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). | |
16 | * wait if count gets too high, wake when it drops to half. | |
32a7627c N |
17 | */ |
18 | ||
bff61975 | 19 | #include <linux/blkdev.h> |
32a7627c | 20 | #include <linux/module.h> |
32a7627c N |
21 | #include <linux/errno.h> |
22 | #include <linux/slab.h> | |
23 | #include <linux/init.h> | |
32a7627c N |
24 | #include <linux/timer.h> |
25 | #include <linux/sched.h> | |
26 | #include <linux/list.h> | |
27 | #include <linux/file.h> | |
28 | #include <linux/mount.h> | |
29 | #include <linux/buffer_head.h> | |
43b2e5d8 | 30 | #include "md.h" |
ef740c37 | 31 | #include "bitmap.h" |
32a7627c N |
32 | |
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 | ||
54 | //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */ | |
55 | #define DPRINTK(x...) do { } while(0) | |
56 | ||
57 | #ifndef PRINTK | |
58 | # if DEBUG > 0 | |
59 | # define PRINTK(x...) printk(KERN_DEBUG x) | |
60 | # else | |
61 | # define PRINTK(x...) | |
62 | # endif | |
63 | #endif | |
64 | ||
65 | static inline char * bmname(struct bitmap *bitmap) | |
66 | { | |
67 | return bitmap->mddev ? mdname(bitmap->mddev) : "mdX"; | |
68 | } | |
69 | ||
32a7627c N |
70 | |
71 | /* | |
72 | * just a placeholder - calls kmalloc for bitmap pages | |
73 | */ | |
74 | static unsigned char *bitmap_alloc_page(struct bitmap *bitmap) | |
75 | { | |
76 | unsigned char *page; | |
77 | ||
44456d37 | 78 | #ifdef INJECT_FAULTS_1 |
32a7627c N |
79 | page = NULL; |
80 | #else | |
81 | page = kmalloc(PAGE_SIZE, GFP_NOIO); | |
82 | #endif | |
83 | if (!page) | |
84 | printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap)); | |
85 | else | |
a654b9d8 | 86 | PRINTK("%s: bitmap_alloc_page: allocated page at %p\n", |
32a7627c N |
87 | bmname(bitmap), page); |
88 | return page; | |
89 | } | |
90 | ||
91 | /* | |
92 | * for now just a placeholder -- just calls kfree for bitmap pages | |
93 | */ | |
94 | static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page) | |
95 | { | |
96 | PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page); | |
97 | kfree(page); | |
98 | } | |
99 | ||
100 | /* | |
101 | * check a page and, if necessary, allocate it (or hijack it if the alloc fails) | |
102 | * | |
103 | * 1) check to see if this page is allocated, if it's not then try to alloc | |
104 | * 2) if the alloc fails, set the page's hijacked flag so we'll use the | |
105 | * page pointer directly as a counter | |
106 | * | |
107 | * if we find our page, we increment the page's refcount so that it stays | |
108 | * allocated while we're using it | |
109 | */ | |
110 | static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create) | |
ee305ace N |
111 | __releases(bitmap->lock) |
112 | __acquires(bitmap->lock) | |
32a7627c N |
113 | { |
114 | unsigned char *mappage; | |
115 | ||
116 | if (page >= bitmap->pages) { | |
1187cf0a N |
117 | /* This can happen if bitmap_start_sync goes beyond |
118 | * End-of-device while looking for a whole page. | |
119 | * It is harmless. | |
120 | */ | |
32a7627c N |
121 | return -EINVAL; |
122 | } | |
123 | ||
124 | ||
125 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ | |
126 | return 0; | |
127 | ||
128 | if (bitmap->bp[page].map) /* page is already allocated, just return */ | |
129 | return 0; | |
130 | ||
131 | if (!create) | |
132 | return -ENOENT; | |
133 | ||
134 | spin_unlock_irq(&bitmap->lock); | |
135 | ||
136 | /* this page has not been allocated yet */ | |
137 | ||
138 | if ((mappage = bitmap_alloc_page(bitmap)) == NULL) { | |
139 | PRINTK("%s: bitmap map page allocation failed, hijacking\n", | |
140 | bmname(bitmap)); | |
141 | /* failed - set the hijacked flag so that we can use the | |
142 | * pointer as a counter */ | |
143 | spin_lock_irq(&bitmap->lock); | |
144 | if (!bitmap->bp[page].map) | |
145 | bitmap->bp[page].hijacked = 1; | |
146 | goto out; | |
147 | } | |
148 | ||
149 | /* got a page */ | |
150 | ||
151 | spin_lock_irq(&bitmap->lock); | |
152 | ||
153 | /* recheck the page */ | |
154 | ||
155 | if (bitmap->bp[page].map || bitmap->bp[page].hijacked) { | |
156 | /* somebody beat us to getting the page */ | |
157 | bitmap_free_page(bitmap, mappage); | |
158 | return 0; | |
159 | } | |
160 | ||
161 | /* no page was in place and we have one, so install it */ | |
162 | ||
163 | memset(mappage, 0, PAGE_SIZE); | |
164 | bitmap->bp[page].map = mappage; | |
165 | bitmap->missing_pages--; | |
166 | out: | |
167 | return 0; | |
168 | } | |
169 | ||
170 | ||
171 | /* if page is completely empty, put it back on the free list, or dealloc it */ | |
172 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
173 | /* Note: lock should be held when calling this */ | |
858119e1 | 174 | static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page) |
32a7627c N |
175 | { |
176 | char *ptr; | |
177 | ||
178 | if (bitmap->bp[page].count) /* page is still busy */ | |
179 | return; | |
180 | ||
181 | /* page is no longer in use, it can be released */ | |
182 | ||
183 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
184 | bitmap->bp[page].hijacked = 0; | |
185 | bitmap->bp[page].map = NULL; | |
186 | return; | |
187 | } | |
188 | ||
189 | /* normal case, free the page */ | |
190 | ||
191 | #if 0 | |
192 | /* actually ... let's not. We will probably need the page again exactly when | |
193 | * memory is tight and we are flusing to disk | |
194 | */ | |
195 | return; | |
196 | #else | |
197 | ptr = bitmap->bp[page].map; | |
198 | bitmap->bp[page].map = NULL; | |
199 | bitmap->missing_pages++; | |
200 | bitmap_free_page(bitmap, ptr); | |
201 | return; | |
202 | #endif | |
203 | } | |
204 | ||
205 | ||
206 | /* | |
207 | * bitmap file handling - read and write the bitmap file and its superblock | |
208 | */ | |
209 | ||
32a7627c N |
210 | /* |
211 | * basic page I/O operations | |
212 | */ | |
213 | ||
a654b9d8 | 214 | /* IO operations when bitmap is stored near all superblocks */ |
f6af949c | 215 | static struct page *read_sb_page(mddev_t *mddev, loff_t offset, |
a2ed9615 N |
216 | struct page *page, |
217 | unsigned long index, int size) | |
a654b9d8 N |
218 | { |
219 | /* choose a good rdev and read the page from there */ | |
220 | ||
221 | mdk_rdev_t *rdev; | |
a654b9d8 N |
222 | sector_t target; |
223 | ||
a2ed9615 N |
224 | if (!page) |
225 | page = alloc_page(GFP_KERNEL); | |
a654b9d8 N |
226 | if (!page) |
227 | return ERR_PTR(-ENOMEM); | |
a654b9d8 | 228 | |
159ec1fc | 229 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 N |
230 | if (! test_bit(In_sync, &rdev->flags) |
231 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
232 | continue; |
233 | ||
0f420358 | 234 | target = rdev->sb_start + offset + index * (PAGE_SIZE/512); |
a654b9d8 | 235 | |
a2ed9615 | 236 | if (sync_page_io(rdev->bdev, target, |
e1defc4f | 237 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
a2ed9615 | 238 | page, READ)) { |
ab904d63 | 239 | page->index = index; |
ce25c31b N |
240 | attach_page_buffers(page, NULL); /* so that free_buffer will |
241 | * quietly no-op */ | |
ab904d63 N |
242 | return page; |
243 | } | |
244 | } | |
245 | return ERR_PTR(-EIO); | |
a654b9d8 | 246 | |
a654b9d8 N |
247 | } |
248 | ||
b2d2c4ce N |
249 | static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
250 | { | |
251 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
252 | * linked list, and raising the refcount of devices we return to ensure | |
253 | * they don't disappear while in use. | |
254 | * As devices are only added or removed when raid_disk is < 0 and | |
255 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
256 | * still be in the same position on the list when we re-enter | |
257 | * list_for_each_continue_rcu. | |
258 | */ | |
259 | struct list_head *pos; | |
260 | rcu_read_lock(); | |
261 | if (rdev == NULL) | |
262 | /* start at the beginning */ | |
263 | pos = &mddev->disks; | |
264 | else { | |
265 | /* release the previous rdev and start from there. */ | |
266 | rdev_dec_pending(rdev, mddev); | |
267 | pos = &rdev->same_set; | |
268 | } | |
269 | list_for_each_continue_rcu(pos, &mddev->disks) { | |
270 | rdev = list_entry(pos, mdk_rdev_t, same_set); | |
271 | if (rdev->raid_disk >= 0 && | |
b2d2c4ce N |
272 | !test_bit(Faulty, &rdev->flags)) { |
273 | /* this is a usable devices */ | |
274 | atomic_inc(&rdev->nr_pending); | |
275 | rcu_read_unlock(); | |
276 | return rdev; | |
277 | } | |
278 | } | |
279 | rcu_read_unlock(); | |
280 | return NULL; | |
281 | } | |
282 | ||
ab6085c7 | 283 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 284 | { |
b2d2c4ce | 285 | mdk_rdev_t *rdev = NULL; |
ab6085c7 | 286 | mddev_t *mddev = bitmap->mddev; |
a654b9d8 | 287 | |
b2d2c4ce | 288 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ab6085c7 | 289 | int size = PAGE_SIZE; |
f6af949c | 290 | loff_t offset = mddev->bitmap_info.offset; |
ab6085c7 N |
291 | if (page->index == bitmap->file_pages-1) |
292 | size = roundup(bitmap->last_page_size, | |
e1defc4f | 293 | bdev_logical_block_size(rdev->bdev)); |
f0d76d70 N |
294 | /* Just make sure we aren't corrupting data or |
295 | * metadata | |
296 | */ | |
f6af949c N |
297 | if (mddev->external) { |
298 | /* Bitmap could be anywhere. */ | |
299 | if (rdev->sb_start + offset + (page->index *(PAGE_SIZE/512)) > | |
300 | rdev->data_offset && | |
301 | rdev->sb_start + offset < | |
302 | rdev->data_offset + mddev->dev_sectors + | |
303 | (PAGE_SIZE/512)) | |
304 | goto bad_alignment; | |
305 | } else if (offset < 0) { | |
f0d76d70 | 306 | /* DATA BITMAP METADATA */ |
42a04b50 | 307 | if (offset |
85bfb4da | 308 | + (long)(page->index * (PAGE_SIZE/512)) |
f0d76d70 N |
309 | + size/512 > 0) |
310 | /* bitmap runs in to metadata */ | |
4b80991c | 311 | goto bad_alignment; |
58c0fed4 | 312 | if (rdev->data_offset + mddev->dev_sectors |
42a04b50 | 313 | > rdev->sb_start + offset) |
f0d76d70 | 314 | /* data runs in to bitmap */ |
4b80991c | 315 | goto bad_alignment; |
0f420358 | 316 | } else if (rdev->sb_start < rdev->data_offset) { |
f0d76d70 | 317 | /* METADATA BITMAP DATA */ |
0f420358 | 318 | if (rdev->sb_start |
42a04b50 | 319 | + offset |
f0d76d70 N |
320 | + page->index*(PAGE_SIZE/512) + size/512 |
321 | > rdev->data_offset) | |
322 | /* bitmap runs in to data */ | |
4b80991c | 323 | goto bad_alignment; |
f0d76d70 N |
324 | } else { |
325 | /* DATA METADATA BITMAP - no problems */ | |
326 | } | |
a654b9d8 | 327 | md_super_write(mddev, rdev, |
42a04b50 | 328 | rdev->sb_start + offset |
a654b9d8 | 329 | + page->index * (PAGE_SIZE/512), |
ab6085c7 | 330 | size, |
a654b9d8 | 331 | page); |
b2d2c4ce | 332 | } |
a654b9d8 N |
333 | |
334 | if (wait) | |
a9701a30 | 335 | md_super_wait(mddev); |
a654b9d8 | 336 | return 0; |
4b80991c N |
337 | |
338 | bad_alignment: | |
4b80991c | 339 | return -EINVAL; |
a654b9d8 N |
340 | } |
341 | ||
4ad13663 | 342 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 343 | /* |
a654b9d8 | 344 | * write out a page to a file |
32a7627c | 345 | */ |
4ad13663 | 346 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 347 | { |
d785a06a | 348 | struct buffer_head *bh; |
32a7627c | 349 | |
f0d76d70 N |
350 | if (bitmap->file == NULL) { |
351 | switch (write_sb_page(bitmap, page, wait)) { | |
352 | case -EINVAL: | |
353 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
f0d76d70 | 354 | } |
4ad13663 | 355 | } else { |
a654b9d8 | 356 | |
4ad13663 | 357 | bh = page_buffers(page); |
c708443c | 358 | |
4ad13663 N |
359 | while (bh && bh->b_blocknr) { |
360 | atomic_inc(&bitmap->pending_writes); | |
361 | set_buffer_locked(bh); | |
362 | set_buffer_mapped(bh); | |
363 | submit_bh(WRITE, bh); | |
364 | bh = bh->b_this_page; | |
365 | } | |
d785a06a | 366 | |
4ad13663 N |
367 | if (wait) { |
368 | wait_event(bitmap->write_wait, | |
369 | atomic_read(&bitmap->pending_writes)==0); | |
370 | } | |
8a5e9cf1 | 371 | } |
4ad13663 N |
372 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
373 | bitmap_file_kick(bitmap); | |
d785a06a N |
374 | } |
375 | ||
376 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
377 | { | |
378 | struct bitmap *bitmap = bh->b_private; | |
379 | unsigned long flags; | |
32a7627c | 380 | |
d785a06a N |
381 | if (!uptodate) { |
382 | spin_lock_irqsave(&bitmap->lock, flags); | |
383 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
384 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
32a7627c | 385 | } |
d785a06a N |
386 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
387 | wake_up(&bitmap->write_wait); | |
388 | } | |
32a7627c | 389 | |
d785a06a N |
390 | /* copied from buffer.c */ |
391 | static void | |
392 | __clear_page_buffers(struct page *page) | |
393 | { | |
394 | ClearPagePrivate(page); | |
395 | set_page_private(page, 0); | |
396 | page_cache_release(page); | |
397 | } | |
398 | static void free_buffers(struct page *page) | |
399 | { | |
400 | struct buffer_head *bh = page_buffers(page); | |
77ad4bc7 | 401 | |
d785a06a N |
402 | while (bh) { |
403 | struct buffer_head *next = bh->b_this_page; | |
404 | free_buffer_head(bh); | |
405 | bh = next; | |
77ad4bc7 | 406 | } |
d785a06a N |
407 | __clear_page_buffers(page); |
408 | put_page(page); | |
32a7627c N |
409 | } |
410 | ||
d785a06a N |
411 | /* read a page from a file. |
412 | * We both read the page, and attach buffers to the page to record the | |
413 | * address of each block (using bmap). These addresses will be used | |
414 | * to write the block later, completely bypassing the filesystem. | |
415 | * This usage is similar to how swap files are handled, and allows us | |
416 | * to write to a file with no concerns of memory allocation failing. | |
417 | */ | |
32a7627c | 418 | static struct page *read_page(struct file *file, unsigned long index, |
d785a06a N |
419 | struct bitmap *bitmap, |
420 | unsigned long count) | |
32a7627c | 421 | { |
32a7627c | 422 | struct page *page = NULL; |
c649bb9c | 423 | struct inode *inode = file->f_path.dentry->d_inode; |
d785a06a N |
424 | struct buffer_head *bh; |
425 | sector_t block; | |
32a7627c | 426 | |
2d1f3b5d N |
427 | PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE, |
428 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 429 | |
d785a06a N |
430 | page = alloc_page(GFP_KERNEL); |
431 | if (!page) | |
432 | page = ERR_PTR(-ENOMEM); | |
32a7627c N |
433 | if (IS_ERR(page)) |
434 | goto out; | |
d785a06a | 435 | |
d785a06a N |
436 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
437 | if (!bh) { | |
2d1f3b5d | 438 | put_page(page); |
d785a06a | 439 | page = ERR_PTR(-ENOMEM); |
32a7627c N |
440 | goto out; |
441 | } | |
d785a06a N |
442 | attach_page_buffers(page, bh); |
443 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
444 | while (bh) { | |
445 | if (count == 0) | |
446 | bh->b_blocknr = 0; | |
447 | else { | |
448 | bh->b_blocknr = bmap(inode, block); | |
449 | if (bh->b_blocknr == 0) { | |
450 | /* Cannot use this file! */ | |
451 | free_buffers(page); | |
452 | page = ERR_PTR(-EINVAL); | |
453 | goto out; | |
454 | } | |
455 | bh->b_bdev = inode->i_sb->s_bdev; | |
456 | if (count < (1<<inode->i_blkbits)) | |
457 | count = 0; | |
458 | else | |
459 | count -= (1<<inode->i_blkbits); | |
460 | ||
461 | bh->b_end_io = end_bitmap_write; | |
462 | bh->b_private = bitmap; | |
ce25c31b N |
463 | atomic_inc(&bitmap->pending_writes); |
464 | set_buffer_locked(bh); | |
465 | set_buffer_mapped(bh); | |
466 | submit_bh(READ, bh); | |
d785a06a N |
467 | } |
468 | block++; | |
469 | bh = bh->b_this_page; | |
470 | } | |
d785a06a | 471 | page->index = index; |
ce25c31b N |
472 | |
473 | wait_event(bitmap->write_wait, | |
474 | atomic_read(&bitmap->pending_writes)==0); | |
475 | if (bitmap->flags & BITMAP_WRITE_ERROR) { | |
476 | free_buffers(page); | |
477 | page = ERR_PTR(-EIO); | |
478 | } | |
32a7627c N |
479 | out: |
480 | if (IS_ERR(page)) | |
481 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n", | |
2d1f3b5d N |
482 | (int)PAGE_SIZE, |
483 | (unsigned long long)index << PAGE_SHIFT, | |
32a7627c N |
484 | PTR_ERR(page)); |
485 | return page; | |
486 | } | |
487 | ||
488 | /* | |
489 | * bitmap file superblock operations | |
490 | */ | |
491 | ||
492 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 493 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
494 | { |
495 | bitmap_super_t *sb; | |
496 | unsigned long flags; | |
497 | ||
498 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 499 | return; |
ece5cff0 N |
500 | if (bitmap->mddev->bitmap_info.external) |
501 | return; | |
32a7627c N |
502 | spin_lock_irqsave(&bitmap->lock, flags); |
503 | if (!bitmap->sb_page) { /* no superblock */ | |
504 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 505 | return; |
32a7627c | 506 | } |
32a7627c | 507 | spin_unlock_irqrestore(&bitmap->lock, flags); |
ea03aff9 | 508 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 509 | sb->events = cpu_to_le64(bitmap->mddev->events); |
a0da84f3 NB |
510 | if (bitmap->mddev->events < bitmap->events_cleared) { |
511 | /* rocking back to read-only */ | |
512 | bitmap->events_cleared = bitmap->mddev->events; | |
513 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); | |
514 | } | |
43a70507 N |
515 | /* Just in case these have been changed via sysfs: */ |
516 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
517 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
ea03aff9 | 518 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 519 | write_page(bitmap, bitmap->sb_page, 1); |
32a7627c N |
520 | } |
521 | ||
522 | /* print out the bitmap file superblock */ | |
523 | void bitmap_print_sb(struct bitmap *bitmap) | |
524 | { | |
525 | bitmap_super_t *sb; | |
526 | ||
527 | if (!bitmap || !bitmap->sb_page) | |
528 | return; | |
ea03aff9 | 529 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 530 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
531 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
532 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
533 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
534 | *(__u32 *)(sb->uuid+0), |
535 | *(__u32 *)(sb->uuid+4), | |
536 | *(__u32 *)(sb->uuid+8), | |
537 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 538 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 539 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 540 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 541 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
542 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
543 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
544 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
545 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
546 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 547 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
ea03aff9 | 548 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
549 | } |
550 | ||
551 | /* read the superblock from the bitmap file and initialize some bitmap fields */ | |
552 | static int bitmap_read_sb(struct bitmap *bitmap) | |
553 | { | |
554 | char *reason = NULL; | |
555 | bitmap_super_t *sb; | |
4b6d287f | 556 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c N |
557 | unsigned long long events; |
558 | int err = -EINVAL; | |
559 | ||
560 | /* page 0 is the superblock, read it... */ | |
f49d5e62 N |
561 | if (bitmap->file) { |
562 | loff_t isize = i_size_read(bitmap->file->f_mapping->host); | |
563 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; | |
564 | ||
565 | bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes); | |
566 | } else { | |
42a04b50 N |
567 | bitmap->sb_page = read_sb_page(bitmap->mddev, |
568 | bitmap->mddev->bitmap_info.offset, | |
a2ed9615 N |
569 | NULL, |
570 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 571 | } |
32a7627c N |
572 | if (IS_ERR(bitmap->sb_page)) { |
573 | err = PTR_ERR(bitmap->sb_page); | |
574 | bitmap->sb_page = NULL; | |
575 | return err; | |
576 | } | |
577 | ||
ea03aff9 | 578 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 579 | |
32a7627c | 580 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 581 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 582 | write_behind = le32_to_cpu(sb->write_behind); |
32a7627c N |
583 | |
584 | /* verify that the bitmap-specific fields are valid */ | |
585 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
586 | reason = "bad magic"; | |
bd926c63 N |
587 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
588 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 589 | reason = "unrecognized superblock version"; |
1187cf0a | 590 | else if (chunksize < 512) |
7dd5d34c | 591 | reason = "bitmap chunksize too small"; |
32a7627c N |
592 | else if ((1 << ffz(~chunksize)) != chunksize) |
593 | reason = "bitmap chunksize not a power of 2"; | |
1b04be96 | 594 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 595 | reason = "daemon sleep period out of range"; |
4b6d287f N |
596 | else if (write_behind > COUNTER_MAX) |
597 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
598 | if (reason) { |
599 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
600 | bmname(bitmap), reason); | |
601 | goto out; | |
602 | } | |
603 | ||
604 | /* keep the array size field of the bitmap superblock up to date */ | |
605 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
606 | ||
607 | if (!bitmap->mddev->persistent) | |
608 | goto success; | |
609 | ||
610 | /* | |
611 | * if we have a persistent array superblock, compare the | |
612 | * bitmap's UUID and event counter to the mddev's | |
613 | */ | |
614 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
615 | printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n", | |
616 | bmname(bitmap)); | |
617 | goto out; | |
618 | } | |
619 | events = le64_to_cpu(sb->events); | |
620 | if (events < bitmap->mddev->events) { | |
621 | printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) " | |
622 | "-- forcing full recovery\n", bmname(bitmap), events, | |
623 | (unsigned long long) bitmap->mddev->events); | |
4f2e639a | 624 | sb->state |= cpu_to_le32(BITMAP_STALE); |
32a7627c N |
625 | } |
626 | success: | |
627 | /* assign fields using values from superblock */ | |
42a04b50 N |
628 | bitmap->mddev->bitmap_info.chunksize = chunksize; |
629 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
42a04b50 | 630 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; |
4f2e639a | 631 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 N |
632 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
633 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
32a7627c | 634 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
4f2e639a | 635 | if (sb->state & cpu_to_le32(BITMAP_STALE)) |
6a07997f | 636 | bitmap->events_cleared = bitmap->mddev->events; |
32a7627c N |
637 | err = 0; |
638 | out: | |
ea03aff9 | 639 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
640 | if (err) |
641 | bitmap_print_sb(bitmap); | |
642 | return err; | |
643 | } | |
644 | ||
645 | enum bitmap_mask_op { | |
646 | MASK_SET, | |
647 | MASK_UNSET | |
648 | }; | |
649 | ||
4ad13663 N |
650 | /* record the state of the bitmap in the superblock. Return the old value */ |
651 | static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits, | |
652 | enum bitmap_mask_op op) | |
32a7627c N |
653 | { |
654 | bitmap_super_t *sb; | |
655 | unsigned long flags; | |
4ad13663 | 656 | int old; |
32a7627c N |
657 | |
658 | spin_lock_irqsave(&bitmap->lock, flags); | |
7e317655 | 659 | if (!bitmap->sb_page) { /* can't set the state */ |
32a7627c | 660 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 661 | return 0; |
32a7627c | 662 | } |
32a7627c | 663 | spin_unlock_irqrestore(&bitmap->lock, flags); |
ea03aff9 | 664 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
4ad13663 | 665 | old = le32_to_cpu(sb->state) & bits; |
32a7627c | 666 | switch (op) { |
4f2e639a | 667 | case MASK_SET: sb->state |= cpu_to_le32(bits); |
32a7627c | 668 | break; |
4f2e639a | 669 | case MASK_UNSET: sb->state &= cpu_to_le32(~bits); |
32a7627c N |
670 | break; |
671 | default: BUG(); | |
672 | } | |
ea03aff9 | 673 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 674 | return old; |
32a7627c N |
675 | } |
676 | ||
677 | /* | |
678 | * general bitmap file operations | |
679 | */ | |
680 | ||
ece5cff0 N |
681 | /* |
682 | * on-disk bitmap: | |
683 | * | |
684 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
685 | * file a page at a time. There's a superblock at the start of the file. | |
686 | */ | |
32a7627c | 687 | /* calculate the index of the page that contains this bit */ |
ece5cff0 | 688 | static inline unsigned long file_page_index(struct bitmap *bitmap, unsigned long chunk) |
32a7627c | 689 | { |
ece5cff0 N |
690 | if (!bitmap->mddev->bitmap_info.external) |
691 | chunk += sizeof(bitmap_super_t) << 3; | |
692 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
693 | } |
694 | ||
695 | /* calculate the (bit) offset of this bit within a page */ | |
ece5cff0 | 696 | static inline unsigned long file_page_offset(struct bitmap *bitmap, unsigned long chunk) |
32a7627c | 697 | { |
ece5cff0 N |
698 | if (!bitmap->mddev->bitmap_info.external) |
699 | chunk += sizeof(bitmap_super_t) << 3; | |
700 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
701 | } |
702 | ||
703 | /* | |
704 | * return a pointer to the page in the filemap that contains the given bit | |
705 | * | |
706 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
707 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
708 | * 0 or page 1 | |
709 | */ | |
710 | static inline struct page *filemap_get_page(struct bitmap *bitmap, | |
711 | unsigned long chunk) | |
712 | { | |
ece5cff0 N |
713 | if (file_page_index(bitmap, chunk) >= bitmap->file_pages) return NULL; |
714 | return bitmap->filemap[file_page_index(bitmap, chunk) | |
715 | - file_page_index(bitmap, 0)]; | |
32a7627c N |
716 | } |
717 | ||
718 | ||
719 | static void bitmap_file_unmap(struct bitmap *bitmap) | |
720 | { | |
721 | struct page **map, *sb_page; | |
722 | unsigned long *attr; | |
723 | int pages; | |
724 | unsigned long flags; | |
725 | ||
726 | spin_lock_irqsave(&bitmap->lock, flags); | |
727 | map = bitmap->filemap; | |
728 | bitmap->filemap = NULL; | |
729 | attr = bitmap->filemap_attr; | |
730 | bitmap->filemap_attr = NULL; | |
731 | pages = bitmap->file_pages; | |
732 | bitmap->file_pages = 0; | |
733 | sb_page = bitmap->sb_page; | |
734 | bitmap->sb_page = NULL; | |
735 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
736 | ||
737 | while (pages--) | |
ece5cff0 | 738 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 739 | free_buffers(map[pages]); |
32a7627c N |
740 | kfree(map); |
741 | kfree(attr); | |
742 | ||
d785a06a N |
743 | if (sb_page) |
744 | free_buffers(sb_page); | |
32a7627c N |
745 | } |
746 | ||
747 | static void bitmap_file_put(struct bitmap *bitmap) | |
748 | { | |
749 | struct file *file; | |
32a7627c N |
750 | unsigned long flags; |
751 | ||
752 | spin_lock_irqsave(&bitmap->lock, flags); | |
753 | file = bitmap->file; | |
754 | bitmap->file = NULL; | |
755 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
756 | ||
d785a06a N |
757 | if (file) |
758 | wait_event(bitmap->write_wait, | |
759 | atomic_read(&bitmap->pending_writes)==0); | |
32a7627c N |
760 | bitmap_file_unmap(bitmap); |
761 | ||
d785a06a | 762 | if (file) { |
c649bb9c | 763 | struct inode *inode = file->f_path.dentry->d_inode; |
fc0ecff6 | 764 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 765 | fput(file); |
d785a06a | 766 | } |
32a7627c N |
767 | } |
768 | ||
769 | ||
770 | /* | |
771 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
772 | * then it is no longer reliable, so we stop using it and we mark the file | |
773 | * as failed in the superblock | |
774 | */ | |
775 | static void bitmap_file_kick(struct bitmap *bitmap) | |
776 | { | |
777 | char *path, *ptr = NULL; | |
778 | ||
4ad13663 N |
779 | if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) { |
780 | bitmap_update_sb(bitmap); | |
32a7627c | 781 | |
4ad13663 N |
782 | if (bitmap->file) { |
783 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
784 | if (path) | |
6bcfd601 CH |
785 | ptr = d_path(&bitmap->file->f_path, path, |
786 | PAGE_SIZE); | |
787 | ||
32a7627c | 788 | |
4ad13663 N |
789 | printk(KERN_ALERT |
790 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 791 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 792 | |
4ad13663 N |
793 | kfree(path); |
794 | } else | |
795 | printk(KERN_ALERT | |
796 | "%s: disabling internal bitmap due to errors\n", | |
797 | bmname(bitmap)); | |
a654b9d8 | 798 | } |
32a7627c N |
799 | |
800 | bitmap_file_put(bitmap); | |
801 | ||
802 | return; | |
803 | } | |
804 | ||
805 | enum bitmap_page_attr { | |
e16b68b6 N |
806 | BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced |
807 | BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared | |
808 | BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced | |
32a7627c N |
809 | }; |
810 | ||
811 | static inline void set_page_attr(struct bitmap *bitmap, struct page *page, | |
812 | enum bitmap_page_attr attr) | |
813 | { | |
e16b68b6 | 814 | __set_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
815 | } |
816 | ||
817 | static inline void clear_page_attr(struct bitmap *bitmap, struct page *page, | |
818 | enum bitmap_page_attr attr) | |
819 | { | |
e16b68b6 | 820 | __clear_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
821 | } |
822 | ||
ec7a3197 N |
823 | static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page, |
824 | enum bitmap_page_attr attr) | |
32a7627c | 825 | { |
e16b68b6 | 826 | return test_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
827 | } |
828 | ||
829 | /* | |
830 | * bitmap_file_set_bit -- called before performing a write to the md device | |
831 | * to set (and eventually sync) a particular bit in the bitmap file | |
832 | * | |
833 | * we set the bit immediately, then we record the page number so that | |
834 | * when an unplug occurs, we can flush the dirty pages out to disk | |
835 | */ | |
836 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
837 | { | |
838 | unsigned long bit; | |
839 | struct page *page; | |
840 | void *kaddr; | |
841 | unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap); | |
842 | ||
a654b9d8 | 843 | if (!bitmap->filemap) { |
32a7627c N |
844 | return; |
845 | } | |
846 | ||
847 | page = filemap_get_page(bitmap, chunk); | |
9b1d1dac | 848 | if (!page) return; |
ece5cff0 | 849 | bit = file_page_offset(bitmap, chunk); |
32a7627c | 850 | |
32a7627c N |
851 | /* set the bit */ |
852 | kaddr = kmap_atomic(page, KM_USER0); | |
bd926c63 N |
853 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
854 | set_bit(bit, kaddr); | |
855 | else | |
856 | ext2_set_bit(bit, kaddr); | |
32a7627c N |
857 | kunmap_atomic(kaddr, KM_USER0); |
858 | PRINTK("set file bit %lu page %lu\n", bit, page->index); | |
859 | ||
860 | /* record page number so it gets flushed to disk when unplug occurs */ | |
861 | set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); | |
862 | ||
863 | } | |
864 | ||
865 | /* this gets called when the md device is ready to unplug its underlying | |
866 | * (slave) device queues -- before we let any writes go down, we need to | |
867 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 868 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 869 | { |
ec7a3197 N |
870 | unsigned long i, flags; |
871 | int dirty, need_write; | |
32a7627c N |
872 | struct page *page; |
873 | int wait = 0; | |
874 | ||
875 | if (!bitmap) | |
4ad13663 | 876 | return; |
32a7627c N |
877 | |
878 | /* look at each page to see if there are any set bits that need to be | |
879 | * flushed out to disk */ | |
880 | for (i = 0; i < bitmap->file_pages; i++) { | |
881 | spin_lock_irqsave(&bitmap->lock, flags); | |
a654b9d8 | 882 | if (!bitmap->filemap) { |
32a7627c | 883 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 884 | return; |
32a7627c N |
885 | } |
886 | page = bitmap->filemap[i]; | |
ec7a3197 N |
887 | dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
888 | need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
32a7627c N |
889 | clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
890 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
ec7a3197 | 891 | if (dirty) |
32a7627c N |
892 | wait = 1; |
893 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
894 | ||
d785a06a | 895 | if (dirty | need_write) |
4ad13663 | 896 | write_page(bitmap, page, 0); |
32a7627c N |
897 | } |
898 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
0b79ccf0 | 899 | if (bitmap->file) |
d785a06a N |
900 | wait_event(bitmap->write_wait, |
901 | atomic_read(&bitmap->pending_writes)==0); | |
0b79ccf0 | 902 | else |
a9701a30 | 903 | md_super_wait(bitmap->mddev); |
32a7627c | 904 | } |
d785a06a N |
905 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
906 | bitmap_file_kick(bitmap); | |
32a7627c N |
907 | } |
908 | ||
6a07997f | 909 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
910 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
911 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
912 | * memory mapping of the bitmap file | |
913 | * Special cases: | |
914 | * if there's no bitmap file, or if the bitmap file had been | |
915 | * previously kicked from the array, we mark all the bits as | |
916 | * 1's in order to cause a full resync. | |
6a07997f N |
917 | * |
918 | * We ignore all bits for sectors that end earlier than 'start'. | |
919 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 920 | */ |
6a07997f | 921 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
922 | { |
923 | unsigned long i, chunks, index, oldindex, bit; | |
924 | struct page *page = NULL, *oldpage = NULL; | |
925 | unsigned long num_pages, bit_cnt = 0; | |
926 | struct file *file; | |
d785a06a | 927 | unsigned long bytes, offset; |
32a7627c N |
928 | int outofdate; |
929 | int ret = -ENOSPC; | |
ea03aff9 | 930 | void *paddr; |
32a7627c N |
931 | |
932 | chunks = bitmap->chunks; | |
933 | file = bitmap->file; | |
934 | ||
42a04b50 | 935 | BUG_ON(!file && !bitmap->mddev->bitmap_info.offset); |
32a7627c | 936 | |
44456d37 | 937 | #ifdef INJECT_FAULTS_3 |
32a7627c N |
938 | outofdate = 1; |
939 | #else | |
940 | outofdate = bitmap->flags & BITMAP_STALE; | |
941 | #endif | |
942 | if (outofdate) | |
943 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
944 | "recovery\n", bmname(bitmap)); | |
945 | ||
946 | bytes = (chunks + 7) / 8; | |
ece5cff0 N |
947 | if (!bitmap->mddev->bitmap_info.external) |
948 | bytes += sizeof(bitmap_super_t); | |
bc7f77de | 949 | |
ece5cff0 N |
950 | |
951 | num_pages = (bytes + PAGE_SIZE - 1) / PAGE_SIZE; | |
bc7f77de | 952 | |
ece5cff0 | 953 | if (file && i_size_read(file->f_mapping->host) < bytes) { |
32a7627c N |
954 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
955 | bmname(bitmap), | |
956 | (unsigned long) i_size_read(file->f_mapping->host), | |
ece5cff0 | 957 | bytes); |
4ad13663 | 958 | goto err; |
32a7627c | 959 | } |
bc7f77de N |
960 | |
961 | ret = -ENOMEM; | |
962 | ||
32a7627c | 963 | bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL); |
bc7f77de | 964 | if (!bitmap->filemap) |
4ad13663 | 965 | goto err; |
32a7627c | 966 | |
e16b68b6 N |
967 | /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */ |
968 | bitmap->filemap_attr = kzalloc( | |
505fa2c4 | 969 | roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), |
e16b68b6 | 970 | GFP_KERNEL); |
bc7f77de | 971 | if (!bitmap->filemap_attr) |
4ad13663 | 972 | goto err; |
32a7627c | 973 | |
32a7627c N |
974 | oldindex = ~0L; |
975 | ||
976 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 977 | int b; |
ece5cff0 N |
978 | index = file_page_index(bitmap, i); |
979 | bit = file_page_offset(bitmap, i); | |
32a7627c | 980 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 981 | int count; |
32a7627c | 982 | /* unmap the old page, we're done with it */ |
d785a06a | 983 | if (index == num_pages-1) |
ece5cff0 | 984 | count = bytes - index * PAGE_SIZE; |
d785a06a N |
985 | else |
986 | count = PAGE_SIZE; | |
ece5cff0 | 987 | if (index == 0 && bitmap->sb_page) { |
32a7627c N |
988 | /* |
989 | * if we're here then the superblock page | |
990 | * contains some bits (PAGE_SIZE != sizeof sb) | |
991 | * we've already read it in, so just use it | |
992 | */ | |
993 | page = bitmap->sb_page; | |
994 | offset = sizeof(bitmap_super_t); | |
53845270 N |
995 | if (!file) |
996 | read_sb_page(bitmap->mddev, | |
42a04b50 | 997 | bitmap->mddev->bitmap_info.offset, |
53845270 N |
998 | page, |
999 | index, count); | |
a654b9d8 | 1000 | } else if (file) { |
d785a06a | 1001 | page = read_page(file, index, bitmap, count); |
a654b9d8 N |
1002 | offset = 0; |
1003 | } else { | |
42a04b50 N |
1004 | page = read_sb_page(bitmap->mddev, |
1005 | bitmap->mddev->bitmap_info.offset, | |
a2ed9615 N |
1006 | NULL, |
1007 | index, count); | |
32a7627c N |
1008 | offset = 0; |
1009 | } | |
a654b9d8 N |
1010 | if (IS_ERR(page)) { /* read error */ |
1011 | ret = PTR_ERR(page); | |
4ad13663 | 1012 | goto err; |
a654b9d8 N |
1013 | } |
1014 | ||
32a7627c N |
1015 | oldindex = index; |
1016 | oldpage = page; | |
32a7627c | 1017 | |
b74fd282 N |
1018 | bitmap->filemap[bitmap->file_pages++] = page; |
1019 | bitmap->last_page_size = count; | |
1020 | ||
32a7627c N |
1021 | if (outofdate) { |
1022 | /* | |
1023 | * if bitmap is out of date, dirty the | |
1024 | * whole page and write it out | |
1025 | */ | |
ea03aff9 N |
1026 | paddr = kmap_atomic(page, KM_USER0); |
1027 | memset(paddr + offset, 0xff, | |
6a07997f | 1028 | PAGE_SIZE - offset); |
ea03aff9 | 1029 | kunmap_atomic(paddr, KM_USER0); |
4ad13663 N |
1030 | write_page(bitmap, page, 1); |
1031 | ||
1032 | ret = -EIO; | |
b74fd282 | 1033 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
4ad13663 | 1034 | goto err; |
32a7627c | 1035 | } |
32a7627c | 1036 | } |
ea03aff9 | 1037 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1038 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1039 | b = test_bit(bit, paddr); |
bd926c63 | 1040 | else |
ea03aff9 N |
1041 | b = ext2_test_bit(bit, paddr); |
1042 | kunmap_atomic(paddr, KM_USER0); | |
bd926c63 | 1043 | if (b) { |
32a7627c | 1044 | /* if the disk bit is set, set the memory bit */ |
db305e50 N |
1045 | int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) |
1046 | >= start); | |
1047 | bitmap_set_memory_bits(bitmap, | |
1048 | (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap), | |
1049 | needed); | |
32a7627c | 1050 | bit_cnt++; |
6a07997f | 1051 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); |
32a7627c | 1052 | } |
32a7627c N |
1053 | } |
1054 | ||
1055 | /* everything went OK */ | |
1056 | ret = 0; | |
1057 | bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET); | |
1058 | ||
32a7627c N |
1059 | if (bit_cnt) { /* Kick recovery if any bits were set */ |
1060 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
1061 | md_wakeup_thread(bitmap->mddev->thread); | |
1062 | } | |
1063 | ||
32a7627c | 1064 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
4ad13663 N |
1065 | "read %lu/%lu pages, set %lu bits\n", |
1066 | bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt); | |
1067 | ||
1068 | return 0; | |
32a7627c | 1069 | |
4ad13663 N |
1070 | err: |
1071 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1072 | bmname(bitmap), ret); | |
32a7627c N |
1073 | return ret; |
1074 | } | |
1075 | ||
a654b9d8 N |
1076 | void bitmap_write_all(struct bitmap *bitmap) |
1077 | { | |
1078 | /* We don't actually write all bitmap blocks here, | |
1079 | * just flag them as needing to be written | |
1080 | */ | |
ec7a3197 | 1081 | int i; |
a654b9d8 | 1082 | |
ec7a3197 N |
1083 | for (i=0; i < bitmap->file_pages; i++) |
1084 | set_page_attr(bitmap, bitmap->filemap[i], | |
1085 | BITMAP_PAGE_NEEDWRITE); | |
a654b9d8 N |
1086 | } |
1087 | ||
32a7627c N |
1088 | |
1089 | static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) | |
1090 | { | |
1091 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1092 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1093 | bitmap->bp[page].count += inc; | |
1094 | /* | |
1095 | if (page == 0) printk("count page 0, offset %llu: %d gives %d\n", | |
1096 | (unsigned long long)offset, inc, bitmap->bp[page].count); | |
1097 | */ | |
1098 | bitmap_checkfree(bitmap, page); | |
1099 | } | |
1100 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, | |
1101 | sector_t offset, int *blocks, | |
1102 | int create); | |
1103 | ||
1104 | /* | |
1105 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1106 | * out to disk | |
1107 | */ | |
1108 | ||
aa5cbd10 | 1109 | void bitmap_daemon_work(mddev_t *mddev) |
32a7627c | 1110 | { |
aa5cbd10 | 1111 | struct bitmap *bitmap; |
aa3163f8 | 1112 | unsigned long j; |
32a7627c N |
1113 | unsigned long flags; |
1114 | struct page *page = NULL, *lastpage = NULL; | |
32a7627c | 1115 | int blocks; |
ea03aff9 | 1116 | void *paddr; |
32a7627c | 1117 | |
aa5cbd10 N |
1118 | /* Use a mutex to guard daemon_work against |
1119 | * bitmap_destroy. | |
1120 | */ | |
c3d9714e | 1121 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1122 | bitmap = mddev->bitmap; |
1123 | if (bitmap == NULL) { | |
c3d9714e | 1124 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1125 | return; |
aa5cbd10 | 1126 | } |
42a04b50 | 1127 | if (time_before(jiffies, bitmap->daemon_lastrun |
1b04be96 | 1128 | + bitmap->mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1129 | goto done; |
1130 | ||
32a7627c | 1131 | bitmap->daemon_lastrun = jiffies; |
8311c29d N |
1132 | if (bitmap->allclean) { |
1133 | bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
aa5cbd10 | 1134 | goto done; |
8311c29d N |
1135 | } |
1136 | bitmap->allclean = 1; | |
32a7627c | 1137 | |
be512691 | 1138 | spin_lock_irqsave(&bitmap->lock, flags); |
32a7627c N |
1139 | for (j = 0; j < bitmap->chunks; j++) { |
1140 | bitmap_counter_t *bmc; | |
be512691 | 1141 | if (!bitmap->filemap) |
32a7627c | 1142 | /* error or shutdown */ |
32a7627c | 1143 | break; |
32a7627c N |
1144 | |
1145 | page = filemap_get_page(bitmap, j); | |
32a7627c N |
1146 | |
1147 | if (page != lastpage) { | |
aa3163f8 | 1148 | /* skip this page unless it's marked as needing cleaning */ |
ec7a3197 N |
1149 | if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) { |
1150 | int need_write = test_page_attr(bitmap, page, | |
1151 | BITMAP_PAGE_NEEDWRITE); | |
a647e4bc | 1152 | if (need_write) |
aa3163f8 | 1153 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); |
a647e4bc | 1154 | |
aa3163f8 | 1155 | spin_unlock_irqrestore(&bitmap->lock, flags); |
8311c29d | 1156 | if (need_write) { |
4ad13663 | 1157 | write_page(bitmap, page, 0); |
8311c29d N |
1158 | bitmap->allclean = 0; |
1159 | } | |
be512691 N |
1160 | spin_lock_irqsave(&bitmap->lock, flags); |
1161 | j |= (PAGE_BITS - 1); | |
aa3163f8 N |
1162 | continue; |
1163 | } | |
1164 | ||
32a7627c | 1165 | /* grab the new page, sync and release the old */ |
32a7627c | 1166 | if (lastpage != NULL) { |
ec7a3197 | 1167 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1168 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1169 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 1170 | write_page(bitmap, lastpage, 0); |
32a7627c N |
1171 | } else { |
1172 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1173 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1174 | } | |
32a7627c N |
1175 | } else |
1176 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1177 | lastpage = page; | |
a0da84f3 NB |
1178 | |
1179 | /* We are possibly going to clear some bits, so make | |
1180 | * sure that events_cleared is up-to-date. | |
1181 | */ | |
ece5cff0 N |
1182 | if (bitmap->need_sync && |
1183 | bitmap->mddev->bitmap_info.external == 0) { | |
a0da84f3 NB |
1184 | bitmap_super_t *sb; |
1185 | bitmap->need_sync = 0; | |
1186 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); | |
1187 | sb->events_cleared = | |
1188 | cpu_to_le64(bitmap->events_cleared); | |
1189 | kunmap_atomic(sb, KM_USER0); | |
1190 | write_page(bitmap, bitmap->sb_page, 1); | |
1191 | } | |
32a7627c | 1192 | spin_lock_irqsave(&bitmap->lock, flags); |
ece5cff0 N |
1193 | if (!bitmap->need_sync) |
1194 | clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
32a7627c | 1195 | } |
db305e50 N |
1196 | bmc = bitmap_get_counter(bitmap, |
1197 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
1198 | &blocks, 0); | |
32a7627c N |
1199 | if (bmc) { |
1200 | /* | |
1201 | if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc); | |
1202 | */ | |
8311c29d N |
1203 | if (*bmc) |
1204 | bitmap->allclean = 0; | |
1205 | ||
32a7627c N |
1206 | if (*bmc == 2) { |
1207 | *bmc=1; /* maybe clear the bit next time */ | |
1208 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
ece5cff0 | 1209 | } else if (*bmc == 1 && !bitmap->need_sync) { |
32a7627c N |
1210 | /* we can clear the bit */ |
1211 | *bmc = 0; | |
db305e50 N |
1212 | bitmap_count_page(bitmap, |
1213 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
32a7627c N |
1214 | -1); |
1215 | ||
1216 | /* clear the bit */ | |
ea03aff9 | 1217 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1218 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ece5cff0 N |
1219 | clear_bit(file_page_offset(bitmap, j), |
1220 | paddr); | |
bd926c63 | 1221 | else |
ece5cff0 N |
1222 | ext2_clear_bit(file_page_offset(bitmap, j), |
1223 | paddr); | |
ea03aff9 | 1224 | kunmap_atomic(paddr, KM_USER0); |
32a7627c | 1225 | } |
be512691 N |
1226 | } else |
1227 | j |= PAGE_COUNTER_MASK; | |
32a7627c | 1228 | } |
be512691 | 1229 | spin_unlock_irqrestore(&bitmap->lock, flags); |
32a7627c N |
1230 | |
1231 | /* now sync the final page */ | |
1232 | if (lastpage != NULL) { | |
32a7627c | 1233 | spin_lock_irqsave(&bitmap->lock, flags); |
ec7a3197 | 1234 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1235 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1236 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 1237 | write_page(bitmap, lastpage, 0); |
32a7627c N |
1238 | } else { |
1239 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1240 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1241 | } | |
32a7627c N |
1242 | } |
1243 | ||
7be3dfec | 1244 | done: |
8311c29d | 1245 | if (bitmap->allclean == 0) |
42a04b50 | 1246 | bitmap->mddev->thread->timeout = |
1b04be96 | 1247 | bitmap->mddev->bitmap_info.daemon_sleep; |
c3d9714e | 1248 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1249 | } |
1250 | ||
32a7627c N |
1251 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
1252 | sector_t offset, int *blocks, | |
1253 | int create) | |
ee305ace N |
1254 | __releases(bitmap->lock) |
1255 | __acquires(bitmap->lock) | |
32a7627c N |
1256 | { |
1257 | /* If 'create', we might release the lock and reclaim it. | |
1258 | * The lock must have been taken with interrupts enabled. | |
1259 | * If !create, we don't release the lock. | |
1260 | */ | |
1261 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1262 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1263 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1264 | sector_t csize; | |
1265 | ||
1266 | if (bitmap_checkpage(bitmap, page, create) < 0) { | |
1267 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); | |
1268 | *blocks = csize - (offset & (csize- 1)); | |
1269 | return NULL; | |
1270 | } | |
1271 | /* now locked ... */ | |
1272 | ||
1273 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1274 | /* should we use the first or second counter field | |
1275 | * of the hijacked pointer? */ | |
1276 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
1277 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) + | |
1278 | PAGE_COUNTER_SHIFT - 1); | |
1279 | *blocks = csize - (offset & (csize- 1)); | |
1280 | return &((bitmap_counter_t *) | |
1281 | &bitmap->bp[page].map)[hi]; | |
1282 | } else { /* page is allocated */ | |
1283 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); | |
1284 | *blocks = csize - (offset & (csize- 1)); | |
1285 | return (bitmap_counter_t *) | |
1286 | &(bitmap->bp[page].map[pageoff]); | |
1287 | } | |
1288 | } | |
1289 | ||
4b6d287f | 1290 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c N |
1291 | { |
1292 | if (!bitmap) return 0; | |
4b6d287f N |
1293 | |
1294 | if (behind) { | |
1295 | atomic_inc(&bitmap->behind_writes); | |
1296 | PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n", | |
1297 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1298 | } | |
1299 | ||
32a7627c N |
1300 | while (sectors) { |
1301 | int blocks; | |
1302 | bitmap_counter_t *bmc; | |
1303 | ||
1304 | spin_lock_irq(&bitmap->lock); | |
1305 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); | |
1306 | if (!bmc) { | |
1307 | spin_unlock_irq(&bitmap->lock); | |
1308 | return 0; | |
1309 | } | |
1310 | ||
da6e1a32 NB |
1311 | if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) { |
1312 | DEFINE_WAIT(__wait); | |
1313 | /* note that it is safe to do the prepare_to_wait | |
1314 | * after the test as long as we do it before dropping | |
1315 | * the spinlock. | |
1316 | */ | |
1317 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1318 | TASK_UNINTERRUPTIBLE); | |
1319 | spin_unlock_irq(&bitmap->lock); | |
2ad8b1ef | 1320 | blk_unplug(bitmap->mddev->queue); |
da6e1a32 NB |
1321 | schedule(); |
1322 | finish_wait(&bitmap->overflow_wait, &__wait); | |
1323 | continue; | |
1324 | } | |
1325 | ||
32a7627c N |
1326 | switch(*bmc) { |
1327 | case 0: | |
1328 | bitmap_file_set_bit(bitmap, offset); | |
1329 | bitmap_count_page(bitmap,offset, 1); | |
93769f58 | 1330 | blk_plug_device_unlocked(bitmap->mddev->queue); |
32a7627c N |
1331 | /* fall through */ |
1332 | case 1: | |
1333 | *bmc = 2; | |
1334 | } | |
da6e1a32 | 1335 | |
32a7627c N |
1336 | (*bmc)++; |
1337 | ||
1338 | spin_unlock_irq(&bitmap->lock); | |
1339 | ||
1340 | offset += blocks; | |
1341 | if (sectors > blocks) | |
1342 | sectors -= blocks; | |
1343 | else sectors = 0; | |
1344 | } | |
8311c29d | 1345 | bitmap->allclean = 0; |
32a7627c N |
1346 | return 0; |
1347 | } | |
1348 | ||
1349 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1350 | int success, int behind) |
32a7627c N |
1351 | { |
1352 | if (!bitmap) return; | |
4b6d287f N |
1353 | if (behind) { |
1354 | atomic_dec(&bitmap->behind_writes); | |
1355 | PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n", | |
1356 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1357 | } | |
d0a4bb49 N |
1358 | if (bitmap->mddev->degraded) |
1359 | /* Never clear bits or update events_cleared when degraded */ | |
1360 | success = 0; | |
4b6d287f | 1361 | |
32a7627c N |
1362 | while (sectors) { |
1363 | int blocks; | |
1364 | unsigned long flags; | |
1365 | bitmap_counter_t *bmc; | |
1366 | ||
1367 | spin_lock_irqsave(&bitmap->lock, flags); | |
1368 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); | |
1369 | if (!bmc) { | |
1370 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1371 | return; | |
1372 | } | |
1373 | ||
a0da84f3 NB |
1374 | if (success && |
1375 | bitmap->events_cleared < bitmap->mddev->events) { | |
1376 | bitmap->events_cleared = bitmap->mddev->events; | |
1377 | bitmap->need_sync = 1; | |
ece5cff0 | 1378 | sysfs_notify_dirent(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1379 | } |
1380 | ||
32a7627c N |
1381 | if (!success && ! (*bmc & NEEDED_MASK)) |
1382 | *bmc |= NEEDED_MASK; | |
1383 | ||
da6e1a32 NB |
1384 | if ((*bmc & COUNTER_MAX) == COUNTER_MAX) |
1385 | wake_up(&bitmap->overflow_wait); | |
1386 | ||
32a7627c N |
1387 | (*bmc)--; |
1388 | if (*bmc <= 2) { | |
1389 | set_page_attr(bitmap, | |
1390 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1391 | BITMAP_PAGE_CLEAN); | |
1392 | } | |
1393 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1394 | offset += blocks; | |
1395 | if (sectors > blocks) | |
1396 | sectors -= blocks; | |
1397 | else sectors = 0; | |
1398 | } | |
1399 | } | |
1400 | ||
1187cf0a N |
1401 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, |
1402 | int degraded) | |
32a7627c N |
1403 | { |
1404 | bitmap_counter_t *bmc; | |
1405 | int rv; | |
1406 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1407 | *blocks = 1024; | |
1408 | return 1; /* always resync if no bitmap */ | |
1409 | } | |
1410 | spin_lock_irq(&bitmap->lock); | |
1411 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1412 | rv = 0; | |
1413 | if (bmc) { | |
1414 | /* locked */ | |
1415 | if (RESYNC(*bmc)) | |
1416 | rv = 1; | |
1417 | else if (NEEDED(*bmc)) { | |
1418 | rv = 1; | |
6a806c51 N |
1419 | if (!degraded) { /* don't set/clear bits if degraded */ |
1420 | *bmc |= RESYNC_MASK; | |
1421 | *bmc &= ~NEEDED_MASK; | |
1422 | } | |
32a7627c N |
1423 | } |
1424 | } | |
1425 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1426 | bitmap->allclean = 0; |
32a7627c N |
1427 | return rv; |
1428 | } | |
1429 | ||
1187cf0a N |
1430 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, |
1431 | int degraded) | |
1432 | { | |
1433 | /* bitmap_start_sync must always report on multiples of whole | |
1434 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1435 | * get confused. | |
1436 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1437 | * At least PAGE_SIZE>>9 blocks are covered. | |
1438 | * Return the 'or' of the result. | |
1439 | */ | |
1440 | int rv = 0; | |
1441 | int blocks1; | |
1442 | ||
1443 | *blocks = 0; | |
1444 | while (*blocks < (PAGE_SIZE>>9)) { | |
1445 | rv |= __bitmap_start_sync(bitmap, offset, | |
1446 | &blocks1, degraded); | |
1447 | offset += blocks1; | |
1448 | *blocks += blocks1; | |
1449 | } | |
1450 | return rv; | |
1451 | } | |
1452 | ||
32a7627c N |
1453 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted) |
1454 | { | |
1455 | bitmap_counter_t *bmc; | |
1456 | unsigned long flags; | |
1457 | /* | |
1458 | if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted); | |
1459 | */ if (bitmap == NULL) { | |
1460 | *blocks = 1024; | |
1461 | return; | |
1462 | } | |
1463 | spin_lock_irqsave(&bitmap->lock, flags); | |
1464 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1465 | if (bmc == NULL) | |
1466 | goto unlock; | |
1467 | /* locked */ | |
1468 | /* | |
1469 | if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks); | |
1470 | */ | |
1471 | if (RESYNC(*bmc)) { | |
1472 | *bmc &= ~RESYNC_MASK; | |
1473 | ||
1474 | if (!NEEDED(*bmc) && aborted) | |
1475 | *bmc |= NEEDED_MASK; | |
1476 | else { | |
1477 | if (*bmc <= 2) { | |
1478 | set_page_attr(bitmap, | |
1479 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1480 | BITMAP_PAGE_CLEAN); | |
1481 | } | |
1482 | } | |
1483 | } | |
1484 | unlock: | |
1485 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
8311c29d | 1486 | bitmap->allclean = 0; |
32a7627c N |
1487 | } |
1488 | ||
1489 | void bitmap_close_sync(struct bitmap *bitmap) | |
1490 | { | |
1491 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1492 | * properly have been aborted. It remains to us to clear the | |
1493 | * RESYNC bit wherever it is still on | |
1494 | */ | |
1495 | sector_t sector = 0; | |
1496 | int blocks; | |
b47490c9 N |
1497 | if (!bitmap) |
1498 | return; | |
32a7627c N |
1499 | while (sector < bitmap->mddev->resync_max_sectors) { |
1500 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1501 | sector += blocks; |
1502 | } | |
1503 | } | |
1504 | ||
1505 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1506 | { | |
1507 | sector_t s = 0; | |
1508 | int blocks; | |
1509 | ||
1510 | if (!bitmap) | |
1511 | return; | |
1512 | if (sector == 0) { | |
1513 | bitmap->last_end_sync = jiffies; | |
1514 | return; | |
1515 | } | |
1516 | if (time_before(jiffies, (bitmap->last_end_sync | |
1b04be96 | 1517 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1518 | return; |
1519 | wait_event(bitmap->mddev->recovery_wait, | |
1520 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1521 | ||
97e4f42d N |
1522 | bitmap->mddev->curr_resync_completed = bitmap->mddev->curr_resync; |
1523 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); | |
b47490c9 N |
1524 | sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1); |
1525 | s = 0; | |
1526 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1527 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1528 | s += blocks; | |
32a7627c | 1529 | } |
b47490c9 | 1530 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1531 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c N |
1532 | } |
1533 | ||
6a07997f | 1534 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1535 | { |
1536 | /* For each chunk covered by any of these sectors, set the | |
193f1c93 | 1537 | * counter to 1 and set resync_needed. They should all |
32a7627c N |
1538 | * be 0 at this point |
1539 | */ | |
193f1c93 N |
1540 | |
1541 | int secs; | |
1542 | bitmap_counter_t *bmc; | |
1543 | spin_lock_irq(&bitmap->lock); | |
1544 | bmc = bitmap_get_counter(bitmap, offset, &secs, 1); | |
1545 | if (!bmc) { | |
32a7627c | 1546 | spin_unlock_irq(&bitmap->lock); |
193f1c93 | 1547 | return; |
32a7627c | 1548 | } |
193f1c93 N |
1549 | if (! *bmc) { |
1550 | struct page *page; | |
6a07997f | 1551 | *bmc = 1 | (needed?NEEDED_MASK:0); |
193f1c93 N |
1552 | bitmap_count_page(bitmap, offset, 1); |
1553 | page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)); | |
1554 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1555 | } | |
1556 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1557 | bitmap->allclean = 0; |
32a7627c N |
1558 | } |
1559 | ||
9b1d1dac PC |
1560 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1561 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1562 | { | |
1563 | unsigned long chunk; | |
1564 | ||
1565 | for (chunk = s; chunk <= e; chunk++) { | |
db305e50 | 1566 | sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap); |
9b1d1dac PC |
1567 | bitmap_set_memory_bits(bitmap, sec, 1); |
1568 | bitmap_file_set_bit(bitmap, sec); | |
ffa23322 N |
1569 | if (sec < bitmap->mddev->recovery_cp) |
1570 | /* We are asserting that the array is dirty, | |
1571 | * so move the recovery_cp address back so | |
1572 | * that it is obvious that it is dirty | |
1573 | */ | |
1574 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1575 | } |
1576 | } | |
1577 | ||
6b8b3e8a N |
1578 | /* |
1579 | * flush out any pending updates | |
1580 | */ | |
1581 | void bitmap_flush(mddev_t *mddev) | |
1582 | { | |
1583 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1584 | long sleep; |
6b8b3e8a N |
1585 | |
1586 | if (!bitmap) /* there was no bitmap */ | |
1587 | return; | |
1588 | ||
1589 | /* run the daemon_work three time to ensure everything is flushed | |
1590 | * that can be | |
1591 | */ | |
1b04be96 | 1592 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1593 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1594 | bitmap_daemon_work(mddev); |
42a04b50 | 1595 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1596 | bitmap_daemon_work(mddev); |
42a04b50 | 1597 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1598 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1599 | bitmap_update_sb(bitmap); |
1600 | } | |
1601 | ||
32a7627c N |
1602 | /* |
1603 | * free memory that was allocated | |
1604 | */ | |
3178b0db | 1605 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1606 | { |
1607 | unsigned long k, pages; | |
1608 | struct bitmap_page *bp; | |
32a7627c N |
1609 | |
1610 | if (!bitmap) /* there was no bitmap */ | |
1611 | return; | |
1612 | ||
32a7627c N |
1613 | /* release the bitmap file and kill the daemon */ |
1614 | bitmap_file_put(bitmap); | |
1615 | ||
1616 | bp = bitmap->bp; | |
1617 | pages = bitmap->pages; | |
1618 | ||
1619 | /* free all allocated memory */ | |
1620 | ||
32a7627c N |
1621 | if (bp) /* deallocate the page memory */ |
1622 | for (k = 0; k < pages; k++) | |
1623 | if (bp[k].map && !bp[k].hijacked) | |
1624 | kfree(bp[k].map); | |
1625 | kfree(bp); | |
1626 | kfree(bitmap); | |
1627 | } | |
aa5cbd10 | 1628 | |
3178b0db N |
1629 | void bitmap_destroy(mddev_t *mddev) |
1630 | { | |
1631 | struct bitmap *bitmap = mddev->bitmap; | |
1632 | ||
1633 | if (!bitmap) /* there was no bitmap */ | |
1634 | return; | |
1635 | ||
c3d9714e | 1636 | mutex_lock(&mddev->bitmap_info.mutex); |
3178b0db | 1637 | mddev->bitmap = NULL; /* disconnect from the md device */ |
c3d9714e | 1638 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1639 | if (mddev->thread) |
1640 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db | 1641 | |
ece5cff0 N |
1642 | if (bitmap->sysfs_can_clear) |
1643 | sysfs_put(bitmap->sysfs_can_clear); | |
1644 | ||
3178b0db N |
1645 | bitmap_free(bitmap); |
1646 | } | |
32a7627c N |
1647 | |
1648 | /* | |
1649 | * initialize the bitmap structure | |
1650 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1651 | */ | |
1652 | int bitmap_create(mddev_t *mddev) | |
1653 | { | |
1654 | struct bitmap *bitmap; | |
1f593903 | 1655 | sector_t blocks = mddev->resync_max_sectors; |
32a7627c N |
1656 | unsigned long chunks; |
1657 | unsigned long pages; | |
c3d9714e | 1658 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1659 | int err; |
6a07997f | 1660 | sector_t start; |
ece5cff0 | 1661 | struct sysfs_dirent *bm; |
32a7627c | 1662 | |
5f6e3c83 | 1663 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1664 | |
c3d9714e | 1665 | if (!file && !mddev->bitmap_info.offset) /* bitmap disabled, nothing to do */ |
32a7627c N |
1666 | return 0; |
1667 | ||
c3d9714e | 1668 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1669 | |
9ffae0cf | 1670 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1671 | if (!bitmap) |
1672 | return -ENOMEM; | |
1673 | ||
32a7627c | 1674 | spin_lock_init(&bitmap->lock); |
ce25c31b N |
1675 | atomic_set(&bitmap->pending_writes, 0); |
1676 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1677 | init_waitqueue_head(&bitmap->overflow_wait); |
ce25c31b | 1678 | |
32a7627c | 1679 | bitmap->mddev = mddev; |
32a7627c | 1680 | |
ece5cff0 N |
1681 | bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap"); |
1682 | if (bm) { | |
1683 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear"); | |
1684 | sysfs_put(bm); | |
1685 | } else | |
1686 | bitmap->sysfs_can_clear = NULL; | |
1687 | ||
32a7627c | 1688 | bitmap->file = file; |
ce25c31b N |
1689 | if (file) { |
1690 | get_file(file); | |
ae8fa283 N |
1691 | /* As future accesses to this file will use bmap, |
1692 | * and bypass the page cache, we must sync the file | |
1693 | * first. | |
1694 | */ | |
1695 | vfs_fsync(file, file->f_dentry, 1); | |
ce25c31b | 1696 | } |
42a04b50 | 1697 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
ece5cff0 N |
1698 | if (!mddev->bitmap_info.external) |
1699 | err = bitmap_read_sb(bitmap); | |
1700 | else { | |
1701 | err = 0; | |
1702 | if (mddev->bitmap_info.chunksize == 0 || | |
1703 | mddev->bitmap_info.daemon_sleep == 0) | |
1704 | /* chunksize and time_base need to be | |
1705 | * set first. */ | |
1706 | err = -EINVAL; | |
1707 | } | |
32a7627c | 1708 | if (err) |
3178b0db | 1709 | goto error; |
32a7627c | 1710 | |
624ce4f5 | 1711 | bitmap->daemon_lastrun = jiffies; |
42a04b50 | 1712 | bitmap->chunkshift = ffz(~mddev->bitmap_info.chunksize); |
32a7627c N |
1713 | |
1714 | /* now that chunksize and chunkshift are set, we can use these macros */ | |
1f593903 N |
1715 | chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >> |
1716 | CHUNK_BLOCK_SHIFT(bitmap); | |
32a7627c N |
1717 | pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; |
1718 | ||
1719 | BUG_ON(!pages); | |
1720 | ||
1721 | bitmap->chunks = chunks; | |
1722 | bitmap->pages = pages; | |
1723 | bitmap->missing_pages = pages; | |
1724 | bitmap->counter_bits = COUNTER_BITS; | |
1725 | ||
1726 | bitmap->syncchunk = ~0UL; | |
1727 | ||
44456d37 | 1728 | #ifdef INJECT_FATAL_FAULT_1 |
32a7627c N |
1729 | bitmap->bp = NULL; |
1730 | #else | |
9ffae0cf | 1731 | bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); |
32a7627c | 1732 | #endif |
3178b0db | 1733 | err = -ENOMEM; |
32a7627c | 1734 | if (!bitmap->bp) |
3178b0db | 1735 | goto error; |
32a7627c | 1736 | |
32a7627c N |
1737 | /* now that we have some pages available, initialize the in-memory |
1738 | * bitmap from the on-disk bitmap */ | |
6a07997f N |
1739 | start = 0; |
1740 | if (mddev->degraded == 0 | |
1741 | || bitmap->events_cleared == mddev->events) | |
1742 | /* no need to keep dirty bits to optimise a re-add of a missing device */ | |
1743 | start = mddev->recovery_cp; | |
1744 | err = bitmap_init_from_disk(bitmap, start); | |
193f1c93 | 1745 | |
32a7627c | 1746 | if (err) |
3178b0db | 1747 | goto error; |
32a7627c N |
1748 | |
1749 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", | |
1750 | pages, bmname(bitmap)); | |
1751 | ||
3178b0db N |
1752 | mddev->bitmap = bitmap; |
1753 | ||
1b04be96 | 1754 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1755 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1756 | |
4ad13663 N |
1757 | bitmap_update_sb(bitmap); |
1758 | ||
1759 | return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0; | |
3178b0db N |
1760 | |
1761 | error: | |
1762 | bitmap_free(bitmap); | |
1763 | return err; | |
32a7627c N |
1764 | } |
1765 | ||
43a70507 N |
1766 | static ssize_t |
1767 | location_show(mddev_t *mddev, char *page) | |
1768 | { | |
1769 | ssize_t len; | |
1770 | if (mddev->bitmap_info.file) { | |
1771 | len = sprintf(page, "file"); | |
1772 | } else if (mddev->bitmap_info.offset) { | |
1773 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); | |
1774 | } else | |
1775 | len = sprintf(page, "none"); | |
1776 | len += sprintf(page+len, "\n"); | |
1777 | return len; | |
1778 | } | |
1779 | ||
1780 | static ssize_t | |
1781 | location_store(mddev_t *mddev, const char *buf, size_t len) | |
1782 | { | |
1783 | ||
1784 | if (mddev->pers) { | |
1785 | if (!mddev->pers->quiesce) | |
1786 | return -EBUSY; | |
1787 | if (mddev->recovery || mddev->sync_thread) | |
1788 | return -EBUSY; | |
1789 | } | |
1790 | ||
1791 | if (mddev->bitmap || mddev->bitmap_info.file || | |
1792 | mddev->bitmap_info.offset) { | |
1793 | /* bitmap already configured. Only option is to clear it */ | |
1794 | if (strncmp(buf, "none", 4) != 0) | |
1795 | return -EBUSY; | |
1796 | if (mddev->pers) { | |
1797 | mddev->pers->quiesce(mddev, 1); | |
1798 | bitmap_destroy(mddev); | |
1799 | mddev->pers->quiesce(mddev, 0); | |
1800 | } | |
1801 | mddev->bitmap_info.offset = 0; | |
1802 | if (mddev->bitmap_info.file) { | |
1803 | struct file *f = mddev->bitmap_info.file; | |
1804 | mddev->bitmap_info.file = NULL; | |
1805 | restore_bitmap_write_access(f); | |
1806 | fput(f); | |
1807 | } | |
1808 | } else { | |
1809 | /* No bitmap, OK to set a location */ | |
1810 | long long offset; | |
1811 | if (strncmp(buf, "none", 4) == 0) | |
1812 | /* nothing to be done */; | |
1813 | else if (strncmp(buf, "file:", 5) == 0) { | |
1814 | /* Not supported yet */ | |
1815 | return -EINVAL; | |
1816 | } else { | |
1817 | int rv; | |
1818 | if (buf[0] == '+') | |
1819 | rv = strict_strtoll(buf+1, 10, &offset); | |
1820 | else | |
1821 | rv = strict_strtoll(buf, 10, &offset); | |
1822 | if (rv) | |
1823 | return rv; | |
1824 | if (offset == 0) | |
1825 | return -EINVAL; | |
ece5cff0 N |
1826 | if (mddev->bitmap_info.external == 0 && |
1827 | mddev->major_version == 0 && | |
43a70507 N |
1828 | offset != mddev->bitmap_info.default_offset) |
1829 | return -EINVAL; | |
1830 | mddev->bitmap_info.offset = offset; | |
1831 | if (mddev->pers) { | |
1832 | mddev->pers->quiesce(mddev, 1); | |
1833 | rv = bitmap_create(mddev); | |
1834 | if (rv) { | |
1835 | bitmap_destroy(mddev); | |
1836 | mddev->bitmap_info.offset = 0; | |
1837 | } | |
1838 | mddev->pers->quiesce(mddev, 0); | |
1839 | if (rv) | |
1840 | return rv; | |
1841 | } | |
1842 | } | |
1843 | } | |
1844 | if (!mddev->external) { | |
1845 | /* Ensure new bitmap info is stored in | |
1846 | * metadata promptly. | |
1847 | */ | |
1848 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
1849 | md_wakeup_thread(mddev->thread); | |
1850 | } | |
1851 | return len; | |
1852 | } | |
1853 | ||
1854 | static struct md_sysfs_entry bitmap_location = | |
1855 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
1856 | ||
1857 | static ssize_t | |
1858 | timeout_show(mddev_t *mddev, char *page) | |
1859 | { | |
1860 | ssize_t len; | |
1861 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
1862 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
1863 | ||
1864 | len = sprintf(page, "%lu", secs); | |
1865 | if (jifs) | |
1866 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
1867 | len += sprintf(page+len, "\n"); | |
1868 | return len; | |
1869 | } | |
1870 | ||
1871 | static ssize_t | |
1872 | timeout_store(mddev_t *mddev, const char *buf, size_t len) | |
1873 | { | |
1874 | /* timeout can be set at any time */ | |
1875 | unsigned long timeout; | |
1876 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
1877 | if (rv) | |
1878 | return rv; | |
1879 | ||
1880 | /* just to make sure we don't overflow... */ | |
1881 | if (timeout >= LONG_MAX / HZ) | |
1882 | return -EINVAL; | |
1883 | ||
1884 | timeout = timeout * HZ / 10000; | |
1885 | ||
1886 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
1887 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
1888 | if (timeout < 1) | |
1889 | timeout = 1; | |
1890 | mddev->bitmap_info.daemon_sleep = timeout; | |
1891 | if (mddev->thread) { | |
1892 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
1893 | * the bitmap is all clean and we don't need to | |
1894 | * adjust the timeout right now | |
1895 | */ | |
1896 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
1897 | mddev->thread->timeout = timeout; | |
1898 | md_wakeup_thread(mddev->thread); | |
1899 | } | |
1900 | } | |
1901 | return len; | |
1902 | } | |
1903 | ||
1904 | static struct md_sysfs_entry bitmap_timeout = | |
1905 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
1906 | ||
1907 | static ssize_t | |
1908 | backlog_show(mddev_t *mddev, char *page) | |
1909 | { | |
1910 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
1911 | } | |
1912 | ||
1913 | static ssize_t | |
1914 | backlog_store(mddev_t *mddev, const char *buf, size_t len) | |
1915 | { | |
1916 | unsigned long backlog; | |
1917 | int rv = strict_strtoul(buf, 10, &backlog); | |
1918 | if (rv) | |
1919 | return rv; | |
1920 | if (backlog > COUNTER_MAX) | |
1921 | return -EINVAL; | |
1922 | mddev->bitmap_info.max_write_behind = backlog; | |
1923 | return len; | |
1924 | } | |
1925 | ||
1926 | static struct md_sysfs_entry bitmap_backlog = | |
1927 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
1928 | ||
1929 | static ssize_t | |
1930 | chunksize_show(mddev_t *mddev, char *page) | |
1931 | { | |
1932 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
1933 | } | |
1934 | ||
1935 | static ssize_t | |
1936 | chunksize_store(mddev_t *mddev, const char *buf, size_t len) | |
1937 | { | |
1938 | /* Can only be changed when no bitmap is active */ | |
1939 | int rv; | |
1940 | unsigned long csize; | |
1941 | if (mddev->bitmap) | |
1942 | return -EBUSY; | |
1943 | rv = strict_strtoul(buf, 10, &csize); | |
1944 | if (rv) | |
1945 | return rv; | |
1946 | if (csize < 512 || | |
1947 | !is_power_of_2(csize)) | |
1948 | return -EINVAL; | |
1949 | mddev->bitmap_info.chunksize = csize; | |
1950 | return len; | |
1951 | } | |
1952 | ||
1953 | static struct md_sysfs_entry bitmap_chunksize = | |
1954 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
1955 | ||
ece5cff0 N |
1956 | static ssize_t metadata_show(mddev_t *mddev, char *page) |
1957 | { | |
1958 | return sprintf(page, "%s\n", (mddev->bitmap_info.external | |
1959 | ? "external" : "internal")); | |
1960 | } | |
1961 | ||
1962 | static ssize_t metadata_store(mddev_t *mddev, const char *buf, size_t len) | |
1963 | { | |
1964 | if (mddev->bitmap || | |
1965 | mddev->bitmap_info.file || | |
1966 | mddev->bitmap_info.offset) | |
1967 | return -EBUSY; | |
1968 | if (strncmp(buf, "external", 8) == 0) | |
1969 | mddev->bitmap_info.external = 1; | |
1970 | else if (strncmp(buf, "internal", 8) == 0) | |
1971 | mddev->bitmap_info.external = 0; | |
1972 | else | |
1973 | return -EINVAL; | |
1974 | return len; | |
1975 | } | |
1976 | ||
1977 | static struct md_sysfs_entry bitmap_metadata = | |
1978 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
1979 | ||
1980 | static ssize_t can_clear_show(mddev_t *mddev, char *page) | |
1981 | { | |
1982 | int len; | |
1983 | if (mddev->bitmap) | |
1984 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
1985 | "false" : "true")); | |
1986 | else | |
1987 | len = sprintf(page, "\n"); | |
1988 | return len; | |
1989 | } | |
1990 | ||
1991 | static ssize_t can_clear_store(mddev_t *mddev, const char *buf, size_t len) | |
1992 | { | |
1993 | if (mddev->bitmap == NULL) | |
1994 | return -ENOENT; | |
1995 | if (strncmp(buf, "false", 5) == 0) | |
1996 | mddev->bitmap->need_sync = 1; | |
1997 | else if (strncmp(buf, "true", 4) == 0) { | |
1998 | if (mddev->degraded) | |
1999 | return -EBUSY; | |
2000 | mddev->bitmap->need_sync = 0; | |
2001 | } else | |
2002 | return -EINVAL; | |
2003 | return len; | |
2004 | } | |
2005 | ||
2006 | static struct md_sysfs_entry bitmap_can_clear = | |
2007 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2008 | ||
43a70507 N |
2009 | static struct attribute *md_bitmap_attrs[] = { |
2010 | &bitmap_location.attr, | |
2011 | &bitmap_timeout.attr, | |
2012 | &bitmap_backlog.attr, | |
2013 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2014 | &bitmap_metadata.attr, |
2015 | &bitmap_can_clear.attr, | |
43a70507 N |
2016 | NULL |
2017 | }; | |
2018 | struct attribute_group md_bitmap_group = { | |
2019 | .name = "bitmap", | |
2020 | .attrs = md_bitmap_attrs, | |
2021 | }; | |
2022 | ||
2023 | ||
32a7627c N |
2024 | /* the bitmap API -- for raid personalities */ |
2025 | EXPORT_SYMBOL(bitmap_startwrite); | |
2026 | EXPORT_SYMBOL(bitmap_endwrite); | |
2027 | EXPORT_SYMBOL(bitmap_start_sync); | |
2028 | EXPORT_SYMBOL(bitmap_end_sync); | |
2029 | EXPORT_SYMBOL(bitmap_unplug); | |
2030 | EXPORT_SYMBOL(bitmap_close_sync); | |
b47490c9 | 2031 | EXPORT_SYMBOL(bitmap_cond_end_sync); |