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 | * just a placeholder - calls kmalloc for bitmap pages | |
40 | */ | |
41 | static unsigned char *bitmap_alloc_page(struct bitmap *bitmap) | |
42 | { | |
43 | unsigned char *page; | |
44 | ||
ac2f40be | 45 | page = kzalloc(PAGE_SIZE, GFP_NOIO); |
32a7627c N |
46 | if (!page) |
47 | printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap)); | |
48 | else | |
36a4e1fe N |
49 | pr_debug("%s: bitmap_alloc_page: allocated page at %p\n", |
50 | bmname(bitmap), page); | |
32a7627c N |
51 | return page; |
52 | } | |
53 | ||
54 | /* | |
55 | * for now just a placeholder -- just calls kfree for bitmap pages | |
56 | */ | |
57 | static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page) | |
58 | { | |
36a4e1fe | 59 | pr_debug("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page); |
32a7627c N |
60 | kfree(page); |
61 | } | |
62 | ||
63 | /* | |
64 | * check a page and, if necessary, allocate it (or hijack it if the alloc fails) | |
65 | * | |
66 | * 1) check to see if this page is allocated, if it's not then try to alloc | |
67 | * 2) if the alloc fails, set the page's hijacked flag so we'll use the | |
68 | * page pointer directly as a counter | |
69 | * | |
70 | * if we find our page, we increment the page's refcount so that it stays | |
71 | * allocated while we're using it | |
72 | */ | |
ac2f40be N |
73 | static int bitmap_checkpage(struct bitmap *bitmap, |
74 | unsigned long page, int create) | |
ee305ace N |
75 | __releases(bitmap->lock) |
76 | __acquires(bitmap->lock) | |
32a7627c N |
77 | { |
78 | unsigned char *mappage; | |
79 | ||
80 | if (page >= bitmap->pages) { | |
1187cf0a N |
81 | /* This can happen if bitmap_start_sync goes beyond |
82 | * End-of-device while looking for a whole page. | |
83 | * It is harmless. | |
84 | */ | |
32a7627c N |
85 | return -EINVAL; |
86 | } | |
87 | ||
32a7627c N |
88 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ |
89 | return 0; | |
90 | ||
91 | if (bitmap->bp[page].map) /* page is already allocated, just return */ | |
92 | return 0; | |
93 | ||
94 | if (!create) | |
95 | return -ENOENT; | |
96 | ||
32a7627c N |
97 | /* this page has not been allocated yet */ |
98 | ||
ac2f40be N |
99 | spin_unlock_irq(&bitmap->lock); |
100 | mappage = bitmap_alloc_page(bitmap); | |
101 | spin_lock_irq(&bitmap->lock); | |
102 | ||
103 | if (mappage == NULL) { | |
36a4e1fe N |
104 | pr_debug("%s: bitmap map page allocation failed, hijacking\n", |
105 | bmname(bitmap)); | |
32a7627c N |
106 | /* failed - set the hijacked flag so that we can use the |
107 | * pointer as a counter */ | |
32a7627c N |
108 | if (!bitmap->bp[page].map) |
109 | bitmap->bp[page].hijacked = 1; | |
ac2f40be N |
110 | } else if (bitmap->bp[page].map || |
111 | bitmap->bp[page].hijacked) { | |
32a7627c N |
112 | /* somebody beat us to getting the page */ |
113 | bitmap_free_page(bitmap, mappage); | |
114 | return 0; | |
ac2f40be | 115 | } else { |
32a7627c | 116 | |
ac2f40be | 117 | /* no page was in place and we have one, so install it */ |
32a7627c | 118 | |
ac2f40be N |
119 | bitmap->bp[page].map = mappage; |
120 | bitmap->missing_pages--; | |
121 | } | |
32a7627c N |
122 | return 0; |
123 | } | |
124 | ||
32a7627c N |
125 | /* if page is completely empty, put it back on the free list, or dealloc it */ |
126 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
127 | /* Note: lock should be held when calling this */ | |
858119e1 | 128 | static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page) |
32a7627c N |
129 | { |
130 | char *ptr; | |
131 | ||
132 | if (bitmap->bp[page].count) /* page is still busy */ | |
133 | return; | |
134 | ||
135 | /* page is no longer in use, it can be released */ | |
136 | ||
137 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
138 | bitmap->bp[page].hijacked = 0; | |
139 | bitmap->bp[page].map = NULL; | |
ac2f40be N |
140 | } else { |
141 | /* normal case, free the page */ | |
142 | ptr = bitmap->bp[page].map; | |
143 | bitmap->bp[page].map = NULL; | |
144 | bitmap->missing_pages++; | |
145 | bitmap_free_page(bitmap, ptr); | |
32a7627c | 146 | } |
32a7627c N |
147 | } |
148 | ||
32a7627c N |
149 | /* |
150 | * bitmap file handling - read and write the bitmap file and its superblock | |
151 | */ | |
152 | ||
32a7627c N |
153 | /* |
154 | * basic page I/O operations | |
155 | */ | |
156 | ||
a654b9d8 | 157 | /* IO operations when bitmap is stored near all superblocks */ |
fd01b88c | 158 | static struct page *read_sb_page(struct mddev *mddev, loff_t offset, |
a2ed9615 N |
159 | struct page *page, |
160 | unsigned long index, int size) | |
a654b9d8 N |
161 | { |
162 | /* choose a good rdev and read the page from there */ | |
163 | ||
3cb03002 | 164 | struct md_rdev *rdev; |
a654b9d8 | 165 | sector_t target; |
ac2f40be | 166 | int did_alloc = 0; |
a654b9d8 | 167 | |
ac2f40be | 168 | if (!page) { |
a2ed9615 | 169 | page = alloc_page(GFP_KERNEL); |
ac2f40be N |
170 | if (!page) |
171 | return ERR_PTR(-ENOMEM); | |
172 | did_alloc = 1; | |
173 | } | |
a654b9d8 | 174 | |
dafb20fa | 175 | rdev_for_each(rdev, mddev) { |
b2d444d7 N |
176 | if (! test_bit(In_sync, &rdev->flags) |
177 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
178 | continue; |
179 | ||
ccebd4c4 | 180 | target = offset + index * (PAGE_SIZE/512); |
a654b9d8 | 181 | |
2b193363 | 182 | if (sync_page_io(rdev, target, |
e1defc4f | 183 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
ccebd4c4 | 184 | page, READ, true)) { |
ab904d63 | 185 | page->index = index; |
ce25c31b N |
186 | attach_page_buffers(page, NULL); /* so that free_buffer will |
187 | * quietly no-op */ | |
ab904d63 N |
188 | return page; |
189 | } | |
190 | } | |
ac2f40be N |
191 | if (did_alloc) |
192 | put_page(page); | |
ab904d63 | 193 | return ERR_PTR(-EIO); |
a654b9d8 | 194 | |
a654b9d8 N |
195 | } |
196 | ||
fd01b88c | 197 | static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) |
b2d2c4ce N |
198 | { |
199 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
200 | * linked list, and raising the refcount of devices we return to ensure | |
201 | * they don't disappear while in use. | |
202 | * As devices are only added or removed when raid_disk is < 0 and | |
203 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
204 | * still be in the same position on the list when we re-enter | |
205 | * list_for_each_continue_rcu. | |
206 | */ | |
207 | struct list_head *pos; | |
208 | rcu_read_lock(); | |
209 | if (rdev == NULL) | |
210 | /* start at the beginning */ | |
211 | pos = &mddev->disks; | |
212 | else { | |
213 | /* release the previous rdev and start from there. */ | |
214 | rdev_dec_pending(rdev, mddev); | |
215 | pos = &rdev->same_set; | |
216 | } | |
217 | list_for_each_continue_rcu(pos, &mddev->disks) { | |
3cb03002 | 218 | rdev = list_entry(pos, struct md_rdev, same_set); |
b2d2c4ce | 219 | if (rdev->raid_disk >= 0 && |
b2d2c4ce N |
220 | !test_bit(Faulty, &rdev->flags)) { |
221 | /* this is a usable devices */ | |
222 | atomic_inc(&rdev->nr_pending); | |
223 | rcu_read_unlock(); | |
224 | return rdev; | |
225 | } | |
226 | } | |
227 | rcu_read_unlock(); | |
228 | return NULL; | |
229 | } | |
230 | ||
ab6085c7 | 231 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 232 | { |
3cb03002 | 233 | struct md_rdev *rdev = NULL; |
a6ff7e08 | 234 | struct block_device *bdev; |
fd01b88c | 235 | struct mddev *mddev = bitmap->mddev; |
a654b9d8 | 236 | |
b2d2c4ce | 237 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ac2f40be N |
238 | int size = PAGE_SIZE; |
239 | loff_t offset = mddev->bitmap_info.offset; | |
a6ff7e08 JB |
240 | |
241 | bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; | |
242 | ||
ac2f40be N |
243 | if (page->index == bitmap->file_pages-1) |
244 | size = roundup(bitmap->last_page_size, | |
a6ff7e08 | 245 | bdev_logical_block_size(bdev)); |
ac2f40be N |
246 | /* Just make sure we aren't corrupting data or |
247 | * metadata | |
248 | */ | |
249 | if (mddev->external) { | |
250 | /* Bitmap could be anywhere. */ | |
251 | if (rdev->sb_start + offset + (page->index | |
252 | * (PAGE_SIZE/512)) | |
253 | > rdev->data_offset | |
254 | && | |
255 | rdev->sb_start + offset | |
256 | < (rdev->data_offset + mddev->dev_sectors | |
257 | + (PAGE_SIZE/512))) | |
258 | goto bad_alignment; | |
259 | } else if (offset < 0) { | |
260 | /* DATA BITMAP METADATA */ | |
261 | if (offset | |
262 | + (long)(page->index * (PAGE_SIZE/512)) | |
263 | + size/512 > 0) | |
264 | /* bitmap runs in to metadata */ | |
265 | goto bad_alignment; | |
266 | if (rdev->data_offset + mddev->dev_sectors | |
267 | > rdev->sb_start + offset) | |
268 | /* data runs in to bitmap */ | |
269 | goto bad_alignment; | |
270 | } else if (rdev->sb_start < rdev->data_offset) { | |
271 | /* METADATA BITMAP DATA */ | |
272 | if (rdev->sb_start | |
273 | + offset | |
274 | + page->index*(PAGE_SIZE/512) + size/512 | |
275 | > rdev->data_offset) | |
276 | /* bitmap runs in to data */ | |
277 | goto bad_alignment; | |
278 | } else { | |
279 | /* DATA METADATA BITMAP - no problems */ | |
280 | } | |
281 | md_super_write(mddev, rdev, | |
282 | rdev->sb_start + offset | |
283 | + page->index * (PAGE_SIZE/512), | |
284 | size, | |
285 | page); | |
b2d2c4ce | 286 | } |
a654b9d8 N |
287 | |
288 | if (wait) | |
a9701a30 | 289 | md_super_wait(mddev); |
a654b9d8 | 290 | return 0; |
4b80991c N |
291 | |
292 | bad_alignment: | |
4b80991c | 293 | return -EINVAL; |
a654b9d8 N |
294 | } |
295 | ||
4ad13663 | 296 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 297 | /* |
a654b9d8 | 298 | * write out a page to a file |
32a7627c | 299 | */ |
4ad13663 | 300 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 301 | { |
d785a06a | 302 | struct buffer_head *bh; |
32a7627c | 303 | |
f0d76d70 N |
304 | if (bitmap->file == NULL) { |
305 | switch (write_sb_page(bitmap, page, wait)) { | |
306 | case -EINVAL: | |
307 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
f0d76d70 | 308 | } |
4ad13663 | 309 | } else { |
a654b9d8 | 310 | |
4ad13663 | 311 | bh = page_buffers(page); |
c708443c | 312 | |
4ad13663 N |
313 | while (bh && bh->b_blocknr) { |
314 | atomic_inc(&bitmap->pending_writes); | |
315 | set_buffer_locked(bh); | |
316 | set_buffer_mapped(bh); | |
721a9602 | 317 | submit_bh(WRITE | REQ_SYNC, bh); |
4ad13663 N |
318 | bh = bh->b_this_page; |
319 | } | |
d785a06a | 320 | |
ac2f40be | 321 | if (wait) |
4ad13663 N |
322 | wait_event(bitmap->write_wait, |
323 | atomic_read(&bitmap->pending_writes)==0); | |
8a5e9cf1 | 324 | } |
4ad13663 N |
325 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
326 | bitmap_file_kick(bitmap); | |
d785a06a N |
327 | } |
328 | ||
329 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
330 | { | |
331 | struct bitmap *bitmap = bh->b_private; | |
332 | unsigned long flags; | |
32a7627c | 333 | |
d785a06a N |
334 | if (!uptodate) { |
335 | spin_lock_irqsave(&bitmap->lock, flags); | |
336 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
337 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
32a7627c | 338 | } |
d785a06a N |
339 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
340 | wake_up(&bitmap->write_wait); | |
341 | } | |
32a7627c | 342 | |
d785a06a N |
343 | /* copied from buffer.c */ |
344 | static void | |
345 | __clear_page_buffers(struct page *page) | |
346 | { | |
347 | ClearPagePrivate(page); | |
348 | set_page_private(page, 0); | |
349 | page_cache_release(page); | |
350 | } | |
351 | static void free_buffers(struct page *page) | |
352 | { | |
353 | struct buffer_head *bh = page_buffers(page); | |
77ad4bc7 | 354 | |
d785a06a N |
355 | while (bh) { |
356 | struct buffer_head *next = bh->b_this_page; | |
357 | free_buffer_head(bh); | |
358 | bh = next; | |
77ad4bc7 | 359 | } |
d785a06a N |
360 | __clear_page_buffers(page); |
361 | put_page(page); | |
32a7627c N |
362 | } |
363 | ||
d785a06a N |
364 | /* read a page from a file. |
365 | * We both read the page, and attach buffers to the page to record the | |
366 | * address of each block (using bmap). These addresses will be used | |
367 | * to write the block later, completely bypassing the filesystem. | |
368 | * This usage is similar to how swap files are handled, and allows us | |
369 | * to write to a file with no concerns of memory allocation failing. | |
370 | */ | |
32a7627c | 371 | static struct page *read_page(struct file *file, unsigned long index, |
d785a06a N |
372 | struct bitmap *bitmap, |
373 | unsigned long count) | |
32a7627c | 374 | { |
32a7627c | 375 | struct page *page = NULL; |
c649bb9c | 376 | struct inode *inode = file->f_path.dentry->d_inode; |
d785a06a N |
377 | struct buffer_head *bh; |
378 | sector_t block; | |
32a7627c | 379 | |
36a4e1fe N |
380 | pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
381 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 382 | |
d785a06a N |
383 | page = alloc_page(GFP_KERNEL); |
384 | if (!page) | |
385 | page = ERR_PTR(-ENOMEM); | |
32a7627c N |
386 | if (IS_ERR(page)) |
387 | goto out; | |
d785a06a | 388 | |
d785a06a N |
389 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
390 | if (!bh) { | |
2d1f3b5d | 391 | put_page(page); |
d785a06a | 392 | page = ERR_PTR(-ENOMEM); |
32a7627c N |
393 | goto out; |
394 | } | |
d785a06a N |
395 | attach_page_buffers(page, bh); |
396 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
397 | while (bh) { | |
398 | if (count == 0) | |
399 | bh->b_blocknr = 0; | |
400 | else { | |
401 | bh->b_blocknr = bmap(inode, block); | |
402 | if (bh->b_blocknr == 0) { | |
403 | /* Cannot use this file! */ | |
404 | free_buffers(page); | |
405 | page = ERR_PTR(-EINVAL); | |
406 | goto out; | |
407 | } | |
408 | bh->b_bdev = inode->i_sb->s_bdev; | |
409 | if (count < (1<<inode->i_blkbits)) | |
410 | count = 0; | |
411 | else | |
412 | count -= (1<<inode->i_blkbits); | |
413 | ||
414 | bh->b_end_io = end_bitmap_write; | |
415 | bh->b_private = bitmap; | |
ce25c31b N |
416 | atomic_inc(&bitmap->pending_writes); |
417 | set_buffer_locked(bh); | |
418 | set_buffer_mapped(bh); | |
419 | submit_bh(READ, bh); | |
d785a06a N |
420 | } |
421 | block++; | |
422 | bh = bh->b_this_page; | |
423 | } | |
d785a06a | 424 | page->index = index; |
ce25c31b N |
425 | |
426 | wait_event(bitmap->write_wait, | |
427 | atomic_read(&bitmap->pending_writes)==0); | |
428 | if (bitmap->flags & BITMAP_WRITE_ERROR) { | |
429 | free_buffers(page); | |
430 | page = ERR_PTR(-EIO); | |
431 | } | |
32a7627c N |
432 | out: |
433 | if (IS_ERR(page)) | |
ac2f40be | 434 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %ld\n", |
2d1f3b5d N |
435 | (int)PAGE_SIZE, |
436 | (unsigned long long)index << PAGE_SHIFT, | |
32a7627c N |
437 | PTR_ERR(page)); |
438 | return page; | |
439 | } | |
440 | ||
441 | /* | |
442 | * bitmap file superblock operations | |
443 | */ | |
444 | ||
445 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 446 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
447 | { |
448 | bitmap_super_t *sb; | |
449 | unsigned long flags; | |
450 | ||
451 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 452 | return; |
ece5cff0 N |
453 | if (bitmap->mddev->bitmap_info.external) |
454 | return; | |
32a7627c N |
455 | spin_lock_irqsave(&bitmap->lock, flags); |
456 | if (!bitmap->sb_page) { /* no superblock */ | |
457 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 458 | return; |
32a7627c | 459 | } |
32a7627c | 460 | spin_unlock_irqrestore(&bitmap->lock, flags); |
7b92813c | 461 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 462 | sb->events = cpu_to_le64(bitmap->mddev->events); |
8258c532 | 463 | if (bitmap->mddev->events < bitmap->events_cleared) |
a0da84f3 NB |
464 | /* rocking back to read-only */ |
465 | bitmap->events_cleared = bitmap->mddev->events; | |
8258c532 N |
466 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); |
467 | sb->state = cpu_to_le32(bitmap->flags); | |
43a70507 N |
468 | /* Just in case these have been changed via sysfs: */ |
469 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
470 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
ea03aff9 | 471 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 472 | write_page(bitmap, bitmap->sb_page, 1); |
32a7627c N |
473 | } |
474 | ||
475 | /* print out the bitmap file superblock */ | |
476 | void bitmap_print_sb(struct bitmap *bitmap) | |
477 | { | |
478 | bitmap_super_t *sb; | |
479 | ||
480 | if (!bitmap || !bitmap->sb_page) | |
481 | return; | |
7b92813c | 482 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 483 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
484 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
485 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
486 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
487 | *(__u32 *)(sb->uuid+0), |
488 | *(__u32 *)(sb->uuid+4), | |
489 | *(__u32 *)(sb->uuid+8), | |
490 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 491 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 492 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 493 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 494 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
495 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
496 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
497 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
498 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
499 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 500 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
ea03aff9 | 501 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
502 | } |
503 | ||
9c81075f JB |
504 | /* |
505 | * bitmap_new_disk_sb | |
506 | * @bitmap | |
507 | * | |
508 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb | |
509 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. | |
510 | * This function verifies 'bitmap_info' and populates the on-disk bitmap | |
511 | * structure, which is to be written to disk. | |
512 | * | |
513 | * Returns: 0 on success, -Exxx on error | |
514 | */ | |
515 | static int bitmap_new_disk_sb(struct bitmap *bitmap) | |
516 | { | |
517 | bitmap_super_t *sb; | |
518 | unsigned long chunksize, daemon_sleep, write_behind; | |
519 | int err = -EINVAL; | |
520 | ||
521 | bitmap->sb_page = alloc_page(GFP_KERNEL); | |
522 | if (IS_ERR(bitmap->sb_page)) { | |
523 | err = PTR_ERR(bitmap->sb_page); | |
524 | bitmap->sb_page = NULL; | |
525 | return err; | |
526 | } | |
527 | bitmap->sb_page->index = 0; | |
528 | ||
529 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); | |
530 | ||
531 | sb->magic = cpu_to_le32(BITMAP_MAGIC); | |
532 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); | |
533 | ||
534 | chunksize = bitmap->mddev->bitmap_info.chunksize; | |
535 | BUG_ON(!chunksize); | |
536 | if (!is_power_of_2(chunksize)) { | |
537 | kunmap_atomic(sb, KM_USER0); | |
538 | printk(KERN_ERR "bitmap chunksize not a power of 2\n"); | |
539 | return -EINVAL; | |
540 | } | |
541 | sb->chunksize = cpu_to_le32(chunksize); | |
542 | ||
543 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; | |
544 | if (!daemon_sleep || | |
545 | (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { | |
546 | printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n"); | |
547 | daemon_sleep = 5 * HZ; | |
548 | } | |
549 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); | |
550 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
551 | ||
552 | /* | |
553 | * FIXME: write_behind for RAID1. If not specified, what | |
554 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. | |
555 | */ | |
556 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; | |
557 | if (write_behind > COUNTER_MAX) | |
558 | write_behind = COUNTER_MAX / 2; | |
559 | sb->write_behind = cpu_to_le32(write_behind); | |
560 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
561 | ||
562 | /* keep the array size field of the bitmap superblock up to date */ | |
563 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
564 | ||
565 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); | |
566 | ||
567 | bitmap->flags |= BITMAP_STALE; | |
568 | sb->state |= cpu_to_le32(BITMAP_STALE); | |
569 | bitmap->events_cleared = bitmap->mddev->events; | |
570 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
571 | ||
572 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
573 | sb->version = cpu_to_le32(BITMAP_MAJOR_HOSTENDIAN); | |
574 | ||
575 | kunmap_atomic(sb, KM_USER0); | |
576 | ||
577 | return 0; | |
578 | } | |
579 | ||
32a7627c N |
580 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
581 | static int bitmap_read_sb(struct bitmap *bitmap) | |
582 | { | |
583 | char *reason = NULL; | |
584 | bitmap_super_t *sb; | |
4b6d287f | 585 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c N |
586 | unsigned long long events; |
587 | int err = -EINVAL; | |
588 | ||
589 | /* page 0 is the superblock, read it... */ | |
f49d5e62 N |
590 | if (bitmap->file) { |
591 | loff_t isize = i_size_read(bitmap->file->f_mapping->host); | |
592 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; | |
593 | ||
594 | bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes); | |
595 | } else { | |
42a04b50 N |
596 | bitmap->sb_page = read_sb_page(bitmap->mddev, |
597 | bitmap->mddev->bitmap_info.offset, | |
a2ed9615 N |
598 | NULL, |
599 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 600 | } |
32a7627c N |
601 | if (IS_ERR(bitmap->sb_page)) { |
602 | err = PTR_ERR(bitmap->sb_page); | |
603 | bitmap->sb_page = NULL; | |
604 | return err; | |
605 | } | |
606 | ||
7b92813c | 607 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 608 | |
32a7627c | 609 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 610 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 611 | write_behind = le32_to_cpu(sb->write_behind); |
32a7627c N |
612 | |
613 | /* verify that the bitmap-specific fields are valid */ | |
614 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
615 | reason = "bad magic"; | |
bd926c63 N |
616 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
617 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 618 | reason = "unrecognized superblock version"; |
1187cf0a | 619 | else if (chunksize < 512) |
7dd5d34c | 620 | reason = "bitmap chunksize too small"; |
d744540c | 621 | else if (!is_power_of_2(chunksize)) |
32a7627c | 622 | reason = "bitmap chunksize not a power of 2"; |
1b04be96 | 623 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 624 | reason = "daemon sleep period out of range"; |
4b6d287f N |
625 | else if (write_behind > COUNTER_MAX) |
626 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
627 | if (reason) { |
628 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
629 | bmname(bitmap), reason); | |
630 | goto out; | |
631 | } | |
632 | ||
633 | /* keep the array size field of the bitmap superblock up to date */ | |
634 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
635 | ||
636 | if (!bitmap->mddev->persistent) | |
637 | goto success; | |
638 | ||
639 | /* | |
640 | * if we have a persistent array superblock, compare the | |
641 | * bitmap's UUID and event counter to the mddev's | |
642 | */ | |
643 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
644 | printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n", | |
645 | bmname(bitmap)); | |
646 | goto out; | |
647 | } | |
648 | events = le64_to_cpu(sb->events); | |
649 | if (events < bitmap->mddev->events) { | |
650 | printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) " | |
651 | "-- forcing full recovery\n", bmname(bitmap), events, | |
652 | (unsigned long long) bitmap->mddev->events); | |
4f2e639a | 653 | sb->state |= cpu_to_le32(BITMAP_STALE); |
32a7627c N |
654 | } |
655 | success: | |
656 | /* assign fields using values from superblock */ | |
42a04b50 N |
657 | bitmap->mddev->bitmap_info.chunksize = chunksize; |
658 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
42a04b50 | 659 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; |
4f2e639a | 660 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 N |
661 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
662 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
32a7627c | 663 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
8258c532 | 664 | if (bitmap->flags & BITMAP_STALE) |
6a07997f | 665 | bitmap->events_cleared = bitmap->mddev->events; |
32a7627c N |
666 | err = 0; |
667 | out: | |
ea03aff9 | 668 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
669 | if (err) |
670 | bitmap_print_sb(bitmap); | |
671 | return err; | |
672 | } | |
673 | ||
674 | enum bitmap_mask_op { | |
675 | MASK_SET, | |
676 | MASK_UNSET | |
677 | }; | |
678 | ||
4ad13663 N |
679 | /* record the state of the bitmap in the superblock. Return the old value */ |
680 | static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits, | |
681 | enum bitmap_mask_op op) | |
32a7627c N |
682 | { |
683 | bitmap_super_t *sb; | |
684 | unsigned long flags; | |
4ad13663 | 685 | int old; |
32a7627c N |
686 | |
687 | spin_lock_irqsave(&bitmap->lock, flags); | |
7e317655 | 688 | if (!bitmap->sb_page) { /* can't set the state */ |
32a7627c | 689 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 690 | return 0; |
32a7627c | 691 | } |
32a7627c | 692 | spin_unlock_irqrestore(&bitmap->lock, flags); |
7b92813c | 693 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
4ad13663 | 694 | old = le32_to_cpu(sb->state) & bits; |
32a7627c | 695 | switch (op) { |
ac2f40be N |
696 | case MASK_SET: |
697 | sb->state |= cpu_to_le32(bits); | |
8258c532 | 698 | bitmap->flags |= bits; |
ac2f40be N |
699 | break; |
700 | case MASK_UNSET: | |
701 | sb->state &= cpu_to_le32(~bits); | |
8258c532 | 702 | bitmap->flags &= ~bits; |
ac2f40be N |
703 | break; |
704 | default: | |
705 | BUG(); | |
32a7627c | 706 | } |
ea03aff9 | 707 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 708 | return old; |
32a7627c N |
709 | } |
710 | ||
711 | /* | |
712 | * general bitmap file operations | |
713 | */ | |
714 | ||
ece5cff0 N |
715 | /* |
716 | * on-disk bitmap: | |
717 | * | |
718 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
719 | * file a page at a time. There's a superblock at the start of the file. | |
720 | */ | |
32a7627c | 721 | /* calculate the index of the page that contains this bit */ |
ece5cff0 | 722 | static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk) |
32a7627c | 723 | { |
ece5cff0 N |
724 | if (!bitmap->mddev->bitmap_info.external) |
725 | chunk += sizeof(bitmap_super_t) << 3; | |
726 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
727 | } |
728 | ||
729 | /* calculate the (bit) offset of this bit within a page */ | |
ece5cff0 | 730 | static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk) |
32a7627c | 731 | { |
ece5cff0 N |
732 | if (!bitmap->mddev->bitmap_info.external) |
733 | chunk += sizeof(bitmap_super_t) << 3; | |
734 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
735 | } |
736 | ||
737 | /* | |
738 | * return a pointer to the page in the filemap that contains the given bit | |
739 | * | |
740 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
741 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
742 | * 0 or page 1 | |
743 | */ | |
744 | static inline struct page *filemap_get_page(struct bitmap *bitmap, | |
3520fa4d | 745 | unsigned long chunk) |
32a7627c | 746 | { |
ac2f40be N |
747 | if (file_page_index(bitmap, chunk) >= bitmap->file_pages) |
748 | return NULL; | |
ece5cff0 N |
749 | return bitmap->filemap[file_page_index(bitmap, chunk) |
750 | - file_page_index(bitmap, 0)]; | |
32a7627c N |
751 | } |
752 | ||
32a7627c N |
753 | static void bitmap_file_unmap(struct bitmap *bitmap) |
754 | { | |
755 | struct page **map, *sb_page; | |
756 | unsigned long *attr; | |
757 | int pages; | |
758 | unsigned long flags; | |
759 | ||
760 | spin_lock_irqsave(&bitmap->lock, flags); | |
761 | map = bitmap->filemap; | |
762 | bitmap->filemap = NULL; | |
763 | attr = bitmap->filemap_attr; | |
764 | bitmap->filemap_attr = NULL; | |
765 | pages = bitmap->file_pages; | |
766 | bitmap->file_pages = 0; | |
767 | sb_page = bitmap->sb_page; | |
768 | bitmap->sb_page = NULL; | |
769 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
770 | ||
771 | while (pages--) | |
ece5cff0 | 772 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 773 | free_buffers(map[pages]); |
32a7627c N |
774 | kfree(map); |
775 | kfree(attr); | |
776 | ||
d785a06a N |
777 | if (sb_page) |
778 | free_buffers(sb_page); | |
32a7627c N |
779 | } |
780 | ||
781 | static void bitmap_file_put(struct bitmap *bitmap) | |
782 | { | |
783 | struct file *file; | |
32a7627c N |
784 | unsigned long flags; |
785 | ||
786 | spin_lock_irqsave(&bitmap->lock, flags); | |
787 | file = bitmap->file; | |
788 | bitmap->file = NULL; | |
789 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
790 | ||
d785a06a N |
791 | if (file) |
792 | wait_event(bitmap->write_wait, | |
793 | atomic_read(&bitmap->pending_writes)==0); | |
32a7627c N |
794 | bitmap_file_unmap(bitmap); |
795 | ||
d785a06a | 796 | if (file) { |
c649bb9c | 797 | struct inode *inode = file->f_path.dentry->d_inode; |
fc0ecff6 | 798 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 799 | fput(file); |
d785a06a | 800 | } |
32a7627c N |
801 | } |
802 | ||
32a7627c N |
803 | /* |
804 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
805 | * then it is no longer reliable, so we stop using it and we mark the file | |
806 | * as failed in the superblock | |
807 | */ | |
808 | static void bitmap_file_kick(struct bitmap *bitmap) | |
809 | { | |
810 | char *path, *ptr = NULL; | |
811 | ||
4ad13663 N |
812 | if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) { |
813 | bitmap_update_sb(bitmap); | |
32a7627c | 814 | |
4ad13663 N |
815 | if (bitmap->file) { |
816 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
817 | if (path) | |
6bcfd601 CH |
818 | ptr = d_path(&bitmap->file->f_path, path, |
819 | PAGE_SIZE); | |
820 | ||
4ad13663 N |
821 | printk(KERN_ALERT |
822 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 823 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 824 | |
4ad13663 N |
825 | kfree(path); |
826 | } else | |
827 | printk(KERN_ALERT | |
828 | "%s: disabling internal bitmap due to errors\n", | |
829 | bmname(bitmap)); | |
a654b9d8 | 830 | } |
32a7627c N |
831 | |
832 | bitmap_file_put(bitmap); | |
833 | ||
834 | return; | |
835 | } | |
836 | ||
837 | enum bitmap_page_attr { | |
ac2f40be | 838 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
5a537df4 N |
839 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
840 | * i.e. counter is 1 or 2. */ | |
ac2f40be | 841 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
32a7627c N |
842 | }; |
843 | ||
844 | static inline void set_page_attr(struct bitmap *bitmap, struct page *page, | |
845 | enum bitmap_page_attr attr) | |
846 | { | |
3520fa4d | 847 | __set_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
848 | } |
849 | ||
850 | static inline void clear_page_attr(struct bitmap *bitmap, struct page *page, | |
851 | enum bitmap_page_attr attr) | |
852 | { | |
3520fa4d | 853 | __clear_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
854 | } |
855 | ||
ec7a3197 N |
856 | static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page, |
857 | enum bitmap_page_attr attr) | |
32a7627c | 858 | { |
3520fa4d | 859 | return test_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
860 | } |
861 | ||
862 | /* | |
863 | * bitmap_file_set_bit -- called before performing a write to the md device | |
864 | * to set (and eventually sync) a particular bit in the bitmap file | |
865 | * | |
866 | * we set the bit immediately, then we record the page number so that | |
867 | * when an unplug occurs, we can flush the dirty pages out to disk | |
868 | */ | |
869 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
870 | { | |
871 | unsigned long bit; | |
3520fa4d | 872 | struct page *page; |
32a7627c N |
873 | void *kaddr; |
874 | unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap); | |
875 | ||
3520fa4d JB |
876 | if (!bitmap->filemap) |
877 | return; | |
32a7627c | 878 | |
3520fa4d JB |
879 | page = filemap_get_page(bitmap, chunk); |
880 | if (!page) | |
881 | return; | |
882 | bit = file_page_offset(bitmap, chunk); | |
32a7627c | 883 | |
3520fa4d JB |
884 | /* set the bit */ |
885 | kaddr = kmap_atomic(page, KM_USER0); | |
886 | if (bitmap->flags & BITMAP_HOSTENDIAN) | |
887 | set_bit(bit, kaddr); | |
888 | else | |
889 | __set_bit_le(bit, kaddr); | |
890 | kunmap_atomic(kaddr, KM_USER0); | |
36a4e1fe | 891 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
32a7627c N |
892 | /* record page number so it gets flushed to disk when unplug occurs */ |
893 | set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); | |
32a7627c N |
894 | } |
895 | ||
896 | /* this gets called when the md device is ready to unplug its underlying | |
897 | * (slave) device queues -- before we let any writes go down, we need to | |
898 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 899 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 900 | { |
ec7a3197 N |
901 | unsigned long i, flags; |
902 | int dirty, need_write; | |
32a7627c N |
903 | struct page *page; |
904 | int wait = 0; | |
905 | ||
906 | if (!bitmap) | |
4ad13663 | 907 | return; |
32a7627c N |
908 | |
909 | /* look at each page to see if there are any set bits that need to be | |
910 | * flushed out to disk */ | |
911 | for (i = 0; i < bitmap->file_pages; i++) { | |
912 | spin_lock_irqsave(&bitmap->lock, flags); | |
a654b9d8 | 913 | if (!bitmap->filemap) { |
32a7627c | 914 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 915 | return; |
32a7627c N |
916 | } |
917 | page = bitmap->filemap[i]; | |
ec7a3197 N |
918 | dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
919 | need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
32a7627c N |
920 | clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
921 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
ec7a3197 | 922 | if (dirty) |
32a7627c N |
923 | wait = 1; |
924 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
925 | ||
ac2f40be | 926 | if (dirty || need_write) |
4ad13663 | 927 | write_page(bitmap, page, 0); |
32a7627c N |
928 | } |
929 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
0b79ccf0 | 930 | if (bitmap->file) |
d785a06a N |
931 | wait_event(bitmap->write_wait, |
932 | atomic_read(&bitmap->pending_writes)==0); | |
0b79ccf0 | 933 | else |
a9701a30 | 934 | md_super_wait(bitmap->mddev); |
32a7627c | 935 | } |
d785a06a N |
936 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
937 | bitmap_file_kick(bitmap); | |
32a7627c | 938 | } |
ac2f40be | 939 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 940 | |
6a07997f | 941 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
942 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
943 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
944 | * memory mapping of the bitmap file | |
945 | * Special cases: | |
946 | * if there's no bitmap file, or if the bitmap file had been | |
947 | * previously kicked from the array, we mark all the bits as | |
948 | * 1's in order to cause a full resync. | |
6a07997f N |
949 | * |
950 | * We ignore all bits for sectors that end earlier than 'start'. | |
951 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 952 | */ |
6a07997f | 953 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
954 | { |
955 | unsigned long i, chunks, index, oldindex, bit; | |
956 | struct page *page = NULL, *oldpage = NULL; | |
957 | unsigned long num_pages, bit_cnt = 0; | |
958 | struct file *file; | |
d785a06a | 959 | unsigned long bytes, offset; |
32a7627c N |
960 | int outofdate; |
961 | int ret = -ENOSPC; | |
ea03aff9 | 962 | void *paddr; |
32a7627c N |
963 | |
964 | chunks = bitmap->chunks; | |
965 | file = bitmap->file; | |
966 | ||
42a04b50 | 967 | BUG_ON(!file && !bitmap->mddev->bitmap_info.offset); |
32a7627c | 968 | |
32a7627c | 969 | outofdate = bitmap->flags & BITMAP_STALE; |
32a7627c N |
970 | if (outofdate) |
971 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
972 | "recovery\n", bmname(bitmap)); | |
973 | ||
e384e585 | 974 | bytes = DIV_ROUND_UP(bitmap->chunks, 8); |
ece5cff0 N |
975 | if (!bitmap->mddev->bitmap_info.external) |
976 | bytes += sizeof(bitmap_super_t); | |
bc7f77de | 977 | |
e384e585 | 978 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); |
bc7f77de | 979 | |
ece5cff0 | 980 | if (file && i_size_read(file->f_mapping->host) < bytes) { |
32a7627c N |
981 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
982 | bmname(bitmap), | |
983 | (unsigned long) i_size_read(file->f_mapping->host), | |
ece5cff0 | 984 | bytes); |
4ad13663 | 985 | goto err; |
32a7627c | 986 | } |
bc7f77de N |
987 | |
988 | ret = -ENOMEM; | |
989 | ||
32a7627c | 990 | bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL); |
bc7f77de | 991 | if (!bitmap->filemap) |
4ad13663 | 992 | goto err; |
32a7627c | 993 | |
e16b68b6 N |
994 | /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */ |
995 | bitmap->filemap_attr = kzalloc( | |
ac2f40be | 996 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), |
e16b68b6 | 997 | GFP_KERNEL); |
bc7f77de | 998 | if (!bitmap->filemap_attr) |
4ad13663 | 999 | goto err; |
32a7627c | 1000 | |
32a7627c N |
1001 | oldindex = ~0L; |
1002 | ||
1003 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 1004 | int b; |
ece5cff0 N |
1005 | index = file_page_index(bitmap, i); |
1006 | bit = file_page_offset(bitmap, i); | |
32a7627c | 1007 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 1008 | int count; |
32a7627c | 1009 | /* unmap the old page, we're done with it */ |
d785a06a | 1010 | if (index == num_pages-1) |
ece5cff0 | 1011 | count = bytes - index * PAGE_SIZE; |
d785a06a N |
1012 | else |
1013 | count = PAGE_SIZE; | |
ece5cff0 | 1014 | if (index == 0 && bitmap->sb_page) { |
32a7627c N |
1015 | /* |
1016 | * if we're here then the superblock page | |
1017 | * contains some bits (PAGE_SIZE != sizeof sb) | |
1018 | * we've already read it in, so just use it | |
1019 | */ | |
1020 | page = bitmap->sb_page; | |
1021 | offset = sizeof(bitmap_super_t); | |
53845270 | 1022 | if (!file) |
5c04f551 VK |
1023 | page = read_sb_page( |
1024 | bitmap->mddev, | |
1025 | bitmap->mddev->bitmap_info.offset, | |
1026 | page, | |
1027 | index, count); | |
a654b9d8 | 1028 | } else if (file) { |
d785a06a | 1029 | page = read_page(file, index, bitmap, count); |
a654b9d8 N |
1030 | offset = 0; |
1031 | } else { | |
42a04b50 N |
1032 | page = read_sb_page(bitmap->mddev, |
1033 | bitmap->mddev->bitmap_info.offset, | |
a2ed9615 N |
1034 | NULL, |
1035 | index, count); | |
32a7627c N |
1036 | offset = 0; |
1037 | } | |
a654b9d8 N |
1038 | if (IS_ERR(page)) { /* read error */ |
1039 | ret = PTR_ERR(page); | |
4ad13663 | 1040 | goto err; |
a654b9d8 N |
1041 | } |
1042 | ||
32a7627c N |
1043 | oldindex = index; |
1044 | oldpage = page; | |
32a7627c | 1045 | |
b74fd282 N |
1046 | bitmap->filemap[bitmap->file_pages++] = page; |
1047 | bitmap->last_page_size = count; | |
1048 | ||
32a7627c N |
1049 | if (outofdate) { |
1050 | /* | |
1051 | * if bitmap is out of date, dirty the | |
ac2f40be | 1052 | * whole page and write it out |
32a7627c | 1053 | */ |
ea03aff9 N |
1054 | paddr = kmap_atomic(page, KM_USER0); |
1055 | memset(paddr + offset, 0xff, | |
6a07997f | 1056 | PAGE_SIZE - offset); |
ea03aff9 | 1057 | kunmap_atomic(paddr, KM_USER0); |
4ad13663 N |
1058 | write_page(bitmap, page, 1); |
1059 | ||
1060 | ret = -EIO; | |
b74fd282 | 1061 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
4ad13663 | 1062 | goto err; |
32a7627c | 1063 | } |
32a7627c | 1064 | } |
ea03aff9 | 1065 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1066 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1067 | b = test_bit(bit, paddr); |
bd926c63 | 1068 | else |
6b33aff3 | 1069 | b = test_bit_le(bit, paddr); |
ea03aff9 | 1070 | kunmap_atomic(paddr, KM_USER0); |
bd926c63 | 1071 | if (b) { |
32a7627c | 1072 | /* if the disk bit is set, set the memory bit */ |
db305e50 N |
1073 | int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) |
1074 | >= start); | |
1075 | bitmap_set_memory_bits(bitmap, | |
1076 | (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap), | |
1077 | needed); | |
32a7627c N |
1078 | bit_cnt++; |
1079 | } | |
32a7627c N |
1080 | } |
1081 | ||
ac2f40be | 1082 | /* everything went OK */ |
32a7627c N |
1083 | ret = 0; |
1084 | bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET); | |
1085 | ||
32a7627c N |
1086 | if (bit_cnt) { /* Kick recovery if any bits were set */ |
1087 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
1088 | md_wakeup_thread(bitmap->mddev->thread); | |
1089 | } | |
1090 | ||
32a7627c | 1091 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
9c81075f JB |
1092 | "read %lu/%lu pages, set %lu of %lu bits\n", |
1093 | bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, chunks); | |
4ad13663 N |
1094 | |
1095 | return 0; | |
32a7627c | 1096 | |
4ad13663 N |
1097 | err: |
1098 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1099 | bmname(bitmap), ret); | |
32a7627c N |
1100 | return ret; |
1101 | } | |
1102 | ||
a654b9d8 N |
1103 | void bitmap_write_all(struct bitmap *bitmap) |
1104 | { | |
1105 | /* We don't actually write all bitmap blocks here, | |
1106 | * just flag them as needing to be written | |
1107 | */ | |
ec7a3197 | 1108 | int i; |
a654b9d8 | 1109 | |
7c8f4247 | 1110 | spin_lock_irq(&bitmap->lock); |
ac2f40be | 1111 | for (i = 0; i < bitmap->file_pages; i++) |
ec7a3197 N |
1112 | set_page_attr(bitmap, bitmap->filemap[i], |
1113 | BITMAP_PAGE_NEEDWRITE); | |
2585f3ef | 1114 | bitmap->allclean = 0; |
7c8f4247 | 1115 | spin_unlock_irq(&bitmap->lock); |
a654b9d8 N |
1116 | } |
1117 | ||
32a7627c N |
1118 | static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) |
1119 | { | |
1120 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1121 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1122 | bitmap->bp[page].count += inc; | |
32a7627c N |
1123 | bitmap_checkfree(bitmap, page); |
1124 | } | |
1125 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, | |
57dab0bd | 1126 | sector_t offset, sector_t *blocks, |
32a7627c N |
1127 | int create); |
1128 | ||
1129 | /* | |
1130 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1131 | * out to disk | |
1132 | */ | |
1133 | ||
fd01b88c | 1134 | void bitmap_daemon_work(struct mddev *mddev) |
32a7627c | 1135 | { |
aa5cbd10 | 1136 | struct bitmap *bitmap; |
aa3163f8 | 1137 | unsigned long j; |
32a7627c N |
1138 | unsigned long flags; |
1139 | struct page *page = NULL, *lastpage = NULL; | |
57dab0bd | 1140 | sector_t blocks; |
ea03aff9 | 1141 | void *paddr; |
32a7627c | 1142 | |
aa5cbd10 N |
1143 | /* Use a mutex to guard daemon_work against |
1144 | * bitmap_destroy. | |
1145 | */ | |
c3d9714e | 1146 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1147 | bitmap = mddev->bitmap; |
1148 | if (bitmap == NULL) { | |
c3d9714e | 1149 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1150 | return; |
aa5cbd10 | 1151 | } |
42a04b50 | 1152 | if (time_before(jiffies, bitmap->daemon_lastrun |
2e61ebbc | 1153 | + mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1154 | goto done; |
1155 | ||
32a7627c | 1156 | bitmap->daemon_lastrun = jiffies; |
8311c29d | 1157 | if (bitmap->allclean) { |
2e61ebbc | 1158 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
aa5cbd10 | 1159 | goto done; |
8311c29d N |
1160 | } |
1161 | bitmap->allclean = 1; | |
32a7627c | 1162 | |
be512691 | 1163 | spin_lock_irqsave(&bitmap->lock, flags); |
32a7627c N |
1164 | for (j = 0; j < bitmap->chunks; j++) { |
1165 | bitmap_counter_t *bmc; | |
3520fa4d JB |
1166 | if (!bitmap->filemap) |
1167 | /* error or shutdown */ | |
1168 | break; | |
1169 | ||
1170 | page = filemap_get_page(bitmap, j); | |
32a7627c N |
1171 | |
1172 | if (page != lastpage) { | |
aa3163f8 | 1173 | /* skip this page unless it's marked as needing cleaning */ |
5a537df4 | 1174 | if (!test_page_attr(bitmap, page, BITMAP_PAGE_PENDING)) { |
ec7a3197 N |
1175 | int need_write = test_page_attr(bitmap, page, |
1176 | BITMAP_PAGE_NEEDWRITE); | |
a647e4bc | 1177 | if (need_write) |
aa3163f8 | 1178 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); |
a647e4bc | 1179 | |
aa3163f8 | 1180 | spin_unlock_irqrestore(&bitmap->lock, flags); |
2585f3ef | 1181 | if (need_write) |
4ad13663 | 1182 | write_page(bitmap, page, 0); |
be512691 N |
1183 | spin_lock_irqsave(&bitmap->lock, flags); |
1184 | j |= (PAGE_BITS - 1); | |
aa3163f8 N |
1185 | continue; |
1186 | } | |
1187 | ||
32a7627c | 1188 | /* grab the new page, sync and release the old */ |
32a7627c | 1189 | if (lastpage != NULL) { |
2585f3ef N |
1190 | if (test_page_attr(bitmap, lastpage, |
1191 | BITMAP_PAGE_NEEDWRITE)) { | |
1192 | clear_page_attr(bitmap, lastpage, | |
1193 | BITMAP_PAGE_NEEDWRITE); | |
32a7627c | 1194 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 1195 | write_page(bitmap, lastpage, 0); |
32a7627c | 1196 | } else { |
2585f3ef N |
1197 | set_page_attr(bitmap, lastpage, |
1198 | BITMAP_PAGE_NEEDWRITE); | |
1199 | bitmap->allclean = 0; | |
32a7627c N |
1200 | spin_unlock_irqrestore(&bitmap->lock, flags); |
1201 | } | |
32a7627c N |
1202 | } else |
1203 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1204 | lastpage = page; | |
a0da84f3 NB |
1205 | |
1206 | /* We are possibly going to clear some bits, so make | |
1207 | * sure that events_cleared is up-to-date. | |
1208 | */ | |
ece5cff0 | 1209 | if (bitmap->need_sync && |
2e61ebbc | 1210 | mddev->bitmap_info.external == 0) { |
a0da84f3 NB |
1211 | bitmap_super_t *sb; |
1212 | bitmap->need_sync = 0; | |
1213 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); | |
1214 | sb->events_cleared = | |
1215 | cpu_to_le64(bitmap->events_cleared); | |
1216 | kunmap_atomic(sb, KM_USER0); | |
1217 | write_page(bitmap, bitmap->sb_page, 1); | |
1218 | } | |
32a7627c | 1219 | spin_lock_irqsave(&bitmap->lock, flags); |
ece5cff0 | 1220 | if (!bitmap->need_sync) |
5a537df4 | 1221 | clear_page_attr(bitmap, page, BITMAP_PAGE_PENDING); |
2585f3ef N |
1222 | else |
1223 | bitmap->allclean = 0; | |
32a7627c | 1224 | } |
db305e50 N |
1225 | bmc = bitmap_get_counter(bitmap, |
1226 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
1227 | &blocks, 0); | |
5a537df4 N |
1228 | if (!bmc) |
1229 | j |= PAGE_COUNTER_MASK; | |
1230 | else if (*bmc) { | |
5a537df4 | 1231 | if (*bmc == 1 && !bitmap->need_sync) { |
32a7627c N |
1232 | /* we can clear the bit */ |
1233 | *bmc = 0; | |
db305e50 N |
1234 | bitmap_count_page(bitmap, |
1235 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
32a7627c N |
1236 | -1); |
1237 | ||
1238 | /* clear the bit */ | |
3520fa4d JB |
1239 | paddr = kmap_atomic(page, KM_USER0); |
1240 | if (bitmap->flags & BITMAP_HOSTENDIAN) | |
1241 | clear_bit(file_page_offset(bitmap, j), | |
1242 | paddr); | |
1243 | else | |
1244 | __clear_bit_le( | |
5a537df4 N |
1245 | file_page_offset(bitmap, |
1246 | j), | |
1247 | paddr); | |
3520fa4d | 1248 | kunmap_atomic(paddr, KM_USER0); |
5a537df4 N |
1249 | } else if (*bmc <= 2) { |
1250 | *bmc = 1; /* maybe clear the bit next time */ | |
1251 | set_page_attr(bitmap, page, BITMAP_PAGE_PENDING); | |
2585f3ef | 1252 | bitmap->allclean = 0; |
32a7627c | 1253 | } |
5a537df4 | 1254 | } |
32a7627c | 1255 | } |
be512691 | 1256 | spin_unlock_irqrestore(&bitmap->lock, flags); |
32a7627c N |
1257 | |
1258 | /* now sync the final page */ | |
3520fa4d | 1259 | if (lastpage != NULL) { |
32a7627c | 1260 | spin_lock_irqsave(&bitmap->lock, flags); |
ec7a3197 | 1261 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1262 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1263 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
3520fa4d | 1264 | write_page(bitmap, lastpage, 0); |
32a7627c N |
1265 | } else { |
1266 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
2585f3ef | 1267 | bitmap->allclean = 0; |
32a7627c N |
1268 | spin_unlock_irqrestore(&bitmap->lock, flags); |
1269 | } | |
32a7627c N |
1270 | } |
1271 | ||
7be3dfec | 1272 | done: |
8311c29d | 1273 | if (bitmap->allclean == 0) |
2e61ebbc N |
1274 | mddev->thread->timeout = |
1275 | mddev->bitmap_info.daemon_sleep; | |
c3d9714e | 1276 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1277 | } |
1278 | ||
32a7627c | 1279 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
57dab0bd | 1280 | sector_t offset, sector_t *blocks, |
32a7627c | 1281 | int create) |
ee305ace N |
1282 | __releases(bitmap->lock) |
1283 | __acquires(bitmap->lock) | |
32a7627c N |
1284 | { |
1285 | /* If 'create', we might release the lock and reclaim it. | |
1286 | * The lock must have been taken with interrupts enabled. | |
1287 | * If !create, we don't release the lock. | |
1288 | */ | |
1289 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1290 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1291 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1292 | sector_t csize; | |
ef425673 | 1293 | int err; |
32a7627c | 1294 | |
ef425673 N |
1295 | err = bitmap_checkpage(bitmap, page, create); |
1296 | ||
1297 | if (bitmap->bp[page].hijacked || | |
1298 | bitmap->bp[page].map == NULL) | |
1299 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) + | |
1300 | PAGE_COUNTER_SHIFT - 1); | |
1301 | else | |
32a7627c | 1302 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); |
ef425673 N |
1303 | *blocks = csize - (offset & (csize - 1)); |
1304 | ||
1305 | if (err < 0) | |
32a7627c | 1306 | return NULL; |
ef425673 | 1307 | |
32a7627c N |
1308 | /* now locked ... */ |
1309 | ||
1310 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1311 | /* should we use the first or second counter field | |
1312 | * of the hijacked pointer? */ | |
1313 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1314 | return &((bitmap_counter_t *) |
1315 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1316 | } else /* page is allocated */ |
32a7627c N |
1317 | return (bitmap_counter_t *) |
1318 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1319 | } |
1320 | ||
4b6d287f | 1321 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1322 | { |
ac2f40be N |
1323 | if (!bitmap) |
1324 | return 0; | |
4b6d287f N |
1325 | |
1326 | if (behind) { | |
696fcd53 | 1327 | int bw; |
4b6d287f | 1328 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1329 | bw = atomic_read(&bitmap->behind_writes); |
1330 | if (bw > bitmap->behind_writes_used) | |
1331 | bitmap->behind_writes_used = bw; | |
1332 | ||
36a4e1fe N |
1333 | pr_debug("inc write-behind count %d/%lu\n", |
1334 | bw, bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1335 | } |
1336 | ||
32a7627c | 1337 | while (sectors) { |
57dab0bd | 1338 | sector_t blocks; |
32a7627c N |
1339 | bitmap_counter_t *bmc; |
1340 | ||
1341 | spin_lock_irq(&bitmap->lock); | |
1342 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); | |
1343 | if (!bmc) { | |
1344 | spin_unlock_irq(&bitmap->lock); | |
1345 | return 0; | |
1346 | } | |
1347 | ||
27d5ea04 | 1348 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
da6e1a32 NB |
1349 | DEFINE_WAIT(__wait); |
1350 | /* note that it is safe to do the prepare_to_wait | |
1351 | * after the test as long as we do it before dropping | |
1352 | * the spinlock. | |
1353 | */ | |
1354 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1355 | TASK_UNINTERRUPTIBLE); | |
1356 | spin_unlock_irq(&bitmap->lock); | |
7eaceacc | 1357 | io_schedule(); |
da6e1a32 NB |
1358 | finish_wait(&bitmap->overflow_wait, &__wait); |
1359 | continue; | |
1360 | } | |
1361 | ||
ac2f40be | 1362 | switch (*bmc) { |
32a7627c N |
1363 | case 0: |
1364 | bitmap_file_set_bit(bitmap, offset); | |
ac2f40be | 1365 | bitmap_count_page(bitmap, offset, 1); |
32a7627c N |
1366 | /* fall through */ |
1367 | case 1: | |
1368 | *bmc = 2; | |
1369 | } | |
da6e1a32 | 1370 | |
32a7627c N |
1371 | (*bmc)++; |
1372 | ||
1373 | spin_unlock_irq(&bitmap->lock); | |
1374 | ||
1375 | offset += blocks; | |
1376 | if (sectors > blocks) | |
1377 | sectors -= blocks; | |
ac2f40be N |
1378 | else |
1379 | sectors = 0; | |
32a7627c N |
1380 | } |
1381 | return 0; | |
1382 | } | |
ac2f40be | 1383 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1384 | |
1385 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1386 | int success, int behind) |
32a7627c | 1387 | { |
ac2f40be N |
1388 | if (!bitmap) |
1389 | return; | |
4b6d287f | 1390 | if (behind) { |
e555190d N |
1391 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1392 | wake_up(&bitmap->behind_wait); | |
36a4e1fe N |
1393 | pr_debug("dec write-behind count %d/%lu\n", |
1394 | atomic_read(&bitmap->behind_writes), | |
1395 | bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1396 | } |
1397 | ||
32a7627c | 1398 | while (sectors) { |
57dab0bd | 1399 | sector_t blocks; |
32a7627c N |
1400 | unsigned long flags; |
1401 | bitmap_counter_t *bmc; | |
1402 | ||
1403 | spin_lock_irqsave(&bitmap->lock, flags); | |
1404 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); | |
1405 | if (!bmc) { | |
1406 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1407 | return; | |
1408 | } | |
1409 | ||
961902c0 | 1410 | if (success && !bitmap->mddev->degraded && |
a0da84f3 NB |
1411 | bitmap->events_cleared < bitmap->mddev->events) { |
1412 | bitmap->events_cleared = bitmap->mddev->events; | |
1413 | bitmap->need_sync = 1; | |
5ff5afff | 1414 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1415 | } |
1416 | ||
27d5ea04 | 1417 | if (!success && !NEEDED(*bmc)) |
32a7627c N |
1418 | *bmc |= NEEDED_MASK; |
1419 | ||
27d5ea04 | 1420 | if (COUNTER(*bmc) == COUNTER_MAX) |
da6e1a32 NB |
1421 | wake_up(&bitmap->overflow_wait); |
1422 | ||
32a7627c | 1423 | (*bmc)--; |
2585f3ef | 1424 | if (*bmc <= 2) { |
32a7627c | 1425 | set_page_attr(bitmap, |
e384e585 N |
1426 | filemap_get_page( |
1427 | bitmap, | |
1428 | offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
5a537df4 | 1429 | BITMAP_PAGE_PENDING); |
2585f3ef N |
1430 | bitmap->allclean = 0; |
1431 | } | |
32a7627c N |
1432 | spin_unlock_irqrestore(&bitmap->lock, flags); |
1433 | offset += blocks; | |
1434 | if (sectors > blocks) | |
1435 | sectors -= blocks; | |
ac2f40be N |
1436 | else |
1437 | sectors = 0; | |
32a7627c N |
1438 | } |
1439 | } | |
ac2f40be | 1440 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1441 | |
57dab0bd | 1442 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1443 | int degraded) |
32a7627c N |
1444 | { |
1445 | bitmap_counter_t *bmc; | |
1446 | int rv; | |
1447 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1448 | *blocks = 1024; | |
1449 | return 1; /* always resync if no bitmap */ | |
1450 | } | |
1451 | spin_lock_irq(&bitmap->lock); | |
1452 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1453 | rv = 0; | |
1454 | if (bmc) { | |
1455 | /* locked */ | |
1456 | if (RESYNC(*bmc)) | |
1457 | rv = 1; | |
1458 | else if (NEEDED(*bmc)) { | |
1459 | rv = 1; | |
6a806c51 N |
1460 | if (!degraded) { /* don't set/clear bits if degraded */ |
1461 | *bmc |= RESYNC_MASK; | |
1462 | *bmc &= ~NEEDED_MASK; | |
1463 | } | |
32a7627c N |
1464 | } |
1465 | } | |
1466 | spin_unlock_irq(&bitmap->lock); | |
1467 | return rv; | |
1468 | } | |
1469 | ||
57dab0bd | 1470 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a N |
1471 | int degraded) |
1472 | { | |
1473 | /* bitmap_start_sync must always report on multiples of whole | |
1474 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1475 | * get confused. | |
1476 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1477 | * At least PAGE_SIZE>>9 blocks are covered. | |
1478 | * Return the 'or' of the result. | |
1479 | */ | |
1480 | int rv = 0; | |
57dab0bd | 1481 | sector_t blocks1; |
1187cf0a N |
1482 | |
1483 | *blocks = 0; | |
1484 | while (*blocks < (PAGE_SIZE>>9)) { | |
1485 | rv |= __bitmap_start_sync(bitmap, offset, | |
1486 | &blocks1, degraded); | |
1487 | offset += blocks1; | |
1488 | *blocks += blocks1; | |
1489 | } | |
1490 | return rv; | |
1491 | } | |
ac2f40be | 1492 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1493 | |
57dab0bd | 1494 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1495 | { |
1496 | bitmap_counter_t *bmc; | |
1497 | unsigned long flags; | |
ac2f40be N |
1498 | |
1499 | if (bitmap == NULL) { | |
32a7627c N |
1500 | *blocks = 1024; |
1501 | return; | |
1502 | } | |
1503 | spin_lock_irqsave(&bitmap->lock, flags); | |
1504 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1505 | if (bmc == NULL) | |
1506 | goto unlock; | |
1507 | /* locked */ | |
32a7627c N |
1508 | if (RESYNC(*bmc)) { |
1509 | *bmc &= ~RESYNC_MASK; | |
1510 | ||
1511 | if (!NEEDED(*bmc) && aborted) | |
1512 | *bmc |= NEEDED_MASK; | |
1513 | else { | |
2585f3ef | 1514 | if (*bmc <= 2) { |
32a7627c N |
1515 | set_page_attr(bitmap, |
1516 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
5a537df4 | 1517 | BITMAP_PAGE_PENDING); |
2585f3ef N |
1518 | bitmap->allclean = 0; |
1519 | } | |
32a7627c N |
1520 | } |
1521 | } | |
1522 | unlock: | |
1523 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1524 | } | |
ac2f40be | 1525 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1526 | |
1527 | void bitmap_close_sync(struct bitmap *bitmap) | |
1528 | { | |
1529 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1530 | * properly have been aborted. It remains to us to clear the | |
1531 | * RESYNC bit wherever it is still on | |
1532 | */ | |
1533 | sector_t sector = 0; | |
57dab0bd | 1534 | sector_t blocks; |
b47490c9 N |
1535 | if (!bitmap) |
1536 | return; | |
32a7627c N |
1537 | while (sector < bitmap->mddev->resync_max_sectors) { |
1538 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1539 | sector += blocks; |
1540 | } | |
1541 | } | |
ac2f40be | 1542 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 N |
1543 | |
1544 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1545 | { | |
1546 | sector_t s = 0; | |
57dab0bd | 1547 | sector_t blocks; |
b47490c9 N |
1548 | |
1549 | if (!bitmap) | |
1550 | return; | |
1551 | if (sector == 0) { | |
1552 | bitmap->last_end_sync = jiffies; | |
1553 | return; | |
1554 | } | |
1555 | if (time_before(jiffies, (bitmap->last_end_sync | |
1b04be96 | 1556 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1557 | return; |
1558 | wait_event(bitmap->mddev->recovery_wait, | |
1559 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1560 | ||
75d3da43 | 1561 | bitmap->mddev->curr_resync_completed = sector; |
070dc6dd | 1562 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
b47490c9 N |
1563 | sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1); |
1564 | s = 0; | |
1565 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1566 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1567 | s += blocks; | |
32a7627c | 1568 | } |
b47490c9 | 1569 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1570 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1571 | } |
ac2f40be | 1572 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1573 | |
6a07997f | 1574 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1575 | { |
1576 | /* For each chunk covered by any of these sectors, set the | |
193f1c93 | 1577 | * counter to 1 and set resync_needed. They should all |
32a7627c N |
1578 | * be 0 at this point |
1579 | */ | |
193f1c93 | 1580 | |
57dab0bd | 1581 | sector_t secs; |
193f1c93 N |
1582 | bitmap_counter_t *bmc; |
1583 | spin_lock_irq(&bitmap->lock); | |
1584 | bmc = bitmap_get_counter(bitmap, offset, &secs, 1); | |
1585 | if (!bmc) { | |
32a7627c | 1586 | spin_unlock_irq(&bitmap->lock); |
193f1c93 | 1587 | return; |
32a7627c | 1588 | } |
ac2f40be | 1589 | if (!*bmc) { |
193f1c93 | 1590 | struct page *page; |
915c420d | 1591 | *bmc = 2 | (needed ? NEEDED_MASK : 0); |
193f1c93 N |
1592 | bitmap_count_page(bitmap, offset, 1); |
1593 | page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)); | |
5a537df4 | 1594 | set_page_attr(bitmap, page, BITMAP_PAGE_PENDING); |
2585f3ef | 1595 | bitmap->allclean = 0; |
193f1c93 N |
1596 | } |
1597 | spin_unlock_irq(&bitmap->lock); | |
32a7627c N |
1598 | } |
1599 | ||
9b1d1dac PC |
1600 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1601 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1602 | { | |
1603 | unsigned long chunk; | |
1604 | ||
1605 | for (chunk = s; chunk <= e; chunk++) { | |
db305e50 | 1606 | sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap); |
9b1d1dac | 1607 | bitmap_set_memory_bits(bitmap, sec, 1); |
7c8f4247 | 1608 | spin_lock_irq(&bitmap->lock); |
9b1d1dac | 1609 | bitmap_file_set_bit(bitmap, sec); |
7c8f4247 | 1610 | spin_unlock_irq(&bitmap->lock); |
ffa23322 N |
1611 | if (sec < bitmap->mddev->recovery_cp) |
1612 | /* We are asserting that the array is dirty, | |
1613 | * so move the recovery_cp address back so | |
1614 | * that it is obvious that it is dirty | |
1615 | */ | |
1616 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1617 | } |
1618 | } | |
1619 | ||
6b8b3e8a N |
1620 | /* |
1621 | * flush out any pending updates | |
1622 | */ | |
fd01b88c | 1623 | void bitmap_flush(struct mddev *mddev) |
6b8b3e8a N |
1624 | { |
1625 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1626 | long sleep; |
6b8b3e8a N |
1627 | |
1628 | if (!bitmap) /* there was no bitmap */ | |
1629 | return; | |
1630 | ||
1631 | /* run the daemon_work three time to ensure everything is flushed | |
1632 | * that can be | |
1633 | */ | |
1b04be96 | 1634 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1635 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1636 | bitmap_daemon_work(mddev); |
42a04b50 | 1637 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1638 | bitmap_daemon_work(mddev); |
42a04b50 | 1639 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1640 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1641 | bitmap_update_sb(bitmap); |
1642 | } | |
1643 | ||
32a7627c N |
1644 | /* |
1645 | * free memory that was allocated | |
1646 | */ | |
3178b0db | 1647 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1648 | { |
1649 | unsigned long k, pages; | |
1650 | struct bitmap_page *bp; | |
32a7627c N |
1651 | |
1652 | if (!bitmap) /* there was no bitmap */ | |
1653 | return; | |
1654 | ||
32a7627c N |
1655 | /* release the bitmap file and kill the daemon */ |
1656 | bitmap_file_put(bitmap); | |
1657 | ||
1658 | bp = bitmap->bp; | |
1659 | pages = bitmap->pages; | |
1660 | ||
1661 | /* free all allocated memory */ | |
1662 | ||
32a7627c N |
1663 | if (bp) /* deallocate the page memory */ |
1664 | for (k = 0; k < pages; k++) | |
1665 | if (bp[k].map && !bp[k].hijacked) | |
1666 | kfree(bp[k].map); | |
1667 | kfree(bp); | |
1668 | kfree(bitmap); | |
1669 | } | |
aa5cbd10 | 1670 | |
fd01b88c | 1671 | void bitmap_destroy(struct mddev *mddev) |
3178b0db N |
1672 | { |
1673 | struct bitmap *bitmap = mddev->bitmap; | |
1674 | ||
1675 | if (!bitmap) /* there was no bitmap */ | |
1676 | return; | |
1677 | ||
c3d9714e | 1678 | mutex_lock(&mddev->bitmap_info.mutex); |
3178b0db | 1679 | mddev->bitmap = NULL; /* disconnect from the md device */ |
c3d9714e | 1680 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1681 | if (mddev->thread) |
1682 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db | 1683 | |
ece5cff0 N |
1684 | if (bitmap->sysfs_can_clear) |
1685 | sysfs_put(bitmap->sysfs_can_clear); | |
1686 | ||
3178b0db N |
1687 | bitmap_free(bitmap); |
1688 | } | |
32a7627c N |
1689 | |
1690 | /* | |
1691 | * initialize the bitmap structure | |
1692 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1693 | */ | |
fd01b88c | 1694 | int bitmap_create(struct mddev *mddev) |
32a7627c N |
1695 | { |
1696 | struct bitmap *bitmap; | |
1f593903 | 1697 | sector_t blocks = mddev->resync_max_sectors; |
32a7627c N |
1698 | unsigned long chunks; |
1699 | unsigned long pages; | |
c3d9714e | 1700 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1701 | int err; |
5ff5afff | 1702 | struct sysfs_dirent *bm = NULL; |
32a7627c | 1703 | |
5f6e3c83 | 1704 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1705 | |
e384e585 | 1706 | if (!file |
3520fa4d | 1707 | && !mddev->bitmap_info.offset) /* bitmap disabled, nothing to do */ |
32a7627c N |
1708 | return 0; |
1709 | ||
c3d9714e | 1710 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1711 | |
9ffae0cf | 1712 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1713 | if (!bitmap) |
1714 | return -ENOMEM; | |
1715 | ||
32a7627c | 1716 | spin_lock_init(&bitmap->lock); |
ce25c31b N |
1717 | atomic_set(&bitmap->pending_writes, 0); |
1718 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1719 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1720 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1721 | |
32a7627c | 1722 | bitmap->mddev = mddev; |
32a7627c | 1723 | |
5ff5afff N |
1724 | if (mddev->kobj.sd) |
1725 | bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap"); | |
ece5cff0 | 1726 | if (bm) { |
3ff195b0 | 1727 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear"); |
ece5cff0 N |
1728 | sysfs_put(bm); |
1729 | } else | |
1730 | bitmap->sysfs_can_clear = NULL; | |
1731 | ||
32a7627c | 1732 | bitmap->file = file; |
ce25c31b N |
1733 | if (file) { |
1734 | get_file(file); | |
ae8fa283 N |
1735 | /* As future accesses to this file will use bmap, |
1736 | * and bypass the page cache, we must sync the file | |
1737 | * first. | |
1738 | */ | |
8018ab05 | 1739 | vfs_fsync(file, 1); |
ce25c31b | 1740 | } |
42a04b50 | 1741 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
9c81075f JB |
1742 | if (!mddev->bitmap_info.external) { |
1743 | /* | |
1744 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is | |
1745 | * instructing us to create a new on-disk bitmap instance. | |
1746 | */ | |
1747 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) | |
1748 | err = bitmap_new_disk_sb(bitmap); | |
1749 | else | |
1750 | err = bitmap_read_sb(bitmap); | |
1751 | } else { | |
ece5cff0 N |
1752 | err = 0; |
1753 | if (mddev->bitmap_info.chunksize == 0 || | |
1754 | mddev->bitmap_info.daemon_sleep == 0) | |
1755 | /* chunksize and time_base need to be | |
1756 | * set first. */ | |
1757 | err = -EINVAL; | |
1758 | } | |
32a7627c | 1759 | if (err) |
3178b0db | 1760 | goto error; |
32a7627c | 1761 | |
624ce4f5 | 1762 | bitmap->daemon_lastrun = jiffies; |
42a04b50 | 1763 | bitmap->chunkshift = ffz(~mddev->bitmap_info.chunksize); |
32a7627c N |
1764 | |
1765 | /* now that chunksize and chunkshift are set, we can use these macros */ | |
ac2f40be | 1766 | chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >> |
1f593903 | 1767 | CHUNK_BLOCK_SHIFT(bitmap); |
ac2f40be | 1768 | pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; |
32a7627c N |
1769 | |
1770 | BUG_ON(!pages); | |
1771 | ||
1772 | bitmap->chunks = chunks; | |
1773 | bitmap->pages = pages; | |
1774 | bitmap->missing_pages = pages; | |
32a7627c | 1775 | |
9ffae0cf | 1776 | bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); |
29d3247e | 1777 | |
3178b0db | 1778 | err = -ENOMEM; |
32a7627c | 1779 | if (!bitmap->bp) |
3178b0db | 1780 | goto error; |
32a7627c | 1781 | |
69e51b44 N |
1782 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", |
1783 | pages, bmname(bitmap)); | |
1784 | ||
1785 | mddev->bitmap = bitmap; | |
1786 | ||
1787 | ||
1788 | return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0; | |
1789 | ||
1790 | error: | |
1791 | bitmap_free(bitmap); | |
1792 | return err; | |
1793 | } | |
1794 | ||
fd01b88c | 1795 | int bitmap_load(struct mddev *mddev) |
69e51b44 N |
1796 | { |
1797 | int err = 0; | |
3520fa4d | 1798 | sector_t start = 0; |
69e51b44 N |
1799 | sector_t sector = 0; |
1800 | struct bitmap *bitmap = mddev->bitmap; | |
1801 | ||
1802 | if (!bitmap) | |
1803 | goto out; | |
1804 | ||
1805 | /* Clear out old bitmap info first: Either there is none, or we | |
1806 | * are resuming after someone else has possibly changed things, | |
1807 | * so we should forget old cached info. | |
1808 | * All chunks should be clean, but some might need_sync. | |
1809 | */ | |
1810 | while (sector < mddev->resync_max_sectors) { | |
57dab0bd | 1811 | sector_t blocks; |
69e51b44 N |
1812 | bitmap_start_sync(bitmap, sector, &blocks, 0); |
1813 | sector += blocks; | |
1814 | } | |
1815 | bitmap_close_sync(bitmap); | |
1816 | ||
3520fa4d JB |
1817 | if (mddev->degraded == 0 |
1818 | || bitmap->events_cleared == mddev->events) | |
1819 | /* no need to keep dirty bits to optimise a | |
1820 | * re-add of a missing device */ | |
1821 | start = mddev->recovery_cp; | |
1822 | ||
1823 | err = bitmap_init_from_disk(bitmap, start); | |
1824 | ||
32a7627c | 1825 | if (err) |
69e51b44 | 1826 | goto out; |
3178b0db | 1827 | |
1b04be96 | 1828 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1829 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1830 | |
4ad13663 N |
1831 | bitmap_update_sb(bitmap); |
1832 | ||
69e51b44 N |
1833 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
1834 | err = -EIO; | |
1835 | out: | |
3178b0db | 1836 | return err; |
32a7627c | 1837 | } |
69e51b44 | 1838 | EXPORT_SYMBOL_GPL(bitmap_load); |
32a7627c | 1839 | |
57148964 N |
1840 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap) |
1841 | { | |
1842 | unsigned long chunk_kb; | |
1843 | unsigned long flags; | |
1844 | ||
1845 | if (!bitmap) | |
1846 | return; | |
1847 | ||
1848 | spin_lock_irqsave(&bitmap->lock, flags); | |
1849 | chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; | |
1850 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
1851 | "%lu%s chunk", | |
1852 | bitmap->pages - bitmap->missing_pages, | |
1853 | bitmap->pages, | |
1854 | (bitmap->pages - bitmap->missing_pages) | |
1855 | << (PAGE_SHIFT - 10), | |
1856 | chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, | |
1857 | chunk_kb ? "KB" : "B"); | |
1858 | if (bitmap->file) { | |
1859 | seq_printf(seq, ", file: "); | |
1860 | seq_path(seq, &bitmap->file->f_path, " \t\n"); | |
1861 | } | |
1862 | ||
1863 | seq_printf(seq, "\n"); | |
1864 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1865 | } | |
1866 | ||
43a70507 | 1867 | static ssize_t |
fd01b88c | 1868 | location_show(struct mddev *mddev, char *page) |
43a70507 N |
1869 | { |
1870 | ssize_t len; | |
ac2f40be | 1871 | if (mddev->bitmap_info.file) |
43a70507 | 1872 | len = sprintf(page, "file"); |
ac2f40be | 1873 | else if (mddev->bitmap_info.offset) |
43a70507 | 1874 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 1875 | else |
43a70507 N |
1876 | len = sprintf(page, "none"); |
1877 | len += sprintf(page+len, "\n"); | |
1878 | return len; | |
1879 | } | |
1880 | ||
1881 | static ssize_t | |
fd01b88c | 1882 | location_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
1883 | { |
1884 | ||
1885 | if (mddev->pers) { | |
1886 | if (!mddev->pers->quiesce) | |
1887 | return -EBUSY; | |
1888 | if (mddev->recovery || mddev->sync_thread) | |
1889 | return -EBUSY; | |
1890 | } | |
1891 | ||
1892 | if (mddev->bitmap || mddev->bitmap_info.file || | |
1893 | mddev->bitmap_info.offset) { | |
1894 | /* bitmap already configured. Only option is to clear it */ | |
1895 | if (strncmp(buf, "none", 4) != 0) | |
1896 | return -EBUSY; | |
1897 | if (mddev->pers) { | |
1898 | mddev->pers->quiesce(mddev, 1); | |
1899 | bitmap_destroy(mddev); | |
1900 | mddev->pers->quiesce(mddev, 0); | |
1901 | } | |
1902 | mddev->bitmap_info.offset = 0; | |
1903 | if (mddev->bitmap_info.file) { | |
1904 | struct file *f = mddev->bitmap_info.file; | |
1905 | mddev->bitmap_info.file = NULL; | |
1906 | restore_bitmap_write_access(f); | |
1907 | fput(f); | |
1908 | } | |
1909 | } else { | |
1910 | /* No bitmap, OK to set a location */ | |
1911 | long long offset; | |
1912 | if (strncmp(buf, "none", 4) == 0) | |
1913 | /* nothing to be done */; | |
1914 | else if (strncmp(buf, "file:", 5) == 0) { | |
1915 | /* Not supported yet */ | |
1916 | return -EINVAL; | |
1917 | } else { | |
1918 | int rv; | |
1919 | if (buf[0] == '+') | |
1920 | rv = strict_strtoll(buf+1, 10, &offset); | |
1921 | else | |
1922 | rv = strict_strtoll(buf, 10, &offset); | |
1923 | if (rv) | |
1924 | return rv; | |
1925 | if (offset == 0) | |
1926 | return -EINVAL; | |
ece5cff0 N |
1927 | if (mddev->bitmap_info.external == 0 && |
1928 | mddev->major_version == 0 && | |
43a70507 N |
1929 | offset != mddev->bitmap_info.default_offset) |
1930 | return -EINVAL; | |
1931 | mddev->bitmap_info.offset = offset; | |
1932 | if (mddev->pers) { | |
1933 | mddev->pers->quiesce(mddev, 1); | |
1934 | rv = bitmap_create(mddev); | |
4474ca42 N |
1935 | if (!rv) |
1936 | rv = bitmap_load(mddev); | |
43a70507 N |
1937 | if (rv) { |
1938 | bitmap_destroy(mddev); | |
1939 | mddev->bitmap_info.offset = 0; | |
1940 | } | |
1941 | mddev->pers->quiesce(mddev, 0); | |
1942 | if (rv) | |
1943 | return rv; | |
1944 | } | |
1945 | } | |
1946 | } | |
1947 | if (!mddev->external) { | |
1948 | /* Ensure new bitmap info is stored in | |
1949 | * metadata promptly. | |
1950 | */ | |
1951 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1952 | md_wakeup_thread(mddev->thread); | |
1953 | } | |
1954 | return len; | |
1955 | } | |
1956 | ||
1957 | static struct md_sysfs_entry bitmap_location = | |
1958 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
1959 | ||
1960 | static ssize_t | |
fd01b88c | 1961 | timeout_show(struct mddev *mddev, char *page) |
43a70507 N |
1962 | { |
1963 | ssize_t len; | |
1964 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
1965 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 1966 | |
43a70507 N |
1967 | len = sprintf(page, "%lu", secs); |
1968 | if (jifs) | |
1969 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
1970 | len += sprintf(page+len, "\n"); | |
1971 | return len; | |
1972 | } | |
1973 | ||
1974 | static ssize_t | |
fd01b88c | 1975 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
1976 | { |
1977 | /* timeout can be set at any time */ | |
1978 | unsigned long timeout; | |
1979 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
1980 | if (rv) | |
1981 | return rv; | |
1982 | ||
1983 | /* just to make sure we don't overflow... */ | |
1984 | if (timeout >= LONG_MAX / HZ) | |
1985 | return -EINVAL; | |
1986 | ||
1987 | timeout = timeout * HZ / 10000; | |
1988 | ||
1989 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
1990 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
1991 | if (timeout < 1) | |
1992 | timeout = 1; | |
1993 | mddev->bitmap_info.daemon_sleep = timeout; | |
1994 | if (mddev->thread) { | |
1995 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
1996 | * the bitmap is all clean and we don't need to | |
1997 | * adjust the timeout right now | |
1998 | */ | |
1999 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2000 | mddev->thread->timeout = timeout; | |
2001 | md_wakeup_thread(mddev->thread); | |
2002 | } | |
2003 | } | |
2004 | return len; | |
2005 | } | |
2006 | ||
2007 | static struct md_sysfs_entry bitmap_timeout = | |
2008 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2009 | ||
2010 | static ssize_t | |
fd01b88c | 2011 | backlog_show(struct mddev *mddev, char *page) |
43a70507 N |
2012 | { |
2013 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2014 | } | |
2015 | ||
2016 | static ssize_t | |
fd01b88c | 2017 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2018 | { |
2019 | unsigned long backlog; | |
2020 | int rv = strict_strtoul(buf, 10, &backlog); | |
2021 | if (rv) | |
2022 | return rv; | |
2023 | if (backlog > COUNTER_MAX) | |
2024 | return -EINVAL; | |
2025 | mddev->bitmap_info.max_write_behind = backlog; | |
2026 | return len; | |
2027 | } | |
2028 | ||
2029 | static struct md_sysfs_entry bitmap_backlog = | |
2030 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2031 | ||
2032 | static ssize_t | |
fd01b88c | 2033 | chunksize_show(struct mddev *mddev, char *page) |
43a70507 N |
2034 | { |
2035 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2036 | } | |
2037 | ||
2038 | static ssize_t | |
fd01b88c | 2039 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2040 | { |
2041 | /* Can only be changed when no bitmap is active */ | |
2042 | int rv; | |
2043 | unsigned long csize; | |
2044 | if (mddev->bitmap) | |
2045 | return -EBUSY; | |
2046 | rv = strict_strtoul(buf, 10, &csize); | |
2047 | if (rv) | |
2048 | return rv; | |
2049 | if (csize < 512 || | |
2050 | !is_power_of_2(csize)) | |
2051 | return -EINVAL; | |
2052 | mddev->bitmap_info.chunksize = csize; | |
2053 | return len; | |
2054 | } | |
2055 | ||
2056 | static struct md_sysfs_entry bitmap_chunksize = | |
2057 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2058 | ||
fd01b88c | 2059 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2060 | { |
2061 | return sprintf(page, "%s\n", (mddev->bitmap_info.external | |
2062 | ? "external" : "internal")); | |
2063 | } | |
2064 | ||
fd01b88c | 2065 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2066 | { |
2067 | if (mddev->bitmap || | |
2068 | mddev->bitmap_info.file || | |
2069 | mddev->bitmap_info.offset) | |
2070 | return -EBUSY; | |
2071 | if (strncmp(buf, "external", 8) == 0) | |
2072 | mddev->bitmap_info.external = 1; | |
2073 | else if (strncmp(buf, "internal", 8) == 0) | |
2074 | mddev->bitmap_info.external = 0; | |
2075 | else | |
2076 | return -EINVAL; | |
2077 | return len; | |
2078 | } | |
2079 | ||
2080 | static struct md_sysfs_entry bitmap_metadata = | |
2081 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2082 | ||
fd01b88c | 2083 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2084 | { |
2085 | int len; | |
2086 | if (mddev->bitmap) | |
2087 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2088 | "false" : "true")); | |
2089 | else | |
2090 | len = sprintf(page, "\n"); | |
2091 | return len; | |
2092 | } | |
2093 | ||
fd01b88c | 2094 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2095 | { |
2096 | if (mddev->bitmap == NULL) | |
2097 | return -ENOENT; | |
2098 | if (strncmp(buf, "false", 5) == 0) | |
2099 | mddev->bitmap->need_sync = 1; | |
2100 | else if (strncmp(buf, "true", 4) == 0) { | |
2101 | if (mddev->degraded) | |
2102 | return -EBUSY; | |
2103 | mddev->bitmap->need_sync = 0; | |
2104 | } else | |
2105 | return -EINVAL; | |
2106 | return len; | |
2107 | } | |
2108 | ||
2109 | static struct md_sysfs_entry bitmap_can_clear = | |
2110 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2111 | ||
696fcd53 | 2112 | static ssize_t |
fd01b88c | 2113 | behind_writes_used_show(struct mddev *mddev, char *page) |
696fcd53 PC |
2114 | { |
2115 | if (mddev->bitmap == NULL) | |
2116 | return sprintf(page, "0\n"); | |
2117 | return sprintf(page, "%lu\n", | |
2118 | mddev->bitmap->behind_writes_used); | |
2119 | } | |
2120 | ||
2121 | static ssize_t | |
fd01b88c | 2122 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
696fcd53 PC |
2123 | { |
2124 | if (mddev->bitmap) | |
2125 | mddev->bitmap->behind_writes_used = 0; | |
2126 | return len; | |
2127 | } | |
2128 | ||
2129 | static struct md_sysfs_entry max_backlog_used = | |
2130 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2131 | behind_writes_used_show, behind_writes_used_reset); | |
2132 | ||
43a70507 N |
2133 | static struct attribute *md_bitmap_attrs[] = { |
2134 | &bitmap_location.attr, | |
2135 | &bitmap_timeout.attr, | |
2136 | &bitmap_backlog.attr, | |
2137 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2138 | &bitmap_metadata.attr, |
2139 | &bitmap_can_clear.attr, | |
696fcd53 | 2140 | &max_backlog_used.attr, |
43a70507 N |
2141 | NULL |
2142 | }; | |
2143 | struct attribute_group md_bitmap_group = { | |
2144 | .name = "bitmap", | |
2145 | .attrs = md_bitmap_attrs, | |
2146 | }; | |
2147 |