 | if (rv) | |
1903 | return rv; | |
1904 | if (offset == 0) | |
1905 | return -EINVAL; | |
ece5cff0 N |
1906 | if (mddev->bitmap_info.external == 0 && |
1907 | mddev->major_version == 0 && | |
43a70507 N |
1908 | offset != mddev->bitmap_info.default_offset) |
1909 | return -EINVAL; | |
1910 | mddev->bitmap_info.offset = offset; | |
1911 | if (mddev->pers) { | |
1912 | mddev->pers->quiesce(mddev, 1); | |
1913 | rv = bitmap_create(mddev); | |
4474ca42 N |
1914 | if (!rv) |
1915 | rv = bitmap_load(mddev); | |
43a70507 N |
1916 | if (rv) { |
1917 | bitmap_destroy(mddev); | |
1918 | mddev->bitmap_info.offset = 0; | |
1919 | } | |
1920 | mddev->pers->quiesce(mddev, 0); | |
1921 | if (rv) | |
1922 | return rv; | |
1923 | } | |
1924 | } | |
1925 | } | |
1926 | if (!mddev->external) { | |
1927 | /* Ensure new bitmap info is stored in | |
1928 | * metadata promptly. | |
1929 | */ | |
1930 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1931 | md_wakeup_thread(mddev->thread); | |
1932 | } | |
1933 | return len; | |
1934 | } | |
1935 | ||
1936 | static struct md_sysfs_entry bitmap_location = | |
1937 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
1938 | ||
6409bb05 N |
1939 | /* 'bitmap/space' is the space available at 'location' for the |
1940 | * bitmap. This allows the kernel to know when it is safe to | |
1941 | * resize the bitmap to match a resized array. | |
1942 | */ | |
1943 | static ssize_t | |
1944 | space_show(struct mddev *mddev, char *page) | |
1945 | { | |
1946 | return sprintf(page, "%lu\n", mddev->bitmap_info.space); | |
1947 | } | |
1948 | ||
1949 | static ssize_t | |
1950 | space_store(struct mddev *mddev, const char *buf, size_t len) | |
1951 | { | |
1952 | unsigned long sectors; | |
1953 | int rv; | |
1954 | ||
1955 | rv = kstrtoul(buf, 10, §ors); | |
1956 | if (rv) | |
1957 | return rv; | |
1958 | ||
1959 | if (sectors == 0) | |
1960 | return -EINVAL; | |
1961 | ||
1962 | if (mddev->bitmap && | |
9b1215c1 | 1963 | sectors < (mddev->bitmap->storage.bytes + 511) >> 9) |
6409bb05 N |
1964 | return -EFBIG; /* Bitmap is too big for this small space */ |
1965 | ||
1966 | /* could make sure it isn't too big, but that isn't really | |
1967 | * needed - user-space should be careful. | |
1968 | */ | |
1969 | mddev->bitmap_info.space = sectors; | |
1970 | return len; | |
1971 | } | |
1972 | ||
1973 | static struct md_sysfs_entry bitmap_space = | |
1974 | __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); | |
1975 | ||
43a70507 | 1976 | static ssize_t |
fd01b88c | 1977 | timeout_show(struct mddev *mddev, char *page) |
43a70507 N |
1978 | { |
1979 | ssize_t len; | |
1980 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
1981 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 1982 | |
43a70507 N |
1983 | len = sprintf(page, "%lu", secs); |
1984 | if (jifs) | |
1985 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
1986 | len += sprintf(page+len, "\n"); | |
1987 | return len; | |
1988 | } | |
1989 | ||
1990 | static ssize_t | |
fd01b88c | 1991 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
1992 | { |
1993 | /* timeout can be set at any time */ | |
1994 | unsigned long timeout; | |
1995 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
1996 | if (rv) | |
1997 | return rv; | |
1998 | ||
1999 | /* just to make sure we don't overflow... */ | |
2000 | if (timeout >= LONG_MAX / HZ) | |
2001 | return -EINVAL; | |
2002 | ||
2003 | timeout = timeout * HZ / 10000; | |
2004 | ||
2005 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
2006 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
2007 | if (timeout < 1) | |
2008 | timeout = 1; | |
2009 | mddev->bitmap_info.daemon_sleep = timeout; | |
2010 | if (mddev->thread) { | |
2011 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
2012 | * the bitmap is all clean and we don't need to | |
2013 | * adjust the timeout right now | |
2014 | */ | |
2015 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2016 | mddev->thread->timeout = timeout; | |
2017 | md_wakeup_thread(mddev->thread); | |
2018 | } | |
2019 | } | |
2020 | return len; | |
2021 | } | |
2022 | ||
2023 | static struct md_sysfs_entry bitmap_timeout = | |
2024 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2025 | ||
2026 | static ssize_t | |
fd01b88c | 2027 | backlog_show(struct mddev *mddev, char *page) |
43a70507 N |
2028 | { |
2029 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2030 | } | |
2031 | ||
2032 | static ssize_t | |
fd01b88c | 2033 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2034 | { |
2035 | unsigned long backlog; | |
2036 | int rv = strict_strtoul(buf, 10, &backlog); | |
2037 | if (rv) | |
2038 | return rv; | |
2039 | if (backlog > COUNTER_MAX) | |
2040 | return -EINVAL; | |
2041 | mddev->bitmap_info.max_write_behind = backlog; | |
2042 | return len; | |
2043 | } | |
2044 | ||
2045 | static struct md_sysfs_entry bitmap_backlog = | |
2046 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2047 | ||
2048 | static ssize_t | |
fd01b88c | 2049 | chunksize_show(struct mddev *mddev, char *page) |
43a70507 N |
2050 | { |
2051 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2052 | } | |
2053 | ||
2054 | static ssize_t | |
fd01b88c | 2055 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2056 | { |
2057 | /* Can only be changed when no bitmap is active */ | |
2058 | int rv; | |
2059 | unsigned long csize; | |
2060 | if (mddev->bitmap) | |
2061 | return -EBUSY; | |
2062 | rv = strict_strtoul(buf, 10, &csize); | |
2063 | if (rv) | |
2064 | return rv; | |
2065 | if (csize < 512 || | |
2066 | !is_power_of_2(csize)) | |
2067 | return -EINVAL; | |
2068 | mddev->bitmap_info.chunksize = csize; | |
2069 | return len; | |
2070 | } | |
2071 | ||
2072 | static struct md_sysfs_entry bitmap_chunksize = | |
2073 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2074 | ||
fd01b88c | 2075 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2076 | { |
2077 | return sprintf(page, "%s\n", (mddev->bitmap_info.external | |
2078 | ? "external" : "internal")); | |
2079 | } | |
2080 | ||
fd01b88c | 2081 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2082 | { |
2083 | if (mddev->bitmap || | |
2084 | mddev->bitmap_info.file || | |
2085 | mddev->bitmap_info.offset) | |
2086 | return -EBUSY; | |
2087 | if (strncmp(buf, "external", 8) == 0) | |
2088 | mddev->bitmap_info.external = 1; | |
2089 | else if (strncmp(buf, "internal", 8) == 0) | |
2090 | mddev->bitmap_info.external = 0; | |
2091 | else | |
2092 | return -EINVAL; | |
2093 | return len; | |
2094 | } | |
2095 | ||
2096 | static struct md_sysfs_entry bitmap_metadata = | |
2097 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2098 | ||
fd01b88c | 2099 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2100 | { |
2101 | int len; | |
2102 | if (mddev->bitmap) | |
2103 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2104 | "false" : "true")); | |
2105 | else | |
2106 | len = sprintf(page, "\n"); | |
2107 | return len; | |
2108 | } | |
2109 | ||
fd01b88c | 2110 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2111 | { |
2112 | if (mddev->bitmap == NULL) | |
2113 | return -ENOENT; | |
2114 | if (strncmp(buf, "false", 5) == 0) | |
2115 | mddev->bitmap->need_sync = 1; | |
2116 | else if (strncmp(buf, "true", 4) == 0) { | |
2117 | if (mddev->degraded) | |
2118 | return -EBUSY; | |
2119 | mddev->bitmap->need_sync = 0; | |
2120 | } else | |
2121 | return -EINVAL; | |
2122 | return len; | |
2123 | } | |
2124 | ||
2125 | static struct md_sysfs_entry bitmap_can_clear = | |
2126 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2127 | ||
696fcd53 | 2128 | static ssize_t |
fd01b88c | 2129 | behind_writes_used_show(struct mddev *mddev, char *page) |
696fcd53 PC |
2130 | { |
2131 | if (mddev->bitmap == NULL) | |
2132 | return sprintf(page, "0\n"); | |
2133 | return sprintf(page, "%lu\n", | |
2134 | mddev->bitmap->behind_writes_used); | |
2135 | } | |
2136 | ||
2137 | static ssize_t | |
fd01b88c | 2138 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
696fcd53 PC |
2139 | { |
2140 | if (mddev->bitmap) | |
2141 | mddev->bitmap->behind_writes_used = 0; | |
2142 | return len; | |
2143 | } | |
2144 | ||
2145 | static struct md_sysfs_entry max_backlog_used = | |
2146 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2147 | behind_writes_used_show, behind_writes_used_reset); | |
2148 | ||
43a70507 N |
2149 | static struct attribute *md_bitmap_attrs[] = { |
2150 | &bitmap_location.attr, | |
6409bb05 | 2151 | &bitmap_space.attr, |
43a70507 N |
2152 | &bitmap_timeout.attr, |
2153 | &bitmap_backlog.attr, | |
2154 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2155 | &bitmap_metadata.attr, |
2156 | &bitmap_can_clear.attr, | |
696fcd53 | 2157 | &max_backlog_used.attr, |
43a70507 N |
2158 | NULL |
2159 | }; | |
2160 | struct attribute_group md_bitmap_group = { | |
2161 | .name = "bitmap", | |
2162 | .attrs = md_bitmap_attrs, | |
2163 | }; | |
2164 |