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 | ||
0f420358 | 213 | target = rdev->sb_start + offset + index * (PAGE_SIZE/512); |
a654b9d8 | 214 | |
a2ed9615 | 215 | if (sync_page_io(rdev->bdev, target, |
e1defc4f | 216 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
a2ed9615 | 217 | page, READ)) { |
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; |
ab6085c7 | 267 | mddev_t *mddev = bitmap->mddev; |
a654b9d8 | 268 | |
b2d2c4ce | 269 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ac2f40be N |
270 | int size = PAGE_SIZE; |
271 | loff_t offset = mddev->bitmap_info.offset; | |
272 | if (page->index == bitmap->file_pages-1) | |
273 | size = roundup(bitmap->last_page_size, | |
274 | bdev_logical_block_size(rdev->bdev)); | |
275 | /* Just make sure we aren't corrupting data or | |
276 | * metadata | |
277 | */ | |
278 | if (mddev->external) { | |
279 | /* Bitmap could be anywhere. */ | |
280 | if (rdev->sb_start + offset + (page->index | |
281 | * (PAGE_SIZE/512)) | |
282 | > rdev->data_offset | |
283 | && | |
284 | rdev->sb_start + offset | |
285 | < (rdev->data_offset + mddev->dev_sectors | |
286 | + (PAGE_SIZE/512))) | |
287 | goto bad_alignment; | |
288 | } else if (offset < 0) { | |
289 | /* DATA BITMAP METADATA */ | |
290 | if (offset | |
291 | + (long)(page->index * (PAGE_SIZE/512)) | |
292 | + size/512 > 0) | |
293 | /* bitmap runs in to metadata */ | |
294 | goto bad_alignment; | |
295 | if (rdev->data_offset + mddev->dev_sectors | |
296 | > rdev->sb_start + offset) | |
297 | /* data runs in to bitmap */ | |
298 | goto bad_alignment; | |
299 | } else if (rdev->sb_start < rdev->data_offset) { | |
300 | /* METADATA BITMAP DATA */ | |
301 | if (rdev->sb_start | |
302 | + offset | |
303 | + page->index*(PAGE_SIZE/512) + size/512 | |
304 | > rdev->data_offset) | |
305 | /* bitmap runs in to data */ | |
306 | goto bad_alignment; | |
307 | } else { | |
308 | /* DATA METADATA BITMAP - no problems */ | |
309 | } | |
310 | md_super_write(mddev, rdev, | |
311 | rdev->sb_start + offset | |
312 | + page->index * (PAGE_SIZE/512), | |
313 | size, | |
314 | page); | |
b2d2c4ce | 315 | } |
a654b9d8 N |
316 | |
317 | if (wait) | |
a9701a30 | 318 | md_super_wait(mddev); |
a654b9d8 | 319 | return 0; |
4b80991c N |
320 | |
321 | bad_alignment: | |
4b80991c | 322 | return -EINVAL; |
a654b9d8 N |
323 | } |
324 | ||
4ad13663 | 325 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 326 | /* |
a654b9d8 | 327 | * write out a page to a file |
32a7627c | 328 | */ |
4ad13663 | 329 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 330 | { |
d785a06a | 331 | struct buffer_head *bh; |
32a7627c | 332 | |
f0d76d70 N |
333 | if (bitmap->file == NULL) { |
334 | switch (write_sb_page(bitmap, page, wait)) { | |
335 | case -EINVAL: | |
336 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
f0d76d70 | 337 | } |
4ad13663 | 338 | } else { |
a654b9d8 | 339 | |
4ad13663 | 340 | bh = page_buffers(page); |
c708443c | 341 | |
4ad13663 N |
342 | while (bh && bh->b_blocknr) { |
343 | atomic_inc(&bitmap->pending_writes); | |
344 | set_buffer_locked(bh); | |
345 | set_buffer_mapped(bh); | |
346 | submit_bh(WRITE, bh); | |
347 | bh = bh->b_this_page; | |
348 | } | |
d785a06a | 349 | |
ac2f40be | 350 | if (wait) |
4ad13663 N |
351 | wait_event(bitmap->write_wait, |
352 | atomic_read(&bitmap->pending_writes)==0); | |
8a5e9cf1 | 353 | } |
4ad13663 N |
354 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
355 | bitmap_file_kick(bitmap); | |
d785a06a N |
356 | } |
357 | ||
358 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
359 | { | |
360 | struct bitmap *bitmap = bh->b_private; | |
361 | unsigned long flags; | |
32a7627c | 362 | |
d785a06a N |
363 | if (!uptodate) { |
364 | spin_lock_irqsave(&bitmap->lock, flags); | |
365 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
366 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
32a7627c | 367 | } |
d785a06a N |
368 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
369 | wake_up(&bitmap->write_wait); | |
370 | } | |
32a7627c | 371 | |
d785a06a N |
372 | /* copied from buffer.c */ |
373 | static void | |
374 | __clear_page_buffers(struct page *page) | |
375 | { | |
376 | ClearPagePrivate(page); | |
377 | set_page_private(page, 0); | |
378 | page_cache_release(page); | |
379 | } | |
380 | static void free_buffers(struct page *page) | |
381 | { | |
382 | struct buffer_head *bh = page_buffers(page); | |
77ad4bc7 | 383 | |
d785a06a N |
384 | while (bh) { |
385 | struct buffer_head *next = bh->b_this_page; | |
386 | free_buffer_head(bh); | |
387 | bh = next; | |
77ad4bc7 | 388 | } |
d785a06a N |
389 | __clear_page_buffers(page); |
390 | put_page(page); | |
32a7627c N |
391 | } |
392 | ||
d785a06a N |
393 | /* read a page from a file. |
394 | * We both read the page, and attach buffers to the page to record the | |
395 | * address of each block (using bmap). These addresses will be used | |
396 | * to write the block later, completely bypassing the filesystem. | |
397 | * This usage is similar to how swap files are handled, and allows us | |
398 | * to write to a file with no concerns of memory allocation failing. | |
399 | */ | |
32a7627c | 400 | static struct page *read_page(struct file *file, unsigned long index, |
d785a06a N |
401 | struct bitmap *bitmap, |
402 | unsigned long count) | |
32a7627c | 403 | { |
32a7627c | 404 | struct page *page = NULL; |
c649bb9c | 405 | struct inode *inode = file->f_path.dentry->d_inode; |
d785a06a N |
406 | struct buffer_head *bh; |
407 | sector_t block; | |
32a7627c | 408 | |
ac2f40be | 409 | PRINTK("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
2d1f3b5d | 410 | (unsigned long long)index << PAGE_SHIFT); |
32a7627c | 411 | |
d785a06a N |
412 | page = alloc_page(GFP_KERNEL); |
413 | if (!page) | |
414 | page = ERR_PTR(-ENOMEM); | |
32a7627c N |
415 | if (IS_ERR(page)) |
416 | goto out; | |
d785a06a | 417 | |
d785a06a N |
418 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
419 | if (!bh) { | |
2d1f3b5d | 420 | put_page(page); |
d785a06a | 421 | page = ERR_PTR(-ENOMEM); |
32a7627c N |
422 | goto out; |
423 | } | |
d785a06a N |
424 | attach_page_buffers(page, bh); |
425 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
426 | while (bh) { | |
427 | if (count == 0) | |
428 | bh->b_blocknr = 0; | |
429 | else { | |
430 | bh->b_blocknr = bmap(inode, block); | |
431 | if (bh->b_blocknr == 0) { | |
432 | /* Cannot use this file! */ | |
433 | free_buffers(page); | |
434 | page = ERR_PTR(-EINVAL); | |
435 | goto out; | |
436 | } | |
437 | bh->b_bdev = inode->i_sb->s_bdev; | |
438 | if (count < (1<<inode->i_blkbits)) | |
439 | count = 0; | |
440 | else | |
441 | count -= (1<<inode->i_blkbits); | |
442 | ||
443 | bh->b_end_io = end_bitmap_write; | |
444 | bh->b_private = bitmap; | |
ce25c31b N |
445 | atomic_inc(&bitmap->pending_writes); |
446 | set_buffer_locked(bh); | |
447 | set_buffer_mapped(bh); | |
448 | submit_bh(READ, bh); | |
d785a06a N |
449 | } |
450 | block++; | |
451 | bh = bh->b_this_page; | |
452 | } | |
d785a06a | 453 | page->index = index; |
ce25c31b N |
454 | |
455 | wait_event(bitmap->write_wait, | |
456 | atomic_read(&bitmap->pending_writes)==0); | |
457 | if (bitmap->flags & BITMAP_WRITE_ERROR) { | |
458 | free_buffers(page); | |
459 | page = ERR_PTR(-EIO); | |
460 | } | |
32a7627c N |
461 | out: |
462 | if (IS_ERR(page)) | |
ac2f40be | 463 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %ld\n", |
2d1f3b5d N |
464 | (int)PAGE_SIZE, |
465 | (unsigned long long)index << PAGE_SHIFT, | |
32a7627c N |
466 | PTR_ERR(page)); |
467 | return page; | |
468 | } | |
469 | ||
470 | /* | |
471 | * bitmap file superblock operations | |
472 | */ | |
473 | ||
474 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 475 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
476 | { |
477 | bitmap_super_t *sb; | |
478 | unsigned long flags; | |
479 | ||
480 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 481 | return; |
ece5cff0 N |
482 | if (bitmap->mddev->bitmap_info.external) |
483 | return; | |
32a7627c N |
484 | spin_lock_irqsave(&bitmap->lock, flags); |
485 | if (!bitmap->sb_page) { /* no superblock */ | |
486 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 487 | return; |
32a7627c | 488 | } |
32a7627c | 489 | spin_unlock_irqrestore(&bitmap->lock, flags); |
7b92813c | 490 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 491 | sb->events = cpu_to_le64(bitmap->mddev->events); |
a0da84f3 NB |
492 | if (bitmap->mddev->events < bitmap->events_cleared) { |
493 | /* rocking back to read-only */ | |
494 | bitmap->events_cleared = bitmap->mddev->events; | |
495 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); | |
496 | } | |
43a70507 N |
497 | /* Just in case these have been changed via sysfs: */ |
498 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
499 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
ea03aff9 | 500 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 501 | write_page(bitmap, bitmap->sb_page, 1); |
32a7627c N |
502 | } |
503 | ||
504 | /* print out the bitmap file superblock */ | |
505 | void bitmap_print_sb(struct bitmap *bitmap) | |
506 | { | |
507 | bitmap_super_t *sb; | |
508 | ||
509 | if (!bitmap || !bitmap->sb_page) | |
510 | return; | |
7b92813c | 511 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 512 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
513 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
514 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
515 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
516 | *(__u32 *)(sb->uuid+0), |
517 | *(__u32 *)(sb->uuid+4), | |
518 | *(__u32 *)(sb->uuid+8), | |
519 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 520 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 521 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 522 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 523 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
524 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
525 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
526 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
527 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
528 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 529 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
ea03aff9 | 530 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
531 | } |
532 | ||
533 | /* read the superblock from the bitmap file and initialize some bitmap fields */ | |
534 | static int bitmap_read_sb(struct bitmap *bitmap) | |
535 | { | |
536 | char *reason = NULL; | |
537 | bitmap_super_t *sb; | |
4b6d287f | 538 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c N |
539 | unsigned long long events; |
540 | int err = -EINVAL; | |
541 | ||
542 | /* page 0 is the superblock, read it... */ | |
f49d5e62 N |
543 | if (bitmap->file) { |
544 | loff_t isize = i_size_read(bitmap->file->f_mapping->host); | |
545 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; | |
546 | ||
547 | bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes); | |
548 | } else { | |
42a04b50 N |
549 | bitmap->sb_page = read_sb_page(bitmap->mddev, |
550 | bitmap->mddev->bitmap_info.offset, | |
a2ed9615 N |
551 | NULL, |
552 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 553 | } |
32a7627c N |
554 | if (IS_ERR(bitmap->sb_page)) { |
555 | err = PTR_ERR(bitmap->sb_page); | |
556 | bitmap->sb_page = NULL; | |
557 | return err; | |
558 | } | |
559 | ||
7b92813c | 560 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 561 | |
32a7627c | 562 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 563 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 564 | write_behind = le32_to_cpu(sb->write_behind); |
32a7627c N |
565 | |
566 | /* verify that the bitmap-specific fields are valid */ | |
567 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
568 | reason = "bad magic"; | |
bd926c63 N |
569 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
570 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 571 | reason = "unrecognized superblock version"; |
1187cf0a | 572 | else if (chunksize < 512) |
7dd5d34c | 573 | reason = "bitmap chunksize too small"; |
32a7627c N |
574 | else if ((1 << ffz(~chunksize)) != chunksize) |
575 | reason = "bitmap chunksize not a power of 2"; | |
1b04be96 | 576 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 577 | reason = "daemon sleep period out of range"; |
4b6d287f N |
578 | else if (write_behind > COUNTER_MAX) |
579 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
580 | if (reason) { |
581 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
582 | bmname(bitmap), reason); | |
583 | goto out; | |
584 | } | |
585 | ||
586 | /* keep the array size field of the bitmap superblock up to date */ | |
587 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
588 | ||
589 | if (!bitmap->mddev->persistent) | |
590 | goto success; | |
591 | ||
592 | /* | |
593 | * if we have a persistent array superblock, compare the | |
594 | * bitmap's UUID and event counter to the mddev's | |
595 | */ | |
596 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
597 | printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n", | |
598 | bmname(bitmap)); | |
599 | goto out; | |
600 | } | |
601 | events = le64_to_cpu(sb->events); | |
602 | if (events < bitmap->mddev->events) { | |
603 | printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) " | |
604 | "-- forcing full recovery\n", bmname(bitmap), events, | |
605 | (unsigned long long) bitmap->mddev->events); | |
4f2e639a | 606 | sb->state |= cpu_to_le32(BITMAP_STALE); |
32a7627c N |
607 | } |
608 | success: | |
609 | /* assign fields using values from superblock */ | |
42a04b50 N |
610 | bitmap->mddev->bitmap_info.chunksize = chunksize; |
611 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
42a04b50 | 612 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; |
4f2e639a | 613 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 N |
614 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
615 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
32a7627c | 616 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
4f2e639a | 617 | if (sb->state & cpu_to_le32(BITMAP_STALE)) |
6a07997f | 618 | bitmap->events_cleared = bitmap->mddev->events; |
32a7627c N |
619 | err = 0; |
620 | out: | |
ea03aff9 | 621 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
622 | if (err) |
623 | bitmap_print_sb(bitmap); | |
624 | return err; | |
625 | } | |
626 | ||
627 | enum bitmap_mask_op { | |
628 | MASK_SET, | |
629 | MASK_UNSET | |
630 | }; | |
631 | ||
4ad13663 N |
632 | /* record the state of the bitmap in the superblock. Return the old value */ |
633 | static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits, | |
634 | enum bitmap_mask_op op) | |
32a7627c N |
635 | { |
636 | bitmap_super_t *sb; | |
637 | unsigned long flags; | |
4ad13663 | 638 | int old; |
32a7627c N |
639 | |
640 | spin_lock_irqsave(&bitmap->lock, flags); | |
7e317655 | 641 | if (!bitmap->sb_page) { /* can't set the state */ |
32a7627c | 642 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 643 | return 0; |
32a7627c | 644 | } |
32a7627c | 645 | spin_unlock_irqrestore(&bitmap->lock, flags); |
7b92813c | 646 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); |
4ad13663 | 647 | old = le32_to_cpu(sb->state) & bits; |
32a7627c | 648 | switch (op) { |
ac2f40be N |
649 | case MASK_SET: |
650 | sb->state |= cpu_to_le32(bits); | |
651 | break; | |
652 | case MASK_UNSET: | |
653 | sb->state &= cpu_to_le32(~bits); | |
654 | break; | |
655 | default: | |
656 | BUG(); | |
32a7627c | 657 | } |
ea03aff9 | 658 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 659 | return old; |
32a7627c N |
660 | } |
661 | ||
662 | /* | |
663 | * general bitmap file operations | |
664 | */ | |
665 | ||
ece5cff0 N |
666 | /* |
667 | * on-disk bitmap: | |
668 | * | |
669 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
670 | * file a page at a time. There's a superblock at the start of the file. | |
671 | */ | |
32a7627c | 672 | /* calculate the index of the page that contains this bit */ |
ece5cff0 | 673 | static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk) |
32a7627c | 674 | { |
ece5cff0 N |
675 | if (!bitmap->mddev->bitmap_info.external) |
676 | chunk += sizeof(bitmap_super_t) << 3; | |
677 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
678 | } |
679 | ||
680 | /* calculate the (bit) offset of this bit within a page */ | |
ece5cff0 | 681 | static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk) |
32a7627c | 682 | { |
ece5cff0 N |
683 | if (!bitmap->mddev->bitmap_info.external) |
684 | chunk += sizeof(bitmap_super_t) << 3; | |
685 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
686 | } |
687 | ||
688 | /* | |
689 | * return a pointer to the page in the filemap that contains the given bit | |
690 | * | |
691 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
692 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
693 | * 0 or page 1 | |
694 | */ | |
695 | static inline struct page *filemap_get_page(struct bitmap *bitmap, | |
696 | unsigned long chunk) | |
697 | { | |
e384e585 N |
698 | if (bitmap->filemap == NULL) |
699 | return NULL; | |
ac2f40be N |
700 | if (file_page_index(bitmap, chunk) >= bitmap->file_pages) |
701 | return NULL; | |
ece5cff0 N |
702 | return bitmap->filemap[file_page_index(bitmap, chunk) |
703 | - file_page_index(bitmap, 0)]; | |
32a7627c N |
704 | } |
705 | ||
32a7627c N |
706 | static void bitmap_file_unmap(struct bitmap *bitmap) |
707 | { | |
708 | struct page **map, *sb_page; | |
709 | unsigned long *attr; | |
710 | int pages; | |
711 | unsigned long flags; | |
712 | ||
713 | spin_lock_irqsave(&bitmap->lock, flags); | |
714 | map = bitmap->filemap; | |
715 | bitmap->filemap = NULL; | |
716 | attr = bitmap->filemap_attr; | |
717 | bitmap->filemap_attr = NULL; | |
718 | pages = bitmap->file_pages; | |
719 | bitmap->file_pages = 0; | |
720 | sb_page = bitmap->sb_page; | |
721 | bitmap->sb_page = NULL; | |
722 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
723 | ||
724 | while (pages--) | |
ece5cff0 | 725 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 726 | free_buffers(map[pages]); |
32a7627c N |
727 | kfree(map); |
728 | kfree(attr); | |
729 | ||
d785a06a N |
730 | if (sb_page) |
731 | free_buffers(sb_page); | |
32a7627c N |
732 | } |
733 | ||
734 | static void bitmap_file_put(struct bitmap *bitmap) | |
735 | { | |
736 | struct file *file; | |
32a7627c N |
737 | unsigned long flags; |
738 | ||
739 | spin_lock_irqsave(&bitmap->lock, flags); | |
740 | file = bitmap->file; | |
741 | bitmap->file = NULL; | |
742 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
743 | ||
d785a06a N |
744 | if (file) |
745 | wait_event(bitmap->write_wait, | |
746 | atomic_read(&bitmap->pending_writes)==0); | |
32a7627c N |
747 | bitmap_file_unmap(bitmap); |
748 | ||
d785a06a | 749 | if (file) { |
c649bb9c | 750 | struct inode *inode = file->f_path.dentry->d_inode; |
fc0ecff6 | 751 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 752 | fput(file); |
d785a06a | 753 | } |
32a7627c N |
754 | } |
755 | ||
32a7627c N |
756 | /* |
757 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
758 | * then it is no longer reliable, so we stop using it and we mark the file | |
759 | * as failed in the superblock | |
760 | */ | |
761 | static void bitmap_file_kick(struct bitmap *bitmap) | |
762 | { | |
763 | char *path, *ptr = NULL; | |
764 | ||
4ad13663 N |
765 | if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) { |
766 | bitmap_update_sb(bitmap); | |
32a7627c | 767 | |
4ad13663 N |
768 | if (bitmap->file) { |
769 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
770 | if (path) | |
6bcfd601 CH |
771 | ptr = d_path(&bitmap->file->f_path, path, |
772 | PAGE_SIZE); | |
773 | ||
4ad13663 N |
774 | printk(KERN_ALERT |
775 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 776 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 777 | |
4ad13663 N |
778 | kfree(path); |
779 | } else | |
780 | printk(KERN_ALERT | |
781 | "%s: disabling internal bitmap due to errors\n", | |
782 | bmname(bitmap)); | |
a654b9d8 | 783 | } |
32a7627c N |
784 | |
785 | bitmap_file_put(bitmap); | |
786 | ||
787 | return; | |
788 | } | |
789 | ||
790 | enum bitmap_page_attr { | |
ac2f40be N |
791 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
792 | BITMAP_PAGE_CLEAN = 1, /* there are bits that might need to be cleared */ | |
793 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ | |
32a7627c N |
794 | }; |
795 | ||
796 | static inline void set_page_attr(struct bitmap *bitmap, struct page *page, | |
797 | enum bitmap_page_attr attr) | |
798 | { | |
e384e585 N |
799 | if (page) |
800 | __set_bit((page->index<<2) + attr, bitmap->filemap_attr); | |
801 | else | |
802 | __set_bit(attr, &bitmap->logattrs); | |
32a7627c N |
803 | } |
804 | ||
805 | static inline void clear_page_attr(struct bitmap *bitmap, struct page *page, | |
806 | enum bitmap_page_attr attr) | |
807 | { | |
e384e585 N |
808 | if (page) |
809 | __clear_bit((page->index<<2) + attr, bitmap->filemap_attr); | |
810 | else | |
811 | __clear_bit(attr, &bitmap->logattrs); | |
32a7627c N |
812 | } |
813 | ||
ec7a3197 N |
814 | static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page, |
815 | enum bitmap_page_attr attr) | |
32a7627c | 816 | { |
e384e585 N |
817 | if (page) |
818 | return test_bit((page->index<<2) + attr, bitmap->filemap_attr); | |
819 | else | |
820 | return test_bit(attr, &bitmap->logattrs); | |
32a7627c N |
821 | } |
822 | ||
823 | /* | |
824 | * bitmap_file_set_bit -- called before performing a write to the md device | |
825 | * to set (and eventually sync) a particular bit in the bitmap file | |
826 | * | |
827 | * we set the bit immediately, then we record the page number so that | |
828 | * when an unplug occurs, we can flush the dirty pages out to disk | |
829 | */ | |
830 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
831 | { | |
832 | unsigned long bit; | |
e384e585 | 833 | struct page *page = NULL; |
32a7627c N |
834 | void *kaddr; |
835 | unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap); | |
836 | ||
e384e585 N |
837 | if (!bitmap->filemap) { |
838 | struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log; | |
839 | if (log) | |
840 | log->type->mark_region(log, chunk); | |
841 | } else { | |
32a7627c | 842 | |
e384e585 N |
843 | page = filemap_get_page(bitmap, chunk); |
844 | if (!page) | |
845 | return; | |
846 | bit = file_page_offset(bitmap, chunk); | |
32a7627c | 847 | |
e384e585 N |
848 | /* set the bit */ |
849 | kaddr = kmap_atomic(page, KM_USER0); | |
850 | if (bitmap->flags & BITMAP_HOSTENDIAN) | |
851 | set_bit(bit, kaddr); | |
852 | else | |
853 | ext2_set_bit(bit, kaddr); | |
854 | kunmap_atomic(kaddr, KM_USER0); | |
855 | PRINTK("set file bit %lu page %lu\n", bit, page->index); | |
856 | } | |
32a7627c N |
857 | /* record page number so it gets flushed to disk when unplug occurs */ |
858 | set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); | |
32a7627c N |
859 | } |
860 | ||
861 | /* this gets called when the md device is ready to unplug its underlying | |
862 | * (slave) device queues -- before we let any writes go down, we need to | |
863 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 864 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 865 | { |
ec7a3197 N |
866 | unsigned long i, flags; |
867 | int dirty, need_write; | |
32a7627c N |
868 | struct page *page; |
869 | int wait = 0; | |
870 | ||
871 | if (!bitmap) | |
4ad13663 | 872 | return; |
e384e585 N |
873 | if (!bitmap->filemap) { |
874 | /* Must be using a dirty_log */ | |
875 | struct dm_dirty_log *log = bitmap->mddev->bitmap_info.log; | |
876 | dirty = test_and_clear_bit(BITMAP_PAGE_DIRTY, &bitmap->logattrs); | |
877 | need_write = test_and_clear_bit(BITMAP_PAGE_NEEDWRITE, &bitmap->logattrs); | |
878 | if (dirty || need_write) | |
879 | if (log->type->flush(log)) | |
880 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
881 | goto out; | |
882 | } | |
32a7627c N |
883 | |
884 | /* look at each page to see if there are any set bits that need to be | |
885 | * flushed out to disk */ | |
886 | for (i = 0; i < bitmap->file_pages; i++) { | |
887 | spin_lock_irqsave(&bitmap->lock, flags); | |
a654b9d8 | 888 | if (!bitmap->filemap) { |
32a7627c | 889 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 890 | return; |
32a7627c N |
891 | } |
892 | page = bitmap->filemap[i]; | |
ec7a3197 N |
893 | dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
894 | need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
32a7627c N |
895 | clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
896 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
ec7a3197 | 897 | if (dirty) |
32a7627c N |
898 | wait = 1; |
899 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
900 | ||
ac2f40be | 901 | if (dirty || need_write) |
4ad13663 | 902 | write_page(bitmap, page, 0); |
32a7627c N |
903 | } |
904 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
0b79ccf0 | 905 | if (bitmap->file) |
d785a06a N |
906 | wait_event(bitmap->write_wait, |
907 | atomic_read(&bitmap->pending_writes)==0); | |
0b79ccf0 | 908 | else |
a9701a30 | 909 | md_super_wait(bitmap->mddev); |
32a7627c | 910 | } |
e384e585 | 911 | out: |
d785a06a N |
912 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
913 | bitmap_file_kick(bitmap); | |
32a7627c | 914 | } |
ac2f40be | 915 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 916 | |
6a07997f | 917 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
918 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
919 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
920 | * memory mapping of the bitmap file | |
921 | * Special cases: | |
922 | * if there's no bitmap file, or if the bitmap file had been | |
923 | * previously kicked from the array, we mark all the bits as | |
924 | * 1's in order to cause a full resync. | |
6a07997f N |
925 | * |
926 | * We ignore all bits for sectors that end earlier than 'start'. | |
927 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 928 | */ |
6a07997f | 929 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
930 | { |
931 | unsigned long i, chunks, index, oldindex, bit; | |
932 | struct page *page = NULL, *oldpage = NULL; | |
933 | unsigned long num_pages, bit_cnt = 0; | |
934 | struct file *file; | |
d785a06a | 935 | unsigned long bytes, offset; |
32a7627c N |
936 | int outofdate; |
937 | int ret = -ENOSPC; | |
ea03aff9 | 938 | void *paddr; |
32a7627c N |
939 | |
940 | chunks = bitmap->chunks; | |
941 | file = bitmap->file; | |
942 | ||
42a04b50 | 943 | BUG_ON(!file && !bitmap->mddev->bitmap_info.offset); |
32a7627c | 944 | |
44456d37 | 945 | #ifdef INJECT_FAULTS_3 |
32a7627c N |
946 | outofdate = 1; |
947 | #else | |
948 | outofdate = bitmap->flags & BITMAP_STALE; | |
949 | #endif | |
950 | if (outofdate) | |
951 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
952 | "recovery\n", bmname(bitmap)); | |
953 | ||
e384e585 | 954 | bytes = DIV_ROUND_UP(bitmap->chunks, 8); |
ece5cff0 N |
955 | if (!bitmap->mddev->bitmap_info.external) |
956 | bytes += sizeof(bitmap_super_t); | |
bc7f77de | 957 | |
e384e585 | 958 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); |
bc7f77de | 959 | |
ece5cff0 | 960 | if (file && i_size_read(file->f_mapping->host) < bytes) { |
32a7627c N |
961 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
962 | bmname(bitmap), | |
963 | (unsigned long) i_size_read(file->f_mapping->host), | |
ece5cff0 | 964 | bytes); |
4ad13663 | 965 | goto err; |
32a7627c | 966 | } |
bc7f77de N |
967 | |
968 | ret = -ENOMEM; | |
969 | ||
32a7627c | 970 | bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL); |
bc7f77de | 971 | if (!bitmap->filemap) |
4ad13663 | 972 | goto err; |
32a7627c | 973 | |
e16b68b6 N |
974 | /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */ |
975 | bitmap->filemap_attr = kzalloc( | |
ac2f40be | 976 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), |
e16b68b6 | 977 | GFP_KERNEL); |
bc7f77de | 978 | if (!bitmap->filemap_attr) |
4ad13663 | 979 | goto err; |
32a7627c | 980 | |
32a7627c N |
981 | oldindex = ~0L; |
982 | ||
983 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 984 | int b; |
ece5cff0 N |
985 | index = file_page_index(bitmap, i); |
986 | bit = file_page_offset(bitmap, i); | |
32a7627c | 987 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 988 | int count; |
32a7627c | 989 | /* unmap the old page, we're done with it */ |
d785a06a | 990 | if (index == num_pages-1) |
ece5cff0 | 991 | count = bytes - index * PAGE_SIZE; |
d785a06a N |
992 | else |
993 | count = PAGE_SIZE; | |
ece5cff0 | 994 | if (index == 0 && bitmap->sb_page) { |
32a7627c N |
995 | /* |
996 | * if we're here then the superblock page | |
997 | * contains some bits (PAGE_SIZE != sizeof sb) | |
998 | * we've already read it in, so just use it | |
999 | */ | |
1000 | page = bitmap->sb_page; | |
1001 | offset = sizeof(bitmap_super_t); | |
53845270 N |
1002 | if (!file) |
1003 | read_sb_page(bitmap->mddev, | |
42a04b50 | 1004 | bitmap->mddev->bitmap_info.offset, |
53845270 N |
1005 | page, |
1006 | index, count); | |
a654b9d8 | 1007 | } else if (file) { |
d785a06a | 1008 | page = read_page(file, index, bitmap, count); |
a654b9d8 N |
1009 | offset = 0; |
1010 | } else { | |
42a04b50 N |
1011 | page = read_sb_page(bitmap->mddev, |
1012 | bitmap->mddev->bitmap_info.offset, | |
a2ed9615 N |
1013 | NULL, |
1014 | index, count); | |
32a7627c N |
1015 | offset = 0; |
1016 | } | |
a654b9d8 N |
1017 | if (IS_ERR(page)) { /* read error */ |
1018 | ret = PTR_ERR(page); | |
4ad13663 | 1019 | goto err; |
a654b9d8 N |
1020 | } |
1021 | ||
32a7627c N |
1022 | oldindex = index; |
1023 | oldpage = page; | |
32a7627c | 1024 | |
b74fd282 N |
1025 | bitmap->filemap[bitmap->file_pages++] = page; |
1026 | bitmap->last_page_size = count; | |
1027 | ||
32a7627c N |
1028 | if (outofdate) { |
1029 | /* | |
1030 | * if bitmap is out of date, dirty the | |
ac2f40be | 1031 | * whole page and write it out |
32a7627c | 1032 | */ |
ea03aff9 N |
1033 | paddr = kmap_atomic(page, KM_USER0); |
1034 | memset(paddr + offset, 0xff, | |
6a07997f | 1035 | PAGE_SIZE - offset); |
ea03aff9 | 1036 | kunmap_atomic(paddr, KM_USER0); |
4ad13663 N |
1037 | write_page(bitmap, page, 1); |
1038 | ||
1039 | ret = -EIO; | |
b74fd282 | 1040 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
4ad13663 | 1041 | goto err; |
32a7627c | 1042 | } |
32a7627c | 1043 | } |
ea03aff9 | 1044 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1045 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1046 | b = test_bit(bit, paddr); |
bd926c63 | 1047 | else |
ea03aff9 N |
1048 | b = ext2_test_bit(bit, paddr); |
1049 | kunmap_atomic(paddr, KM_USER0); | |
bd926c63 | 1050 | if (b) { |
32a7627c | 1051 | /* if the disk bit is set, set the memory bit */ |
db305e50 N |
1052 | int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) |
1053 | >= start); | |
1054 | bitmap_set_memory_bits(bitmap, | |
1055 | (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap), | |
1056 | needed); | |
32a7627c | 1057 | bit_cnt++; |
6a07997f | 1058 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); |
32a7627c | 1059 | } |
32a7627c N |
1060 | } |
1061 | ||
ac2f40be | 1062 | /* everything went OK */ |
32a7627c N |
1063 | ret = 0; |
1064 | bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET); | |
1065 | ||
32a7627c N |
1066 | if (bit_cnt) { /* Kick recovery if any bits were set */ |
1067 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
1068 | md_wakeup_thread(bitmap->mddev->thread); | |
1069 | } | |
1070 | ||
32a7627c | 1071 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
4ad13663 N |
1072 | "read %lu/%lu pages, set %lu bits\n", |
1073 | bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt); | |
1074 | ||
1075 | return 0; | |
32a7627c | 1076 | |
4ad13663 N |
1077 | err: |
1078 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1079 | bmname(bitmap), ret); | |
32a7627c N |
1080 | return ret; |
1081 | } | |
1082 | ||
a654b9d8 N |
1083 | void bitmap_write_all(struct bitmap *bitmap) |
1084 | { | |
1085 | /* We don't actually write all bitmap blocks here, | |
1086 | * just flag them as needing to be written | |
1087 | */ | |
ec7a3197 | 1088 | int i; |
a654b9d8 | 1089 | |
ac2f40be | 1090 | for (i = 0; i < bitmap->file_pages; i++) |
ec7a3197 N |
1091 | set_page_attr(bitmap, bitmap->filemap[i], |
1092 | BITMAP_PAGE_NEEDWRITE); | |
a654b9d8 N |
1093 | } |
1094 | ||
32a7627c N |
1095 | static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) |
1096 | { | |
1097 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1098 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1099 | bitmap->bp[page].count += inc; | |
32a7627c N |
1100 | bitmap_checkfree(bitmap, page); |
1101 | } | |
1102 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, | |
1103 | sector_t offset, int *blocks, | |
1104 | int create); | |
1105 | ||
1106 | /* | |
1107 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1108 | * out to disk | |
1109 | */ | |
1110 | ||
aa5cbd10 | 1111 | void bitmap_daemon_work(mddev_t *mddev) |
32a7627c | 1112 | { |
aa5cbd10 | 1113 | struct bitmap *bitmap; |
aa3163f8 | 1114 | unsigned long j; |
32a7627c N |
1115 | unsigned long flags; |
1116 | struct page *page = NULL, *lastpage = NULL; | |
32a7627c | 1117 | int blocks; |
ea03aff9 | 1118 | void *paddr; |
e384e585 | 1119 | struct dm_dirty_log *log = mddev->bitmap_info.log; |
32a7627c | 1120 | |
aa5cbd10 N |
1121 | /* Use a mutex to guard daemon_work against |
1122 | * bitmap_destroy. | |
1123 | */ | |
c3d9714e | 1124 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1125 | bitmap = mddev->bitmap; |
1126 | if (bitmap == NULL) { | |
c3d9714e | 1127 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1128 | return; |
aa5cbd10 | 1129 | } |
42a04b50 | 1130 | if (time_before(jiffies, bitmap->daemon_lastrun |
1b04be96 | 1131 | + bitmap->mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1132 | goto done; |
1133 | ||
32a7627c | 1134 | bitmap->daemon_lastrun = jiffies; |
8311c29d N |
1135 | if (bitmap->allclean) { |
1136 | bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
aa5cbd10 | 1137 | goto done; |
8311c29d N |
1138 | } |
1139 | bitmap->allclean = 1; | |
32a7627c | 1140 | |
be512691 | 1141 | spin_lock_irqsave(&bitmap->lock, flags); |
32a7627c N |
1142 | for (j = 0; j < bitmap->chunks; j++) { |
1143 | bitmap_counter_t *bmc; | |
e384e585 N |
1144 | if (!bitmap->filemap) { |
1145 | if (!log) | |
1146 | /* error or shutdown */ | |
1147 | break; | |
1148 | } else | |
1149 | page = filemap_get_page(bitmap, j); | |
32a7627c N |
1150 | |
1151 | if (page != lastpage) { | |
aa3163f8 | 1152 | /* skip this page unless it's marked as needing cleaning */ |
ec7a3197 N |
1153 | if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) { |
1154 | int need_write = test_page_attr(bitmap, page, | |
1155 | BITMAP_PAGE_NEEDWRITE); | |
a647e4bc | 1156 | if (need_write) |
aa3163f8 | 1157 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); |
a647e4bc | 1158 | |
aa3163f8 | 1159 | spin_unlock_irqrestore(&bitmap->lock, flags); |
8311c29d | 1160 | if (need_write) { |
4ad13663 | 1161 | write_page(bitmap, page, 0); |
8311c29d N |
1162 | bitmap->allclean = 0; |
1163 | } | |
be512691 N |
1164 | spin_lock_irqsave(&bitmap->lock, flags); |
1165 | j |= (PAGE_BITS - 1); | |
aa3163f8 N |
1166 | continue; |
1167 | } | |
1168 | ||
32a7627c | 1169 | /* grab the new page, sync and release the old */ |
32a7627c | 1170 | if (lastpage != NULL) { |
ec7a3197 | 1171 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1172 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1173 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 1174 | write_page(bitmap, lastpage, 0); |
32a7627c N |
1175 | } else { |
1176 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1177 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1178 | } | |
32a7627c N |
1179 | } else |
1180 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1181 | lastpage = page; | |
a0da84f3 NB |
1182 | |
1183 | /* We are possibly going to clear some bits, so make | |
1184 | * sure that events_cleared is up-to-date. | |
1185 | */ | |
ece5cff0 N |
1186 | if (bitmap->need_sync && |
1187 | bitmap->mddev->bitmap_info.external == 0) { | |
a0da84f3 NB |
1188 | bitmap_super_t *sb; |
1189 | bitmap->need_sync = 0; | |
1190 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); | |
1191 | sb->events_cleared = | |
1192 | cpu_to_le64(bitmap->events_cleared); | |
1193 | kunmap_atomic(sb, KM_USER0); | |
1194 | write_page(bitmap, bitmap->sb_page, 1); | |
1195 | } | |
32a7627c | 1196 | spin_lock_irqsave(&bitmap->lock, flags); |
ece5cff0 N |
1197 | if (!bitmap->need_sync) |
1198 | clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
32a7627c | 1199 | } |
db305e50 N |
1200 | bmc = bitmap_get_counter(bitmap, |
1201 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
1202 | &blocks, 0); | |
32a7627c | 1203 | if (bmc) { |
8311c29d N |
1204 | if (*bmc) |
1205 | bitmap->allclean = 0; | |
1206 | ||
32a7627c | 1207 | if (*bmc == 2) { |
ac2f40be | 1208 | *bmc = 1; /* maybe clear the bit next time */ |
32a7627c | 1209 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); |
ece5cff0 | 1210 | } else if (*bmc == 1 && !bitmap->need_sync) { |
32a7627c N |
1211 | /* we can clear the bit */ |
1212 | *bmc = 0; | |
db305e50 N |
1213 | bitmap_count_page(bitmap, |
1214 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
32a7627c N |
1215 | -1); |
1216 | ||
1217 | /* clear the bit */ | |
e384e585 N |
1218 | if (page) { |
1219 | paddr = kmap_atomic(page, KM_USER0); | |
1220 | if (bitmap->flags & BITMAP_HOSTENDIAN) | |
1221 | clear_bit(file_page_offset(bitmap, j), | |
1222 | paddr); | |
1223 | else | |
1224 | ext2_clear_bit(file_page_offset(bitmap, j), | |
1225 | paddr); | |
1226 | kunmap_atomic(paddr, KM_USER0); | |
1227 | } else | |
1228 | log->type->clear_region(log, j); | |
32a7627c | 1229 | } |
be512691 N |
1230 | } else |
1231 | j |= PAGE_COUNTER_MASK; | |
32a7627c | 1232 | } |
be512691 | 1233 | spin_unlock_irqrestore(&bitmap->lock, flags); |
32a7627c N |
1234 | |
1235 | /* now sync the final page */ | |
e384e585 | 1236 | if (lastpage != NULL || log != NULL) { |
32a7627c | 1237 | spin_lock_irqsave(&bitmap->lock, flags); |
ec7a3197 | 1238 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1239 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1240 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
e384e585 N |
1241 | if (lastpage) |
1242 | write_page(bitmap, lastpage, 0); | |
1243 | else | |
1244 | if (log->type->flush(log)) | |
1245 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
32a7627c N |
1246 | } else { |
1247 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1248 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1249 | } | |
32a7627c N |
1250 | } |
1251 | ||
7be3dfec | 1252 | done: |
8311c29d | 1253 | if (bitmap->allclean == 0) |
ac2f40be | 1254 | bitmap->mddev->thread->timeout = |
1b04be96 | 1255 | bitmap->mddev->bitmap_info.daemon_sleep; |
c3d9714e | 1256 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1257 | } |
1258 | ||
32a7627c N |
1259 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
1260 | sector_t offset, int *blocks, | |
1261 | int create) | |
ee305ace N |
1262 | __releases(bitmap->lock) |
1263 | __acquires(bitmap->lock) | |
32a7627c N |
1264 | { |
1265 | /* If 'create', we might release the lock and reclaim it. | |
1266 | * The lock must have been taken with interrupts enabled. | |
1267 | * If !create, we don't release the lock. | |
1268 | */ | |
1269 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1270 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1271 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1272 | sector_t csize; | |
ef425673 | 1273 | int err; |
32a7627c | 1274 | |
ef425673 N |
1275 | err = bitmap_checkpage(bitmap, page, create); |
1276 | ||
1277 | if (bitmap->bp[page].hijacked || | |
1278 | bitmap->bp[page].map == NULL) | |
1279 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) + | |
1280 | PAGE_COUNTER_SHIFT - 1); | |
1281 | else | |
32a7627c | 1282 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); |
ef425673 N |
1283 | *blocks = csize - (offset & (csize - 1)); |
1284 | ||
1285 | if (err < 0) | |
32a7627c | 1286 | return NULL; |
ef425673 | 1287 | |
32a7627c N |
1288 | /* now locked ... */ |
1289 | ||
1290 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1291 | /* should we use the first or second counter field | |
1292 | * of the hijacked pointer? */ | |
1293 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1294 | return &((bitmap_counter_t *) |
1295 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1296 | } else /* page is allocated */ |
32a7627c N |
1297 | return (bitmap_counter_t *) |
1298 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1299 | } |
1300 | ||
4b6d287f | 1301 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1302 | { |
ac2f40be N |
1303 | if (!bitmap) |
1304 | return 0; | |
4b6d287f N |
1305 | |
1306 | if (behind) { | |
696fcd53 | 1307 | int bw; |
4b6d287f | 1308 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1309 | bw = atomic_read(&bitmap->behind_writes); |
1310 | if (bw > bitmap->behind_writes_used) | |
1311 | bitmap->behind_writes_used = bw; | |
1312 | ||
4b6d287f | 1313 | PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n", |
696fcd53 | 1314 | bw, bitmap->max_write_behind); |
4b6d287f N |
1315 | } |
1316 | ||
32a7627c N |
1317 | while (sectors) { |
1318 | int blocks; | |
1319 | bitmap_counter_t *bmc; | |
1320 | ||
1321 | spin_lock_irq(&bitmap->lock); | |
1322 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); | |
1323 | if (!bmc) { | |
1324 | spin_unlock_irq(&bitmap->lock); | |
1325 | return 0; | |
1326 | } | |
1327 | ||
da6e1a32 NB |
1328 | if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) { |
1329 | DEFINE_WAIT(__wait); | |
1330 | /* note that it is safe to do the prepare_to_wait | |
1331 | * after the test as long as we do it before dropping | |
1332 | * the spinlock. | |
1333 | */ | |
1334 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1335 | TASK_UNINTERRUPTIBLE); | |
1336 | spin_unlock_irq(&bitmap->lock); | |
b63d7c2e | 1337 | md_unplug(bitmap->mddev); |
da6e1a32 NB |
1338 | schedule(); |
1339 | finish_wait(&bitmap->overflow_wait, &__wait); | |
1340 | continue; | |
1341 | } | |
1342 | ||
ac2f40be | 1343 | switch (*bmc) { |
32a7627c N |
1344 | case 0: |
1345 | bitmap_file_set_bit(bitmap, offset); | |
ac2f40be | 1346 | bitmap_count_page(bitmap, offset, 1); |
32a7627c N |
1347 | /* fall through */ |
1348 | case 1: | |
1349 | *bmc = 2; | |
1350 | } | |
da6e1a32 | 1351 | |
32a7627c N |
1352 | (*bmc)++; |
1353 | ||
1354 | spin_unlock_irq(&bitmap->lock); | |
1355 | ||
1356 | offset += blocks; | |
1357 | if (sectors > blocks) | |
1358 | sectors -= blocks; | |
ac2f40be N |
1359 | else |
1360 | sectors = 0; | |
32a7627c | 1361 | } |
8311c29d | 1362 | bitmap->allclean = 0; |
32a7627c N |
1363 | return 0; |
1364 | } | |
ac2f40be | 1365 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1366 | |
1367 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1368 | int success, int behind) |
32a7627c | 1369 | { |
ac2f40be N |
1370 | if (!bitmap) |
1371 | return; | |
4b6d287f | 1372 | if (behind) { |
e555190d N |
1373 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1374 | wake_up(&bitmap->behind_wait); | |
4b6d287f N |
1375 | PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n", |
1376 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1377 | } | |
d0a4bb49 N |
1378 | if (bitmap->mddev->degraded) |
1379 | /* Never clear bits or update events_cleared when degraded */ | |
1380 | success = 0; | |
4b6d287f | 1381 | |
32a7627c N |
1382 | while (sectors) { |
1383 | int blocks; | |
1384 | unsigned long flags; | |
1385 | bitmap_counter_t *bmc; | |
1386 | ||
1387 | spin_lock_irqsave(&bitmap->lock, flags); | |
1388 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); | |
1389 | if (!bmc) { | |
1390 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1391 | return; | |
1392 | } | |
1393 | ||
a0da84f3 NB |
1394 | if (success && |
1395 | bitmap->events_cleared < bitmap->mddev->events) { | |
1396 | bitmap->events_cleared = bitmap->mddev->events; | |
1397 | bitmap->need_sync = 1; | |
5ff5afff | 1398 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1399 | } |
1400 | ||
32a7627c N |
1401 | if (!success && ! (*bmc & NEEDED_MASK)) |
1402 | *bmc |= NEEDED_MASK; | |
1403 | ||
da6e1a32 NB |
1404 | if ((*bmc & COUNTER_MAX) == COUNTER_MAX) |
1405 | wake_up(&bitmap->overflow_wait); | |
1406 | ||
32a7627c | 1407 | (*bmc)--; |
ac2f40be | 1408 | if (*bmc <= 2) |
32a7627c | 1409 | set_page_attr(bitmap, |
e384e585 N |
1410 | filemap_get_page( |
1411 | bitmap, | |
1412 | offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
32a7627c | 1413 | BITMAP_PAGE_CLEAN); |
ac2f40be | 1414 | |
32a7627c N |
1415 | spin_unlock_irqrestore(&bitmap->lock, flags); |
1416 | offset += blocks; | |
1417 | if (sectors > blocks) | |
1418 | sectors -= blocks; | |
ac2f40be N |
1419 | else |
1420 | sectors = 0; | |
32a7627c N |
1421 | } |
1422 | } | |
ac2f40be | 1423 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1424 | |
1187cf0a N |
1425 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, |
1426 | int degraded) | |
32a7627c N |
1427 | { |
1428 | bitmap_counter_t *bmc; | |
1429 | int rv; | |
1430 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1431 | *blocks = 1024; | |
1432 | return 1; /* always resync if no bitmap */ | |
1433 | } | |
1434 | spin_lock_irq(&bitmap->lock); | |
1435 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1436 | rv = 0; | |
1437 | if (bmc) { | |
1438 | /* locked */ | |
1439 | if (RESYNC(*bmc)) | |
1440 | rv = 1; | |
1441 | else if (NEEDED(*bmc)) { | |
1442 | rv = 1; | |
6a806c51 N |
1443 | if (!degraded) { /* don't set/clear bits if degraded */ |
1444 | *bmc |= RESYNC_MASK; | |
1445 | *bmc &= ~NEEDED_MASK; | |
1446 | } | |
32a7627c N |
1447 | } |
1448 | } | |
1449 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1450 | bitmap->allclean = 0; |
32a7627c N |
1451 | return rv; |
1452 | } | |
1453 | ||
1187cf0a N |
1454 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, |
1455 | int degraded) | |
1456 | { | |
1457 | /* bitmap_start_sync must always report on multiples of whole | |
1458 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1459 | * get confused. | |
1460 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1461 | * At least PAGE_SIZE>>9 blocks are covered. | |
1462 | * Return the 'or' of the result. | |
1463 | */ | |
1464 | int rv = 0; | |
1465 | int blocks1; | |
1466 | ||
1467 | *blocks = 0; | |
1468 | while (*blocks < (PAGE_SIZE>>9)) { | |
1469 | rv |= __bitmap_start_sync(bitmap, offset, | |
1470 | &blocks1, degraded); | |
1471 | offset += blocks1; | |
1472 | *blocks += blocks1; | |
1473 | } | |
1474 | return rv; | |
1475 | } | |
ac2f40be | 1476 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1477 | |
32a7627c N |
1478 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted) |
1479 | { | |
1480 | bitmap_counter_t *bmc; | |
1481 | unsigned long flags; | |
ac2f40be N |
1482 | |
1483 | if (bitmap == NULL) { | |
32a7627c N |
1484 | *blocks = 1024; |
1485 | return; | |
1486 | } | |
1487 | spin_lock_irqsave(&bitmap->lock, flags); | |
1488 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1489 | if (bmc == NULL) | |
1490 | goto unlock; | |
1491 | /* locked */ | |
32a7627c N |
1492 | if (RESYNC(*bmc)) { |
1493 | *bmc &= ~RESYNC_MASK; | |
1494 | ||
1495 | if (!NEEDED(*bmc) && aborted) | |
1496 | *bmc |= NEEDED_MASK; | |
1497 | else { | |
ac2f40be | 1498 | if (*bmc <= 2) |
32a7627c N |
1499 | set_page_attr(bitmap, |
1500 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1501 | BITMAP_PAGE_CLEAN); | |
32a7627c N |
1502 | } |
1503 | } | |
1504 | unlock: | |
1505 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
8311c29d | 1506 | bitmap->allclean = 0; |
32a7627c | 1507 | } |
ac2f40be | 1508 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1509 | |
1510 | void bitmap_close_sync(struct bitmap *bitmap) | |
1511 | { | |
1512 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1513 | * properly have been aborted. It remains to us to clear the | |
1514 | * RESYNC bit wherever it is still on | |
1515 | */ | |
1516 | sector_t sector = 0; | |
1517 | int blocks; | |
b47490c9 N |
1518 | if (!bitmap) |
1519 | return; | |
32a7627c N |
1520 | while (sector < bitmap->mddev->resync_max_sectors) { |
1521 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1522 | sector += blocks; |
1523 | } | |
1524 | } | |
ac2f40be | 1525 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 N |
1526 | |
1527 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1528 | { | |
1529 | sector_t s = 0; | |
1530 | int blocks; | |
1531 | ||
1532 | if (!bitmap) | |
1533 | return; | |
1534 | if (sector == 0) { | |
1535 | bitmap->last_end_sync = jiffies; | |
1536 | return; | |
1537 | } | |
1538 | if (time_before(jiffies, (bitmap->last_end_sync | |
1b04be96 | 1539 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1540 | return; |
1541 | wait_event(bitmap->mddev->recovery_wait, | |
1542 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1543 | ||
97e4f42d | 1544 | bitmap->mddev->curr_resync_completed = bitmap->mddev->curr_resync; |
070dc6dd | 1545 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
b47490c9 N |
1546 | sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1); |
1547 | s = 0; | |
1548 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1549 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1550 | s += blocks; | |
32a7627c | 1551 | } |
b47490c9 | 1552 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1553 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1554 | } |
ac2f40be | 1555 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1556 | |
6a07997f | 1557 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1558 | { |
1559 | /* For each chunk covered by any of these sectors, set the | |
193f1c93 | 1560 | * counter to 1 and set resync_needed. They should all |
32a7627c N |
1561 | * be 0 at this point |
1562 | */ | |
193f1c93 N |
1563 | |
1564 | int secs; | |
1565 | bitmap_counter_t *bmc; | |
1566 | spin_lock_irq(&bitmap->lock); | |
1567 | bmc = bitmap_get_counter(bitmap, offset, &secs, 1); | |
1568 | if (!bmc) { | |
32a7627c | 1569 | spin_unlock_irq(&bitmap->lock); |
193f1c93 | 1570 | return; |
32a7627c | 1571 | } |
ac2f40be | 1572 | if (!*bmc) { |
193f1c93 | 1573 | struct page *page; |
ac2f40be | 1574 | *bmc = 1 | (needed ? NEEDED_MASK : 0); |
193f1c93 N |
1575 | bitmap_count_page(bitmap, offset, 1); |
1576 | page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)); | |
1577 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1578 | } | |
1579 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1580 | bitmap->allclean = 0; |
32a7627c N |
1581 | } |
1582 | ||
9b1d1dac PC |
1583 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1584 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1585 | { | |
1586 | unsigned long chunk; | |
1587 | ||
1588 | for (chunk = s; chunk <= e; chunk++) { | |
db305e50 | 1589 | sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap); |
9b1d1dac PC |
1590 | bitmap_set_memory_bits(bitmap, sec, 1); |
1591 | bitmap_file_set_bit(bitmap, sec); | |
ffa23322 N |
1592 | if (sec < bitmap->mddev->recovery_cp) |
1593 | /* We are asserting that the array is dirty, | |
1594 | * so move the recovery_cp address back so | |
1595 | * that it is obvious that it is dirty | |
1596 | */ | |
1597 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1598 | } |
1599 | } | |
1600 | ||
6b8b3e8a N |
1601 | /* |
1602 | * flush out any pending updates | |
1603 | */ | |
1604 | void bitmap_flush(mddev_t *mddev) | |
1605 | { | |
1606 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1607 | long sleep; |
6b8b3e8a N |
1608 | |
1609 | if (!bitmap) /* there was no bitmap */ | |
1610 | return; | |
1611 | ||
1612 | /* run the daemon_work three time to ensure everything is flushed | |
1613 | * that can be | |
1614 | */ | |
1b04be96 | 1615 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1616 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1617 | bitmap_daemon_work(mddev); |
42a04b50 | 1618 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1619 | bitmap_daemon_work(mddev); |
42a04b50 | 1620 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1621 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1622 | bitmap_update_sb(bitmap); |
1623 | } | |
1624 | ||
32a7627c N |
1625 | /* |
1626 | * free memory that was allocated | |
1627 | */ | |
3178b0db | 1628 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1629 | { |
1630 | unsigned long k, pages; | |
1631 | struct bitmap_page *bp; | |
32a7627c N |
1632 | |
1633 | if (!bitmap) /* there was no bitmap */ | |
1634 | return; | |
1635 | ||
32a7627c N |
1636 | /* release the bitmap file and kill the daemon */ |
1637 | bitmap_file_put(bitmap); | |
1638 | ||
1639 | bp = bitmap->bp; | |
1640 | pages = bitmap->pages; | |
1641 | ||
1642 | /* free all allocated memory */ | |
1643 | ||
32a7627c N |
1644 | if (bp) /* deallocate the page memory */ |
1645 | for (k = 0; k < pages; k++) | |
1646 | if (bp[k].map && !bp[k].hijacked) | |
1647 | kfree(bp[k].map); | |
1648 | kfree(bp); | |
1649 | kfree(bitmap); | |
1650 | } | |
aa5cbd10 | 1651 | |
3178b0db N |
1652 | void bitmap_destroy(mddev_t *mddev) |
1653 | { | |
1654 | struct bitmap *bitmap = mddev->bitmap; | |
1655 | ||
1656 | if (!bitmap) /* there was no bitmap */ | |
1657 | return; | |
1658 | ||
c3d9714e | 1659 | mutex_lock(&mddev->bitmap_info.mutex); |
3178b0db | 1660 | mddev->bitmap = NULL; /* disconnect from the md device */ |
c3d9714e | 1661 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1662 | if (mddev->thread) |
1663 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db | 1664 | |
ece5cff0 N |
1665 | if (bitmap->sysfs_can_clear) |
1666 | sysfs_put(bitmap->sysfs_can_clear); | |
1667 | ||
3178b0db N |
1668 | bitmap_free(bitmap); |
1669 | } | |
32a7627c N |
1670 | |
1671 | /* | |
1672 | * initialize the bitmap structure | |
1673 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1674 | */ | |
1675 | int bitmap_create(mddev_t *mddev) | |
1676 | { | |
1677 | struct bitmap *bitmap; | |
1f593903 | 1678 | sector_t blocks = mddev->resync_max_sectors; |
32a7627c N |
1679 | unsigned long chunks; |
1680 | unsigned long pages; | |
c3d9714e | 1681 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1682 | int err; |
5ff5afff | 1683 | struct sysfs_dirent *bm = NULL; |
32a7627c | 1684 | |
5f6e3c83 | 1685 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1686 | |
e384e585 N |
1687 | if (!file |
1688 | && !mddev->bitmap_info.offset | |
1689 | && !mddev->bitmap_info.log) /* bitmap disabled, nothing to do */ | |
32a7627c N |
1690 | return 0; |
1691 | ||
c3d9714e | 1692 | BUG_ON(file && mddev->bitmap_info.offset); |
e384e585 | 1693 | BUG_ON(mddev->bitmap_info.offset && mddev->bitmap_info.log); |
a654b9d8 | 1694 | |
9ffae0cf | 1695 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1696 | if (!bitmap) |
1697 | return -ENOMEM; | |
1698 | ||
32a7627c | 1699 | spin_lock_init(&bitmap->lock); |
ce25c31b N |
1700 | atomic_set(&bitmap->pending_writes, 0); |
1701 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1702 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1703 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1704 | |
32a7627c | 1705 | bitmap->mddev = mddev; |
32a7627c | 1706 | |
5ff5afff N |
1707 | if (mddev->kobj.sd) |
1708 | bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap"); | |
ece5cff0 | 1709 | if (bm) { |
3ff195b0 | 1710 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear"); |
ece5cff0 N |
1711 | sysfs_put(bm); |
1712 | } else | |
1713 | bitmap->sysfs_can_clear = NULL; | |
1714 | ||
32a7627c | 1715 | bitmap->file = file; |
ce25c31b N |
1716 | if (file) { |
1717 | get_file(file); | |
ae8fa283 N |
1718 | /* As future accesses to this file will use bmap, |
1719 | * and bypass the page cache, we must sync the file | |
1720 | * first. | |
1721 | */ | |
8018ab05 | 1722 | vfs_fsync(file, 1); |
ce25c31b | 1723 | } |
42a04b50 | 1724 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
ece5cff0 N |
1725 | if (!mddev->bitmap_info.external) |
1726 | err = bitmap_read_sb(bitmap); | |
1727 | else { | |
1728 | err = 0; | |
1729 | if (mddev->bitmap_info.chunksize == 0 || | |
1730 | mddev->bitmap_info.daemon_sleep == 0) | |
1731 | /* chunksize and time_base need to be | |
1732 | * set first. */ | |
1733 | err = -EINVAL; | |
1734 | } | |
32a7627c | 1735 | if (err) |
3178b0db | 1736 | goto error; |
32a7627c | 1737 | |
624ce4f5 | 1738 | bitmap->daemon_lastrun = jiffies; |
42a04b50 | 1739 | bitmap->chunkshift = ffz(~mddev->bitmap_info.chunksize); |
32a7627c N |
1740 | |
1741 | /* now that chunksize and chunkshift are set, we can use these macros */ | |
ac2f40be | 1742 | chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >> |
1f593903 | 1743 | CHUNK_BLOCK_SHIFT(bitmap); |
ac2f40be | 1744 | pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; |
32a7627c N |
1745 | |
1746 | BUG_ON(!pages); | |
1747 | ||
1748 | bitmap->chunks = chunks; | |
1749 | bitmap->pages = pages; | |
1750 | bitmap->missing_pages = pages; | |
1751 | bitmap->counter_bits = COUNTER_BITS; | |
1752 | ||
1753 | bitmap->syncchunk = ~0UL; | |
1754 | ||
44456d37 | 1755 | #ifdef INJECT_FATAL_FAULT_1 |
32a7627c N |
1756 | bitmap->bp = NULL; |
1757 | #else | |
9ffae0cf | 1758 | bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); |
32a7627c | 1759 | #endif |
3178b0db | 1760 | err = -ENOMEM; |
32a7627c | 1761 | if (!bitmap->bp) |
3178b0db | 1762 | goto error; |
32a7627c | 1763 | |
69e51b44 N |
1764 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", |
1765 | pages, bmname(bitmap)); | |
1766 | ||
1767 | mddev->bitmap = bitmap; | |
1768 | ||
1769 | ||
1770 | return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0; | |
1771 | ||
1772 | error: | |
1773 | bitmap_free(bitmap); | |
1774 | return err; | |
1775 | } | |
1776 | ||
1777 | int bitmap_load(mddev_t *mddev) | |
1778 | { | |
1779 | int err = 0; | |
1780 | sector_t sector = 0; | |
1781 | struct bitmap *bitmap = mddev->bitmap; | |
1782 | ||
1783 | if (!bitmap) | |
1784 | goto out; | |
1785 | ||
1786 | /* Clear out old bitmap info first: Either there is none, or we | |
1787 | * are resuming after someone else has possibly changed things, | |
1788 | * so we should forget old cached info. | |
1789 | * All chunks should be clean, but some might need_sync. | |
1790 | */ | |
1791 | while (sector < mddev->resync_max_sectors) { | |
1792 | int blocks; | |
1793 | bitmap_start_sync(bitmap, sector, &blocks, 0); | |
1794 | sector += blocks; | |
1795 | } | |
1796 | bitmap_close_sync(bitmap); | |
1797 | ||
e384e585 N |
1798 | if (mddev->bitmap_info.log) { |
1799 | unsigned long i; | |
1800 | struct dm_dirty_log *log = mddev->bitmap_info.log; | |
1801 | for (i = 0; i < bitmap->chunks; i++) | |
1802 | if (!log->type->in_sync(log, i, 1)) | |
1803 | bitmap_set_memory_bits(bitmap, | |
1804 | (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap), | |
1805 | 1); | |
69e51b44 N |
1806 | } else { |
1807 | sector_t start = 0; | |
1808 | if (mddev->degraded == 0 | |
1809 | || bitmap->events_cleared == mddev->events) | |
1810 | /* no need to keep dirty bits to optimise a | |
1811 | * re-add of a missing device */ | |
1812 | start = mddev->recovery_cp; | |
193f1c93 | 1813 | |
69e51b44 N |
1814 | err = bitmap_init_from_disk(bitmap, start); |
1815 | } | |
32a7627c | 1816 | if (err) |
69e51b44 | 1817 | goto out; |
3178b0db | 1818 | |
1b04be96 | 1819 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1820 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1821 | |
4ad13663 N |
1822 | bitmap_update_sb(bitmap); |
1823 | ||
69e51b44 N |
1824 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
1825 | err = -EIO; | |
1826 | out: | |
3178b0db | 1827 | return err; |
32a7627c | 1828 | } |
69e51b44 | 1829 | EXPORT_SYMBOL_GPL(bitmap_load); |
32a7627c | 1830 | |
43a70507 N |
1831 | static ssize_t |
1832 | location_show(mddev_t *mddev, char *page) | |
1833 | { | |
1834 | ssize_t len; | |
ac2f40be | 1835 | if (mddev->bitmap_info.file) |
43a70507 | 1836 | len = sprintf(page, "file"); |
ac2f40be | 1837 | else if (mddev->bitmap_info.offset) |
43a70507 | 1838 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 1839 | else |
43a70507 N |
1840 | len = sprintf(page, "none"); |
1841 | len += sprintf(page+len, "\n"); | |
1842 | return len; | |
1843 | } | |
1844 | ||
1845 | static ssize_t | |
1846 | location_store(mddev_t *mddev, const char *buf, size_t len) | |
1847 | { | |
1848 | ||
1849 | if (mddev->pers) { | |
1850 | if (!mddev->pers->quiesce) | |
1851 | return -EBUSY; | |
1852 | if (mddev->recovery || mddev->sync_thread) | |
1853 | return -EBUSY; | |
1854 | } | |
1855 | ||
1856 | if (mddev->bitmap || mddev->bitmap_info.file || | |
1857 | mddev->bitmap_info.offset) { | |
1858 | /* bitmap already configured. Only option is to clear it */ | |
1859 | if (strncmp(buf, "none", 4) != 0) | |
1860 | return -EBUSY; | |
1861 | if (mddev->pers) { | |
1862 | mddev->pers->quiesce(mddev, 1); | |
1863 | bitmap_destroy(mddev); | |
1864 | mddev->pers->quiesce(mddev, 0); | |
1865 | } | |
1866 | mddev->bitmap_info.offset = 0; | |
1867 | if (mddev->bitmap_info.file) { | |
1868 | struct file *f = mddev->bitmap_info.file; | |
1869 | mddev->bitmap_info.file = NULL; | |
1870 | restore_bitmap_write_access(f); | |
1871 | fput(f); | |
1872 | } | |
1873 | } else { | |
1874 | /* No bitmap, OK to set a location */ | |
1875 | long long offset; | |
1876 | if (strncmp(buf, "none", 4) == 0) | |
1877 | /* nothing to be done */; | |
1878 | else if (strncmp(buf, "file:", 5) == 0) { | |
1879 | /* Not supported yet */ | |
1880 | return -EINVAL; | |
1881 | } else { | |
1882 | int rv; | |
1883 | if (buf[0] == '+') | |
1884 | rv = strict_strtoll(buf+1, 10, &offset); | |
1885 | else | |
1886 | rv = strict_strtoll(buf, 10, &offset); | |
1887 | if (rv) | |
1888 | return rv; | |
1889 | if (offset == 0) | |
1890 | return -EINVAL; | |
ece5cff0 N |
1891 | if (mddev->bitmap_info.external == 0 && |
1892 | mddev->major_version == 0 && | |
43a70507 N |
1893 | offset != mddev->bitmap_info.default_offset) |
1894 | return -EINVAL; | |
1895 | mddev->bitmap_info.offset = offset; | |
1896 | if (mddev->pers) { | |
1897 | mddev->pers->quiesce(mddev, 1); | |
1898 | rv = bitmap_create(mddev); | |
1899 | if (rv) { | |
1900 | bitmap_destroy(mddev); | |
1901 | mddev->bitmap_info.offset = 0; | |
1902 | } | |
1903 | mddev->pers->quiesce(mddev, 0); | |
1904 | if (rv) | |
1905 | return rv; | |
1906 | } | |
1907 | } | |
1908 | } | |
1909 | if (!mddev->external) { | |
1910 | /* Ensure new bitmap info is stored in | |
1911 | * metadata promptly. | |
1912 | */ | |
1913 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1914 | md_wakeup_thread(mddev->thread); | |
1915 | } | |
1916 | return len; | |
1917 | } | |
1918 | ||
1919 | static struct md_sysfs_entry bitmap_location = | |
1920 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
1921 | ||
1922 | static ssize_t | |
1923 | timeout_show(mddev_t *mddev, char *page) | |
1924 | { | |
1925 | ssize_t len; | |
1926 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
1927 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 1928 | |
43a70507 N |
1929 | len = sprintf(page, "%lu", secs); |
1930 | if (jifs) | |
1931 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
1932 | len += sprintf(page+len, "\n"); | |
1933 | return len; | |
1934 | } | |
1935 | ||
1936 | static ssize_t | |
1937 | timeout_store(mddev_t *mddev, const char *buf, size_t len) | |
1938 | { | |
1939 | /* timeout can be set at any time */ | |
1940 | unsigned long timeout; | |
1941 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
1942 | if (rv) | |
1943 | return rv; | |
1944 | ||
1945 | /* just to make sure we don't overflow... */ | |
1946 | if (timeout >= LONG_MAX / HZ) | |
1947 | return -EINVAL; | |
1948 | ||
1949 | timeout = timeout * HZ / 10000; | |
1950 | ||
1951 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
1952 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
1953 | if (timeout < 1) | |
1954 | timeout = 1; | |
1955 | mddev->bitmap_info.daemon_sleep = timeout; | |
1956 | if (mddev->thread) { | |
1957 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
1958 | * the bitmap is all clean and we don't need to | |
1959 | * adjust the timeout right now | |
1960 | */ | |
1961 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
1962 | mddev->thread->timeout = timeout; | |
1963 | md_wakeup_thread(mddev->thread); | |
1964 | } | |
1965 | } | |
1966 | return len; | |
1967 | } | |
1968 | ||
1969 | static struct md_sysfs_entry bitmap_timeout = | |
1970 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
1971 | ||
1972 | static ssize_t | |
1973 | backlog_show(mddev_t *mddev, char *page) | |
1974 | { | |
1975 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
1976 | } | |
1977 | ||
1978 | static ssize_t | |
1979 | backlog_store(mddev_t *mddev, const char *buf, size_t len) | |
1980 | { | |
1981 | unsigned long backlog; | |
1982 | int rv = strict_strtoul(buf, 10, &backlog); | |
1983 | if (rv) | |
1984 | return rv; | |
1985 | if (backlog > COUNTER_MAX) | |
1986 | return -EINVAL; | |
1987 | mddev->bitmap_info.max_write_behind = backlog; | |
1988 | return len; | |
1989 | } | |
1990 | ||
1991 | static struct md_sysfs_entry bitmap_backlog = | |
1992 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
1993 | ||
1994 | static ssize_t | |
1995 | chunksize_show(mddev_t *mddev, char *page) | |
1996 | { | |
1997 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
1998 | } | |
1999 | ||
2000 | static ssize_t | |
2001 | chunksize_store(mddev_t *mddev, const char *buf, size_t len) | |
2002 | { | |
2003 | /* Can only be changed when no bitmap is active */ | |
2004 | int rv; | |
2005 | unsigned long csize; | |
2006 | if (mddev->bitmap) | |
2007 | return -EBUSY; | |
2008 | rv = strict_strtoul(buf, 10, &csize); | |
2009 | if (rv) | |
2010 | return rv; | |
2011 | if (csize < 512 || | |
2012 | !is_power_of_2(csize)) | |
2013 | return -EINVAL; | |
2014 | mddev->bitmap_info.chunksize = csize; | |
2015 | return len; | |
2016 | } | |
2017 | ||
2018 | static struct md_sysfs_entry bitmap_chunksize = | |
2019 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2020 | ||
ece5cff0 N |
2021 | static ssize_t metadata_show(mddev_t *mddev, char *page) |
2022 | { | |
2023 | return sprintf(page, "%s\n", (mddev->bitmap_info.external | |
2024 | ? "external" : "internal")); | |
2025 | } | |
2026 | ||
2027 | static ssize_t metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
2028 | { | |
2029 | if (mddev->bitmap || | |
2030 | mddev->bitmap_info.file || | |
2031 | mddev->bitmap_info.offset) | |
2032 | return -EBUSY; | |
2033 | if (strncmp(buf, "external", 8) == 0) | |
2034 | mddev->bitmap_info.external = 1; | |
2035 | else if (strncmp(buf, "internal", 8) == 0) | |
2036 | mddev->bitmap_info.external = 0; | |
2037 | else | |
2038 | return -EINVAL; | |
2039 | return len; | |
2040 | } | |
2041 | ||
2042 | static struct md_sysfs_entry bitmap_metadata = | |
2043 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2044 | ||
2045 | static ssize_t can_clear_show(mddev_t *mddev, char *page) | |
2046 | { | |
2047 | int len; | |
2048 | if (mddev->bitmap) | |
2049 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2050 | "false" : "true")); | |
2051 | else | |
2052 | len = sprintf(page, "\n"); | |
2053 | return len; | |
2054 | } | |
2055 | ||
2056 | static ssize_t can_clear_store(mddev_t *mddev, const char *buf, size_t len) | |
2057 | { | |
2058 | if (mddev->bitmap == NULL) | |
2059 | return -ENOENT; | |
2060 | if (strncmp(buf, "false", 5) == 0) | |
2061 | mddev->bitmap->need_sync = 1; | |
2062 | else if (strncmp(buf, "true", 4) == 0) { | |
2063 | if (mddev->degraded) | |
2064 | return -EBUSY; | |
2065 | mddev->bitmap->need_sync = 0; | |
2066 | } else | |
2067 | return -EINVAL; | |
2068 | return len; | |
2069 | } | |
2070 | ||
2071 | static struct md_sysfs_entry bitmap_can_clear = | |
2072 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2073 | ||
696fcd53 PC |
2074 | static ssize_t |
2075 | behind_writes_used_show(mddev_t *mddev, char *page) | |
2076 | { | |
2077 | if (mddev->bitmap == NULL) | |
2078 | return sprintf(page, "0\n"); | |
2079 | return sprintf(page, "%lu\n", | |
2080 | mddev->bitmap->behind_writes_used); | |
2081 | } | |
2082 | ||
2083 | static ssize_t | |
2084 | behind_writes_used_reset(mddev_t *mddev, const char *buf, size_t len) | |
2085 | { | |
2086 | if (mddev->bitmap) | |
2087 | mddev->bitmap->behind_writes_used = 0; | |
2088 | return len; | |
2089 | } | |
2090 | ||
2091 | static struct md_sysfs_entry max_backlog_used = | |
2092 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2093 | behind_writes_used_show, behind_writes_used_reset); | |
2094 | ||
43a70507 N |
2095 | static struct attribute *md_bitmap_attrs[] = { |
2096 | &bitmap_location.attr, | |
2097 | &bitmap_timeout.attr, | |
2098 | &bitmap_backlog.attr, | |
2099 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2100 | &bitmap_metadata.attr, |
2101 | &bitmap_can_clear.attr, | |
696fcd53 | 2102 | &max_backlog_used.attr, |
43a70507 N |
2103 | NULL |
2104 | }; | |
2105 | struct attribute_group md_bitmap_group = { | |
2106 | .name = "bitmap", | |
2107 | .attrs = md_bitmap_attrs, | |
2108 | }; | |
2109 |