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> | |
43b2e5d8 | 29 | #include "md.h" |
ef740c37 | 30 | #include "bitmap.h" |
32a7627c | 31 | |
e384e585 | 32 | #include <linux/dm-dirty-log.h> |
32a7627c N |
33 | /* debug macros */ |
34 | ||
35 | #define DEBUG 0 | |
36 | ||
37 | #if DEBUG | |
38 | /* these are for debugging purposes only! */ | |
39 | ||
40 | /* define one and only one of these */ | |
41 | #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */ | |
42 | #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/ | |
43 | #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */ | |
44 | #define INJECT_FAULTS_4 0 /* undef */ | |
45 | #define INJECT_FAULTS_5 0 /* undef */ | |
46 | #define INJECT_FAULTS_6 0 | |
47 | ||
48 | /* if these are defined, the driver will fail! debug only */ | |
49 | #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */ | |
50 | #define INJECT_FATAL_FAULT_2 0 /* undef */ | |
51 | #define INJECT_FATAL_FAULT_3 0 /* undef */ | |
52 | #endif | |
53 | ||
32a7627c N |
54 | #ifndef PRINTK |
55 | # if DEBUG > 0 | |
56 | # define PRINTK(x...) printk(KERN_DEBUG x) | |
57 | # else | |
58 | # define PRINTK(x...) | |
59 | # endif | |
60 | #endif | |
61 | ||
ac2f40be | 62 | static inline char *bmname(struct bitmap *bitmap) |
32a7627c N |
63 | { |
64 | return bitmap->mddev ? mdname(bitmap->mddev) : "mdX"; | |
65 | } | |
66 | ||
32a7627c N |
67 | /* |
68 | * just a placeholder - calls kmalloc for bitmap pages | |
69 | */ | |
70 | static unsigned char *bitmap_alloc_page(struct bitmap *bitmap) | |
71 | { | |
72 | unsigned char *page; | |
73 | ||
44456d37 | 74 | #ifdef INJECT_FAULTS_1 |
32a7627c N |
75 | page = NULL; |
76 | #else | |
ac2f40be | 77 | page = kzalloc(PAGE_SIZE, GFP_NOIO); |
32a7627c N |
78 | #endif |
79 | if (!page) | |
80 | printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap)); | |
81 | else | |
a654b9d8 | 82 | PRINTK("%s: bitmap_alloc_page: allocated page at %p\n", |
32a7627c N |
83 | bmname(bitmap), page); |
84 | return page; | |
85 | } | |
86 | ||
87 | /* | |
88 | * for now just a placeholder -- just calls kfree for bitmap pages | |
89 | */ | |
90 | static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page) | |
91 | { | |
92 | PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page); | |
93 | kfree(page); | |
94 | } | |
95 | ||
96 | /* | |
97 | * check a page and, if necessary, allocate it (or hijack it if the alloc fails) | |
98 | * | |
99 | * 1) check to see if this page is allocated, if it's not then try to alloc | |
100 | * 2) if the alloc fails, set the page's hijacked flag so we'll use the | |
101 | * page pointer directly as a counter | |
102 | * | |
103 | * if we find our page, we increment the page's refcount so that it stays | |
104 | * allocated while we're using it | |
105 | */ | |
ac2f40be N |
106 | static int bitmap_checkpage(struct bitmap *bitmap, |
107 | unsigned long page, int create) | |
ee305ace N |
108 | __releases(bitmap->lock) |
109 | __acquires(bitmap->lock) | |
32a7627c N |
110 | { |
111 | unsigned char *mappage; | |
112 | ||
113 | if (page >= bitmap->pages) { | |
1187cf0a N |
114 | /* This can happen if bitmap_start_sync goes beyond |
115 | * End-of-device while looking for a whole page. | |
116 | * It is harmless. | |
117 | */ | |
32a7627c N |
118 | return -EINVAL; |
119 | } | |
120 | ||
32a7627c N |
121 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ |
122 | return 0; | |
123 | ||
124 | if (bitmap->bp[page].map) /* page is already allocated, just return */ | |
125 | return 0; | |
126 | ||
127 | if (!create) | |
128 | return -ENOENT; | |
129 | ||
32a7627c N |
130 | /* this page has not been allocated yet */ |
131 | ||
ac2f40be N |
132 | spin_unlock_irq(&bitmap->lock); |
133 | mappage = bitmap_alloc_page(bitmap); | |
134 | spin_lock_irq(&bitmap->lock); | |
135 | ||
136 | if (mappage == NULL) { | |
32a7627c N |
137 | PRINTK("%s: bitmap map page allocation failed, hijacking\n", |
138 | bmname(bitmap)); | |
139 | /* failed - set the hijacked flag so that we can use the | |
140 | * pointer as a counter */ | |
32a7627c N |
141 | if (!bitmap->bp[page].map) |
142 | bitmap->bp[page].hijacked = 1; | |
ac2f40be N |
143 | } else if (bitmap->bp[page].map || |
144 | bitmap->bp[page].hijacked) { | |
32a7627c N |
145 | /* somebody beat us to getting the page */ |
146 | bitmap_free_page(bitmap, mappage); | |
147 | return 0; | |
ac2f40be | 148 | } else { |
32a7627c | 149 | |
ac2f40be | 150 | /* no page was in place and we have one, so install it */ |
32a7627c | 151 | |
ac2f40be N |
152 | bitmap->bp[page].map = mappage; |
153 | bitmap->missing_pages--; | |
154 | } | |
32a7627c N |
155 | return 0; |
156 | } | |
157 | ||
32a7627c N |
158 | /* if page is completely empty, put it back on the free list, or dealloc it */ |
159 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
160 | /* Note: lock should be held when calling this */ | |
858119e1 | 161 | static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page) |
32a7627c N |
162 | { |
163 | char *ptr; | |
164 | ||
165 | if (bitmap->bp[page].count) /* page is still busy */ | |
166 | return; | |
167 | ||
168 | /* page is no longer in use, it can be released */ | |
169 | ||
170 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
171 | bitmap->bp[page].hijacked = 0; | |
172 | bitmap->bp[page].map = NULL; | |
ac2f40be N |
173 | } else { |
174 | /* normal case, free the page */ | |
175 | ptr = bitmap->bp[page].map; | |
176 | bitmap->bp[page].map = NULL; | |
177 | bitmap->missing_pages++; | |
178 | bitmap_free_page(bitmap, ptr); | |
32a7627c | 179 | } |
32a7627c N |
180 | } |
181 | ||
32a7627c N |
182 | /* |
183 | * bitmap file handling - read and write the bitmap file and its superblock | |
184 | */ | |
185 | ||
32a7627c N |
186 | /* |
187 | * basic page I/O operations | |
188 | */ | |
189 | ||
a654b9d8 | 190 | /* IO operations when bitmap is stored near all superblocks */ |
f6af949c | 191 | static struct page *read_sb_page(mddev_t *mddev, loff_t offset, |
a2ed9615 N |
192 | struct page *page, |
193 | unsigned long index, int size) | |
a654b9d8 N |
194 | { |
195 | /* choose a good rdev and read the page from there */ | |
196 | ||
197 | mdk_rdev_t *rdev; | |
a654b9d8 | 198 | sector_t target; |
ac2f40be | 199 | int did_alloc = 0; |
a654b9d8 | 200 | |
ac2f40be | 201 | if (!page) { |
a2ed9615 | 202 | page = alloc_page(GFP_KERNEL); |
ac2f40be N |
203 | if (!page) |
204 | return ERR_PTR(-ENOMEM); | |
205 | did_alloc = 1; | |
206 | } | |
a654b9d8 | 207 | |
159ec1fc | 208 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 N |
209 | if (! test_bit(In_sync, &rdev->flags) |
210 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
211 | continue; |
212 | ||
ccebd4c4 | 213 | target = offset + index * (PAGE_SIZE/512); |
a654b9d8 | 214 | |
2b193363 | 215 | if (sync_page_io(rdev, target, |
e1defc4f | 216 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
ccebd4c4 | 217 | page, READ, true)) { |
ab904d63 | 218 | page->index = index; |
ce25c31b N |
219 | attach_page_buffers(page, NULL); /* so that free_buffer will |
220 | * quietly no-op */ | |
ab904d63 N |
221 | return page; |
222 | } | |
223 | } | |
ac2f40be N |
224 | if (did_alloc) |
225 | put_page(page); | |
ab904d63 | 226 | return ERR_PTR(-EIO); |
a654b9d8 | 227 | |
a654b9d8 N |
228 | } |
229 | ||
b2d2c4ce N |
230 | static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
231 | { | |
232 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
233 | * linked list, and raising the refcount of devices we return to ensure | |
234 | * they don't disappear while in use. | |
235 | * As devices are only added or removed when raid_disk is < 0 and | |
236 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
237 | * still be in the same position on the list when we re-enter | |
238 | * list_for_each_continue_rcu. | |
239 | */ | |
240 | struct list_head *pos; | |
241 | rcu_read_lock(); | |
242 | if (rdev == NULL) | |
243 | /* start at the beginning */ | |
244 | pos = &mddev->disks; | |
245 | else { | |
246 | /* release the previous rdev and start from there. */ | |
247 | rdev_dec_pending(rdev, mddev); | |
248 | pos = &rdev->same_set; | |
249 | } | |
250 | list_for_each_continue_rcu(pos, &mddev->disks) { | |
251 | rdev = list_entry(pos, mdk_rdev_t, same_set); | |
252 | if (rdev->raid_disk >= 0 && | |
b2d2c4ce N |
253 | !test_bit(Faulty, &rdev->flags)) { |
254 | /* this is a usable devices */ | |
255 | atomic_inc(&rdev->nr_pending); | |
256 | rcu_read_unlock(); | |
257 | return rdev; | |
258 | } | |
259 | } | |
260 | rcu_read_unlock(); | |
261 | return NULL; | |
262 | } | |
263 | ||
ab6085c7 | 264 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 265 | { |
b2d2c4ce | 266 | mdk_rdev_t *rdev = NULL; |
a6ff7e08 | 267 | struct block_device *bdev; |
ab6085c7 | 268 | mddev_t *mddev = bitmap->mddev; |
a654b9d8 | 269 | |
b2d2c4ce | 270 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ac2f40be N |
271 | int size = PAGE_SIZE; |
272 | loff_t offset = mddev->bitmap_info.offset; | |
a6ff7e08 JB |
273 | |
274 | bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; | |
275 | ||
ac2f40be N |
276 | if (page->index == bitmap->file_pages-1) |
277 | size = roundup(bitmap->last_page_size, | |
a6ff7e08 | 278 | bdev_logical_block_size(bdev)); |
ac2f40be N |
279 | /* Just make sure we aren't corrupting data or |
280 | * metadata | |
281 | */ | |
282 | if (mddev->external) { | |
283 | /* Bitmap could be anywhere. */ | |
284 | if (rdev->sb_start + offset + (page->index | |
285 | * (PAGE_SIZE/512)) | |
286 | > rdev->data_offset | |
287 | && | |
288 | rdev->sb_start + offset | |
289 | < (rdev->data_offset + mddev->dev_sectors | |
290 | + (PAGE_SIZE/512))) | |
291 | goto bad_alignment; | |
292 | } else if (offset < 0) { | |
293 | /* DATA BITMAP METADATA */ | |
294 | if (offset | |
295 | + (long)(page->index * (PAGE_SIZE/512)) | |
296 | + size/512 > 0) | |
297 | /* bitmap runs in to metadata */ | |
298 | goto bad_alignment; | |
299 | if (rdev->data_offset + mddev->dev_sectors | |
300 | > rdev->sb_start + offset) | |
301 | /* data runs in to bitmap */ | |
302 | goto bad_alignment; | |
303 | } else if (rdev->sb_start < rdev->data_offset) { | |
304 | /* METADATA BITMAP DATA */ | |
305 | if (rdev->sb_start | |
306 | + offset | |
307 | + page->index*(PAGE_SIZE/512) + size/512 | |
308 | > rdev->data_offset) | |
309 | /* bitmap runs in to data */ | |
310 | goto bad_alignment; | |
311 | } else { | |
312 | /* DATA METADATA BITMAP - no problems */ | |
313 | } | |
314 | md_super_write(mddev, rdev, | |
315 | rdev->sb_start + offset | |
316 | + page->index * (PAGE_SIZE/512), | |
317 | size, | |
318 | page); | |
b2d2c4ce | 319 | } |
a654b9d8 N |
320 | |
321 | if (wait) | |
a9701a30 | 322 | md_super_wait(mddev); |
a654b9d8 | 323 | return 0; |
4b80991c N |
324 | |
325 | bad_alignment: | |
4b80991c | 326 | return -EINVAL; |
a654b9d8 N |
327 | } |
328 | ||
4ad13663 | 329 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 330 | /* |
a654b9d8 | 331 | * write out a page to a file |
32a7627c | 332 | */ |
4ad13663 | 333 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 334 | { |
d785a06a | 335 | struct buffer_head *bh; |
32a7627c | 336 | |
f0d76d70 N |
337 | if (bitmap->file == NULL) { |
338 | switch (write_sb_page(bitmap, page, wait)) { | |
339 | case -EINVAL: | |
340 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
f0d76d70 | 341 | } |
4ad13663 | 342 | } else { |
a654b9d8 | 343 | |
4ad13663 | 344 | bh = page_buffers(page); |
c708443c | 345 | |
4ad13663 N |
346 | while (bh && bh->b_blocknr) { |
347 | atomic_inc(&bitmap->pending_writes); | |
348 | set_buffer_locked(bh); | |
349 | set_buffer_mapped(bh); | |
721a9602 | 350 | submit_bh(WRITE | REQ_SYNC, bh); |
4ad13663 N |
351 | bh = bh->b_this_page; |
352 | } | |
d785a06a | 353 | |
ac2f40be | 354 | if (wait) |
4ad13663 N |
355 | wait_event(bitmap->write_wait, |
356 | atomic_read(&bitmap->pending_writes)==0); | |
8a5e9cf1 | 357 | } |
4ad13663 N |
358 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
359 | bitmap_file_kick(bitmap); | |
d785a06a N |
360 | } |
361 | ||
362 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
363 | { | |
364 | struct bitmap *bitmap = bh->b_private; | |
365 | unsigned long flags; | |
32a7627c | 366 | |
d785a06a N |
367 | if (!uptodate) { |
368 | spin_lock_irqsave(&bitmap->lock, flags); | |
369 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
370 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
32a7627c | 371 | } |
d785a06a N |
372 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
373 | wake_up(&bitmap->write_wait); | |
374 | } | |
32a7627c | 375 | |
d785a06a N |
376 | /* copied from buffer.c */ |
377 | static void | |
378 | __clear_page_buffers(struct page *page) | |
379 | { | |
380 | ClearPagePrivate(page); | |
381 | set_page_private(page, 0); | |
382 | page_cache_release(page); | |
383 | } | |
384 | static void free_buffers(struct page *page) | |
385 | { | |
386 | struct buffer_head *bh = page_buffers(page); | |
77ad4bc7 | 387 | |
d785a06a N |
388 | while (bh) { |
389 | struct buffer_head *next = bh->b_this_page; | |
390 | free_buffer_head(bh); | |
391 | bh = next; | |
77ad4bc7 | 392 | } |
d785a06a N |
393 | __clear_page_buffers(page); |
394 | put_page(page); | |
32a7627c N |
395 | } |
396 | ||
d785a06a N |
397 | /* read a page from a file. |
398 | * We both read the page, and attach buffers to the page to record the | |
399 | * address of each block (using bmap). These addresses will be used | |
400 | * to write the block later, completely bypassing the filesystem. | |
401 | * This usage is similar to how swap files are handled, and allows us | |
402 | * to write to a file with no concerns of memory allocation failing. | |
403 | */ | |
32a7627c | 404 | static struct page *read_page(struct file *file, unsigned long index, |
d785a06a N |
405 | struct bitmap *bitmap, |
406 | unsigned long count) | |
32a7627c | 407 | { |
32a7627c | 408 | struct page *page = NULL; |
c649bb9c | 409 | struct inode *inode = file->f_path.dentry->d_inode; |
d785a06a N |
410 | struct buffer_head *bh; |
411 | sector_t block; | |
32a7627c | 412 | |
ac2f40be | 413 | PRINTK("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
2d1f3b5d | 414 | (unsigned long long)index << PAGE_SHIFT); |
32a7627c | 415 | |
d785a06a N |
416 | page = alloc_page(GFP_KERNEL); |
417 | if (!page) | |
418 | page = ERR_PTR(-ENOMEM); | |
32a7627c N |
419 | if (IS_ERR(page)) |
420 | goto out; | |
d785a06a | 421 | |
d785a06a N |
422 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
423 | if (!bh) { | |
2d1f3b5d | 424 | put_page(page); |
d785a06a | 425 | page = ERR_PTR(-ENOMEM); |
32a7627c N |
426 | goto out; |
427 | } | |
d785a06a N |
428 | attach_page_buffers(page, bh); |
429 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
430 | while (bh) { | |
431 | if (count == 0) | |
432 | bh->b_blocknr = 0; | |
433 | else { | |
434 | bh->b_blocknr = bmap(inode, block); | |
435 | if (bh->b_blocknr == 0) { | |
436 | /* Cannot use this file! */ | |
437 | free_buffers(page); | |
438 | page = ERR_PTR(-EINVAL); | |
439 | goto out; | |
440 | } | |
441 | bh->b_bdev = inode->i_sb->s_bdev; | |
442 | if (count < (1<<inode->i_blkbits)) | |
443 | count = 0; | |
444 | else | |
445 | count -= (1<<inode->i_blkbits); | |
446 | ||
447 | bh->b_end_io = end_bitmap_write; | |
448 | bh->b_private = bitmap; | |
ce25c31b N |
449 | atomic_inc(&bitmap->pending_writes); |
450 | set_buffer_locked(bh); | |
451 | set_buffer_mapped(bh); | |
452 | submit_bh(READ, bh); | |
d785a06a N |
453 | } |
454 | block++; | |
455 | bh = bh->b_this_page; | |
456 | } | |
d785a06a | 457 | page->index = index; |
ce25c31b N |
458 | |
459 | wait_event(bitmap->write_wait, | |
460 | atomic_read(&bitmap->pending_writes)==0); | |
461 | if (bitmap->flags & BITMAP_WRITE_ERROR) { | |
462 | free_buffers(page); | |
463 | page = ERR_PTR(-EIO); | |
464 | } | |
32a7627c N |
465 | out: |
466 | if (IS_ERR(page)) | |
ac2f40be | 467 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %ld\n", |
2d1f3b5d N |
468 | (int)PAGE_SIZE, |
469 | (unsigned long long)index << PAGE_SHIFT, | |
32a7627c N |
470 | PTR_ERR(page)); |
471 | return page; | |
472 | } | |
473 | ||
474 | /* | |
475 | * bitmap file superblock operations | |
476 | */ | |
477 | ||
478 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 479 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
480 | { |
481 | bitmap_super_t *sb; | |
482 | unsigned long flags; | |
483 | ||
484 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 485 | return; |
ece5cff0 N |
486 | if (bitmap->mddev->bitmap_info.external) |
487 | return; | |
32a7627c N |
488 | spin_lock_irqsave(&bitmap->lock, flags); |
489 | if (!bitmap->sb_page) { /* no superblock */ | |
490 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 491 | return; |
32a7627c | 492 | } |
32a7627c | 493 | spin_unlock_irqrestore(&bitmap->lock, flags); |
7b92813c | 494 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 495 | sb->events = cpu_to_le64(bitmap->mddev->events); |
8258c532 | 496 | if (bitmap->mddev->events < bitmap->events_cleared) |
a0da84f3 NB |
497 | /* rocking back to read-only */ |
498 | bitmap->events_cleared = bitmap->mddev->events; | |
8258c532 N |
499 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); |
500 | sb->state = cpu_to_le32(bitmap->flags); | |
43a70507 N |
501 | /* Just in case these have been changed via sysfs: */ |
502 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
503 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
ea03aff9 | 504 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 505 | write_page(bitmap, bitmap->sb_page, 1); |
32a7627c N |
506 | } |
507 | ||
508 | /* print out the bitmap file superblock */ | |
509 | void bitmap_print_sb(struct bitmap *bitmap) | |
510 | { | |
511 | bitmap_super_t *sb; | |
512 | ||
513 | if (!bitmap || !bitmap->sb_page) | |
514 | return; | |
7b92813c | 515 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 516 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
517 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
518 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
519 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
520 | *(__u32 *)(sb->uuid+0), |
521 | *(__u32 *)(sb->uuid+4), | |
522 | *(__u32 *)(sb->uuid+8), | |
523 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 524 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 525 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 526 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 527 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
528 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
529 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
530 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
531 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
532 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 533 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
ea03aff9 | 534 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
535 | } |
536 | ||
9c81075f JB |
537 | /* |
538 | * bitmap_new_disk_sb | |
539 | * @bitmap | |
540 | * | |
541 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb | |
542 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. | |
543 | * This function verifies 'bitmap_info' and populates the on-disk bitmap | |
544 | * structure, which is to be written to disk. | |
545 | * | |
546 | * Returns: 0 on success, -Exxx on error | |
547 | */ | |
548 | static int bitmap_new_disk_sb(struct bitmap *bitmap) | |
549 | { | |
550 | bitmap_super_t *sb; | |
551 | unsigned long chunksize, daemon_sleep, write_behind; | |
552 | int err = -EINVAL; | |
553 | ||
554 | bitmap->sb_page = alloc_page(GFP_KERNEL); | |
555 | if (IS_ERR(bitmap->sb_page)) { | |
556 | err = PTR_ERR(bitmap->sb_page); | |
557 | bitmap->sb_page = NULL; | |
558 | return err; | |
559 | } | |
560 | bitmap->sb_page->index = 0; | |
561 | ||
562 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); | |
563 | ||
564 | sb->magic = cpu_to_le32(BITMAP_MAGIC); | |
565 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); | |
566 | ||
567 | chunksize = bitmap->mddev->bitmap_info.chunksize; | |
568 | BUG_ON(!chunksize); | |
569 | if (!is_power_of_2(chunksize)) { | |
570 | kunmap_atomic(sb, KM_USER0); | |
571 | printk(KERN_ERR "bitmap chunksize not a power of 2\n"); | |
572 | return -EINVAL; | |
573 | } | |
574 | sb->chunksize = cpu_to_le32(chunksize); | |
575 | ||
576 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; | |
577 | if (!daemon_sleep || | |
578 | (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { | |
579 | printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n"); | |
580 | daemon_sleep = 5 * HZ; | |
581 | } | |
582 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); | |
583 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
584 | ||
585 | /* | |
586 | * FIXME: write_behind for RAID1. If not specified, what | |
587 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. | |
588 | */ | |
589 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; | |
590 | if (write_behind > COUNTER_MAX) | |
591 | write_behind = COUNTER_MAX / 2; | |
592 | sb->write_behind = cpu_to_le32(write_behind); | |
593 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
594 | ||
595 | /* keep the array size field of the bitmap superblock up to date */ | |
596 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
597 | ||
598 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); | |
599 | ||
600 | bitmap->flags |= BITMAP_STALE; | |
601 | sb->state |= cpu_to_le32(BITMAP_STALE); | |
602 | bitmap->events_cleared = bitmap->mddev->events; | |
603 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
604 | ||
605 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
606 | sb->version = cpu_to_le32(BITMAP_MAJOR_HOSTENDIAN); | |
607 | ||
608 | kunmap_atomic(sb, KM_USER0); | |
609 | ||
610 | return 0; | |
611 | } | |
612 | ||
32a7627c N |
613 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
614 | static int bitmap_read_sb(struct bitmap *bitmap) | |
615 | { | |
616 | char *reason = NULL; | |
617 | bitmap_super_t *sb; | |
4b6d287f | 618 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c N |
619 | unsigned long long events; |
620 | int err = -EINVAL; | |
621 | ||
622 | /* page 0 is the superblock, read it... */ | |
f49d5e62 N |
623 | if (bitmap->file) { |
624 | loff_t isize = i_size_read(bitmap->file->f_mapping->host); | |
625 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; | |
626 | ||
627 | bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes); | |
628 | } else { | |
42a04b50 N |
629 | bitmap->sb_page = read_sb_page(bitmap->mddev, |
630 | bitmap->mddev->bitmap_info.offset, | |
a2ed9615 N |
631 | NULL, |
632 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 633 | } |
32a7627c N |
634 | if (IS_ERR(bitmap->sb_page)) { |
635 | err = PTR_ERR(bitmap->sb_page); | |
636 | bitmap->sb_page = NULL; | |
637 | return err; | |
638 | } | |
639 | ||
7b92813c | 640 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 641 | |
32a7627c | 642 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 643 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 644 | write_behind = le32_to_cpu(sb->write_behind); |
32a7627c N |
645 | |
646 | /* verify that the bitmap-specific fields are valid */ | |
647 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
648 | reason = "bad magic"; | |
bd926c63 N |
649 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
650 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 651 | reason = "unrecognized superblock version"; |
1187cf0a | 652 | else if (chunksize < 512) |
7dd5d34c | 653 | reason = "bitmap chunksize too small"; |
d744540c | 654 | else if (!is_power_of_2(chunksize)) |
32a7627c | 655 | reason = "bitmap chunksize not a power of 2"; |
1b04be96 | 656 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 657 | reason = "daemon sleep period out of range"; |
4b6d287f N |
658 | else if (write_behind > COUNTER_MAX) |
659 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
660 | if (reason) { |
661 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
662 | bmname(bitmap), reason); | |
663 | goto out; | |
664 | } | |
665 | ||
666 | /* keep the array size field of the bitmap superblock up to date */ | |
667 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
668 | ||
669 | if (!bitmap->mddev->persistent) | |
670 | goto success; | |
671 | ||
672 | /* | |
673 | * if we have a persistent array superblock, compare the | |
674 | * bitmap's UUID and event counter to the mddev's | |
675 | */ | |
676 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
677 | printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n", | |
678 | bmname(bitmap)); | |
679 | goto out; | |
680 | } | |
681 | events = le64_to_cpu(sb->events); | |
682 | if (events < bitmap->mddev->events) { | |
683 | printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) " | |
684 | "-- forcing full recovery\n", bmname(bitmap), events, | |
685 | (unsigned long long) bitmap->mddev->events); | |
4f2e639a | 686 | sb->state |= cpu_to_le32(BITMAP_STALE); |
32a7627c N |
687 | } |
688 | success: | |
689 | /* assign fields using values from superblock */ | |
42a04b50 N |
690 | bitmap->mddev->bitmap_info.chunksize = chunksize; |
691 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
42a04b50 | 692 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; |
4f2e639a | 693 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 N |
694 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
695 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
32a7627c | 696 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
8258c532 | 697 | if (bitmap->flags & BITMAP_STALE) |
6a07997f | 698 | bitmap->events_cleared = bitmap->mddev->events; |
32a7627c N |
699 | err = 0; |
700 | out: | |
ea03aff9 | 701 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
702 | if (err) |
703 | bitmap_print_sb(bitmap); | |
704 | return err; | |
705 | } | |
706 | ||
707 | enum bitmap_mask_op { | |
708 | MASK_SET, | |
709 | MASK_UNSET | |
710 | }; | |
711 | ||
4ad13663 N |
712 | /* record the state of the bitmap in the superblock. Return the old value */ |
713 | static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits, | |
714 | enum bitmap_mask_op op) | |
32a7627c N |
715 | { |
716 | bitmap_super_t *sb; | |
717 | unsigned long flags; | |
4ad13663 | 718 | int old; |
32a7627c N |
719 | |
720 | spin_lock_irqsave(&bitmap->lock, flags); | |
7e317655 | 721 | if (!bitmap->sb_page) { /* can't set the state */ |
32a7627c | 722 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 723 | return 0; |
32a7627c | 724 | } |
32a7627c | 725 | spin_unlock_irqrestore(&bitmap->lock, flags); |
7b92813c | 726 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
4ad13663 | 727 | old = le32_to_cpu(sb->state) & bits; |
32a7627c | 728 | switch (op) { |
ac2f40be N |
729 | case MASK_SET: |
730 | sb->state |= cpu_to_le32(bits); | |
8258c532 | 731 | bitmap->flags |= bits; |
ac2f40be N |
732 | break; |
733 | case MASK_UNSET: | |
734 | sb->state &= cpu_to_le32(~bits); | |
8258c532 | 735 | bitmap->flags &= ~bits; |
ac2f40be N |
736 | break; |
737 | default: | |
738 | BUG(); | |
32a7627c | 739 | } |
ea03aff9 | 740 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 741 | return old; |
32a7627c N |
742 | } |
743 | ||
744 | /* | |
745 | * general bitmap file operations | |
746 | */ | |
747 | ||
ece5cff0 N |
748 | /* |
749 | * on-disk bitmap: | |
750 | * | |
751 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
752 | * file a page at a time. There's a superblock at the start of the file. | |
753 | */ | |
32a7627c | 754 | /* calculate the index of the page that contains this bit */ |
ece5cff0 | 755 | static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk) |
32a7627c | 756 | { |
ece5cff0 N |
757 | if (!bitmap->mddev->bitmap_info.external) |
758 | chunk += sizeof(bitmap_super_t) << 3; | |
759 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
760 | } |
761 | ||
762 | /* calculate the (bit) offset of this bit within a page */ | |
ece5cff0 | 763 | static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk) |
32a7627c | 764 | { |
ece5cff0 N |
765 | if (!bitmap->mddev->bitmap_info.external) |
766 | chunk += sizeof(bitmap_super_t) << 3; | |
767 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
768 | } |
769 | ||
770 | /* | |
771 | * return a pointer to the page in the filemap that contains the given bit | |
772 | * | |
773 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
774 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
775 | * 0 or page 1 | |
776 | */ | |
777 | static inline struct page *filemap_get_page(struct bitmap *bitmap, | |
778 | unsigned long chunk) | |
779 | { | |
e384e585 N |
780 | if (bitmap->filemap == NULL) |
781 | return NULL; | |
ac2f40be N |
782 | if (file_page_index(bitmap, chunk) >= bitmap->file_pages) |
783 | return NULL; | |
ece5cff0 N |
784 | return bitmap->filemap[file_page_index(bitmap, chunk) |
785 | - file_page_index(bitmap, 0)]; | |
32a7627c N |
786 | } |
787 | ||
32a7627c N |
788 | static void bitmap_file_unmap(struct bitmap *bitmap) |
789 | { | |
790 | struct page **map, *sb_page; | |
791 | unsigned long *attr; | |
792 | int pages; | |
793 | unsigned long flags; | |
794 | ||
795 | spin_lock_irqsave(&bitmap->lock, flags); | |
796 | map = bitmap->filemap; | |
797 | bitmap->filemap = NULL; | |
798 | attr = bitmap->filemap_attr; | |
799 | bitmap->filemap_attr = NULL; | |
800 | pages = bitmap->file_pages; | |
801 | bitmap->file_pages = 0; | |
802 | sb_page = bitmap->sb_page; | |
803 | bitmap->sb_page = NULL; | |
804 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
805 | ||
806 | while (pages--) | |
ece5cff0 | 807 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 808 | free_buffers(map[pages]); |
32a7627c N |
809 | kfree(map); |
810 | kfree(attr); | |
811 | ||
d785a06a N |
812 | if (sb_page) |
813 | free_buffers(sb_page); | |
32a7627c N |
814 | } |
815 | ||
816 | static void bitmap_file_put(struct bitmap *bitmap) | |
817 | { | |
818 | struct file *file; | |
32a7627c N |
819 | unsigned long flags; |
820 | ||
821 | spin_lock_irqsave(&bitmap->lock, flags); | |
822 | file = bitmap->file; | |
823 | bitmap->file = NULL; | |
824 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
825 | ||
d785a06a N |
826 | if (file) |
827 | wait_event(bitmap->write_wait, | |
828 | atomic_read(&bitmap->pending_writes)==0); | |
32a7627c N |
829 | bitmap_file_unmap(bitmap); |
830 | ||
d785a06a | 831 | if (file) { |
c649bb9c | 832 | struct inode *inode = file->f_path.dentry->d_inode; |
fc0ecff6 | 833 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 834 | fput(file); |
d785a06a | 835 | } |
32a7627c N |
836 | } |
837 | ||
32a7627c N |
838 | /* |
839 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
840 | * then it is no longer reliable, so we stop using it and we mark the file | |
841 | * as failed in the superblock | |
842 | */ | |
843 | static void bitmap_file_kick(struct bitmap *bitmap) | |
844 | { | |
845 | char *path, *ptr = NULL; | |
846 | ||
4ad13663 N |
847 | if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) { |
848 | bitmap_update_sb(bitmap); | |
32a7627c | 849 | |
4ad13663 N |
850 | if (bitmap->file) { |
851 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
852 | if (path) | |
6bcfd601 CH |
853 | ptr = d_path(&bitmap->file->f_path, path, |
854 | PAGE_SIZE); | |
855 | ||
4ad13663 N |
856 | printk(KERN_ALERT |
857 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 858 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 859 | |
4ad13663 N |
860 | kfree(path); |
861 | } else | |
862 | printk(KERN_ALERT | |
863 | "%s: disabling internal bitmap due to errors\n", | |
864 | bmname(bitmap)); | |
a654b9d8 | 865 | } |
32a7627c N |
866 | |
867 | bitmap_file_put(bitmap); | |
868 | ||
869 | return; | |
870 | } | |
871 | ||
872 | enum bitmap_page_attr { | |
ac2f40be N |
873 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
874 | BITMAP_PAGE_CLEAN = 1, /* there are bits that might need to be cleared */ | |
875 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ | |
32a7627c N |
876 | }; |
877 | ||
878 | static inline void set_page_attr(struct bitmap *bitmap, struct page *page, | |
879 | enum bitmap_page_attr attr) | |
880 | { | |
e384e585 N |
881 | if (page) |
882 | __set_bit((page->index<<2) + attr, bitmap->filemap_attr); | |
883 | else | |
884 | __set_bit(attr, &bitmap->logattrs); | |
32a7627c N |
885 | } |
886 | ||
887 | static inline void clear_page_attr(struct bitmap *bitmap, struct page *page, | |
888 | enum bitmap_page_attr attr) | |
889 | { | |
e384e585 N |
890 | if (page) |
891 | __clear_bit((page->index<<2) + attr, bitmap->filemap_attr); | |
892 | else | |
893 | __clear_bit(attr, &bitmap->logattrs); | |
32a7627c N |
894 | } |
895 | ||
ec7a3197 N |
896 | static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page, |
897 | enum bitmap_page_attr attr) | |
32a7627c | 898 | { |
e384e585 N |
899 | if (page) |
900 | return test_bit((page->index<<2) + attr, bitmap->filemap_attr); | |
901 | else | |
902 | return test_bit(attr, &bitmap->logattrs); | |
32a7627c N |
903 | } |
904 | ||
905 | /* | |
906 | * bitmap_file_set_bit -- called before performing a write to the md device | |
907 | * to set (and eventually sync) a particular bit in the bitmap file | |
908 | * | |
909 | * we set the bit immediately, then we record the page number so that | |
910 | * when an unplug occurs, we can flush the dirty pages out to disk | |
911 | */ | |
912 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
913 | { | |
914 | unsigned long bit; | |
e384e585 | 915 | struct page *page = NULL; |
32a7627c N |
916 | void *kaddr; |
917 | unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap); | |
918 | ||
e384e585 N |
919 | if (!bitmap->filemap) { |
920 | struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log; | |
921 | if (log) | |
922 | log->type->mark_region(log, chunk); | |
923 | } else { | |
32a7627c | 924 | |
e384e585 N |
925 | page = filemap_get_page(bitmap, chunk); |
926 | if (!page) | |
927 | return; | |
928 | bit = file_page_offset(bitmap, chunk); | |
32a7627c | 929 | |
e384e585 N |
930 | /* set the bit */ |
931 | kaddr = kmap_atomic(page, KM_USER0); | |
932 | if (bitmap->flags & BITMAP_HOSTENDIAN) | |
933 | set_bit(bit, kaddr); | |
934 | else | |
6b33aff3 | 935 | __test_and_set_bit_le(bit, kaddr); |
e384e585 N |
936 | kunmap_atomic(kaddr, KM_USER0); |
937 | PRINTK("set file bit %lu page %lu\n", bit, page->index); | |
938 | } | |
32a7627c N |
939 | /* record page number so it gets flushed to disk when unplug occurs */ |
940 | set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); | |
32a7627c N |
941 | } |
942 | ||
943 | /* this gets called when the md device is ready to unplug its underlying | |
944 | * (slave) device queues -- before we let any writes go down, we need to | |
945 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 946 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 947 | { |
ec7a3197 N |
948 | unsigned long i, flags; |
949 | int dirty, need_write; | |
32a7627c N |
950 | struct page *page; |
951 | int wait = 0; | |
952 | ||
953 | if (!bitmap) | |
4ad13663 | 954 | return; |
e384e585 N |
955 | if (!bitmap->filemap) { |
956 | /* Must be using a dirty_log */ | |
957 | struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log; | |
958 | dirty = test_and_clear_bit(BITMAP_PAGE_DIRTY, &bitmap->logattrs); | |
959 | need_write = test_and_clear_bit(BITMAP_PAGE_NEEDWRITE, &bitmap->logattrs); | |
960 | if (dirty || need_write) | |
961 | if (log->type->flush(log)) | |
962 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
963 | goto out; | |
964 | } | |
32a7627c N |
965 | |
966 | /* look at each page to see if there are any set bits that need to be | |
967 | * flushed out to disk */ | |
968 | for (i = 0; i < bitmap->file_pages; i++) { | |
969 | spin_lock_irqsave(&bitmap->lock, flags); | |
a654b9d8 | 970 | if (!bitmap->filemap) { |
32a7627c | 971 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 972 | return; |
32a7627c N |
973 | } |
974 | page = bitmap->filemap[i]; | |
ec7a3197 N |
975 | dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
976 | need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
32a7627c N |
977 | clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
978 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
ec7a3197 | 979 | if (dirty) |
32a7627c N |
980 | wait = 1; |
981 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
982 | ||
ac2f40be | 983 | if (dirty || need_write) |
4ad13663 | 984 | write_page(bitmap, page, 0); |
32a7627c N |
985 | } |
986 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
0b79ccf0 | 987 | if (bitmap->file) |
d785a06a N |
988 | wait_event(bitmap->write_wait, |
989 | atomic_read(&bitmap->pending_writes)==0); | |
0b79ccf0 | 990 | else |
a9701a30 | 991 | md_super_wait(bitmap->mddev); |
32a7627c | 992 | } |
e384e585 | 993 | out: |
d785a06a N |
994 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
995 | bitmap_file_kick(bitmap); | |
32a7627c | 996 | } |
ac2f40be | 997 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 998 | |
6a07997f | 999 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
1000 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
1001 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
1002 | * memory mapping of the bitmap file | |
1003 | * Special cases: | |
1004 | * if there's no bitmap file, or if the bitmap file had been | |
1005 | * previously kicked from the array, we mark all the bits as | |
1006 | * 1's in order to cause a full resync. | |
6a07997f N |
1007 | * |
1008 | * We ignore all bits for sectors that end earlier than 'start'. | |
1009 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 1010 | */ |
6a07997f | 1011 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
1012 | { |
1013 | unsigned long i, chunks, index, oldindex, bit; | |
1014 | struct page *page = NULL, *oldpage = NULL; | |
1015 | unsigned long num_pages, bit_cnt = 0; | |
1016 | struct file *file; | |
d785a06a | 1017 | unsigned long bytes, offset; |
32a7627c N |
1018 | int outofdate; |
1019 | int ret = -ENOSPC; | |
ea03aff9 | 1020 | void *paddr; |
32a7627c N |
1021 | |
1022 | chunks = bitmap->chunks; | |
1023 | file = bitmap->file; | |
1024 | ||
42a04b50 | 1025 | BUG_ON(!file && !bitmap->mddev->bitmap_info.offset); |
32a7627c | 1026 | |
44456d37 | 1027 | #ifdef INJECT_FAULTS_3 |
32a7627c N |
1028 | outofdate = 1; |
1029 | #else | |
1030 | outofdate = bitmap->flags & BITMAP_STALE; | |
1031 | #endif | |
1032 | if (outofdate) | |
1033 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
1034 | "recovery\n", bmname(bitmap)); | |
1035 | ||
e384e585 | 1036 | bytes = DIV_ROUND_UP(bitmap->chunks, 8); |
ece5cff0 N |
1037 | if (!bitmap->mddev->bitmap_info.external) |
1038 | bytes += sizeof(bitmap_super_t); | |
bc7f77de | 1039 | |
e384e585 | 1040 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); |
bc7f77de | 1041 | |
ece5cff0 | 1042 | if (file && i_size_read(file->f_mapping->host) < bytes) { |
32a7627c N |
1043 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
1044 | bmname(bitmap), | |
1045 | (unsigned long) i_size_read(file->f_mapping->host), | |
ece5cff0 | 1046 | bytes); |
4ad13663 | 1047 | goto err; |
32a7627c | 1048 | } |
bc7f77de N |
1049 | |
1050 | ret = -ENOMEM; | |
1051 | ||
32a7627c | 1052 | bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL); |
bc7f77de | 1053 | if (!bitmap->filemap) |
4ad13663 | 1054 | goto err; |
32a7627c | 1055 | |
e16b68b6 N |
1056 | /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */ |
1057 | bitmap->filemap_attr = kzalloc( | |
ac2f40be | 1058 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), |
e16b68b6 | 1059 | GFP_KERNEL); |
bc7f77de | 1060 | if (!bitmap->filemap_attr) |
4ad13663 | 1061 | goto err; |
32a7627c | 1062 | |
32a7627c N |
1063 | oldindex = ~0L; |
1064 | ||
1065 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 1066 | int b; |
ece5cff0 N |
1067 | index = file_page_index(bitmap, i); |
1068 | bit = file_page_offset(bitmap, i); | |
32a7627c | 1069 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 1070 | int count; |
32a7627c | 1071 | /* unmap the old page, we're done with it */ |
d785a06a | 1072 | if (index == num_pages-1) |
ece5cff0 | 1073 | count = bytes - index * PAGE_SIZE; |
d785a06a N |
1074 | else |
1075 | count = PAGE_SIZE; | |
ece5cff0 | 1076 | if (index == 0 && bitmap->sb_page) { |
32a7627c N |
1077 | /* |
1078 | * if we're here then the superblock page | |
1079 | * contains some bits (PAGE_SIZE != sizeof sb) | |
1080 | * we've already read it in, so just use it | |
1081 | */ | |
1082 | page = bitmap->sb_page; | |
1083 | offset = sizeof(bitmap_super_t); | |
53845270 | 1084 | if (!file) |
5c04f551 VK |
1085 | page = read_sb_page( |
1086 | bitmap->mddev, | |
1087 | bitmap->mddev->bitmap_info.offset, | |
1088 | page, | |
1089 | index, count); | |
a654b9d8 | 1090 | } else if (file) { |
d785a06a | 1091 | page = read_page(file, index, bitmap, count); |
a654b9d8 N |
1092 | offset = 0; |
1093 | } else { | |
42a04b50 N |
1094 | page = read_sb_page(bitmap->mddev, |
1095 | bitmap->mddev->bitmap_info.offset, | |
a2ed9615 N |
1096 | NULL, |
1097 | index, count); | |
32a7627c N |
1098 | offset = 0; |
1099 | } | |
a654b9d8 N |
1100 | if (IS_ERR(page)) { /* read error */ |
1101 | ret = PTR_ERR(page); | |
4ad13663 | 1102 | goto err; |
a654b9d8 N |
1103 | } |
1104 | ||
32a7627c N |
1105 | oldindex = index; |
1106 | oldpage = page; | |
32a7627c | 1107 | |
b74fd282 N |
1108 | bitmap->filemap[bitmap->file_pages++] = page; |
1109 | bitmap->last_page_size = count; | |
1110 | ||
32a7627c N |
1111 | if (outofdate) { |
1112 | /* | |
1113 | * if bitmap is out of date, dirty the | |
ac2f40be | 1114 | * whole page and write it out |
32a7627c | 1115 | */ |
ea03aff9 N |
1116 | paddr = kmap_atomic(page, KM_USER0); |
1117 | memset(paddr + offset, 0xff, | |
6a07997f | 1118 | PAGE_SIZE - offset); |
ea03aff9 | 1119 | kunmap_atomic(paddr, KM_USER0); |
4ad13663 N |
1120 | write_page(bitmap, page, 1); |
1121 | ||
1122 | ret = -EIO; | |
b74fd282 | 1123 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
4ad13663 | 1124 | goto err; |
32a7627c | 1125 | } |
32a7627c | 1126 | } |
ea03aff9 | 1127 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1128 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1129 | b = test_bit(bit, paddr); |
bd926c63 | 1130 | else |
6b33aff3 | 1131 | b = test_bit_le(bit, paddr); |
ea03aff9 | 1132 | kunmap_atomic(paddr, KM_USER0); |
bd926c63 | 1133 | if (b) { |
32a7627c | 1134 | /* if the disk bit is set, set the memory bit */ |
db305e50 N |
1135 | int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) |
1136 | >= start); | |
1137 | bitmap_set_memory_bits(bitmap, | |
1138 | (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap), | |
1139 | needed); | |
32a7627c | 1140 | bit_cnt++; |
6a07997f | 1141 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); |
32a7627c | 1142 | } |
32a7627c N |
1143 | } |
1144 | ||
ac2f40be | 1145 | /* everything went OK */ |
32a7627c N |
1146 | ret = 0; |
1147 | bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET); | |
1148 | ||
32a7627c N |
1149 | if (bit_cnt) { /* Kick recovery if any bits were set */ |
1150 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
1151 | md_wakeup_thread(bitmap->mddev->thread); | |
1152 | } | |
1153 | ||
32a7627c | 1154 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
9c81075f JB |
1155 | "read %lu/%lu pages, set %lu of %lu bits\n", |
1156 | bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt, chunks); | |
4ad13663 N |
1157 | |
1158 | return 0; | |
32a7627c | 1159 | |
4ad13663 N |
1160 | err: |
1161 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1162 | bmname(bitmap), ret); | |
32a7627c N |
1163 | return ret; |
1164 | } | |
1165 | ||
a654b9d8 N |
1166 | void bitmap_write_all(struct bitmap *bitmap) |
1167 | { | |
1168 | /* We don't actually write all bitmap blocks here, | |
1169 | * just flag them as needing to be written | |
1170 | */ | |
ec7a3197 | 1171 | int i; |
a654b9d8 | 1172 | |
ac2f40be | 1173 | for (i = 0; i < bitmap->file_pages; i++) |
ec7a3197 N |
1174 | set_page_attr(bitmap, bitmap->filemap[i], |
1175 | BITMAP_PAGE_NEEDWRITE); | |
a654b9d8 N |
1176 | } |
1177 | ||
32a7627c N |
1178 | static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) |
1179 | { | |
1180 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1181 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1182 | bitmap->bp[page].count += inc; | |
32a7627c N |
1183 | bitmap_checkfree(bitmap, page); |
1184 | } | |
1185 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, | |
57dab0bd | 1186 | sector_t offset, sector_t *blocks, |
32a7627c N |
1187 | int create); |
1188 | ||
1189 | /* | |
1190 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1191 | * out to disk | |
1192 | */ | |
1193 | ||
aa5cbd10 | 1194 | void bitmap_daemon_work(mddev_t *mddev) |
32a7627c | 1195 | { |
aa5cbd10 | 1196 | struct bitmap *bitmap; |
aa3163f8 | 1197 | unsigned long j; |
32a7627c N |
1198 | unsigned long flags; |
1199 | struct page *page = NULL, *lastpage = NULL; | |
57dab0bd | 1200 | sector_t blocks; |
ea03aff9 | 1201 | void *paddr; |
e384e585 | 1202 | struct dm_dirty_log *log = mddev->bitmap_info.log; |
32a7627c | 1203 | |
aa5cbd10 N |
1204 | /* Use a mutex to guard daemon_work against |
1205 | * bitmap_destroy. | |
1206 | */ | |
c3d9714e | 1207 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1208 | bitmap = mddev->bitmap; |
1209 | if (bitmap == NULL) { | |
c3d9714e | 1210 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1211 | return; |
aa5cbd10 | 1212 | } |
42a04b50 | 1213 | if (time_before(jiffies, bitmap->daemon_lastrun |
1b04be96 | 1214 | + bitmap->mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1215 | goto done; |
1216 | ||
32a7627c | 1217 | bitmap->daemon_lastrun = jiffies; |
8311c29d N |
1218 | if (bitmap->allclean) { |
1219 | bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
aa5cbd10 | 1220 | goto done; |
8311c29d N |
1221 | } |
1222 | bitmap->allclean = 1; | |
32a7627c | 1223 | |
be512691 | 1224 | spin_lock_irqsave(&bitmap->lock, flags); |
32a7627c N |
1225 | for (j = 0; j < bitmap->chunks; j++) { |
1226 | bitmap_counter_t *bmc; | |
e384e585 N |
1227 | if (!bitmap->filemap) { |
1228 | if (!log) | |
1229 | /* error or shutdown */ | |
1230 | break; | |
1231 | } else | |
1232 | page = filemap_get_page(bitmap, j); | |
32a7627c N |
1233 | |
1234 | if (page != lastpage) { | |
aa3163f8 | 1235 | /* skip this page unless it's marked as needing cleaning */ |
ec7a3197 N |
1236 | if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) { |
1237 | int need_write = test_page_attr(bitmap, page, | |
1238 | BITMAP_PAGE_NEEDWRITE); | |
a647e4bc | 1239 | if (need_write) |
aa3163f8 | 1240 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); |
a647e4bc | 1241 | |
aa3163f8 | 1242 | spin_unlock_irqrestore(&bitmap->lock, flags); |
8311c29d | 1243 | if (need_write) { |
4ad13663 | 1244 | write_page(bitmap, page, 0); |
8311c29d N |
1245 | bitmap->allclean = 0; |
1246 | } | |
be512691 N |
1247 | spin_lock_irqsave(&bitmap->lock, flags); |
1248 | j |= (PAGE_BITS - 1); | |
aa3163f8 N |
1249 | continue; |
1250 | } | |
1251 | ||
32a7627c | 1252 | /* grab the new page, sync and release the old */ |
32a7627c | 1253 | if (lastpage != NULL) { |
ec7a3197 | 1254 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1255 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1256 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 1257 | write_page(bitmap, lastpage, 0); |
32a7627c N |
1258 | } else { |
1259 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1260 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1261 | } | |
32a7627c N |
1262 | } else |
1263 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1264 | lastpage = page; | |
a0da84f3 NB |
1265 | |
1266 | /* We are possibly going to clear some bits, so make | |
1267 | * sure that events_cleared is up-to-date. | |
1268 | */ | |
ece5cff0 N |
1269 | if (bitmap->need_sync && |
1270 | bitmap->mddev->bitmap_info.external == 0) { | |
a0da84f3 NB |
1271 | bitmap_super_t *sb; |
1272 | bitmap->need_sync = 0; | |
1273 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); | |
1274 | sb->events_cleared = | |
1275 | cpu_to_le64(bitmap->events_cleared); | |
1276 | kunmap_atomic(sb, KM_USER0); | |
1277 | write_page(bitmap, bitmap->sb_page, 1); | |
1278 | } | |
32a7627c | 1279 | spin_lock_irqsave(&bitmap->lock, flags); |
ece5cff0 N |
1280 | if (!bitmap->need_sync) |
1281 | clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
32a7627c | 1282 | } |
db305e50 N |
1283 | bmc = bitmap_get_counter(bitmap, |
1284 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
1285 | &blocks, 0); | |
32a7627c | 1286 | if (bmc) { |
8311c29d N |
1287 | if (*bmc) |
1288 | bitmap->allclean = 0; | |
1289 | ||
32a7627c | 1290 | if (*bmc == 2) { |
ac2f40be | 1291 | *bmc = 1; /* maybe clear the bit next time */ |
32a7627c | 1292 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); |
ece5cff0 | 1293 | } else if (*bmc == 1 && !bitmap->need_sync) { |
32a7627c N |
1294 | /* we can clear the bit */ |
1295 | *bmc = 0; | |
db305e50 N |
1296 | bitmap_count_page(bitmap, |
1297 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
32a7627c N |
1298 | -1); |
1299 | ||
1300 | /* clear the bit */ | |
e384e585 N |
1301 | if (page) { |
1302 | paddr = kmap_atomic(page, KM_USER0); | |
1303 | if (bitmap->flags & BITMAP_HOSTENDIAN) | |
1304 | clear_bit(file_page_offset(bitmap, j), | |
1305 | paddr); | |
1306 | else | |
6b33aff3 | 1307 | __test_and_clear_bit_le(file_page_offset(bitmap, j), |
e384e585 N |
1308 | paddr); |
1309 | kunmap_atomic(paddr, KM_USER0); | |
1310 | } else | |
1311 | log->type->clear_region(log, j); | |
32a7627c | 1312 | } |
be512691 N |
1313 | } else |
1314 | j |= PAGE_COUNTER_MASK; | |
32a7627c | 1315 | } |
be512691 | 1316 | spin_unlock_irqrestore(&bitmap->lock, flags); |
32a7627c N |
1317 | |
1318 | /* now sync the final page */ | |
e384e585 | 1319 | if (lastpage != NULL || log != NULL) { |
32a7627c | 1320 | spin_lock_irqsave(&bitmap->lock, flags); |
ec7a3197 | 1321 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1322 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1323 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
e384e585 N |
1324 | if (lastpage) |
1325 | write_page(bitmap, lastpage, 0); | |
1326 | else | |
1327 | if (log->type->flush(log)) | |
1328 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
32a7627c N |
1329 | } else { |
1330 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1331 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1332 | } | |
32a7627c N |
1333 | } |
1334 | ||
7be3dfec | 1335 | done: |
8311c29d | 1336 | if (bitmap->allclean == 0) |
ac2f40be | 1337 | bitmap->mddev->thread->timeout = |
1b04be96 | 1338 | bitmap->mddev->bitmap_info.daemon_sleep; |
c3d9714e | 1339 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1340 | } |
1341 | ||
32a7627c | 1342 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
57dab0bd | 1343 | sector_t offset, sector_t *blocks, |
32a7627c | 1344 | int create) |
ee305ace N |
1345 | __releases(bitmap->lock) |
1346 | __acquires(bitmap->lock) | |
32a7627c N |
1347 | { |
1348 | /* If 'create', we might release the lock and reclaim it. | |
1349 | * The lock must have been taken with interrupts enabled. | |
1350 | * If !create, we don't release the lock. | |
1351 | */ | |
1352 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1353 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1354 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1355 | sector_t csize; | |
ef425673 | 1356 | int err; |
32a7627c | 1357 | |
ef425673 N |
1358 | err = bitmap_checkpage(bitmap, page, create); |
1359 | ||
1360 | if (bitmap->bp[page].hijacked || | |
1361 | bitmap->bp[page].map == NULL) | |
1362 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) + | |
1363 | PAGE_COUNTER_SHIFT - 1); | |
1364 | else | |
32a7627c | 1365 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); |
ef425673 N |
1366 | *blocks = csize - (offset & (csize - 1)); |
1367 | ||
1368 | if (err < 0) | |
32a7627c | 1369 | return NULL; |
ef425673 | 1370 | |
32a7627c N |
1371 | /* now locked ... */ |
1372 | ||
1373 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1374 | /* should we use the first or second counter field | |
1375 | * of the hijacked pointer? */ | |
1376 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1377 | return &((bitmap_counter_t *) |
1378 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1379 | } else /* page is allocated */ |
32a7627c N |
1380 | return (bitmap_counter_t *) |
1381 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1382 | } |
1383 | ||
4b6d287f | 1384 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1385 | { |
ac2f40be N |
1386 | if (!bitmap) |
1387 | return 0; | |
4b6d287f N |
1388 | |
1389 | if (behind) { | |
696fcd53 | 1390 | int bw; |
4b6d287f | 1391 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1392 | bw = atomic_read(&bitmap->behind_writes); |
1393 | if (bw > bitmap->behind_writes_used) | |
1394 | bitmap->behind_writes_used = bw; | |
1395 | ||
4b6d287f | 1396 | PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n", |
696fcd53 | 1397 | bw, bitmap->max_write_behind); |
4b6d287f N |
1398 | } |
1399 | ||
32a7627c | 1400 | while (sectors) { |
57dab0bd | 1401 | sector_t blocks; |
32a7627c N |
1402 | bitmap_counter_t *bmc; |
1403 | ||
1404 | spin_lock_irq(&bitmap->lock); | |
1405 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); | |
1406 | if (!bmc) { | |
1407 | spin_unlock_irq(&bitmap->lock); | |
1408 | return 0; | |
1409 | } | |
1410 | ||
da6e1a32 NB |
1411 | if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) { |
1412 | DEFINE_WAIT(__wait); | |
1413 | /* note that it is safe to do the prepare_to_wait | |
1414 | * after the test as long as we do it before dropping | |
1415 | * the spinlock. | |
1416 | */ | |
1417 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1418 | TASK_UNINTERRUPTIBLE); | |
1419 | spin_unlock_irq(&bitmap->lock); | |
7eaceacc | 1420 | io_schedule(); |
da6e1a32 NB |
1421 | finish_wait(&bitmap->overflow_wait, &__wait); |
1422 | continue; | |
1423 | } | |
1424 | ||
ac2f40be | 1425 | switch (*bmc) { |
32a7627c N |
1426 | case 0: |
1427 | bitmap_file_set_bit(bitmap, offset); | |
ac2f40be | 1428 | bitmap_count_page(bitmap, offset, 1); |
32a7627c N |
1429 | /* fall through */ |
1430 | case 1: | |
1431 | *bmc = 2; | |
1432 | } | |
da6e1a32 | 1433 | |
32a7627c N |
1434 | (*bmc)++; |
1435 | ||
1436 | spin_unlock_irq(&bitmap->lock); | |
1437 | ||
1438 | offset += blocks; | |
1439 | if (sectors > blocks) | |
1440 | sectors -= blocks; | |
ac2f40be N |
1441 | else |
1442 | sectors = 0; | |
32a7627c | 1443 | } |
8311c29d | 1444 | bitmap->allclean = 0; |
32a7627c N |
1445 | return 0; |
1446 | } | |
ac2f40be | 1447 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1448 | |
1449 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1450 | int success, int behind) |
32a7627c | 1451 | { |
ac2f40be N |
1452 | if (!bitmap) |
1453 | return; | |
4b6d287f | 1454 | if (behind) { |
e555190d N |
1455 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1456 | wake_up(&bitmap->behind_wait); | |
4b6d287f N |
1457 | PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n", |
1458 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1459 | } | |
d0a4bb49 N |
1460 | if (bitmap->mddev->degraded) |
1461 | /* Never clear bits or update events_cleared when degraded */ | |
1462 | success = 0; | |
4b6d287f | 1463 | |
32a7627c | 1464 | while (sectors) { |
57dab0bd | 1465 | sector_t blocks; |
32a7627c N |
1466 | unsigned long flags; |
1467 | bitmap_counter_t *bmc; | |
1468 | ||
1469 | spin_lock_irqsave(&bitmap->lock, flags); | |
1470 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); | |
1471 | if (!bmc) { | |
1472 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1473 | return; | |
1474 | } | |
1475 | ||
a0da84f3 NB |
1476 | if (success && |
1477 | bitmap->events_cleared < bitmap->mddev->events) { | |
1478 | bitmap->events_cleared = bitmap->mddev->events; | |
1479 | bitmap->need_sync = 1; | |
5ff5afff | 1480 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1481 | } |
1482 | ||
32a7627c N |
1483 | if (!success && ! (*bmc & NEEDED_MASK)) |
1484 | *bmc |= NEEDED_MASK; | |
1485 | ||
da6e1a32 NB |
1486 | if ((*bmc & COUNTER_MAX) == COUNTER_MAX) |
1487 | wake_up(&bitmap->overflow_wait); | |
1488 | ||
32a7627c | 1489 | (*bmc)--; |
ac2f40be | 1490 | if (*bmc <= 2) |
32a7627c | 1491 | set_page_attr(bitmap, |
e384e585 N |
1492 | filemap_get_page( |
1493 | bitmap, | |
1494 | offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
32a7627c | 1495 | BITMAP_PAGE_CLEAN); |
ac2f40be | 1496 | |
32a7627c N |
1497 | spin_unlock_irqrestore(&bitmap->lock, flags); |
1498 | offset += blocks; | |
1499 | if (sectors > blocks) | |
1500 | sectors -= blocks; | |
ac2f40be N |
1501 | else |
1502 | sectors = 0; | |
32a7627c N |
1503 | } |
1504 | } | |
ac2f40be | 1505 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1506 | |
57dab0bd | 1507 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1508 | int degraded) |
32a7627c N |
1509 | { |
1510 | bitmap_counter_t *bmc; | |
1511 | int rv; | |
1512 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1513 | *blocks = 1024; | |
1514 | return 1; /* always resync if no bitmap */ | |
1515 | } | |
1516 | spin_lock_irq(&bitmap->lock); | |
1517 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1518 | rv = 0; | |
1519 | if (bmc) { | |
1520 | /* locked */ | |
1521 | if (RESYNC(*bmc)) | |
1522 | rv = 1; | |
1523 | else if (NEEDED(*bmc)) { | |
1524 | rv = 1; | |
6a806c51 N |
1525 | if (!degraded) { /* don't set/clear bits if degraded */ |
1526 | *bmc |= RESYNC_MASK; | |
1527 | *bmc &= ~NEEDED_MASK; | |
1528 | } | |
32a7627c N |
1529 | } |
1530 | } | |
1531 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1532 | bitmap->allclean = 0; |
32a7627c N |
1533 | return rv; |
1534 | } | |
1535 | ||
57dab0bd | 1536 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a N |
1537 | int degraded) |
1538 | { | |
1539 | /* bitmap_start_sync must always report on multiples of whole | |
1540 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1541 | * get confused. | |
1542 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1543 | * At least PAGE_SIZE>>9 blocks are covered. | |
1544 | * Return the 'or' of the result. | |
1545 | */ | |
1546 | int rv = 0; | |
57dab0bd | 1547 | sector_t blocks1; |
1187cf0a N |
1548 | |
1549 | *blocks = 0; | |
1550 | while (*blocks < (PAGE_SIZE>>9)) { | |
1551 | rv |= __bitmap_start_sync(bitmap, offset, | |
1552 | &blocks1, degraded); | |
1553 | offset += blocks1; | |
1554 | *blocks += blocks1; | |
1555 | } | |
1556 | return rv; | |
1557 | } | |
ac2f40be | 1558 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1559 | |
57dab0bd | 1560 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1561 | { |
1562 | bitmap_counter_t *bmc; | |
1563 | unsigned long flags; | |
ac2f40be N |
1564 | |
1565 | if (bitmap == NULL) { | |
32a7627c N |
1566 | *blocks = 1024; |
1567 | return; | |
1568 | } | |
1569 | spin_lock_irqsave(&bitmap->lock, flags); | |
1570 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1571 | if (bmc == NULL) | |
1572 | goto unlock; | |
1573 | /* locked */ | |
32a7627c N |
1574 | if (RESYNC(*bmc)) { |
1575 | *bmc &= ~RESYNC_MASK; | |
1576 | ||
1577 | if (!NEEDED(*bmc) && aborted) | |
1578 | *bmc |= NEEDED_MASK; | |
1579 | else { | |
ac2f40be | 1580 | if (*bmc <= 2) |
32a7627c N |
1581 | set_page_attr(bitmap, |
1582 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1583 | BITMAP_PAGE_CLEAN); | |
32a7627c N |
1584 | } |
1585 | } | |
1586 | unlock: | |
1587 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
8311c29d | 1588 | bitmap->allclean = 0; |
32a7627c | 1589 | } |
ac2f40be | 1590 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1591 | |
1592 | void bitmap_close_sync(struct bitmap *bitmap) | |
1593 | { | |
1594 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1595 | * properly have been aborted. It remains to us to clear the | |
1596 | * RESYNC bit wherever it is still on | |
1597 | */ | |
1598 | sector_t sector = 0; | |
57dab0bd | 1599 | sector_t blocks; |
b47490c9 N |
1600 | if (!bitmap) |
1601 | return; | |
32a7627c N |
1602 | while (sector < bitmap->mddev->resync_max_sectors) { |
1603 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1604 | sector += blocks; |
1605 | } | |
1606 | } | |
ac2f40be | 1607 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 N |
1608 | |
1609 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1610 | { | |
1611 | sector_t s = 0; | |
57dab0bd | 1612 | sector_t blocks; |
b47490c9 N |
1613 | |
1614 | if (!bitmap) | |
1615 | return; | |
1616 | if (sector == 0) { | |
1617 | bitmap->last_end_sync = jiffies; | |
1618 | return; | |
1619 | } | |
1620 | if (time_before(jiffies, (bitmap->last_end_sync | |
1b04be96 | 1621 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1622 | return; |
1623 | wait_event(bitmap->mddev->recovery_wait, | |
1624 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1625 | ||
75d3da43 | 1626 | bitmap->mddev->curr_resync_completed = sector; |
070dc6dd | 1627 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
b47490c9 N |
1628 | sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1); |
1629 | s = 0; | |
1630 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1631 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1632 | s += blocks; | |
32a7627c | 1633 | } |
b47490c9 | 1634 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1635 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1636 | } |
ac2f40be | 1637 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1638 | |
6a07997f | 1639 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1640 | { |
1641 | /* For each chunk covered by any of these sectors, set the | |
193f1c93 | 1642 | * counter to 1 and set resync_needed. They should all |
32a7627c N |
1643 | * be 0 at this point |
1644 | */ | |
193f1c93 | 1645 | |
57dab0bd | 1646 | sector_t secs; |
193f1c93 N |
1647 | bitmap_counter_t *bmc; |
1648 | spin_lock_irq(&bitmap->lock); | |
1649 | bmc = bitmap_get_counter(bitmap, offset, &secs, 1); | |
1650 | if (!bmc) { | |
32a7627c | 1651 | spin_unlock_irq(&bitmap->lock); |
193f1c93 | 1652 | return; |
32a7627c | 1653 | } |
ac2f40be | 1654 | if (!*bmc) { |
193f1c93 | 1655 | struct page *page; |
ac2f40be | 1656 | *bmc = 1 | (needed ? NEEDED_MASK : 0); |
193f1c93 N |
1657 | bitmap_count_page(bitmap, offset, 1); |
1658 | page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)); | |
1659 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1660 | } | |
1661 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1662 | bitmap->allclean = 0; |
32a7627c N |
1663 | } |
1664 | ||
9b1d1dac PC |
1665 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1666 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1667 | { | |
1668 | unsigned long chunk; | |
1669 | ||
1670 | for (chunk = s; chunk <= e; chunk++) { | |
db305e50 | 1671 | sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap); |
9b1d1dac PC |
1672 | bitmap_set_memory_bits(bitmap, sec, 1); |
1673 | bitmap_file_set_bit(bitmap, sec); | |
ffa23322 N |
1674 | if (sec < bitmap->mddev->recovery_cp) |
1675 | /* We are asserting that the array is dirty, | |
1676 | * so move the recovery_cp address back so | |
1677 | * that it is obvious that it is dirty | |
1678 | */ | |
1679 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1680 | } |
1681 | } | |
1682 | ||
6b8b3e8a N |
1683 | /* |
1684 | * flush out any pending updates | |
1685 | */ | |
1686 | void bitmap_flush(mddev_t *mddev) | |
1687 | { | |
1688 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1689 | long sleep; |
6b8b3e8a N |
1690 | |
1691 | if (!bitmap) /* there was no bitmap */ | |
1692 | return; | |
1693 | ||
1694 | /* run the daemon_work three time to ensure everything is flushed | |
1695 | * that can be | |
1696 | */ | |
1b04be96 | 1697 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1698 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1699 | bitmap_daemon_work(mddev); |
42a04b50 | 1700 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1701 | bitmap_daemon_work(mddev); |
42a04b50 | 1702 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1703 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1704 | bitmap_update_sb(bitmap); |
1705 | } | |
1706 | ||
32a7627c N |
1707 | /* |
1708 | * free memory that was allocated | |
1709 | */ | |
3178b0db | 1710 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1711 | { |
1712 | unsigned long k, pages; | |
1713 | struct bitmap_page *bp; | |
32a7627c N |
1714 | |
1715 | if (!bitmap) /* there was no bitmap */ | |
1716 | return; | |
1717 | ||
32a7627c N |
1718 | /* release the bitmap file and kill the daemon */ |
1719 | bitmap_file_put(bitmap); | |
1720 | ||
1721 | bp = bitmap->bp; | |
1722 | pages = bitmap->pages; | |
1723 | ||
1724 | /* free all allocated memory */ | |
1725 | ||
32a7627c N |
1726 | if (bp) /* deallocate the page memory */ |
1727 | for (k = 0; k < pages; k++) | |
1728 | if (bp[k].map && !bp[k].hijacked) | |
1729 | kfree(bp[k].map); | |
1730 | kfree(bp); | |
1731 | kfree(bitmap); | |
1732 | } | |
aa5cbd10 | 1733 | |
3178b0db N |
1734 | void bitmap_destroy(mddev_t *mddev) |
1735 | { | |
1736 | struct bitmap *bitmap = mddev->bitmap; | |
1737 | ||
1738 | if (!bitmap) /* there was no bitmap */ | |
1739 | return; | |
1740 | ||
c3d9714e | 1741 | mutex_lock(&mddev->bitmap_info.mutex); |
3178b0db | 1742 | mddev->bitmap = NULL; /* disconnect from the md device */ |
c3d9714e | 1743 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1744 | if (mddev->thread) |
1745 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db | 1746 | |
ece5cff0 N |
1747 | if (bitmap->sysfs_can_clear) |
1748 | sysfs_put(bitmap->sysfs_can_clear); | |
1749 | ||
3178b0db N |
1750 | bitmap_free(bitmap); |
1751 | } | |
32a7627c N |
1752 | |
1753 | /* | |
1754 | * initialize the bitmap structure | |
1755 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1756 | */ | |
1757 | int bitmap_create(mddev_t *mddev) | |
1758 | { | |
1759 | struct bitmap *bitmap; | |
1f593903 | 1760 | sector_t blocks = mddev->resync_max_sectors; |
32a7627c N |
1761 | unsigned long chunks; |
1762 | unsigned long pages; | |
c3d9714e | 1763 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1764 | int err; |
5ff5afff | 1765 | struct sysfs_dirent *bm = NULL; |
32a7627c | 1766 | |
5f6e3c83 | 1767 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1768 | |
e384e585 N |
1769 | if (!file |
1770 | && !mddev->bitmap_info.offset | |
1771 | && !mddev->bitmap_info.log) /* bitmap disabled, nothing to do */ | |
32a7627c N |
1772 | return 0; |
1773 | ||
c3d9714e | 1774 | BUG_ON(file && mddev->bitmap_info.offset); |
e384e585 | 1775 | BUG_ON(mddev->bitmap_info.offset && mddev->bitmap_info.log); |
a654b9d8 | 1776 | |
9ffae0cf | 1777 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1778 | if (!bitmap) |
1779 | return -ENOMEM; | |
1780 | ||
32a7627c | 1781 | spin_lock_init(&bitmap->lock); |
ce25c31b N |
1782 | atomic_set(&bitmap->pending_writes, 0); |
1783 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1784 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1785 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1786 | |
32a7627c | 1787 | bitmap->mddev = mddev; |
32a7627c | 1788 | |
5ff5afff N |
1789 | if (mddev->kobj.sd) |
1790 | bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap"); | |
ece5cff0 | 1791 | if (bm) { |
3ff195b0 | 1792 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear"); |
ece5cff0 N |
1793 | sysfs_put(bm); |
1794 | } else | |
1795 | bitmap->sysfs_can_clear = NULL; | |
1796 | ||
32a7627c | 1797 | bitmap->file = file; |
ce25c31b N |
1798 | if (file) { |
1799 | get_file(file); | |
ae8fa283 N |
1800 | /* As future accesses to this file will use bmap, |
1801 | * and bypass the page cache, we must sync the file | |
1802 | * first. | |
1803 | */ | |
8018ab05 | 1804 | vfs_fsync(file, 1); |
ce25c31b | 1805 | } |
42a04b50 | 1806 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
9c81075f JB |
1807 | if (!mddev->bitmap_info.external) { |
1808 | /* | |
1809 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is | |
1810 | * instructing us to create a new on-disk bitmap instance. | |
1811 | */ | |
1812 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) | |
1813 | err = bitmap_new_disk_sb(bitmap); | |
1814 | else | |
1815 | err = bitmap_read_sb(bitmap); | |
1816 | } else { | |
ece5cff0 N |
1817 | err = 0; |
1818 | if (mddev->bitmap_info.chunksize == 0 || | |
1819 | mddev->bitmap_info.daemon_sleep == 0) | |
1820 | /* chunksize and time_base need to be | |
1821 | * set first. */ | |
1822 | err = -EINVAL; | |
1823 | } | |
32a7627c | 1824 | if (err) |
3178b0db | 1825 | goto error; |
32a7627c | 1826 | |
624ce4f5 | 1827 | bitmap->daemon_lastrun = jiffies; |
42a04b50 | 1828 | bitmap->chunkshift = ffz(~mddev->bitmap_info.chunksize); |
32a7627c N |
1829 | |
1830 | /* now that chunksize and chunkshift are set, we can use these macros */ | |
ac2f40be | 1831 | chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >> |
1f593903 | 1832 | CHUNK_BLOCK_SHIFT(bitmap); |
ac2f40be | 1833 | pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; |
32a7627c N |
1834 | |
1835 | BUG_ON(!pages); | |
1836 | ||
1837 | bitmap->chunks = chunks; | |
1838 | bitmap->pages = pages; | |
1839 | bitmap->missing_pages = pages; | |
1840 | bitmap->counter_bits = COUNTER_BITS; | |
1841 | ||
1842 | bitmap->syncchunk = ~0UL; | |
1843 | ||
44456d37 | 1844 | #ifdef INJECT_FATAL_FAULT_1 |
32a7627c N |
1845 | bitmap->bp = NULL; |
1846 | #else | |
9ffae0cf | 1847 | bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); |
32a7627c | 1848 | #endif |
3178b0db | 1849 | err = -ENOMEM; |
32a7627c | 1850 | if (!bitmap->bp) |
3178b0db | 1851 | goto error; |
32a7627c | 1852 | |
69e51b44 N |
1853 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", |
1854 | pages, bmname(bitmap)); | |
1855 | ||
1856 | mddev->bitmap = bitmap; | |
1857 | ||
1858 | ||
1859 | return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0; | |
1860 | ||
1861 | error: | |
1862 | bitmap_free(bitmap); | |
1863 | return err; | |
1864 | } | |
1865 | ||
1866 | int bitmap_load(mddev_t *mddev) | |
1867 | { | |
1868 | int err = 0; | |
1869 | sector_t sector = 0; | |
1870 | struct bitmap *bitmap = mddev->bitmap; | |
1871 | ||
1872 | if (!bitmap) | |
1873 | goto out; | |
1874 | ||
1875 | /* Clear out old bitmap info first: Either there is none, or we | |
1876 | * are resuming after someone else has possibly changed things, | |
1877 | * so we should forget old cached info. | |
1878 | * All chunks should be clean, but some might need_sync. | |
1879 | */ | |
1880 | while (sector < mddev->resync_max_sectors) { | |
57dab0bd | 1881 | sector_t blocks; |
69e51b44 N |
1882 | bitmap_start_sync(bitmap, sector, &blocks, 0); |
1883 | sector += blocks; | |
1884 | } | |
1885 | bitmap_close_sync(bitmap); | |
1886 | ||
e384e585 N |
1887 | if (mddev->bitmap_info.log) { |
1888 | unsigned long i; | |
1889 | struct dm_dirty_log *log = mddev->bitmap_info.log; | |
1890 | for (i = 0; i < bitmap->chunks; i++) | |
1891 | if (!log->type->in_sync(log, i, 1)) | |
1892 | bitmap_set_memory_bits(bitmap, | |
1893 | (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap), | |
1894 | 1); | |
69e51b44 N |
1895 | } else { |
1896 | sector_t start = 0; | |
1897 | if (mddev->degraded == 0 | |
1898 | || bitmap->events_cleared == mddev->events) | |
1899 | /* no need to keep dirty bits to optimise a | |
1900 | * re-add of a missing device */ | |
1901 | start = mddev->recovery_cp; | |
193f1c93 | 1902 | |
69e51b44 N |
1903 | err = bitmap_init_from_disk(bitmap, start); |
1904 | } | |
32a7627c | 1905 | if (err) |
69e51b44 | 1906 | goto out; |
3178b0db | 1907 | |
1b04be96 | 1908 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1909 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1910 | |
4ad13663 N |
1911 | bitmap_update_sb(bitmap); |
1912 | ||
69e51b44 N |
1913 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
1914 | err = -EIO; | |
1915 | out: | |
3178b0db | 1916 | return err; |
32a7627c | 1917 | } |
69e51b44 | 1918 | EXPORT_SYMBOL_GPL(bitmap_load); |
32a7627c | 1919 | |
43a70507 N |
1920 | static ssize_t |
1921 | location_show(mddev_t *mddev, char *page) | |
1922 | { | |
1923 | ssize_t len; | |
ac2f40be | 1924 | if (mddev->bitmap_info.file) |
43a70507 | 1925 | len = sprintf(page, "file"); |
ac2f40be | 1926 | else if (mddev->bitmap_info.offset) |
43a70507 | 1927 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 1928 | else |
43a70507 N |
1929 | len = sprintf(page, "none"); |
1930 | len += sprintf(page+len, "\n"); | |
1931 | return len; | |
1932 | } | |
1933 | ||
1934 | static ssize_t | |
1935 | location_store(mddev_t *mddev, const char *buf, size_t len) | |
1936 | { | |
1937 | ||
1938 | if (mddev->pers) { | |
1939 | if (!mddev->pers->quiesce) | |
1940 | return -EBUSY; | |
1941 | if (mddev->recovery || mddev->sync_thread) | |
1942 | return -EBUSY; | |
1943 | } | |
1944 | ||
1945 | if (mddev->bitmap || mddev->bitmap_info.file || | |
1946 | mddev->bitmap_info.offset) { | |
1947 | /* bitmap already configured. Only option is to clear it */ | |
1948 | if (strncmp(buf, "none", 4) != 0) | |
1949 | return -EBUSY; | |
1950 | if (mddev->pers) { | |
1951 | mddev->pers->quiesce(mddev, 1); | |
1952 | bitmap_destroy(mddev); | |
1953 | mddev->pers->quiesce(mddev, 0); | |
1954 | } | |
1955 | mddev->bitmap_info.offset = 0; | |
1956 | if (mddev->bitmap_info.file) { | |
1957 | struct file *f = mddev->bitmap_info.file; | |
1958 | mddev->bitmap_info.file = NULL; | |
1959 | restore_bitmap_write_access(f); | |
1960 | fput(f); | |
1961 | } | |
1962 | } else { | |
1963 | /* No bitmap, OK to set a location */ | |
1964 | long long offset; | |
1965 | if (strncmp(buf, "none", 4) == 0) | |
1966 | /* nothing to be done */; | |
1967 | else if (strncmp(buf, "file:", 5) == 0) { | |
1968 | /* Not supported yet */ | |
1969 | return -EINVAL; | |
1970 | } else { | |
1971 | int rv; | |
1972 | if (buf[0] == '+') | |
1973 | rv = strict_strtoll(buf+1, 10, &offset); | |
1974 | else | |
1975 | rv = strict_strtoll(buf, 10, &offset); | |
1976 | if (rv) | |
1977 | return rv; | |
1978 | if (offset == 0) | |
1979 | return -EINVAL; | |
ece5cff0 N |
1980 | if (mddev->bitmap_info.external == 0 && |
1981 | mddev->major_version == 0 && | |
43a70507 N |
1982 | offset != mddev->bitmap_info.default_offset) |
1983 | return -EINVAL; | |
1984 | mddev->bitmap_info.offset = offset; | |
1985 | if (mddev->pers) { | |
1986 | mddev->pers->quiesce(mddev, 1); | |
1987 | rv = bitmap_create(mddev); | |
1988 | if (rv) { | |
1989 | bitmap_destroy(mddev); | |
1990 | mddev->bitmap_info.offset = 0; | |
1991 | } | |
1992 | mddev->pers->quiesce(mddev, 0); | |
1993 | if (rv) | |
1994 | return rv; | |
1995 | } | |
1996 | } | |
1997 | } | |
1998 | if (!mddev->external) { | |
1999 | /* Ensure new bitmap info is stored in | |
2000 | * metadata promptly. | |
2001 | */ | |
2002 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2003 | md_wakeup_thread(mddev->thread); | |
2004 | } | |
2005 | return len; | |
2006 | } | |
2007 | ||
2008 | static struct md_sysfs_entry bitmap_location = | |
2009 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
2010 | ||
2011 | static ssize_t | |
2012 | timeout_show(mddev_t *mddev, char *page) | |
2013 | { | |
2014 | ssize_t len; | |
2015 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
2016 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 2017 | |
43a70507 N |
2018 | len = sprintf(page, "%lu", secs); |
2019 | if (jifs) | |
2020 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
2021 | len += sprintf(page+len, "\n"); | |
2022 | return len; | |
2023 | } | |
2024 | ||
2025 | static ssize_t | |
2026 | timeout_store(mddev_t *mddev, const char *buf, size_t len) | |
2027 | { | |
2028 | /* timeout can be set at any time */ | |
2029 | unsigned long timeout; | |
2030 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
2031 | if (rv) | |
2032 | return rv; | |
2033 | ||
2034 | /* just to make sure we don't overflow... */ | |
2035 | if (timeout >= LONG_MAX / HZ) | |
2036 | return -EINVAL; | |
2037 | ||
2038 | timeout = timeout * HZ / 10000; | |
2039 | ||
2040 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
2041 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
2042 | if (timeout < 1) | |
2043 | timeout = 1; | |
2044 | mddev->bitmap_info.daemon_sleep = timeout; | |
2045 | if (mddev->thread) { | |
2046 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
2047 | * the bitmap is all clean and we don't need to | |
2048 | * adjust the timeout right now | |
2049 | */ | |
2050 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2051 | mddev->thread->timeout = timeout; | |
2052 | md_wakeup_thread(mddev->thread); | |
2053 | } | |
2054 | } | |
2055 | return len; | |
2056 | } | |
2057 | ||
2058 | static struct md_sysfs_entry bitmap_timeout = | |
2059 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2060 | ||
2061 | static ssize_t | |
2062 | backlog_show(mddev_t *mddev, char *page) | |
2063 | { | |
2064 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2065 | } | |
2066 | ||
2067 | static ssize_t | |
2068 | backlog_store(mddev_t *mddev, const char *buf, size_t len) | |
2069 | { | |
2070 | unsigned long backlog; | |
2071 | int rv = strict_strtoul(buf, 10, &backlog); | |
2072 | if (rv) | |
2073 | return rv; | |
2074 | if (backlog > COUNTER_MAX) | |
2075 | return -EINVAL; | |
2076 | mddev->bitmap_info.max_write_behind = backlog; | |
2077 | return len; | |
2078 | } | |
2079 | ||
2080 | static struct md_sysfs_entry bitmap_backlog = | |
2081 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2082 | ||
2083 | static ssize_t | |
2084 | chunksize_show(mddev_t *mddev, char *page) | |
2085 | { | |
2086 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2087 | } | |
2088 | ||
2089 | static ssize_t | |
2090 | chunksize_store(mddev_t *mddev, const char *buf, size_t len) | |
2091 | { | |
2092 | /* Can only be changed when no bitmap is active */ | |
2093 | int rv; | |
2094 | unsigned long csize; | |
2095 | if (mddev->bitmap) | |
2096 | return -EBUSY; | |
2097 | rv = strict_strtoul(buf, 10, &csize); | |
2098 | if (rv) | |
2099 | return rv; | |
2100 | if (csize < 512 || | |
2101 | !is_power_of_2(csize)) | |
2102 | return -EINVAL; | |
2103 | mddev->bitmap_info.chunksize = csize; | |
2104 | return len; | |
2105 | } | |
2106 | ||
2107 | static struct md_sysfs_entry bitmap_chunksize = | |
2108 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2109 | ||
ece5cff0 N |
2110 | static ssize_t metadata_show(mddev_t *mddev, char *page) |
2111 | { | |
2112 | return sprintf(page, "%s\n", (mddev->bitmap_info.external | |
2113 | ? "external" : "internal")); | |
2114 | } | |
2115 | ||
2116 | static ssize_t metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
2117 | { | |
2118 | if (mddev->bitmap || | |
2119 | mddev->bitmap_info.file || | |
2120 | mddev->bitmap_info.offset) | |
2121 | return -EBUSY; | |
2122 | if (strncmp(buf, "external", 8) == 0) | |
2123 | mddev->bitmap_info.external = 1; | |
2124 | else if (strncmp(buf, "internal", 8) == 0) | |
2125 | mddev->bitmap_info.external = 0; | |
2126 | else | |
2127 | return -EINVAL; | |
2128 | return len; | |
2129 | } | |
2130 | ||
2131 | static struct md_sysfs_entry bitmap_metadata = | |
2132 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2133 | ||
2134 | static ssize_t can_clear_show(mddev_t *mddev, char *page) | |
2135 | { | |
2136 | int len; | |
2137 | if (mddev->bitmap) | |
2138 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2139 | "false" : "true")); | |
2140 | else | |
2141 | len = sprintf(page, "\n"); | |
2142 | return len; | |
2143 | } | |
2144 | ||
2145 | static ssize_t can_clear_store(mddev_t *mddev, const char *buf, size_t len) | |
2146 | { | |
2147 | if (mddev->bitmap == NULL) | |
2148 | return -ENOENT; | |
2149 | if (strncmp(buf, "false", 5) == 0) | |
2150 | mddev->bitmap->need_sync = 1; | |
2151 | else if (strncmp(buf, "true", 4) == 0) { | |
2152 | if (mddev->degraded) | |
2153 | return -EBUSY; | |
2154 | mddev->bitmap->need_sync = 0; | |
2155 | } else | |
2156 | return -EINVAL; | |
2157 | return len; | |
2158 | } | |
2159 | ||
2160 | static struct md_sysfs_entry bitmap_can_clear = | |
2161 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2162 | ||
696fcd53 PC |
2163 | static ssize_t |
2164 | behind_writes_used_show(mddev_t *mddev, char *page) | |
2165 | { | |
2166 | if (mddev->bitmap == NULL) | |
2167 | return sprintf(page, "0\n"); | |
2168 | return sprintf(page, "%lu\n", | |
2169 | mddev->bitmap->behind_writes_used); | |
2170 | } | |
2171 | ||
2172 | static ssize_t | |
2173 | behind_writes_used_reset(mddev_t *mddev, const char *buf, size_t len) | |
2174 | { | |
2175 | if (mddev->bitmap) | |
2176 | mddev->bitmap->behind_writes_used = 0; | |
2177 | return len; | |
2178 | } | |
2179 | ||
2180 | static struct md_sysfs_entry max_backlog_used = | |
2181 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2182 | behind_writes_used_show, behind_writes_used_reset); | |
2183 | ||
43a70507 N |
2184 | static struct attribute *md_bitmap_attrs[] = { |
2185 | &bitmap_location.attr, | |
2186 | &bitmap_timeout.attr, | |
2187 | &bitmap_backlog.attr, | |
2188 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2189 | &bitmap_metadata.attr, |
2190 | &bitmap_can_clear.attr, | |
696fcd53 | 2191 | &max_backlog_used.attr, |
43a70507 N |
2192 | NULL |
2193 | }; | |
2194 | struct attribute_group md_bitmap_group = { | |
2195 | .name = "bitmap", | |
2196 | .attrs = md_bitmap_attrs, | |
2197 | }; | |
2198 |