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 | ||
643 | enum bitmap_mask_op { | |
644 | MASK_SET, | |
645 | MASK_UNSET | |
646 | }; | |
647 | ||
4ad13663 N |
648 | /* record the state of the bitmap in the superblock. Return the old value */ |
649 | static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits, | |
650 | enum bitmap_mask_op op) | |
32a7627c N |
651 | { |
652 | bitmap_super_t *sb; | |
4ad13663 | 653 | int old; |
32a7627c | 654 | |
1ec885cd | 655 | if (!bitmap->storage.sb_page) /* can't set the state */ |
4ad13663 | 656 | return 0; |
1ec885cd | 657 | sb = kmap_atomic(bitmap->storage.sb_page); |
4ad13663 | 658 | old = le32_to_cpu(sb->state) & bits; |
32a7627c | 659 | switch (op) { |
ac2f40be N |
660 | case MASK_SET: |
661 | sb->state |= cpu_to_le32(bits); | |
8258c532 | 662 | bitmap->flags |= bits; |
ac2f40be N |
663 | break; |
664 | case MASK_UNSET: | |
665 | sb->state &= cpu_to_le32(~bits); | |
8258c532 | 666 | bitmap->flags &= ~bits; |
ac2f40be N |
667 | break; |
668 | default: | |
669 | BUG(); | |
32a7627c | 670 | } |
b2f46e68 | 671 | kunmap_atomic(sb); |
4ad13663 | 672 | return old; |
32a7627c N |
673 | } |
674 | ||
675 | /* | |
676 | * general bitmap file operations | |
677 | */ | |
678 | ||
ece5cff0 N |
679 | /* |
680 | * on-disk bitmap: | |
681 | * | |
682 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
683 | * file a page at a time. There's a superblock at the start of the file. | |
684 | */ | |
32a7627c | 685 | /* calculate the index of the page that contains this bit */ |
1ec885cd N |
686 | static inline unsigned long file_page_index(struct bitmap_storage *store, |
687 | unsigned long chunk) | |
32a7627c | 688 | { |
1ec885cd | 689 | if (store->sb_page) |
ece5cff0 N |
690 | chunk += sizeof(bitmap_super_t) << 3; |
691 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
692 | } |
693 | ||
694 | /* calculate the (bit) offset of this bit within a page */ | |
1ec885cd N |
695 | static inline unsigned long file_page_offset(struct bitmap_storage *store, |
696 | unsigned long chunk) | |
32a7627c | 697 | { |
1ec885cd | 698 | if (store->sb_page) |
ece5cff0 N |
699 | chunk += sizeof(bitmap_super_t) << 3; |
700 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
701 | } |
702 | ||
703 | /* | |
704 | * return a pointer to the page in the filemap that contains the given bit | |
705 | * | |
706 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
707 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
708 | * 0 or page 1 | |
709 | */ | |
1ec885cd | 710 | static inline struct page *filemap_get_page(struct bitmap_storage *store, |
3520fa4d | 711 | unsigned long chunk) |
32a7627c | 712 | { |
1ec885cd | 713 | if (file_page_index(store, chunk) >= store->file_pages) |
ac2f40be | 714 | return NULL; |
1ec885cd N |
715 | return store->filemap[file_page_index(store, chunk) |
716 | - file_page_index(store, 0)]; | |
32a7627c N |
717 | } |
718 | ||
32a7627c N |
719 | static void bitmap_file_unmap(struct bitmap *bitmap) |
720 | { | |
721 | struct page **map, *sb_page; | |
722 | unsigned long *attr; | |
723 | int pages; | |
724 | unsigned long flags; | |
1ec885cd | 725 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c N |
726 | |
727 | spin_lock_irqsave(&bitmap->lock, flags); | |
1ec885cd N |
728 | map = store->filemap; |
729 | store->filemap = NULL; | |
730 | attr = store->filemap_attr; | |
731 | store->filemap_attr = NULL; | |
732 | pages = store->file_pages; | |
733 | store->file_pages = 0; | |
734 | sb_page = store->sb_page; | |
735 | store->sb_page = NULL; | |
32a7627c N |
736 | spin_unlock_irqrestore(&bitmap->lock, flags); |
737 | ||
738 | while (pages--) | |
ece5cff0 | 739 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 740 | free_buffers(map[pages]); |
32a7627c N |
741 | kfree(map); |
742 | kfree(attr); | |
743 | ||
d785a06a N |
744 | if (sb_page) |
745 | free_buffers(sb_page); | |
32a7627c N |
746 | } |
747 | ||
748 | static void bitmap_file_put(struct bitmap *bitmap) | |
749 | { | |
750 | struct file *file; | |
32a7627c N |
751 | unsigned long flags; |
752 | ||
753 | spin_lock_irqsave(&bitmap->lock, flags); | |
1ec885cd N |
754 | file = bitmap->storage.file; |
755 | bitmap->storage.file = NULL; | |
32a7627c N |
756 | spin_unlock_irqrestore(&bitmap->lock, flags); |
757 | ||
d785a06a N |
758 | if (file) |
759 | wait_event(bitmap->write_wait, | |
760 | atomic_read(&bitmap->pending_writes)==0); | |
32a7627c N |
761 | bitmap_file_unmap(bitmap); |
762 | ||
d785a06a | 763 | if (file) { |
c649bb9c | 764 | struct inode *inode = file->f_path.dentry->d_inode; |
fc0ecff6 | 765 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 766 | fput(file); |
d785a06a | 767 | } |
32a7627c N |
768 | } |
769 | ||
32a7627c N |
770 | /* |
771 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
772 | * then it is no longer reliable, so we stop using it and we mark the file | |
773 | * as failed in the superblock | |
774 | */ | |
775 | static void bitmap_file_kick(struct bitmap *bitmap) | |
776 | { | |
777 | char *path, *ptr = NULL; | |
778 | ||
4ad13663 N |
779 | if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) { |
780 | bitmap_update_sb(bitmap); | |
32a7627c | 781 | |
1ec885cd | 782 | if (bitmap->storage.file) { |
4ad13663 N |
783 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); |
784 | if (path) | |
1ec885cd N |
785 | ptr = d_path(&bitmap->storage.file->f_path, |
786 | path, PAGE_SIZE); | |
6bcfd601 | 787 | |
4ad13663 N |
788 | printk(KERN_ALERT |
789 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 790 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 791 | |
4ad13663 N |
792 | kfree(path); |
793 | } else | |
794 | printk(KERN_ALERT | |
795 | "%s: disabling internal bitmap due to errors\n", | |
796 | bmname(bitmap)); | |
a654b9d8 | 797 | } |
32a7627c N |
798 | |
799 | bitmap_file_put(bitmap); | |
800 | ||
801 | return; | |
802 | } | |
803 | ||
804 | enum bitmap_page_attr { | |
ac2f40be | 805 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
5a537df4 N |
806 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
807 | * i.e. counter is 1 or 2. */ | |
ac2f40be | 808 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
32a7627c N |
809 | }; |
810 | ||
d189122d N |
811 | static inline void set_page_attr(struct bitmap *bitmap, int pnum, |
812 | enum bitmap_page_attr attr) | |
32a7627c | 813 | { |
1ec885cd | 814 | __set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
815 | } |
816 | ||
d189122d N |
817 | static inline void clear_page_attr(struct bitmap *bitmap, int pnum, |
818 | enum bitmap_page_attr attr) | |
32a7627c | 819 | { |
1ec885cd | 820 | __clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
821 | } |
822 | ||
d189122d | 823 | static inline unsigned long test_page_attr(struct bitmap *bitmap, int pnum, |
ec7a3197 | 824 | enum bitmap_page_attr attr) |
32a7627c | 825 | { |
1ec885cd | 826 | return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
827 | } |
828 | ||
829 | /* | |
830 | * bitmap_file_set_bit -- called before performing a write to the md device | |
831 | * to set (and eventually sync) a particular bit in the bitmap file | |
832 | * | |
833 | * we set the bit immediately, then we record the page number so that | |
834 | * when an unplug occurs, we can flush the dirty pages out to disk | |
835 | */ | |
836 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
837 | { | |
838 | unsigned long bit; | |
3520fa4d | 839 | struct page *page; |
32a7627c | 840 | void *kaddr; |
61a0d80c | 841 | unsigned long chunk = block >> bitmap->chunkshift; |
32a7627c | 842 | |
1ec885cd | 843 | page = filemap_get_page(&bitmap->storage, chunk); |
3520fa4d JB |
844 | if (!page) |
845 | return; | |
1ec885cd | 846 | bit = file_page_offset(&bitmap->storage, chunk); |
32a7627c | 847 | |
3520fa4d | 848 | /* set the bit */ |
b2f46e68 | 849 | kaddr = kmap_atomic(page); |
3520fa4d JB |
850 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
851 | set_bit(bit, kaddr); | |
852 | else | |
853 | __set_bit_le(bit, kaddr); | |
b2f46e68 | 854 | kunmap_atomic(kaddr); |
36a4e1fe | 855 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
32a7627c | 856 | /* record page number so it gets flushed to disk when unplug occurs */ |
d189122d | 857 | set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY); |
32a7627c N |
858 | } |
859 | ||
ef99bf48 N |
860 | static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) |
861 | { | |
862 | unsigned long bit; | |
863 | struct page *page; | |
864 | void *paddr; | |
865 | unsigned long chunk = block >> bitmap->chunkshift; | |
866 | ||
1ec885cd | 867 | page = filemap_get_page(&bitmap->storage, chunk); |
ef99bf48 N |
868 | if (!page) |
869 | return; | |
1ec885cd | 870 | bit = file_page_offset(&bitmap->storage, chunk); |
ef99bf48 N |
871 | paddr = kmap_atomic(page); |
872 | if (bitmap->flags & BITMAP_HOSTENDIAN) | |
873 | clear_bit(bit, paddr); | |
874 | else | |
875 | __clear_bit_le(bit, paddr); | |
876 | kunmap_atomic(paddr); | |
d189122d N |
877 | if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) { |
878 | set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING); | |
ef99bf48 N |
879 | bitmap->allclean = 0; |
880 | } | |
881 | } | |
882 | ||
32a7627c N |
883 | /* this gets called when the md device is ready to unplug its underlying |
884 | * (slave) device queues -- before we let any writes go down, we need to | |
885 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 886 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 887 | { |
ec7a3197 N |
888 | unsigned long i, flags; |
889 | int dirty, need_write; | |
32a7627c N |
890 | int wait = 0; |
891 | ||
1ec885cd | 892 | if (!bitmap || !bitmap->storage.filemap) |
4ad13663 | 893 | return; |
32a7627c N |
894 | |
895 | /* look at each page to see if there are any set bits that need to be | |
896 | * flushed out to disk */ | |
1ec885cd | 897 | for (i = 0; i < bitmap->storage.file_pages; i++) { |
32a7627c | 898 | spin_lock_irqsave(&bitmap->lock, flags); |
1ec885cd | 899 | if (!bitmap->storage.filemap) { |
32a7627c | 900 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 901 | return; |
32a7627c | 902 | } |
d189122d N |
903 | dirty = test_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
904 | need_write = test_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE); | |
905 | clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); | |
906 | clear_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE); | |
bf07bb7d | 907 | if (dirty || need_write) |
d189122d | 908 | clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); |
ec7a3197 | 909 | if (dirty) |
32a7627c N |
910 | wait = 1; |
911 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
912 | ||
ac2f40be | 913 | if (dirty || need_write) |
1ec885cd | 914 | write_page(bitmap, bitmap->storage.filemap[i], 0); |
32a7627c N |
915 | } |
916 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
1ec885cd | 917 | if (bitmap->storage.file) |
d785a06a N |
918 | wait_event(bitmap->write_wait, |
919 | atomic_read(&bitmap->pending_writes)==0); | |
0b79ccf0 | 920 | else |
a9701a30 | 921 | md_super_wait(bitmap->mddev); |
32a7627c | 922 | } |
d785a06a N |
923 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
924 | bitmap_file_kick(bitmap); | |
32a7627c | 925 | } |
ac2f40be | 926 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 927 | |
6a07997f | 928 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
929 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
930 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
931 | * memory mapping of the bitmap file | |
932 | * Special cases: | |
933 | * if there's no bitmap file, or if the bitmap file had been | |
934 | * previously kicked from the array, we mark all the bits as | |
935 | * 1's in order to cause a full resync. | |
6a07997f N |
936 | * |
937 | * We ignore all bits for sectors that end earlier than 'start'. | |
938 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 939 | */ |
6a07997f | 940 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
941 | { |
942 | unsigned long i, chunks, index, oldindex, bit; | |
27581e5a N |
943 | int pnum; |
944 | struct page *page = NULL; | |
32a7627c N |
945 | unsigned long num_pages, bit_cnt = 0; |
946 | struct file *file; | |
d785a06a | 947 | unsigned long bytes, offset; |
32a7627c N |
948 | int outofdate; |
949 | int ret = -ENOSPC; | |
ea03aff9 | 950 | void *paddr; |
1ec885cd | 951 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c N |
952 | |
953 | chunks = bitmap->chunks; | |
1ec885cd | 954 | file = store->file; |
32a7627c | 955 | |
ef99bf48 N |
956 | if (!file && !bitmap->mddev->bitmap_info.offset) { |
957 | /* No permanent bitmap - fill with '1s'. */ | |
1ec885cd N |
958 | store->filemap = NULL; |
959 | store->file_pages = 0; | |
ef99bf48 N |
960 | for (i = 0; i < chunks ; i++) { |
961 | /* if the disk bit is set, set the memory bit */ | |
962 | int needed = ((sector_t)(i+1) << (bitmap->chunkshift) | |
963 | >= start); | |
964 | bitmap_set_memory_bits(bitmap, | |
965 | (sector_t)i << bitmap->chunkshift, | |
966 | needed); | |
967 | } | |
968 | return 0; | |
969 | } | |
32a7627c | 970 | |
32a7627c | 971 | outofdate = bitmap->flags & BITMAP_STALE; |
32a7627c N |
972 | if (outofdate) |
973 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
974 | "recovery\n", bmname(bitmap)); | |
975 | ||
e384e585 | 976 | bytes = DIV_ROUND_UP(bitmap->chunks, 8); |
ece5cff0 N |
977 | if (!bitmap->mddev->bitmap_info.external) |
978 | bytes += sizeof(bitmap_super_t); | |
bc7f77de | 979 | |
9b1215c1 N |
980 | store->bytes = bytes; |
981 | ||
e384e585 | 982 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); |
bc7f77de | 983 | |
ece5cff0 | 984 | if (file && i_size_read(file->f_mapping->host) < bytes) { |
32a7627c N |
985 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
986 | bmname(bitmap), | |
987 | (unsigned long) i_size_read(file->f_mapping->host), | |
ece5cff0 | 988 | bytes); |
4ad13663 | 989 | goto err; |
32a7627c | 990 | } |
bc7f77de N |
991 | |
992 | ret = -ENOMEM; | |
993 | ||
1ec885cd N |
994 | store->filemap = kmalloc(sizeof(struct page *) |
995 | * num_pages, GFP_KERNEL); | |
996 | if (!store->filemap) | |
4ad13663 | 997 | goto err; |
32a7627c | 998 | |
27581e5a N |
999 | pnum = 0; |
1000 | offset = 0; | |
1ec885cd N |
1001 | if (store->sb_page) { |
1002 | store->filemap[0] = store->sb_page; | |
27581e5a N |
1003 | pnum = 1; |
1004 | offset = sizeof(bitmap_super_t); | |
1005 | } | |
1006 | for ( ; pnum < num_pages; pnum++) { | |
1ec885cd N |
1007 | store->filemap[pnum] = alloc_page(GFP_KERNEL); |
1008 | if (!store->filemap[pnum]) { | |
1009 | store->file_pages = pnum; | |
27581e5a N |
1010 | goto err; |
1011 | } | |
1012 | } | |
1ec885cd | 1013 | store->file_pages = pnum; |
27581e5a | 1014 | |
e16b68b6 | 1015 | /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */ |
1ec885cd | 1016 | store->filemap_attr = kzalloc( |
ac2f40be | 1017 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), |
e16b68b6 | 1018 | GFP_KERNEL); |
1ec885cd | 1019 | if (!store->filemap_attr) |
4ad13663 | 1020 | goto err; |
32a7627c | 1021 | |
32a7627c N |
1022 | oldindex = ~0L; |
1023 | ||
1024 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 1025 | int b; |
1ec885cd N |
1026 | index = file_page_index(&bitmap->storage, i); |
1027 | bit = file_page_offset(&bitmap->storage, i); | |
32a7627c | 1028 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 1029 | int count; |
32a7627c | 1030 | /* unmap the old page, we're done with it */ |
d785a06a | 1031 | if (index == num_pages-1) |
ece5cff0 | 1032 | count = bytes - index * PAGE_SIZE; |
d785a06a N |
1033 | else |
1034 | count = PAGE_SIZE; | |
1ec885cd | 1035 | page = store->filemap[index]; |
27581e5a N |
1036 | if (file) |
1037 | ret = read_page(file, index, bitmap, | |
1038 | count, page); | |
1039 | else | |
1040 | ret = read_sb_page( | |
1041 | bitmap->mddev, | |
1042 | bitmap->mddev->bitmap_info.offset, | |
1043 | page, | |
1044 | index, count); | |
1045 | ||
1046 | if (ret) | |
4ad13663 | 1047 | goto err; |
a654b9d8 | 1048 | |
32a7627c | 1049 | oldindex = index; |
32a7627c N |
1050 | |
1051 | if (outofdate) { | |
1052 | /* | |
1053 | * if bitmap is out of date, dirty the | |
ac2f40be | 1054 | * whole page and write it out |
32a7627c | 1055 | */ |
b2f46e68 | 1056 | paddr = kmap_atomic(page); |
ea03aff9 | 1057 | memset(paddr + offset, 0xff, |
6a07997f | 1058 | PAGE_SIZE - offset); |
b2f46e68 | 1059 | kunmap_atomic(paddr); |
4ad13663 N |
1060 | write_page(bitmap, page, 1); |
1061 | ||
1062 | ret = -EIO; | |
b74fd282 | 1063 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
4ad13663 | 1064 | goto err; |
32a7627c | 1065 | } |
32a7627c | 1066 | } |
b2f46e68 | 1067 | paddr = kmap_atomic(page); |
bd926c63 | 1068 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1069 | b = test_bit(bit, paddr); |
bd926c63 | 1070 | else |
6b33aff3 | 1071 | b = test_bit_le(bit, paddr); |
b2f46e68 | 1072 | kunmap_atomic(paddr); |
bd926c63 | 1073 | if (b) { |
32a7627c | 1074 | /* if the disk bit is set, set the memory bit */ |
61a0d80c | 1075 | int needed = ((sector_t)(i+1) << bitmap->chunkshift |
db305e50 N |
1076 | >= start); |
1077 | bitmap_set_memory_bits(bitmap, | |
61a0d80c | 1078 | (sector_t)i << bitmap->chunkshift, |
db305e50 | 1079 | needed); |
32a7627c N |
1080 | bit_cnt++; |
1081 | } | |
27581e5a | 1082 | offset = 0; |
32a7627c N |
1083 | } |
1084 | ||
32a7627c | 1085 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
9c81075f | 1086 | "read %lu/%lu pages, set %lu of %lu bits\n", |
1ec885cd N |
1087 | bmname(bitmap), store->file_pages, |
1088 | num_pages, bit_cnt, chunks); | |
4ad13663 N |
1089 | |
1090 | return 0; | |
32a7627c | 1091 | |
4ad13663 N |
1092 | err: |
1093 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1094 | bmname(bitmap), ret); | |
32a7627c N |
1095 | return ret; |
1096 | } | |
1097 | ||
a654b9d8 N |
1098 | void bitmap_write_all(struct bitmap *bitmap) |
1099 | { | |
1100 | /* We don't actually write all bitmap blocks here, | |
1101 | * just flag them as needing to be written | |
1102 | */ | |
ec7a3197 | 1103 | int i; |
a654b9d8 | 1104 | |
1ec885cd | 1105 | if (!bitmap || !bitmap->storage.filemap) |
ef99bf48 | 1106 | return; |
1ec885cd | 1107 | if (bitmap->storage.file) |
ef99bf48 N |
1108 | /* Only one copy, so nothing needed */ |
1109 | return; | |
1110 | ||
7c8f4247 | 1111 | spin_lock_irq(&bitmap->lock); |
1ec885cd | 1112 | for (i = 0; i < bitmap->storage.file_pages; i++) |
d189122d | 1113 | set_page_attr(bitmap, i, |
ec7a3197 | 1114 | BITMAP_PAGE_NEEDWRITE); |
2585f3ef | 1115 | bitmap->allclean = 0; |
7c8f4247 | 1116 | spin_unlock_irq(&bitmap->lock); |
a654b9d8 N |
1117 | } |
1118 | ||
32a7627c N |
1119 | static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) |
1120 | { | |
61a0d80c | 1121 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1122 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1123 | bitmap->bp[page].count += inc; | |
32a7627c N |
1124 | bitmap_checkfree(bitmap, page); |
1125 | } | |
bf07bb7d N |
1126 | |
1127 | static void bitmap_set_pending(struct bitmap *bitmap, sector_t offset) | |
1128 | { | |
1129 | sector_t chunk = offset >> bitmap->chunkshift; | |
1130 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1131 | struct bitmap_page *bp = &bitmap->bp[page]; | |
1132 | ||
1133 | if (!bp->pending) | |
1134 | bp->pending = 1; | |
1135 | } | |
1136 | ||
32a7627c | 1137 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
57dab0bd | 1138 | sector_t offset, sector_t *blocks, |
32a7627c N |
1139 | int create); |
1140 | ||
1141 | /* | |
1142 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1143 | * out to disk | |
1144 | */ | |
1145 | ||
fd01b88c | 1146 | void bitmap_daemon_work(struct mddev *mddev) |
32a7627c | 1147 | { |
aa5cbd10 | 1148 | struct bitmap *bitmap; |
aa3163f8 | 1149 | unsigned long j; |
bf07bb7d | 1150 | unsigned long nextpage; |
32a7627c | 1151 | unsigned long flags; |
57dab0bd | 1152 | sector_t blocks; |
32a7627c | 1153 | |
aa5cbd10 N |
1154 | /* Use a mutex to guard daemon_work against |
1155 | * bitmap_destroy. | |
1156 | */ | |
c3d9714e | 1157 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1158 | bitmap = mddev->bitmap; |
1159 | if (bitmap == NULL) { | |
c3d9714e | 1160 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1161 | return; |
aa5cbd10 | 1162 | } |
42a04b50 | 1163 | if (time_before(jiffies, bitmap->daemon_lastrun |
2e61ebbc | 1164 | + mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1165 | goto done; |
1166 | ||
32a7627c | 1167 | bitmap->daemon_lastrun = jiffies; |
8311c29d | 1168 | if (bitmap->allclean) { |
2e61ebbc | 1169 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
aa5cbd10 | 1170 | goto done; |
8311c29d N |
1171 | } |
1172 | bitmap->allclean = 1; | |
32a7627c | 1173 | |
bf07bb7d N |
1174 | /* Any file-page which is PENDING now needs to be written. |
1175 | * So set NEEDWRITE now, then after we make any last-minute changes | |
1176 | * we will write it. | |
1177 | */ | |
be512691 | 1178 | spin_lock_irqsave(&bitmap->lock, flags); |
1ec885cd | 1179 | for (j = 0; j < bitmap->storage.file_pages; j++) |
d189122d | 1180 | if (test_page_attr(bitmap, j, |
bf07bb7d | 1181 | BITMAP_PAGE_PENDING)) { |
d189122d | 1182 | set_page_attr(bitmap, j, |
bf07bb7d | 1183 | BITMAP_PAGE_NEEDWRITE); |
d189122d | 1184 | clear_page_attr(bitmap, j, |
bf07bb7d N |
1185 | BITMAP_PAGE_PENDING); |
1186 | } | |
1187 | ||
1188 | if (bitmap->need_sync && | |
1189 | mddev->bitmap_info.external == 0) { | |
1190 | /* Arrange for superblock update as well as | |
1191 | * other changes */ | |
1192 | bitmap_super_t *sb; | |
1193 | bitmap->need_sync = 0; | |
1ec885cd N |
1194 | if (bitmap->storage.filemap) { |
1195 | sb = kmap_atomic(bitmap->storage.sb_page); | |
ef99bf48 N |
1196 | sb->events_cleared = |
1197 | cpu_to_le64(bitmap->events_cleared); | |
1198 | kunmap_atomic(sb); | |
d189122d | 1199 | set_page_attr(bitmap, 0, |
ef99bf48 N |
1200 | BITMAP_PAGE_NEEDWRITE); |
1201 | } | |
bf07bb7d N |
1202 | } |
1203 | /* Now look at the bitmap counters and if any are '2' or '1', | |
1204 | * decrement and handle accordingly. | |
1205 | */ | |
1206 | nextpage = 0; | |
32a7627c N |
1207 | for (j = 0; j < bitmap->chunks; j++) { |
1208 | bitmap_counter_t *bmc; | |
ef99bf48 | 1209 | sector_t block = (sector_t)j << bitmap->chunkshift; |
3520fa4d | 1210 | |
bf07bb7d N |
1211 | if (j == nextpage) { |
1212 | nextpage += PAGE_COUNTER_RATIO; | |
1213 | if (!bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending) { | |
1214 | j |= PAGE_COUNTER_MASK; | |
aa3163f8 N |
1215 | continue; |
1216 | } | |
bf07bb7d | 1217 | bitmap->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; |
32a7627c | 1218 | } |
db305e50 | 1219 | bmc = bitmap_get_counter(bitmap, |
ef99bf48 | 1220 | block, |
db305e50 | 1221 | &blocks, 0); |
bf07bb7d N |
1222 | |
1223 | if (!bmc) { | |
5a537df4 | 1224 | j |= PAGE_COUNTER_MASK; |
bf07bb7d N |
1225 | continue; |
1226 | } | |
1227 | if (*bmc == 1 && !bitmap->need_sync) { | |
1228 | /* We can clear the bit */ | |
bf07bb7d | 1229 | *bmc = 0; |
ef99bf48 N |
1230 | bitmap_count_page(bitmap, block, -1); |
1231 | bitmap_file_clear_bit(bitmap, block); | |
bf07bb7d N |
1232 | } else if (*bmc && *bmc <= 2) { |
1233 | *bmc = 1; | |
ef99bf48 | 1234 | bitmap_set_pending(bitmap, block); |
bf07bb7d | 1235 | bitmap->allclean = 0; |
5a537df4 | 1236 | } |
32a7627c N |
1237 | } |
1238 | ||
bf07bb7d N |
1239 | /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. |
1240 | * DIRTY pages need to be written by bitmap_unplug so it can wait | |
1241 | * for them. | |
1242 | * If we find any DIRTY page we stop there and let bitmap_unplug | |
1243 | * handle all the rest. This is important in the case where | |
1244 | * the first blocking holds the superblock and it has been updated. | |
1245 | * We mustn't write any other blocks before the superblock. | |
1246 | */ | |
1ec885cd | 1247 | for (j = 0; j < bitmap->storage.file_pages; j++) { |
bf07bb7d | 1248 | |
d189122d | 1249 | if (test_page_attr(bitmap, j, |
bf07bb7d N |
1250 | BITMAP_PAGE_DIRTY)) |
1251 | /* bitmap_unplug will handle the rest */ | |
1252 | break; | |
d189122d | 1253 | if (test_page_attr(bitmap, j, |
bf07bb7d | 1254 | BITMAP_PAGE_NEEDWRITE)) { |
d189122d | 1255 | clear_page_attr(bitmap, j, |
bf07bb7d | 1256 | BITMAP_PAGE_NEEDWRITE); |
32a7627c | 1257 | spin_unlock_irqrestore(&bitmap->lock, flags); |
1ec885cd | 1258 | write_page(bitmap, bitmap->storage.filemap[j], 0); |
bf07bb7d | 1259 | spin_lock_irqsave(&bitmap->lock, flags); |
1ec885cd | 1260 | if (!bitmap->storage.filemap) |
bf07bb7d | 1261 | break; |
32a7627c | 1262 | } |
32a7627c | 1263 | } |
bf07bb7d | 1264 | spin_unlock_irqrestore(&bitmap->lock, flags); |
32a7627c | 1265 | |
7be3dfec | 1266 | done: |
8311c29d | 1267 | if (bitmap->allclean == 0) |
2e61ebbc N |
1268 | mddev->thread->timeout = |
1269 | mddev->bitmap_info.daemon_sleep; | |
c3d9714e | 1270 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1271 | } |
1272 | ||
32a7627c | 1273 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
57dab0bd | 1274 | sector_t offset, sector_t *blocks, |
32a7627c | 1275 | int create) |
ee305ace N |
1276 | __releases(bitmap->lock) |
1277 | __acquires(bitmap->lock) | |
32a7627c N |
1278 | { |
1279 | /* If 'create', we might release the lock and reclaim it. | |
1280 | * The lock must have been taken with interrupts enabled. | |
1281 | * If !create, we don't release the lock. | |
1282 | */ | |
61a0d80c | 1283 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1284 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1285 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1286 | sector_t csize; | |
ef425673 | 1287 | int err; |
32a7627c | 1288 | |
ef425673 N |
1289 | err = bitmap_checkpage(bitmap, page, create); |
1290 | ||
1291 | if (bitmap->bp[page].hijacked || | |
1292 | bitmap->bp[page].map == NULL) | |
61a0d80c | 1293 | csize = ((sector_t)1) << (bitmap->chunkshift + |
ef425673 N |
1294 | PAGE_COUNTER_SHIFT - 1); |
1295 | else | |
61a0d80c | 1296 | csize = ((sector_t)1) << bitmap->chunkshift; |
ef425673 N |
1297 | *blocks = csize - (offset & (csize - 1)); |
1298 | ||
1299 | if (err < 0) | |
32a7627c | 1300 | return NULL; |
ef425673 | 1301 | |
32a7627c N |
1302 | /* now locked ... */ |
1303 | ||
1304 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1305 | /* should we use the first or second counter field | |
1306 | * of the hijacked pointer? */ | |
1307 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1308 | return &((bitmap_counter_t *) |
1309 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1310 | } else /* page is allocated */ |
32a7627c N |
1311 | return (bitmap_counter_t *) |
1312 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1313 | } |
1314 | ||
4b6d287f | 1315 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1316 | { |
ac2f40be N |
1317 | if (!bitmap) |
1318 | return 0; | |
4b6d287f N |
1319 | |
1320 | if (behind) { | |
696fcd53 | 1321 | int bw; |
4b6d287f | 1322 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1323 | bw = atomic_read(&bitmap->behind_writes); |
1324 | if (bw > bitmap->behind_writes_used) | |
1325 | bitmap->behind_writes_used = bw; | |
1326 | ||
36a4e1fe N |
1327 | pr_debug("inc write-behind count %d/%lu\n", |
1328 | bw, bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1329 | } |
1330 | ||
32a7627c | 1331 | while (sectors) { |
57dab0bd | 1332 | sector_t blocks; |
32a7627c N |
1333 | bitmap_counter_t *bmc; |
1334 | ||
1335 | spin_lock_irq(&bitmap->lock); | |
1336 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); | |
1337 | if (!bmc) { | |
1338 | spin_unlock_irq(&bitmap->lock); | |
1339 | return 0; | |
1340 | } | |
1341 | ||
27d5ea04 | 1342 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
da6e1a32 NB |
1343 | DEFINE_WAIT(__wait); |
1344 | /* note that it is safe to do the prepare_to_wait | |
1345 | * after the test as long as we do it before dropping | |
1346 | * the spinlock. | |
1347 | */ | |
1348 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1349 | TASK_UNINTERRUPTIBLE); | |
1350 | spin_unlock_irq(&bitmap->lock); | |
7eaceacc | 1351 | io_schedule(); |
da6e1a32 NB |
1352 | finish_wait(&bitmap->overflow_wait, &__wait); |
1353 | continue; | |
1354 | } | |
1355 | ||
ac2f40be | 1356 | switch (*bmc) { |
32a7627c N |
1357 | case 0: |
1358 | bitmap_file_set_bit(bitmap, offset); | |
ac2f40be | 1359 | bitmap_count_page(bitmap, offset, 1); |
32a7627c N |
1360 | /* fall through */ |
1361 | case 1: | |
1362 | *bmc = 2; | |
1363 | } | |
da6e1a32 | 1364 | |
32a7627c N |
1365 | (*bmc)++; |
1366 | ||
1367 | spin_unlock_irq(&bitmap->lock); | |
1368 | ||
1369 | offset += blocks; | |
1370 | if (sectors > blocks) | |
1371 | sectors -= blocks; | |
ac2f40be N |
1372 | else |
1373 | sectors = 0; | |
32a7627c N |
1374 | } |
1375 | return 0; | |
1376 | } | |
ac2f40be | 1377 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1378 | |
1379 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1380 | int success, int behind) |
32a7627c | 1381 | { |
ac2f40be N |
1382 | if (!bitmap) |
1383 | return; | |
4b6d287f | 1384 | if (behind) { |
e555190d N |
1385 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1386 | wake_up(&bitmap->behind_wait); | |
36a4e1fe N |
1387 | pr_debug("dec write-behind count %d/%lu\n", |
1388 | atomic_read(&bitmap->behind_writes), | |
1389 | bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1390 | } |
1391 | ||
32a7627c | 1392 | while (sectors) { |
57dab0bd | 1393 | sector_t blocks; |
32a7627c N |
1394 | unsigned long flags; |
1395 | bitmap_counter_t *bmc; | |
1396 | ||
1397 | spin_lock_irqsave(&bitmap->lock, flags); | |
1398 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); | |
1399 | if (!bmc) { | |
1400 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1401 | return; | |
1402 | } | |
1403 | ||
961902c0 | 1404 | if (success && !bitmap->mddev->degraded && |
a0da84f3 NB |
1405 | bitmap->events_cleared < bitmap->mddev->events) { |
1406 | bitmap->events_cleared = bitmap->mddev->events; | |
1407 | bitmap->need_sync = 1; | |
5ff5afff | 1408 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1409 | } |
1410 | ||
27d5ea04 | 1411 | if (!success && !NEEDED(*bmc)) |
32a7627c N |
1412 | *bmc |= NEEDED_MASK; |
1413 | ||
27d5ea04 | 1414 | if (COUNTER(*bmc) == COUNTER_MAX) |
da6e1a32 NB |
1415 | wake_up(&bitmap->overflow_wait); |
1416 | ||
32a7627c | 1417 | (*bmc)--; |
2585f3ef | 1418 | if (*bmc <= 2) { |
bf07bb7d | 1419 | bitmap_set_pending(bitmap, offset); |
2585f3ef N |
1420 | bitmap->allclean = 0; |
1421 | } | |
32a7627c N |
1422 | spin_unlock_irqrestore(&bitmap->lock, flags); |
1423 | offset += blocks; | |
1424 | if (sectors > blocks) | |
1425 | sectors -= blocks; | |
ac2f40be N |
1426 | else |
1427 | sectors = 0; | |
32a7627c N |
1428 | } |
1429 | } | |
ac2f40be | 1430 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1431 | |
57dab0bd | 1432 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1433 | int degraded) |
32a7627c N |
1434 | { |
1435 | bitmap_counter_t *bmc; | |
1436 | int rv; | |
1437 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1438 | *blocks = 1024; | |
1439 | return 1; /* always resync if no bitmap */ | |
1440 | } | |
1441 | spin_lock_irq(&bitmap->lock); | |
1442 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1443 | rv = 0; | |
1444 | if (bmc) { | |
1445 | /* locked */ | |
1446 | if (RESYNC(*bmc)) | |
1447 | rv = 1; | |
1448 | else if (NEEDED(*bmc)) { | |
1449 | rv = 1; | |
6a806c51 N |
1450 | if (!degraded) { /* don't set/clear bits if degraded */ |
1451 | *bmc |= RESYNC_MASK; | |
1452 | *bmc &= ~NEEDED_MASK; | |
1453 | } | |
32a7627c N |
1454 | } |
1455 | } | |
1456 | spin_unlock_irq(&bitmap->lock); | |
1457 | return rv; | |
1458 | } | |
1459 | ||
57dab0bd | 1460 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a N |
1461 | int degraded) |
1462 | { | |
1463 | /* bitmap_start_sync must always report on multiples of whole | |
1464 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1465 | * get confused. | |
1466 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1467 | * At least PAGE_SIZE>>9 blocks are covered. | |
1468 | * Return the 'or' of the result. | |
1469 | */ | |
1470 | int rv = 0; | |
57dab0bd | 1471 | sector_t blocks1; |
1187cf0a N |
1472 | |
1473 | *blocks = 0; | |
1474 | while (*blocks < (PAGE_SIZE>>9)) { | |
1475 | rv |= __bitmap_start_sync(bitmap, offset, | |
1476 | &blocks1, degraded); | |
1477 | offset += blocks1; | |
1478 | *blocks += blocks1; | |
1479 | } | |
1480 | return rv; | |
1481 | } | |
ac2f40be | 1482 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1483 | |
57dab0bd | 1484 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1485 | { |
1486 | bitmap_counter_t *bmc; | |
1487 | unsigned long flags; | |
ac2f40be N |
1488 | |
1489 | if (bitmap == NULL) { | |
32a7627c N |
1490 | *blocks = 1024; |
1491 | return; | |
1492 | } | |
1493 | spin_lock_irqsave(&bitmap->lock, flags); | |
1494 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1495 | if (bmc == NULL) | |
1496 | goto unlock; | |
1497 | /* locked */ | |
32a7627c N |
1498 | if (RESYNC(*bmc)) { |
1499 | *bmc &= ~RESYNC_MASK; | |
1500 | ||
1501 | if (!NEEDED(*bmc) && aborted) | |
1502 | *bmc |= NEEDED_MASK; | |
1503 | else { | |
2585f3ef | 1504 | if (*bmc <= 2) { |
bf07bb7d | 1505 | bitmap_set_pending(bitmap, offset); |
2585f3ef N |
1506 | bitmap->allclean = 0; |
1507 | } | |
32a7627c N |
1508 | } |
1509 | } | |
1510 | unlock: | |
1511 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1512 | } | |
ac2f40be | 1513 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1514 | |
1515 | void bitmap_close_sync(struct bitmap *bitmap) | |
1516 | { | |
1517 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1518 | * properly have been aborted. It remains to us to clear the | |
1519 | * RESYNC bit wherever it is still on | |
1520 | */ | |
1521 | sector_t sector = 0; | |
57dab0bd | 1522 | sector_t blocks; |
b47490c9 N |
1523 | if (!bitmap) |
1524 | return; | |
32a7627c N |
1525 | while (sector < bitmap->mddev->resync_max_sectors) { |
1526 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1527 | sector += blocks; |
1528 | } | |
1529 | } | |
ac2f40be | 1530 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 N |
1531 | |
1532 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1533 | { | |
1534 | sector_t s = 0; | |
57dab0bd | 1535 | sector_t blocks; |
b47490c9 N |
1536 | |
1537 | if (!bitmap) | |
1538 | return; | |
1539 | if (sector == 0) { | |
1540 | bitmap->last_end_sync = jiffies; | |
1541 | return; | |
1542 | } | |
1543 | if (time_before(jiffies, (bitmap->last_end_sync | |
1b04be96 | 1544 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1545 | return; |
1546 | wait_event(bitmap->mddev->recovery_wait, | |
1547 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1548 | ||
75d3da43 | 1549 | bitmap->mddev->curr_resync_completed = sector; |
070dc6dd | 1550 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
61a0d80c | 1551 | sector &= ~((1ULL << bitmap->chunkshift) - 1); |
b47490c9 N |
1552 | s = 0; |
1553 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1554 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1555 | s += blocks; | |
32a7627c | 1556 | } |
b47490c9 | 1557 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1558 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1559 | } |
ac2f40be | 1560 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1561 | |
6a07997f | 1562 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1563 | { |
1564 | /* For each chunk covered by any of these sectors, set the | |
ef99bf48 | 1565 | * counter to 2 and possibly set resync_needed. They should all |
32a7627c N |
1566 | * be 0 at this point |
1567 | */ | |
193f1c93 | 1568 | |
57dab0bd | 1569 | sector_t secs; |
193f1c93 N |
1570 | bitmap_counter_t *bmc; |
1571 | spin_lock_irq(&bitmap->lock); | |
1572 | bmc = bitmap_get_counter(bitmap, offset, &secs, 1); | |
1573 | if (!bmc) { | |
32a7627c | 1574 | spin_unlock_irq(&bitmap->lock); |
193f1c93 | 1575 | return; |
32a7627c | 1576 | } |
ac2f40be | 1577 | if (!*bmc) { |
915c420d | 1578 | *bmc = 2 | (needed ? NEEDED_MASK : 0); |
193f1c93 | 1579 | bitmap_count_page(bitmap, offset, 1); |
bf07bb7d | 1580 | bitmap_set_pending(bitmap, offset); |
2585f3ef | 1581 | bitmap->allclean = 0; |
193f1c93 N |
1582 | } |
1583 | spin_unlock_irq(&bitmap->lock); | |
32a7627c N |
1584 | } |
1585 | ||
9b1d1dac PC |
1586 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1587 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1588 | { | |
1589 | unsigned long chunk; | |
1590 | ||
1591 | for (chunk = s; chunk <= e; chunk++) { | |
61a0d80c | 1592 | sector_t sec = (sector_t)chunk << bitmap->chunkshift; |
9b1d1dac | 1593 | bitmap_set_memory_bits(bitmap, sec, 1); |
7c8f4247 | 1594 | spin_lock_irq(&bitmap->lock); |
9b1d1dac | 1595 | bitmap_file_set_bit(bitmap, sec); |
7c8f4247 | 1596 | spin_unlock_irq(&bitmap->lock); |
ffa23322 N |
1597 | if (sec < bitmap->mddev->recovery_cp) |
1598 | /* We are asserting that the array is dirty, | |
1599 | * so move the recovery_cp address back so | |
1600 | * that it is obvious that it is dirty | |
1601 | */ | |
1602 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1603 | } |
1604 | } | |
1605 | ||
6b8b3e8a N |
1606 | /* |
1607 | * flush out any pending updates | |
1608 | */ | |
fd01b88c | 1609 | void bitmap_flush(struct mddev *mddev) |
6b8b3e8a N |
1610 | { |
1611 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1612 | long sleep; |
6b8b3e8a N |
1613 | |
1614 | if (!bitmap) /* there was no bitmap */ | |
1615 | return; | |
1616 | ||
1617 | /* run the daemon_work three time to ensure everything is flushed | |
1618 | * that can be | |
1619 | */ | |
1b04be96 | 1620 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1621 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1622 | bitmap_daemon_work(mddev); |
42a04b50 | 1623 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1624 | bitmap_daemon_work(mddev); |
42a04b50 | 1625 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1626 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1627 | bitmap_update_sb(bitmap); |
1628 | } | |
1629 | ||
32a7627c N |
1630 | /* |
1631 | * free memory that was allocated | |
1632 | */ | |
3178b0db | 1633 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1634 | { |
1635 | unsigned long k, pages; | |
1636 | struct bitmap_page *bp; | |
32a7627c N |
1637 | |
1638 | if (!bitmap) /* there was no bitmap */ | |
1639 | return; | |
1640 | ||
32a7627c N |
1641 | /* release the bitmap file and kill the daemon */ |
1642 | bitmap_file_put(bitmap); | |
1643 | ||
1644 | bp = bitmap->bp; | |
1645 | pages = bitmap->pages; | |
1646 | ||
1647 | /* free all allocated memory */ | |
1648 | ||
32a7627c N |
1649 | if (bp) /* deallocate the page memory */ |
1650 | for (k = 0; k < pages; k++) | |
1651 | if (bp[k].map && !bp[k].hijacked) | |
1652 | kfree(bp[k].map); | |
1653 | kfree(bp); | |
1654 | kfree(bitmap); | |
1655 | } | |
aa5cbd10 | 1656 | |
fd01b88c | 1657 | void bitmap_destroy(struct mddev *mddev) |
3178b0db N |
1658 | { |
1659 | struct bitmap *bitmap = mddev->bitmap; | |
1660 | ||
1661 | if (!bitmap) /* there was no bitmap */ | |
1662 | return; | |
1663 | ||
c3d9714e | 1664 | mutex_lock(&mddev->bitmap_info.mutex); |
3178b0db | 1665 | mddev->bitmap = NULL; /* disconnect from the md device */ |
c3d9714e | 1666 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1667 | if (mddev->thread) |
1668 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db | 1669 | |
ece5cff0 N |
1670 | if (bitmap->sysfs_can_clear) |
1671 | sysfs_put(bitmap->sysfs_can_clear); | |
1672 | ||
3178b0db N |
1673 | bitmap_free(bitmap); |
1674 | } | |
32a7627c N |
1675 | |
1676 | /* | |
1677 | * initialize the bitmap structure | |
1678 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1679 | */ | |
fd01b88c | 1680 | int bitmap_create(struct mddev *mddev) |
32a7627c N |
1681 | { |
1682 | struct bitmap *bitmap; | |
1f593903 | 1683 | sector_t blocks = mddev->resync_max_sectors; |
32a7627c N |
1684 | unsigned long chunks; |
1685 | unsigned long pages; | |
c3d9714e | 1686 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1687 | int err; |
5ff5afff | 1688 | struct sysfs_dirent *bm = NULL; |
32a7627c | 1689 | |
5f6e3c83 | 1690 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1691 | |
c3d9714e | 1692 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1693 | |
9ffae0cf | 1694 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1695 | if (!bitmap) |
1696 | return -ENOMEM; | |
1697 | ||
32a7627c | 1698 | spin_lock_init(&bitmap->lock); |
ce25c31b N |
1699 | atomic_set(&bitmap->pending_writes, 0); |
1700 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1701 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1702 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1703 | |
32a7627c | 1704 | bitmap->mddev = mddev; |
32a7627c | 1705 | |
5ff5afff N |
1706 | if (mddev->kobj.sd) |
1707 | bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap"); | |
ece5cff0 | 1708 | if (bm) { |
3ff195b0 | 1709 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear"); |
ece5cff0 N |
1710 | sysfs_put(bm); |
1711 | } else | |
1712 | bitmap->sysfs_can_clear = NULL; | |
1713 | ||
1ec885cd | 1714 | bitmap->storage.file = file; |
ce25c31b N |
1715 | if (file) { |
1716 | get_file(file); | |
ae8fa283 N |
1717 | /* As future accesses to this file will use bmap, |
1718 | * and bypass the page cache, we must sync the file | |
1719 | * first. | |
1720 | */ | |
8018ab05 | 1721 | vfs_fsync(file, 1); |
ce25c31b | 1722 | } |
42a04b50 | 1723 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
9c81075f JB |
1724 | if (!mddev->bitmap_info.external) { |
1725 | /* | |
1726 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is | |
1727 | * instructing us to create a new on-disk bitmap instance. | |
1728 | */ | |
1729 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) | |
1730 | err = bitmap_new_disk_sb(bitmap); | |
1731 | else | |
1732 | err = bitmap_read_sb(bitmap); | |
1733 | } else { | |
ece5cff0 N |
1734 | err = 0; |
1735 | if (mddev->bitmap_info.chunksize == 0 || | |
1736 | mddev->bitmap_info.daemon_sleep == 0) | |
1737 | /* chunksize and time_base need to be | |
1738 | * set first. */ | |
1739 | err = -EINVAL; | |
1740 | } | |
32a7627c | 1741 | if (err) |
3178b0db | 1742 | goto error; |
32a7627c | 1743 | |
624ce4f5 | 1744 | bitmap->daemon_lastrun = jiffies; |
61a0d80c N |
1745 | bitmap->chunkshift = (ffz(~mddev->bitmap_info.chunksize) |
1746 | - BITMAP_BLOCK_SHIFT); | |
32a7627c | 1747 | |
b16b1b6c | 1748 | chunks = (blocks + (1 << bitmap->chunkshift) - 1) >> |
61a0d80c | 1749 | bitmap->chunkshift; |
ac2f40be | 1750 | pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; |
32a7627c N |
1751 | |
1752 | BUG_ON(!pages); | |
1753 | ||
1754 | bitmap->chunks = chunks; | |
1755 | bitmap->pages = pages; | |
1756 | bitmap->missing_pages = pages; | |
32a7627c | 1757 | |
9ffae0cf | 1758 | bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); |
29d3247e | 1759 | |
3178b0db | 1760 | err = -ENOMEM; |
32a7627c | 1761 | if (!bitmap->bp) |
3178b0db | 1762 | goto error; |
32a7627c | 1763 | |
69e51b44 N |
1764 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", |
1765 | pages, bmname(bitmap)); | |
1766 | ||
1767 | mddev->bitmap = bitmap; | |
1768 | ||
1769 | ||
1770 | return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0; | |
1771 | ||
1772 | error: | |
1773 | bitmap_free(bitmap); | |
1774 | return err; | |
1775 | } | |
1776 | ||
fd01b88c | 1777 | int bitmap_load(struct mddev *mddev) |
69e51b44 N |
1778 | { |
1779 | int err = 0; | |
3520fa4d | 1780 | sector_t start = 0; |
69e51b44 N |
1781 | sector_t sector = 0; |
1782 | struct bitmap *bitmap = mddev->bitmap; | |
1783 | ||
1784 | if (!bitmap) | |
1785 | goto out; | |
1786 | ||
1787 | /* Clear out old bitmap info first: Either there is none, or we | |
1788 | * are resuming after someone else has possibly changed things, | |
1789 | * so we should forget old cached info. | |
1790 | * All chunks should be clean, but some might need_sync. | |
1791 | */ | |
1792 | while (sector < mddev->resync_max_sectors) { | |
57dab0bd | 1793 | sector_t blocks; |
69e51b44 N |
1794 | bitmap_start_sync(bitmap, sector, &blocks, 0); |
1795 | sector += blocks; | |
1796 | } | |
1797 | bitmap_close_sync(bitmap); | |
1798 | ||
3520fa4d JB |
1799 | if (mddev->degraded == 0 |
1800 | || bitmap->events_cleared == mddev->events) | |
1801 | /* no need to keep dirty bits to optimise a | |
1802 | * re-add of a missing device */ | |
1803 | start = mddev->recovery_cp; | |
1804 | ||
afbaa90b | 1805 | mutex_lock(&mddev->bitmap_info.mutex); |
3520fa4d | 1806 | err = bitmap_init_from_disk(bitmap, start); |
afbaa90b | 1807 | mutex_unlock(&mddev->bitmap_info.mutex); |
3520fa4d | 1808 | |
32a7627c | 1809 | if (err) |
69e51b44 | 1810 | goto out; |
ef99bf48 N |
1811 | bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET); |
1812 | ||
1813 | /* Kick recovery in case any bits were set */ | |
1814 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
3178b0db | 1815 | |
1b04be96 | 1816 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1817 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1818 | |
4ad13663 N |
1819 | bitmap_update_sb(bitmap); |
1820 | ||
69e51b44 N |
1821 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
1822 | err = -EIO; | |
1823 | out: | |
3178b0db | 1824 | return err; |
32a7627c | 1825 | } |
69e51b44 | 1826 | EXPORT_SYMBOL_GPL(bitmap_load); |
32a7627c | 1827 | |
57148964 N |
1828 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap) |
1829 | { | |
1830 | unsigned long chunk_kb; | |
1831 | unsigned long flags; | |
1832 | ||
1833 | if (!bitmap) | |
1834 | return; | |
1835 | ||
1836 | spin_lock_irqsave(&bitmap->lock, flags); | |
1837 | chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; | |
1838 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
1839 | "%lu%s chunk", | |
1840 | bitmap->pages - bitmap->missing_pages, | |
1841 | bitmap->pages, | |
1842 | (bitmap->pages - bitmap->missing_pages) | |
1843 | << (PAGE_SHIFT - 10), | |
1844 | chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, | |
1845 | chunk_kb ? "KB" : "B"); | |
1ec885cd | 1846 | if (bitmap->storage.file) { |
57148964 | 1847 | seq_printf(seq, ", file: "); |
1ec885cd | 1848 | seq_path(seq, &bitmap->storage.file->f_path, " \t\n"); |
57148964 N |
1849 | } |
1850 | ||
1851 | seq_printf(seq, "\n"); | |
1852 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1853 | } | |
1854 | ||
43a70507 | 1855 | static ssize_t |
fd01b88c | 1856 | location_show(struct mddev *mddev, char *page) |
43a70507 N |
1857 | { |
1858 | ssize_t len; | |
ac2f40be | 1859 | if (mddev->bitmap_info.file) |
43a70507 | 1860 | len = sprintf(page, "file"); |
ac2f40be | 1861 | else if (mddev->bitmap_info.offset) |
43a70507 | 1862 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 1863 | else |
43a70507 N |
1864 | len = sprintf(page, "none"); |
1865 | len += sprintf(page+len, "\n"); | |
1866 | return len; | |
1867 | } | |
1868 | ||
1869 | static ssize_t | |
fd01b88c | 1870 | location_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
1871 | { |
1872 | ||
1873 | if (mddev->pers) { | |
1874 | if (!mddev->pers->quiesce) | |
1875 | return -EBUSY; | |
1876 | if (mddev->recovery || mddev->sync_thread) | |
1877 | return -EBUSY; | |
1878 | } | |
1879 | ||
1880 | if (mddev->bitmap || mddev->bitmap_info.file || | |
1881 | mddev->bitmap_info.offset) { | |
1882 | /* bitmap already configured. Only option is to clear it */ | |
1883 | if (strncmp(buf, "none", 4) != 0) | |
1884 | return -EBUSY; | |
1885 | if (mddev->pers) { | |
1886 | mddev->pers->quiesce(mddev, 1); | |
1887 | bitmap_destroy(mddev); | |
1888 | mddev->pers->quiesce(mddev, 0); | |
1889 | } | |
1890 | mddev->bitmap_info.offset = 0; | |
1891 | if (mddev->bitmap_info.file) { | |
1892 | struct file *f = mddev->bitmap_info.file; | |
1893 | mddev->bitmap_info.file = NULL; | |
1894 | restore_bitmap_write_access(f); | |
1895 | fput(f); | |
1896 | } | |
1897 | } else { | |
1898 | /* No bitmap, OK to set a location */ | |
1899 | long long offset; | |
1900 | if (strncmp(buf, "none", 4) == 0) | |
1901 | /* nothing to be done */; | |
1902 | else if (strncmp(buf, "file:", 5) == 0) { | |
1903 | /* Not supported yet */ | |
1904 | return -EINVAL; | |
1905 | } else { | |
1906 | int rv; | |
1907 | if (buf[0] == '+') | |
1908 | rv = strict_strtoll(buf+1, 10, &offset); | |
1909 | else | |
1910 | rv = strict_strtoll(buf, 10, &offset); | |
1911 | if (rv) | |
1912 | return rv; | |
1913 | if (offset == 0) | |
1914 | return -EINVAL; | |
ece5cff0 N |
1915 | if (mddev->bitmap_info.external == 0 && |
1916 | mddev->major_version == 0 && | |
43a70507 N |
1917 | offset != mddev->bitmap_info.default_offset) |
1918 | return -EINVAL; | |
1919 | mddev->bitmap_info.offset = offset; | |
1920 | if (mddev->pers) { | |
1921 | mddev->pers->quiesce(mddev, 1); | |
1922 | rv = bitmap_create(mddev); | |
4474ca42 N |
1923 | if (!rv) |
1924 | rv = bitmap_load(mddev); | |
43a70507 N |
1925 | if (rv) { |
1926 | bitmap_destroy(mddev); | |
1927 | mddev->bitmap_info.offset = 0; | |
1928 | } | |
1929 | mddev->pers->quiesce(mddev, 0); | |
1930 | if (rv) | |
1931 | return rv; | |
1932 | } | |
1933 | } | |
1934 | } | |
1935 | if (!mddev->external) { | |
1936 | /* Ensure new bitmap info is stored in | |
1937 | * metadata promptly. | |
1938 | */ | |
1939 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1940 | md_wakeup_thread(mddev->thread); | |
1941 | } | |
1942 | return len; | |
1943 | } | |
1944 | ||
1945 | static struct md_sysfs_entry bitmap_location = | |
1946 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
1947 | ||
6409bb05 N |
1948 | /* 'bitmap/space' is the space available at 'location' for the |
1949 | * bitmap. This allows the kernel to know when it is safe to | |
1950 | * resize the bitmap to match a resized array. | |
1951 | */ | |
1952 | static ssize_t | |
1953 | space_show(struct mddev *mddev, char *page) | |
1954 | { | |
1955 | return sprintf(page, "%lu\n", mddev->bitmap_info.space); | |
1956 | } | |
1957 | ||
1958 | static ssize_t | |
1959 | space_store(struct mddev *mddev, const char *buf, size_t len) | |
1960 | { | |
1961 | unsigned long sectors; | |
1962 | int rv; | |
1963 | ||
1964 | rv = kstrtoul(buf, 10, §ors); | |
1965 | if (rv) | |
1966 | return rv; | |
1967 | ||
1968 | if (sectors == 0) | |
1969 | return -EINVAL; | |
1970 | ||
1971 | if (mddev->bitmap && | |
9b1215c1 | 1972 | sectors < (mddev->bitmap->storage.bytes + 511) >> 9) |
6409bb05 N |
1973 | return -EFBIG; /* Bitmap is too big for this small space */ |
1974 | ||
1975 | /* could make sure it isn't too big, but that isn't really | |
1976 | * needed - user-space should be careful. | |
1977 | */ | |
1978 | mddev->bitmap_info.space = sectors; | |
1979 | return len; | |
1980 | } | |
1981 | ||
1982 | static struct md_sysfs_entry bitmap_space = | |
1983 | __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); | |
1984 | ||
43a70507 | 1985 | static ssize_t |
fd01b88c | 1986 | timeout_show(struct mddev *mddev, char *page) |
43a70507 N |
1987 | { |
1988 | ssize_t len; | |
1989 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
1990 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 1991 | |
43a70507 N |
1992 | len = sprintf(page, "%lu", secs); |
1993 | if (jifs) | |
1994 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
1995 | len += sprintf(page+len, "\n"); | |
1996 | return len; | |
1997 | } | |
1998 | ||
1999 | static ssize_t | |
fd01b88c | 2000 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2001 | { |
2002 | /* timeout can be set at any time */ | |
2003 | unsigned long timeout; | |
2004 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
2005 | if (rv) | |
2006 | return rv; | |
2007 | ||
2008 | /* just to make sure we don't overflow... */ | |
2009 | if (timeout >= LONG_MAX / HZ) | |
2010 | return -EINVAL; | |
2011 | ||
2012 | timeout = timeout * HZ / 10000; | |
2013 | ||
2014 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
2015 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
2016 | if (timeout < 1) | |
2017 | timeout = 1; | |
2018 | mddev->bitmap_info.daemon_sleep = timeout; | |
2019 | if (mddev->thread) { | |
2020 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
2021 | * the bitmap is all clean and we don't need to | |
2022 | * adjust the timeout right now | |
2023 | */ | |
2024 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2025 | mddev->thread->timeout = timeout; | |
2026 | md_wakeup_thread(mddev->thread); | |
2027 | } | |
2028 | } | |
2029 | return len; | |
2030 | } | |
2031 | ||
2032 | static struct md_sysfs_entry bitmap_timeout = | |
2033 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2034 | ||
2035 | static ssize_t | |
fd01b88c | 2036 | backlog_show(struct mddev *mddev, char *page) |
43a70507 N |
2037 | { |
2038 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2039 | } | |
2040 | ||
2041 | static ssize_t | |
fd01b88c | 2042 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2043 | { |
2044 | unsigned long backlog; | |
2045 | int rv = strict_strtoul(buf, 10, &backlog); | |
2046 | if (rv) | |
2047 | return rv; | |
2048 | if (backlog > COUNTER_MAX) | |
2049 | return -EINVAL; | |
2050 | mddev->bitmap_info.max_write_behind = backlog; | |
2051 | return len; | |
2052 | } | |
2053 | ||
2054 | static struct md_sysfs_entry bitmap_backlog = | |
2055 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2056 | ||
2057 | static ssize_t | |
fd01b88c | 2058 | chunksize_show(struct mddev *mddev, char *page) |
43a70507 N |
2059 | { |
2060 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2061 | } | |
2062 | ||
2063 | static ssize_t | |
fd01b88c | 2064 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2065 | { |
2066 | /* Can only be changed when no bitmap is active */ | |
2067 | int rv; | |
2068 | unsigned long csize; | |
2069 | if (mddev->bitmap) | |
2070 | return -EBUSY; | |
2071 | rv = strict_strtoul(buf, 10, &csize); | |
2072 | if (rv) | |
2073 | return rv; | |
2074 | if (csize < 512 || | |
2075 | !is_power_of_2(csize)) | |
2076 | return -EINVAL; | |
2077 | mddev->bitmap_info.chunksize = csize; | |
2078 | return len; | |
2079 | } | |
2080 | ||
2081 | static struct md_sysfs_entry bitmap_chunksize = | |
2082 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2083 | ||
fd01b88c | 2084 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2085 | { |
2086 | return sprintf(page, "%s\n", (mddev->bitmap_info.external | |
2087 | ? "external" : "internal")); | |
2088 | } | |
2089 | ||
fd01b88c | 2090 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2091 | { |
2092 | if (mddev->bitmap || | |
2093 | mddev->bitmap_info.file || | |
2094 | mddev->bitmap_info.offset) | |
2095 | return -EBUSY; | |
2096 | if (strncmp(buf, "external", 8) == 0) | |
2097 | mddev->bitmap_info.external = 1; | |
2098 | else if (strncmp(buf, "internal", 8) == 0) | |
2099 | mddev->bitmap_info.external = 0; | |
2100 | else | |
2101 | return -EINVAL; | |
2102 | return len; | |
2103 | } | |
2104 | ||
2105 | static struct md_sysfs_entry bitmap_metadata = | |
2106 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2107 | ||
fd01b88c | 2108 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2109 | { |
2110 | int len; | |
2111 | if (mddev->bitmap) | |
2112 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2113 | "false" : "true")); | |
2114 | else | |
2115 | len = sprintf(page, "\n"); | |
2116 | return len; | |
2117 | } | |
2118 | ||
fd01b88c | 2119 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2120 | { |
2121 | if (mddev->bitmap == NULL) | |
2122 | return -ENOENT; | |
2123 | if (strncmp(buf, "false", 5) == 0) | |
2124 | mddev->bitmap->need_sync = 1; | |
2125 | else if (strncmp(buf, "true", 4) == 0) { | |
2126 | if (mddev->degraded) | |
2127 | return -EBUSY; | |
2128 | mddev->bitmap->need_sync = 0; | |
2129 | } else | |
2130 | return -EINVAL; | |
2131 | return len; | |
2132 | } | |
2133 | ||
2134 | static struct md_sysfs_entry bitmap_can_clear = | |
2135 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2136 | ||
696fcd53 | 2137 | static ssize_t |
fd01b88c | 2138 | behind_writes_used_show(struct mddev *mddev, char *page) |
696fcd53 PC |
2139 | { |
2140 | if (mddev->bitmap == NULL) | |
2141 | return sprintf(page, "0\n"); | |
2142 | return sprintf(page, "%lu\n", | |
2143 | mddev->bitmap->behind_writes_used); | |
2144 | } | |
2145 | ||
2146 | static ssize_t | |
fd01b88c | 2147 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
696fcd53 PC |
2148 | { |
2149 | if (mddev->bitmap) | |
2150 | mddev->bitmap->behind_writes_used = 0; | |
2151 | return len; | |
2152 | } | |
2153 | ||
2154 | static struct md_sysfs_entry max_backlog_used = | |
2155 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2156 | behind_writes_used_show, behind_writes_used_reset); | |
2157 | ||
43a70507 N |
2158 | static struct attribute *md_bitmap_attrs[] = { |
2159 | &bitmap_location.attr, | |
6409bb05 | 2160 | &bitmap_space.attr, |
43a70507 N |
2161 | &bitmap_timeout.attr, |
2162 | &bitmap_backlog.attr, | |
2163 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2164 | &bitmap_metadata.attr, |
2165 | &bitmap_can_clear.attr, | |
696fcd53 | 2166 | &max_backlog_used.attr, |
43a70507 N |
2167 | NULL |
2168 | }; | |
2169 | struct attribute_group md_bitmap_group = { | |
2170 | .name = "bitmap", | |
2171 | .attrs = md_bitmap_attrs, | |
2172 | }; | |
2173 |