]>
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 */ |
a2ed9615 N |
215 | static struct page *read_sb_page(mddev_t *mddev, long offset, |
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 N |
289 | int size = PAGE_SIZE; |
290 | if (page->index == bitmap->file_pages-1) | |
291 | size = roundup(bitmap->last_page_size, | |
e1defc4f | 292 | bdev_logical_block_size(rdev->bdev)); |
f0d76d70 N |
293 | /* Just make sure we aren't corrupting data or |
294 | * metadata | |
295 | */ | |
296 | if (bitmap->offset < 0) { | |
297 | /* DATA BITMAP METADATA */ | |
298 | if (bitmap->offset | |
85bfb4da | 299 | + (long)(page->index * (PAGE_SIZE/512)) |
f0d76d70 N |
300 | + size/512 > 0) |
301 | /* bitmap runs in to metadata */ | |
4b80991c | 302 | goto bad_alignment; |
58c0fed4 | 303 | if (rdev->data_offset + mddev->dev_sectors |
0f420358 | 304 | > rdev->sb_start + bitmap->offset) |
f0d76d70 | 305 | /* data runs in to bitmap */ |
4b80991c | 306 | goto bad_alignment; |
0f420358 | 307 | } else if (rdev->sb_start < rdev->data_offset) { |
f0d76d70 | 308 | /* METADATA BITMAP DATA */ |
0f420358 | 309 | if (rdev->sb_start |
f0d76d70 N |
310 | + bitmap->offset |
311 | + page->index*(PAGE_SIZE/512) + size/512 | |
312 | > rdev->data_offset) | |
313 | /* bitmap runs in to data */ | |
4b80991c | 314 | goto bad_alignment; |
f0d76d70 N |
315 | } else { |
316 | /* DATA METADATA BITMAP - no problems */ | |
317 | } | |
a654b9d8 | 318 | md_super_write(mddev, rdev, |
0f420358 | 319 | rdev->sb_start + bitmap->offset |
a654b9d8 | 320 | + page->index * (PAGE_SIZE/512), |
ab6085c7 | 321 | size, |
a654b9d8 | 322 | page); |
b2d2c4ce | 323 | } |
a654b9d8 N |
324 | |
325 | if (wait) | |
a9701a30 | 326 | md_super_wait(mddev); |
a654b9d8 | 327 | return 0; |
4b80991c N |
328 | |
329 | bad_alignment: | |
4b80991c | 330 | return -EINVAL; |
a654b9d8 N |
331 | } |
332 | ||
4ad13663 | 333 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 334 | /* |
a654b9d8 | 335 | * write out a page to a file |
32a7627c | 336 | */ |
4ad13663 | 337 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 338 | { |
d785a06a | 339 | struct buffer_head *bh; |
32a7627c | 340 | |
f0d76d70 N |
341 | if (bitmap->file == NULL) { |
342 | switch (write_sb_page(bitmap, page, wait)) { | |
343 | case -EINVAL: | |
344 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
f0d76d70 | 345 | } |
4ad13663 | 346 | } else { |
a654b9d8 | 347 | |
4ad13663 | 348 | bh = page_buffers(page); |
c708443c | 349 | |
4ad13663 N |
350 | while (bh && bh->b_blocknr) { |
351 | atomic_inc(&bitmap->pending_writes); | |
352 | set_buffer_locked(bh); | |
353 | set_buffer_mapped(bh); | |
354 | submit_bh(WRITE, bh); | |
355 | bh = bh->b_this_page; | |
356 | } | |
d785a06a | 357 | |
4ad13663 N |
358 | if (wait) { |
359 | wait_event(bitmap->write_wait, | |
360 | atomic_read(&bitmap->pending_writes)==0); | |
361 | } | |
8a5e9cf1 | 362 | } |
4ad13663 N |
363 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
364 | bitmap_file_kick(bitmap); | |
d785a06a N |
365 | } |
366 | ||
367 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
368 | { | |
369 | struct bitmap *bitmap = bh->b_private; | |
370 | unsigned long flags; | |
32a7627c | 371 | |
d785a06a N |
372 | if (!uptodate) { |
373 | spin_lock_irqsave(&bitmap->lock, flags); | |
374 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
375 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
32a7627c | 376 | } |
d785a06a N |
377 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
378 | wake_up(&bitmap->write_wait); | |
379 | } | |
32a7627c | 380 | |
d785a06a N |
381 | /* copied from buffer.c */ |
382 | static void | |
383 | __clear_page_buffers(struct page *page) | |
384 | { | |
385 | ClearPagePrivate(page); | |
386 | set_page_private(page, 0); | |
387 | page_cache_release(page); | |
388 | } | |
389 | static void free_buffers(struct page *page) | |
390 | { | |
391 | struct buffer_head *bh = page_buffers(page); | |
77ad4bc7 | 392 | |
d785a06a N |
393 | while (bh) { |
394 | struct buffer_head *next = bh->b_this_page; | |
395 | free_buffer_head(bh); | |
396 | bh = next; | |
77ad4bc7 | 397 | } |
d785a06a N |
398 | __clear_page_buffers(page); |
399 | put_page(page); | |
32a7627c N |
400 | } |
401 | ||
d785a06a N |
402 | /* read a page from a file. |
403 | * We both read the page, and attach buffers to the page to record the | |
404 | * address of each block (using bmap). These addresses will be used | |
405 | * to write the block later, completely bypassing the filesystem. | |
406 | * This usage is similar to how swap files are handled, and allows us | |
407 | * to write to a file with no concerns of memory allocation failing. | |
408 | */ | |
32a7627c | 409 | static struct page *read_page(struct file *file, unsigned long index, |
d785a06a N |
410 | struct bitmap *bitmap, |
411 | unsigned long count) | |
32a7627c | 412 | { |
32a7627c | 413 | struct page *page = NULL; |
c649bb9c | 414 | struct inode *inode = file->f_path.dentry->d_inode; |
d785a06a N |
415 | struct buffer_head *bh; |
416 | sector_t block; | |
32a7627c | 417 | |
2d1f3b5d N |
418 | PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE, |
419 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 420 | |
d785a06a N |
421 | page = alloc_page(GFP_KERNEL); |
422 | if (!page) | |
423 | page = ERR_PTR(-ENOMEM); | |
32a7627c N |
424 | if (IS_ERR(page)) |
425 | goto out; | |
d785a06a | 426 | |
d785a06a N |
427 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
428 | if (!bh) { | |
2d1f3b5d | 429 | put_page(page); |
d785a06a | 430 | page = ERR_PTR(-ENOMEM); |
32a7627c N |
431 | goto out; |
432 | } | |
d785a06a N |
433 | attach_page_buffers(page, bh); |
434 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
435 | while (bh) { | |
436 | if (count == 0) | |
437 | bh->b_blocknr = 0; | |
438 | else { | |
439 | bh->b_blocknr = bmap(inode, block); | |
440 | if (bh->b_blocknr == 0) { | |
441 | /* Cannot use this file! */ | |
442 | free_buffers(page); | |
443 | page = ERR_PTR(-EINVAL); | |
444 | goto out; | |
445 | } | |
446 | bh->b_bdev = inode->i_sb->s_bdev; | |
447 | if (count < (1<<inode->i_blkbits)) | |
448 | count = 0; | |
449 | else | |
450 | count -= (1<<inode->i_blkbits); | |
451 | ||
452 | bh->b_end_io = end_bitmap_write; | |
453 | bh->b_private = bitmap; | |
ce25c31b N |
454 | atomic_inc(&bitmap->pending_writes); |
455 | set_buffer_locked(bh); | |
456 | set_buffer_mapped(bh); | |
457 | submit_bh(READ, bh); | |
d785a06a N |
458 | } |
459 | block++; | |
460 | bh = bh->b_this_page; | |
461 | } | |
d785a06a | 462 | page->index = index; |
ce25c31b N |
463 | |
464 | wait_event(bitmap->write_wait, | |
465 | atomic_read(&bitmap->pending_writes)==0); | |
466 | if (bitmap->flags & BITMAP_WRITE_ERROR) { | |
467 | free_buffers(page); | |
468 | page = ERR_PTR(-EIO); | |
469 | } | |
32a7627c N |
470 | out: |
471 | if (IS_ERR(page)) | |
472 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n", | |
2d1f3b5d N |
473 | (int)PAGE_SIZE, |
474 | (unsigned long long)index << PAGE_SHIFT, | |
32a7627c N |
475 | PTR_ERR(page)); |
476 | return page; | |
477 | } | |
478 | ||
479 | /* | |
480 | * bitmap file superblock operations | |
481 | */ | |
482 | ||
483 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 484 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
485 | { |
486 | bitmap_super_t *sb; | |
487 | unsigned long flags; | |
488 | ||
489 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 490 | return; |
32a7627c N |
491 | spin_lock_irqsave(&bitmap->lock, flags); |
492 | if (!bitmap->sb_page) { /* no superblock */ | |
493 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 494 | return; |
32a7627c | 495 | } |
32a7627c | 496 | spin_unlock_irqrestore(&bitmap->lock, flags); |
ea03aff9 | 497 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 498 | sb->events = cpu_to_le64(bitmap->mddev->events); |
a0da84f3 NB |
499 | if (bitmap->mddev->events < bitmap->events_cleared) { |
500 | /* rocking back to read-only */ | |
501 | bitmap->events_cleared = bitmap->mddev->events; | |
502 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); | |
503 | } | |
ea03aff9 | 504 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 505 | write_page(bitmap, bitmap->sb_page, 1); |
32a7627c N |
506 | } |
507 | ||
508 | /* print out the bitmap file superblock */ | |
509 | void bitmap_print_sb(struct bitmap *bitmap) | |
510 | { | |
511 | bitmap_super_t *sb; | |
512 | ||
513 | if (!bitmap || !bitmap->sb_page) | |
514 | return; | |
ea03aff9 | 515 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 516 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
517 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
518 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
519 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
520 | *(__u32 *)(sb->uuid+0), |
521 | *(__u32 *)(sb->uuid+4), | |
522 | *(__u32 *)(sb->uuid+8), | |
523 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 524 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 525 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 526 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 527 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
528 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
529 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
530 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
531 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
532 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 533 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
ea03aff9 | 534 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
535 | } |
536 | ||
537 | /* read the superblock from the bitmap file and initialize some bitmap fields */ | |
538 | static int bitmap_read_sb(struct bitmap *bitmap) | |
539 | { | |
540 | char *reason = NULL; | |
541 | bitmap_super_t *sb; | |
4b6d287f | 542 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c N |
543 | unsigned long long events; |
544 | int err = -EINVAL; | |
545 | ||
546 | /* page 0 is the superblock, read it... */ | |
f49d5e62 N |
547 | if (bitmap->file) { |
548 | loff_t isize = i_size_read(bitmap->file->f_mapping->host); | |
549 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; | |
550 | ||
551 | bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes); | |
552 | } else { | |
a2ed9615 N |
553 | bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, |
554 | NULL, | |
555 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 556 | } |
32a7627c N |
557 | if (IS_ERR(bitmap->sb_page)) { |
558 | err = PTR_ERR(bitmap->sb_page); | |
559 | bitmap->sb_page = NULL; | |
560 | return err; | |
561 | } | |
562 | ||
ea03aff9 | 563 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 564 | |
32a7627c N |
565 | chunksize = le32_to_cpu(sb->chunksize); |
566 | daemon_sleep = le32_to_cpu(sb->daemon_sleep); | |
4b6d287f | 567 | write_behind = le32_to_cpu(sb->write_behind); |
32a7627c N |
568 | |
569 | /* verify that the bitmap-specific fields are valid */ | |
570 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
571 | reason = "bad magic"; | |
bd926c63 N |
572 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
573 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 574 | reason = "unrecognized superblock version"; |
1187cf0a | 575 | else if (chunksize < 512) |
7dd5d34c | 576 | reason = "bitmap chunksize too small"; |
32a7627c N |
577 | else if ((1 << ffz(~chunksize)) != chunksize) |
578 | reason = "bitmap chunksize not a power of 2"; | |
7dd5d34c N |
579 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ) |
580 | reason = "daemon sleep period out of range"; | |
4b6d287f N |
581 | else if (write_behind > COUNTER_MAX) |
582 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
583 | if (reason) { |
584 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
585 | bmname(bitmap), reason); | |
586 | goto out; | |
587 | } | |
588 | ||
589 | /* keep the array size field of the bitmap superblock up to date */ | |
590 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
591 | ||
592 | if (!bitmap->mddev->persistent) | |
593 | goto success; | |
594 | ||
595 | /* | |
596 | * if we have a persistent array superblock, compare the | |
597 | * bitmap's UUID and event counter to the mddev's | |
598 | */ | |
599 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
600 | printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n", | |
601 | bmname(bitmap)); | |
602 | goto out; | |
603 | } | |
604 | events = le64_to_cpu(sb->events); | |
605 | if (events < bitmap->mddev->events) { | |
606 | printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) " | |
607 | "-- forcing full recovery\n", bmname(bitmap), events, | |
608 | (unsigned long long) bitmap->mddev->events); | |
4f2e639a | 609 | sb->state |= cpu_to_le32(BITMAP_STALE); |
32a7627c N |
610 | } |
611 | success: | |
612 | /* assign fields using values from superblock */ | |
613 | bitmap->chunksize = chunksize; | |
614 | bitmap->daemon_sleep = daemon_sleep; | |
585f0dd5 | 615 | bitmap->daemon_lastrun = jiffies; |
4b6d287f | 616 | bitmap->max_write_behind = write_behind; |
4f2e639a | 617 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 N |
618 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
619 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
32a7627c | 620 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
4f2e639a | 621 | if (sb->state & cpu_to_le32(BITMAP_STALE)) |
6a07997f | 622 | bitmap->events_cleared = bitmap->mddev->events; |
32a7627c N |
623 | err = 0; |
624 | out: | |
ea03aff9 | 625 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
626 | if (err) |
627 | bitmap_print_sb(bitmap); | |
628 | return err; | |
629 | } | |
630 | ||
631 | enum bitmap_mask_op { | |
632 | MASK_SET, | |
633 | MASK_UNSET | |
634 | }; | |
635 | ||
4ad13663 N |
636 | /* record the state of the bitmap in the superblock. Return the old value */ |
637 | static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits, | |
638 | enum bitmap_mask_op op) | |
32a7627c N |
639 | { |
640 | bitmap_super_t *sb; | |
641 | unsigned long flags; | |
4ad13663 | 642 | int old; |
32a7627c N |
643 | |
644 | spin_lock_irqsave(&bitmap->lock, flags); | |
7e317655 | 645 | if (!bitmap->sb_page) { /* can't set the state */ |
32a7627c | 646 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 647 | return 0; |
32a7627c | 648 | } |
32a7627c | 649 | spin_unlock_irqrestore(&bitmap->lock, flags); |
ea03aff9 | 650 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
4ad13663 | 651 | old = le32_to_cpu(sb->state) & bits; |
32a7627c | 652 | switch (op) { |
4f2e639a | 653 | case MASK_SET: sb->state |= cpu_to_le32(bits); |
32a7627c | 654 | break; |
4f2e639a | 655 | case MASK_UNSET: sb->state &= cpu_to_le32(~bits); |
32a7627c N |
656 | break; |
657 | default: BUG(); | |
658 | } | |
ea03aff9 | 659 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 660 | return old; |
32a7627c N |
661 | } |
662 | ||
663 | /* | |
664 | * general bitmap file operations | |
665 | */ | |
666 | ||
667 | /* calculate the index of the page that contains this bit */ | |
668 | static inline unsigned long file_page_index(unsigned long chunk) | |
669 | { | |
670 | return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT; | |
671 | } | |
672 | ||
673 | /* calculate the (bit) offset of this bit within a page */ | |
674 | static inline unsigned long file_page_offset(unsigned long chunk) | |
675 | { | |
676 | return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1); | |
677 | } | |
678 | ||
679 | /* | |
680 | * return a pointer to the page in the filemap that contains the given bit | |
681 | * | |
682 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
683 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
684 | * 0 or page 1 | |
685 | */ | |
686 | static inline struct page *filemap_get_page(struct bitmap *bitmap, | |
687 | unsigned long chunk) | |
688 | { | |
9b1d1dac | 689 | if (file_page_index(chunk) >= bitmap->file_pages) return NULL; |
32a7627c N |
690 | return bitmap->filemap[file_page_index(chunk) - file_page_index(0)]; |
691 | } | |
692 | ||
693 | ||
694 | static void bitmap_file_unmap(struct bitmap *bitmap) | |
695 | { | |
696 | struct page **map, *sb_page; | |
697 | unsigned long *attr; | |
698 | int pages; | |
699 | unsigned long flags; | |
700 | ||
701 | spin_lock_irqsave(&bitmap->lock, flags); | |
702 | map = bitmap->filemap; | |
703 | bitmap->filemap = NULL; | |
704 | attr = bitmap->filemap_attr; | |
705 | bitmap->filemap_attr = NULL; | |
706 | pages = bitmap->file_pages; | |
707 | bitmap->file_pages = 0; | |
708 | sb_page = bitmap->sb_page; | |
709 | bitmap->sb_page = NULL; | |
710 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
711 | ||
712 | while (pages--) | |
713 | if (map[pages]->index != 0) /* 0 is sb_page, release it below */ | |
d785a06a | 714 | free_buffers(map[pages]); |
32a7627c N |
715 | kfree(map); |
716 | kfree(attr); | |
717 | ||
d785a06a N |
718 | if (sb_page) |
719 | free_buffers(sb_page); | |
32a7627c N |
720 | } |
721 | ||
722 | static void bitmap_file_put(struct bitmap *bitmap) | |
723 | { | |
724 | struct file *file; | |
32a7627c N |
725 | unsigned long flags; |
726 | ||
727 | spin_lock_irqsave(&bitmap->lock, flags); | |
728 | file = bitmap->file; | |
729 | bitmap->file = NULL; | |
730 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
731 | ||
d785a06a N |
732 | if (file) |
733 | wait_event(bitmap->write_wait, | |
734 | atomic_read(&bitmap->pending_writes)==0); | |
32a7627c N |
735 | bitmap_file_unmap(bitmap); |
736 | ||
d785a06a | 737 | if (file) { |
c649bb9c | 738 | struct inode *inode = file->f_path.dentry->d_inode; |
fc0ecff6 | 739 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 740 | fput(file); |
d785a06a | 741 | } |
32a7627c N |
742 | } |
743 | ||
744 | ||
745 | /* | |
746 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
747 | * then it is no longer reliable, so we stop using it and we mark the file | |
748 | * as failed in the superblock | |
749 | */ | |
750 | static void bitmap_file_kick(struct bitmap *bitmap) | |
751 | { | |
752 | char *path, *ptr = NULL; | |
753 | ||
4ad13663 N |
754 | if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) { |
755 | bitmap_update_sb(bitmap); | |
32a7627c | 756 | |
4ad13663 N |
757 | if (bitmap->file) { |
758 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
759 | if (path) | |
6bcfd601 CH |
760 | ptr = d_path(&bitmap->file->f_path, path, |
761 | PAGE_SIZE); | |
762 | ||
32a7627c | 763 | |
4ad13663 N |
764 | printk(KERN_ALERT |
765 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 766 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 767 | |
4ad13663 N |
768 | kfree(path); |
769 | } else | |
770 | printk(KERN_ALERT | |
771 | "%s: disabling internal bitmap due to errors\n", | |
772 | bmname(bitmap)); | |
a654b9d8 | 773 | } |
32a7627c N |
774 | |
775 | bitmap_file_put(bitmap); | |
776 | ||
777 | return; | |
778 | } | |
779 | ||
780 | enum bitmap_page_attr { | |
e16b68b6 N |
781 | BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced |
782 | BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared | |
783 | BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced | |
32a7627c N |
784 | }; |
785 | ||
786 | static inline void set_page_attr(struct bitmap *bitmap, struct page *page, | |
787 | enum bitmap_page_attr attr) | |
788 | { | |
e16b68b6 | 789 | __set_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
790 | } |
791 | ||
792 | static inline void clear_page_attr(struct bitmap *bitmap, struct page *page, | |
793 | enum bitmap_page_attr attr) | |
794 | { | |
e16b68b6 | 795 | __clear_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
796 | } |
797 | ||
ec7a3197 N |
798 | static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page, |
799 | enum bitmap_page_attr attr) | |
32a7627c | 800 | { |
e16b68b6 | 801 | return test_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
802 | } |
803 | ||
804 | /* | |
805 | * bitmap_file_set_bit -- called before performing a write to the md device | |
806 | * to set (and eventually sync) a particular bit in the bitmap file | |
807 | * | |
808 | * we set the bit immediately, then we record the page number so that | |
809 | * when an unplug occurs, we can flush the dirty pages out to disk | |
810 | */ | |
811 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
812 | { | |
813 | unsigned long bit; | |
814 | struct page *page; | |
815 | void *kaddr; | |
816 | unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap); | |
817 | ||
a654b9d8 | 818 | if (!bitmap->filemap) { |
32a7627c N |
819 | return; |
820 | } | |
821 | ||
822 | page = filemap_get_page(bitmap, chunk); | |
9b1d1dac | 823 | if (!page) return; |
32a7627c N |
824 | bit = file_page_offset(chunk); |
825 | ||
32a7627c N |
826 | /* set the bit */ |
827 | kaddr = kmap_atomic(page, KM_USER0); | |
bd926c63 N |
828 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
829 | set_bit(bit, kaddr); | |
830 | else | |
831 | ext2_set_bit(bit, kaddr); | |
32a7627c N |
832 | kunmap_atomic(kaddr, KM_USER0); |
833 | PRINTK("set file bit %lu page %lu\n", bit, page->index); | |
834 | ||
835 | /* record page number so it gets flushed to disk when unplug occurs */ | |
836 | set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); | |
837 | ||
838 | } | |
839 | ||
840 | /* this gets called when the md device is ready to unplug its underlying | |
841 | * (slave) device queues -- before we let any writes go down, we need to | |
842 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 843 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 844 | { |
ec7a3197 N |
845 | unsigned long i, flags; |
846 | int dirty, need_write; | |
32a7627c N |
847 | struct page *page; |
848 | int wait = 0; | |
849 | ||
850 | if (!bitmap) | |
4ad13663 | 851 | return; |
32a7627c N |
852 | |
853 | /* look at each page to see if there are any set bits that need to be | |
854 | * flushed out to disk */ | |
855 | for (i = 0; i < bitmap->file_pages; i++) { | |
856 | spin_lock_irqsave(&bitmap->lock, flags); | |
a654b9d8 | 857 | if (!bitmap->filemap) { |
32a7627c | 858 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 859 | return; |
32a7627c N |
860 | } |
861 | page = bitmap->filemap[i]; | |
ec7a3197 N |
862 | dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
863 | need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
32a7627c N |
864 | clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
865 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
ec7a3197 | 866 | if (dirty) |
32a7627c N |
867 | wait = 1; |
868 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
869 | ||
d785a06a | 870 | if (dirty | need_write) |
4ad13663 | 871 | write_page(bitmap, page, 0); |
32a7627c N |
872 | } |
873 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
0b79ccf0 | 874 | if (bitmap->file) |
d785a06a N |
875 | wait_event(bitmap->write_wait, |
876 | atomic_read(&bitmap->pending_writes)==0); | |
0b79ccf0 | 877 | else |
a9701a30 | 878 | md_super_wait(bitmap->mddev); |
32a7627c | 879 | } |
d785a06a N |
880 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
881 | bitmap_file_kick(bitmap); | |
32a7627c N |
882 | } |
883 | ||
6a07997f | 884 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
885 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
886 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
887 | * memory mapping of the bitmap file | |
888 | * Special cases: | |
889 | * if there's no bitmap file, or if the bitmap file had been | |
890 | * previously kicked from the array, we mark all the bits as | |
891 | * 1's in order to cause a full resync. | |
6a07997f N |
892 | * |
893 | * We ignore all bits for sectors that end earlier than 'start'. | |
894 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 895 | */ |
6a07997f | 896 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
897 | { |
898 | unsigned long i, chunks, index, oldindex, bit; | |
899 | struct page *page = NULL, *oldpage = NULL; | |
900 | unsigned long num_pages, bit_cnt = 0; | |
901 | struct file *file; | |
d785a06a | 902 | unsigned long bytes, offset; |
32a7627c N |
903 | int outofdate; |
904 | int ret = -ENOSPC; | |
ea03aff9 | 905 | void *paddr; |
32a7627c N |
906 | |
907 | chunks = bitmap->chunks; | |
908 | file = bitmap->file; | |
909 | ||
a654b9d8 | 910 | BUG_ON(!file && !bitmap->offset); |
32a7627c | 911 | |
44456d37 | 912 | #ifdef INJECT_FAULTS_3 |
32a7627c N |
913 | outofdate = 1; |
914 | #else | |
915 | outofdate = bitmap->flags & BITMAP_STALE; | |
916 | #endif | |
917 | if (outofdate) | |
918 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
919 | "recovery\n", bmname(bitmap)); | |
920 | ||
921 | bytes = (chunks + 7) / 8; | |
bc7f77de | 922 | |
cdbb4cc2 | 923 | num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE; |
bc7f77de | 924 | |
a654b9d8 | 925 | if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) { |
32a7627c N |
926 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
927 | bmname(bitmap), | |
928 | (unsigned long) i_size_read(file->f_mapping->host), | |
929 | bytes + sizeof(bitmap_super_t)); | |
4ad13663 | 930 | goto err; |
32a7627c | 931 | } |
bc7f77de N |
932 | |
933 | ret = -ENOMEM; | |
934 | ||
32a7627c | 935 | bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL); |
bc7f77de | 936 | if (!bitmap->filemap) |
4ad13663 | 937 | goto err; |
32a7627c | 938 | |
e16b68b6 N |
939 | /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */ |
940 | bitmap->filemap_attr = kzalloc( | |
505fa2c4 | 941 | roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), |
e16b68b6 | 942 | GFP_KERNEL); |
bc7f77de | 943 | if (!bitmap->filemap_attr) |
4ad13663 | 944 | goto err; |
32a7627c | 945 | |
32a7627c N |
946 | oldindex = ~0L; |
947 | ||
948 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 949 | int b; |
32a7627c N |
950 | index = file_page_index(i); |
951 | bit = file_page_offset(i); | |
952 | if (index != oldindex) { /* this is a new page, read it in */ | |
d785a06a | 953 | int count; |
32a7627c | 954 | /* unmap the old page, we're done with it */ |
d785a06a | 955 | if (index == num_pages-1) |
f49d5e62 N |
956 | count = bytes + sizeof(bitmap_super_t) |
957 | - index * PAGE_SIZE; | |
d785a06a N |
958 | else |
959 | count = PAGE_SIZE; | |
32a7627c N |
960 | if (index == 0) { |
961 | /* | |
962 | * if we're here then the superblock page | |
963 | * contains some bits (PAGE_SIZE != sizeof sb) | |
964 | * we've already read it in, so just use it | |
965 | */ | |
966 | page = bitmap->sb_page; | |
967 | offset = sizeof(bitmap_super_t); | |
53845270 N |
968 | if (!file) |
969 | read_sb_page(bitmap->mddev, | |
970 | bitmap->offset, | |
971 | page, | |
972 | index, count); | |
a654b9d8 | 973 | } else if (file) { |
d785a06a | 974 | page = read_page(file, index, bitmap, count); |
a654b9d8 N |
975 | offset = 0; |
976 | } else { | |
a2ed9615 N |
977 | page = read_sb_page(bitmap->mddev, bitmap->offset, |
978 | NULL, | |
979 | index, count); | |
32a7627c N |
980 | offset = 0; |
981 | } | |
a654b9d8 N |
982 | if (IS_ERR(page)) { /* read error */ |
983 | ret = PTR_ERR(page); | |
4ad13663 | 984 | goto err; |
a654b9d8 N |
985 | } |
986 | ||
32a7627c N |
987 | oldindex = index; |
988 | oldpage = page; | |
32a7627c | 989 | |
b74fd282 N |
990 | bitmap->filemap[bitmap->file_pages++] = page; |
991 | bitmap->last_page_size = count; | |
992 | ||
32a7627c N |
993 | if (outofdate) { |
994 | /* | |
995 | * if bitmap is out of date, dirty the | |
996 | * whole page and write it out | |
997 | */ | |
ea03aff9 N |
998 | paddr = kmap_atomic(page, KM_USER0); |
999 | memset(paddr + offset, 0xff, | |
6a07997f | 1000 | PAGE_SIZE - offset); |
ea03aff9 | 1001 | kunmap_atomic(paddr, KM_USER0); |
4ad13663 N |
1002 | write_page(bitmap, page, 1); |
1003 | ||
1004 | ret = -EIO; | |
b74fd282 | 1005 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
4ad13663 | 1006 | goto err; |
32a7627c | 1007 | } |
32a7627c | 1008 | } |
ea03aff9 | 1009 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1010 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1011 | b = test_bit(bit, paddr); |
bd926c63 | 1012 | else |
ea03aff9 N |
1013 | b = ext2_test_bit(bit, paddr); |
1014 | kunmap_atomic(paddr, KM_USER0); | |
bd926c63 | 1015 | if (b) { |
32a7627c | 1016 | /* if the disk bit is set, set the memory bit */ |
db305e50 N |
1017 | int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) |
1018 | >= start); | |
1019 | bitmap_set_memory_bits(bitmap, | |
1020 | (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap), | |
1021 | needed); | |
32a7627c | 1022 | bit_cnt++; |
6a07997f | 1023 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); |
32a7627c | 1024 | } |
32a7627c N |
1025 | } |
1026 | ||
1027 | /* everything went OK */ | |
1028 | ret = 0; | |
1029 | bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET); | |
1030 | ||
32a7627c N |
1031 | if (bit_cnt) { /* Kick recovery if any bits were set */ |
1032 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
1033 | md_wakeup_thread(bitmap->mddev->thread); | |
1034 | } | |
1035 | ||
32a7627c | 1036 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
4ad13663 N |
1037 | "read %lu/%lu pages, set %lu bits\n", |
1038 | bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt); | |
1039 | ||
1040 | return 0; | |
32a7627c | 1041 | |
4ad13663 N |
1042 | err: |
1043 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1044 | bmname(bitmap), ret); | |
32a7627c N |
1045 | return ret; |
1046 | } | |
1047 | ||
a654b9d8 N |
1048 | void bitmap_write_all(struct bitmap *bitmap) |
1049 | { | |
1050 | /* We don't actually write all bitmap blocks here, | |
1051 | * just flag them as needing to be written | |
1052 | */ | |
ec7a3197 | 1053 | int i; |
a654b9d8 | 1054 | |
ec7a3197 N |
1055 | for (i=0; i < bitmap->file_pages; i++) |
1056 | set_page_attr(bitmap, bitmap->filemap[i], | |
1057 | BITMAP_PAGE_NEEDWRITE); | |
a654b9d8 N |
1058 | } |
1059 | ||
32a7627c N |
1060 | |
1061 | static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) | |
1062 | { | |
1063 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1064 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1065 | bitmap->bp[page].count += inc; | |
1066 | /* | |
1067 | if (page == 0) printk("count page 0, offset %llu: %d gives %d\n", | |
1068 | (unsigned long long)offset, inc, bitmap->bp[page].count); | |
1069 | */ | |
1070 | bitmap_checkfree(bitmap, page); | |
1071 | } | |
1072 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, | |
1073 | sector_t offset, int *blocks, | |
1074 | int create); | |
1075 | ||
1076 | /* | |
1077 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1078 | * out to disk | |
1079 | */ | |
1080 | ||
aa5cbd10 | 1081 | void bitmap_daemon_work(mddev_t *mddev) |
32a7627c | 1082 | { |
aa5cbd10 | 1083 | struct bitmap *bitmap; |
aa3163f8 | 1084 | unsigned long j; |
32a7627c N |
1085 | unsigned long flags; |
1086 | struct page *page = NULL, *lastpage = NULL; | |
32a7627c | 1087 | int blocks; |
ea03aff9 | 1088 | void *paddr; |
32a7627c | 1089 | |
aa5cbd10 N |
1090 | /* Use a mutex to guard daemon_work against |
1091 | * bitmap_destroy. | |
1092 | */ | |
c3d9714e | 1093 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1094 | bitmap = mddev->bitmap; |
1095 | if (bitmap == NULL) { | |
c3d9714e | 1096 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1097 | return; |
aa5cbd10 | 1098 | } |
32a7627c | 1099 | if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ)) |
7be3dfec N |
1100 | goto done; |
1101 | ||
32a7627c | 1102 | bitmap->daemon_lastrun = jiffies; |
8311c29d N |
1103 | if (bitmap->allclean) { |
1104 | bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
aa5cbd10 | 1105 | goto done; |
8311c29d N |
1106 | } |
1107 | bitmap->allclean = 1; | |
32a7627c | 1108 | |
be512691 | 1109 | spin_lock_irqsave(&bitmap->lock, flags); |
32a7627c N |
1110 | for (j = 0; j < bitmap->chunks; j++) { |
1111 | bitmap_counter_t *bmc; | |
be512691 | 1112 | if (!bitmap->filemap) |
32a7627c | 1113 | /* error or shutdown */ |
32a7627c | 1114 | break; |
32a7627c N |
1115 | |
1116 | page = filemap_get_page(bitmap, j); | |
32a7627c N |
1117 | |
1118 | if (page != lastpage) { | |
aa3163f8 | 1119 | /* skip this page unless it's marked as needing cleaning */ |
ec7a3197 N |
1120 | if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) { |
1121 | int need_write = test_page_attr(bitmap, page, | |
1122 | BITMAP_PAGE_NEEDWRITE); | |
a647e4bc | 1123 | if (need_write) |
aa3163f8 | 1124 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); |
a647e4bc | 1125 | |
aa3163f8 | 1126 | spin_unlock_irqrestore(&bitmap->lock, flags); |
8311c29d | 1127 | if (need_write) { |
4ad13663 | 1128 | write_page(bitmap, page, 0); |
8311c29d N |
1129 | bitmap->allclean = 0; |
1130 | } | |
be512691 N |
1131 | spin_lock_irqsave(&bitmap->lock, flags); |
1132 | j |= (PAGE_BITS - 1); | |
aa3163f8 N |
1133 | continue; |
1134 | } | |
1135 | ||
32a7627c | 1136 | /* grab the new page, sync and release the old */ |
32a7627c | 1137 | if (lastpage != NULL) { |
ec7a3197 | 1138 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1139 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1140 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 1141 | write_page(bitmap, lastpage, 0); |
32a7627c N |
1142 | } else { |
1143 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1144 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1145 | } | |
32a7627c N |
1146 | } else |
1147 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1148 | lastpage = page; | |
a0da84f3 NB |
1149 | |
1150 | /* We are possibly going to clear some bits, so make | |
1151 | * sure that events_cleared is up-to-date. | |
1152 | */ | |
1153 | if (bitmap->need_sync) { | |
1154 | bitmap_super_t *sb; | |
1155 | bitmap->need_sync = 0; | |
1156 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); | |
1157 | sb->events_cleared = | |
1158 | cpu_to_le64(bitmap->events_cleared); | |
1159 | kunmap_atomic(sb, KM_USER0); | |
1160 | write_page(bitmap, bitmap->sb_page, 1); | |
1161 | } | |
32a7627c N |
1162 | spin_lock_irqsave(&bitmap->lock, flags); |
1163 | clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1164 | } | |
db305e50 N |
1165 | bmc = bitmap_get_counter(bitmap, |
1166 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
1167 | &blocks, 0); | |
32a7627c N |
1168 | if (bmc) { |
1169 | /* | |
1170 | if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc); | |
1171 | */ | |
8311c29d N |
1172 | if (*bmc) |
1173 | bitmap->allclean = 0; | |
1174 | ||
32a7627c N |
1175 | if (*bmc == 2) { |
1176 | *bmc=1; /* maybe clear the bit next time */ | |
1177 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1178 | } else if (*bmc == 1) { | |
1179 | /* we can clear the bit */ | |
1180 | *bmc = 0; | |
db305e50 N |
1181 | bitmap_count_page(bitmap, |
1182 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
32a7627c N |
1183 | -1); |
1184 | ||
1185 | /* clear the bit */ | |
ea03aff9 | 1186 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1187 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1188 | clear_bit(file_page_offset(j), paddr); |
bd926c63 | 1189 | else |
ea03aff9 N |
1190 | ext2_clear_bit(file_page_offset(j), paddr); |
1191 | kunmap_atomic(paddr, KM_USER0); | |
32a7627c | 1192 | } |
be512691 N |
1193 | } else |
1194 | j |= PAGE_COUNTER_MASK; | |
32a7627c | 1195 | } |
be512691 | 1196 | spin_unlock_irqrestore(&bitmap->lock, flags); |
32a7627c N |
1197 | |
1198 | /* now sync the final page */ | |
1199 | if (lastpage != NULL) { | |
32a7627c | 1200 | spin_lock_irqsave(&bitmap->lock, flags); |
ec7a3197 | 1201 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1202 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1203 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 1204 | write_page(bitmap, lastpage, 0); |
32a7627c N |
1205 | } else { |
1206 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1207 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1208 | } | |
32a7627c N |
1209 | } |
1210 | ||
7be3dfec | 1211 | done: |
8311c29d N |
1212 | if (bitmap->allclean == 0) |
1213 | bitmap->mddev->thread->timeout = bitmap->daemon_sleep * HZ; | |
c3d9714e | 1214 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1215 | } |
1216 | ||
32a7627c N |
1217 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
1218 | sector_t offset, int *blocks, | |
1219 | int create) | |
ee305ace N |
1220 | __releases(bitmap->lock) |
1221 | __acquires(bitmap->lock) | |
32a7627c N |
1222 | { |
1223 | /* If 'create', we might release the lock and reclaim it. | |
1224 | * The lock must have been taken with interrupts enabled. | |
1225 | * If !create, we don't release the lock. | |
1226 | */ | |
1227 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1228 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1229 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1230 | sector_t csize; | |
1231 | ||
1232 | if (bitmap_checkpage(bitmap, page, create) < 0) { | |
1233 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); | |
1234 | *blocks = csize - (offset & (csize- 1)); | |
1235 | return NULL; | |
1236 | } | |
1237 | /* now locked ... */ | |
1238 | ||
1239 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1240 | /* should we use the first or second counter field | |
1241 | * of the hijacked pointer? */ | |
1242 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
1243 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) + | |
1244 | PAGE_COUNTER_SHIFT - 1); | |
1245 | *blocks = csize - (offset & (csize- 1)); | |
1246 | return &((bitmap_counter_t *) | |
1247 | &bitmap->bp[page].map)[hi]; | |
1248 | } else { /* page is allocated */ | |
1249 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); | |
1250 | *blocks = csize - (offset & (csize- 1)); | |
1251 | return (bitmap_counter_t *) | |
1252 | &(bitmap->bp[page].map[pageoff]); | |
1253 | } | |
1254 | } | |
1255 | ||
4b6d287f | 1256 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c N |
1257 | { |
1258 | if (!bitmap) return 0; | |
4b6d287f N |
1259 | |
1260 | if (behind) { | |
1261 | atomic_inc(&bitmap->behind_writes); | |
1262 | PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n", | |
1263 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1264 | } | |
1265 | ||
32a7627c N |
1266 | while (sectors) { |
1267 | int blocks; | |
1268 | bitmap_counter_t *bmc; | |
1269 | ||
1270 | spin_lock_irq(&bitmap->lock); | |
1271 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); | |
1272 | if (!bmc) { | |
1273 | spin_unlock_irq(&bitmap->lock); | |
1274 | return 0; | |
1275 | } | |
1276 | ||
da6e1a32 NB |
1277 | if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) { |
1278 | DEFINE_WAIT(__wait); | |
1279 | /* note that it is safe to do the prepare_to_wait | |
1280 | * after the test as long as we do it before dropping | |
1281 | * the spinlock. | |
1282 | */ | |
1283 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1284 | TASK_UNINTERRUPTIBLE); | |
1285 | spin_unlock_irq(&bitmap->lock); | |
2ad8b1ef | 1286 | blk_unplug(bitmap->mddev->queue); |
da6e1a32 NB |
1287 | schedule(); |
1288 | finish_wait(&bitmap->overflow_wait, &__wait); | |
1289 | continue; | |
1290 | } | |
1291 | ||
32a7627c N |
1292 | switch(*bmc) { |
1293 | case 0: | |
1294 | bitmap_file_set_bit(bitmap, offset); | |
1295 | bitmap_count_page(bitmap,offset, 1); | |
93769f58 | 1296 | blk_plug_device_unlocked(bitmap->mddev->queue); |
32a7627c N |
1297 | /* fall through */ |
1298 | case 1: | |
1299 | *bmc = 2; | |
1300 | } | |
da6e1a32 | 1301 | |
32a7627c N |
1302 | (*bmc)++; |
1303 | ||
1304 | spin_unlock_irq(&bitmap->lock); | |
1305 | ||
1306 | offset += blocks; | |
1307 | if (sectors > blocks) | |
1308 | sectors -= blocks; | |
1309 | else sectors = 0; | |
1310 | } | |
8311c29d | 1311 | bitmap->allclean = 0; |
32a7627c N |
1312 | return 0; |
1313 | } | |
1314 | ||
1315 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1316 | int success, int behind) |
32a7627c N |
1317 | { |
1318 | if (!bitmap) return; | |
4b6d287f N |
1319 | if (behind) { |
1320 | atomic_dec(&bitmap->behind_writes); | |
1321 | PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n", | |
1322 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1323 | } | |
d0a4bb49 N |
1324 | if (bitmap->mddev->degraded) |
1325 | /* Never clear bits or update events_cleared when degraded */ | |
1326 | success = 0; | |
4b6d287f | 1327 | |
32a7627c N |
1328 | while (sectors) { |
1329 | int blocks; | |
1330 | unsigned long flags; | |
1331 | bitmap_counter_t *bmc; | |
1332 | ||
1333 | spin_lock_irqsave(&bitmap->lock, flags); | |
1334 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); | |
1335 | if (!bmc) { | |
1336 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1337 | return; | |
1338 | } | |
1339 | ||
a0da84f3 NB |
1340 | if (success && |
1341 | bitmap->events_cleared < bitmap->mddev->events) { | |
1342 | bitmap->events_cleared = bitmap->mddev->events; | |
1343 | bitmap->need_sync = 1; | |
1344 | } | |
1345 | ||
32a7627c N |
1346 | if (!success && ! (*bmc & NEEDED_MASK)) |
1347 | *bmc |= NEEDED_MASK; | |
1348 | ||
da6e1a32 NB |
1349 | if ((*bmc & COUNTER_MAX) == COUNTER_MAX) |
1350 | wake_up(&bitmap->overflow_wait); | |
1351 | ||
32a7627c N |
1352 | (*bmc)--; |
1353 | if (*bmc <= 2) { | |
1354 | set_page_attr(bitmap, | |
1355 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1356 | BITMAP_PAGE_CLEAN); | |
1357 | } | |
1358 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1359 | offset += blocks; | |
1360 | if (sectors > blocks) | |
1361 | sectors -= blocks; | |
1362 | else sectors = 0; | |
1363 | } | |
1364 | } | |
1365 | ||
1187cf0a N |
1366 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, |
1367 | int degraded) | |
32a7627c N |
1368 | { |
1369 | bitmap_counter_t *bmc; | |
1370 | int rv; | |
1371 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1372 | *blocks = 1024; | |
1373 | return 1; /* always resync if no bitmap */ | |
1374 | } | |
1375 | spin_lock_irq(&bitmap->lock); | |
1376 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1377 | rv = 0; | |
1378 | if (bmc) { | |
1379 | /* locked */ | |
1380 | if (RESYNC(*bmc)) | |
1381 | rv = 1; | |
1382 | else if (NEEDED(*bmc)) { | |
1383 | rv = 1; | |
6a806c51 N |
1384 | if (!degraded) { /* don't set/clear bits if degraded */ |
1385 | *bmc |= RESYNC_MASK; | |
1386 | *bmc &= ~NEEDED_MASK; | |
1387 | } | |
32a7627c N |
1388 | } |
1389 | } | |
1390 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1391 | bitmap->allclean = 0; |
32a7627c N |
1392 | return rv; |
1393 | } | |
1394 | ||
1187cf0a N |
1395 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, |
1396 | int degraded) | |
1397 | { | |
1398 | /* bitmap_start_sync must always report on multiples of whole | |
1399 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1400 | * get confused. | |
1401 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1402 | * At least PAGE_SIZE>>9 blocks are covered. | |
1403 | * Return the 'or' of the result. | |
1404 | */ | |
1405 | int rv = 0; | |
1406 | int blocks1; | |
1407 | ||
1408 | *blocks = 0; | |
1409 | while (*blocks < (PAGE_SIZE>>9)) { | |
1410 | rv |= __bitmap_start_sync(bitmap, offset, | |
1411 | &blocks1, degraded); | |
1412 | offset += blocks1; | |
1413 | *blocks += blocks1; | |
1414 | } | |
1415 | return rv; | |
1416 | } | |
1417 | ||
32a7627c N |
1418 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted) |
1419 | { | |
1420 | bitmap_counter_t *bmc; | |
1421 | unsigned long flags; | |
1422 | /* | |
1423 | if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted); | |
1424 | */ if (bitmap == NULL) { | |
1425 | *blocks = 1024; | |
1426 | return; | |
1427 | } | |
1428 | spin_lock_irqsave(&bitmap->lock, flags); | |
1429 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1430 | if (bmc == NULL) | |
1431 | goto unlock; | |
1432 | /* locked */ | |
1433 | /* | |
1434 | if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks); | |
1435 | */ | |
1436 | if (RESYNC(*bmc)) { | |
1437 | *bmc &= ~RESYNC_MASK; | |
1438 | ||
1439 | if (!NEEDED(*bmc) && aborted) | |
1440 | *bmc |= NEEDED_MASK; | |
1441 | else { | |
1442 | if (*bmc <= 2) { | |
1443 | set_page_attr(bitmap, | |
1444 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1445 | BITMAP_PAGE_CLEAN); | |
1446 | } | |
1447 | } | |
1448 | } | |
1449 | unlock: | |
1450 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
8311c29d | 1451 | bitmap->allclean = 0; |
32a7627c N |
1452 | } |
1453 | ||
1454 | void bitmap_close_sync(struct bitmap *bitmap) | |
1455 | { | |
1456 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1457 | * properly have been aborted. It remains to us to clear the | |
1458 | * RESYNC bit wherever it is still on | |
1459 | */ | |
1460 | sector_t sector = 0; | |
1461 | int blocks; | |
b47490c9 N |
1462 | if (!bitmap) |
1463 | return; | |
32a7627c N |
1464 | while (sector < bitmap->mddev->resync_max_sectors) { |
1465 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1466 | sector += blocks; |
1467 | } | |
1468 | } | |
1469 | ||
1470 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1471 | { | |
1472 | sector_t s = 0; | |
1473 | int blocks; | |
1474 | ||
1475 | if (!bitmap) | |
1476 | return; | |
1477 | if (sector == 0) { | |
1478 | bitmap->last_end_sync = jiffies; | |
1479 | return; | |
1480 | } | |
1481 | if (time_before(jiffies, (bitmap->last_end_sync | |
1482 | + bitmap->daemon_sleep * HZ))) | |
1483 | return; | |
1484 | wait_event(bitmap->mddev->recovery_wait, | |
1485 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1486 | ||
97e4f42d N |
1487 | bitmap->mddev->curr_resync_completed = bitmap->mddev->curr_resync; |
1488 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); | |
b47490c9 N |
1489 | sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1); |
1490 | s = 0; | |
1491 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1492 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1493 | s += blocks; | |
32a7627c | 1494 | } |
b47490c9 | 1495 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1496 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c N |
1497 | } |
1498 | ||
6a07997f | 1499 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1500 | { |
1501 | /* For each chunk covered by any of these sectors, set the | |
193f1c93 | 1502 | * counter to 1 and set resync_needed. They should all |
32a7627c N |
1503 | * be 0 at this point |
1504 | */ | |
193f1c93 N |
1505 | |
1506 | int secs; | |
1507 | bitmap_counter_t *bmc; | |
1508 | spin_lock_irq(&bitmap->lock); | |
1509 | bmc = bitmap_get_counter(bitmap, offset, &secs, 1); | |
1510 | if (!bmc) { | |
32a7627c | 1511 | spin_unlock_irq(&bitmap->lock); |
193f1c93 | 1512 | return; |
32a7627c | 1513 | } |
193f1c93 N |
1514 | if (! *bmc) { |
1515 | struct page *page; | |
6a07997f | 1516 | *bmc = 1 | (needed?NEEDED_MASK:0); |
193f1c93 N |
1517 | bitmap_count_page(bitmap, offset, 1); |
1518 | page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)); | |
1519 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1520 | } | |
1521 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1522 | bitmap->allclean = 0; |
32a7627c N |
1523 | } |
1524 | ||
9b1d1dac PC |
1525 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1526 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1527 | { | |
1528 | unsigned long chunk; | |
1529 | ||
1530 | for (chunk = s; chunk <= e; chunk++) { | |
db305e50 | 1531 | sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap); |
9b1d1dac PC |
1532 | bitmap_set_memory_bits(bitmap, sec, 1); |
1533 | bitmap_file_set_bit(bitmap, sec); | |
1534 | } | |
1535 | } | |
1536 | ||
6b8b3e8a N |
1537 | /* |
1538 | * flush out any pending updates | |
1539 | */ | |
1540 | void bitmap_flush(mddev_t *mddev) | |
1541 | { | |
1542 | struct bitmap *bitmap = mddev->bitmap; | |
1543 | int sleep; | |
1544 | ||
1545 | if (!bitmap) /* there was no bitmap */ | |
1546 | return; | |
1547 | ||
1548 | /* run the daemon_work three time to ensure everything is flushed | |
1549 | * that can be | |
1550 | */ | |
1551 | sleep = bitmap->daemon_sleep; | |
1552 | bitmap->daemon_sleep = 0; | |
aa5cbd10 N |
1553 | bitmap_daemon_work(mddev); |
1554 | bitmap_daemon_work(mddev); | |
1555 | bitmap_daemon_work(mddev); | |
6b8b3e8a N |
1556 | bitmap->daemon_sleep = sleep; |
1557 | bitmap_update_sb(bitmap); | |
1558 | } | |
1559 | ||
32a7627c N |
1560 | /* |
1561 | * free memory that was allocated | |
1562 | */ | |
3178b0db | 1563 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1564 | { |
1565 | unsigned long k, pages; | |
1566 | struct bitmap_page *bp; | |
32a7627c N |
1567 | |
1568 | if (!bitmap) /* there was no bitmap */ | |
1569 | return; | |
1570 | ||
32a7627c N |
1571 | /* release the bitmap file and kill the daemon */ |
1572 | bitmap_file_put(bitmap); | |
1573 | ||
1574 | bp = bitmap->bp; | |
1575 | pages = bitmap->pages; | |
1576 | ||
1577 | /* free all allocated memory */ | |
1578 | ||
32a7627c N |
1579 | if (bp) /* deallocate the page memory */ |
1580 | for (k = 0; k < pages; k++) | |
1581 | if (bp[k].map && !bp[k].hijacked) | |
1582 | kfree(bp[k].map); | |
1583 | kfree(bp); | |
1584 | kfree(bitmap); | |
1585 | } | |
aa5cbd10 | 1586 | |
3178b0db N |
1587 | void bitmap_destroy(mddev_t *mddev) |
1588 | { | |
1589 | struct bitmap *bitmap = mddev->bitmap; | |
1590 | ||
1591 | if (!bitmap) /* there was no bitmap */ | |
1592 | return; | |
1593 | ||
c3d9714e | 1594 | mutex_lock(&mddev->bitmap_info.mutex); |
3178b0db | 1595 | mddev->bitmap = NULL; /* disconnect from the md device */ |
c3d9714e | 1596 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1597 | if (mddev->thread) |
1598 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db N |
1599 | |
1600 | bitmap_free(bitmap); | |
1601 | } | |
32a7627c N |
1602 | |
1603 | /* | |
1604 | * initialize the bitmap structure | |
1605 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1606 | */ | |
1607 | int bitmap_create(mddev_t *mddev) | |
1608 | { | |
1609 | struct bitmap *bitmap; | |
1f593903 | 1610 | sector_t blocks = mddev->resync_max_sectors; |
32a7627c N |
1611 | unsigned long chunks; |
1612 | unsigned long pages; | |
c3d9714e | 1613 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1614 | int err; |
6a07997f | 1615 | sector_t start; |
32a7627c | 1616 | |
5f6e3c83 | 1617 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1618 | |
c3d9714e | 1619 | if (!file && !mddev->bitmap_info.offset) /* bitmap disabled, nothing to do */ |
32a7627c N |
1620 | return 0; |
1621 | ||
c3d9714e | 1622 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1623 | |
9ffae0cf | 1624 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1625 | if (!bitmap) |
1626 | return -ENOMEM; | |
1627 | ||
32a7627c | 1628 | spin_lock_init(&bitmap->lock); |
ce25c31b N |
1629 | atomic_set(&bitmap->pending_writes, 0); |
1630 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1631 | init_waitqueue_head(&bitmap->overflow_wait); |
ce25c31b | 1632 | |
32a7627c | 1633 | bitmap->mddev = mddev; |
32a7627c | 1634 | |
32a7627c | 1635 | bitmap->file = file; |
c3d9714e | 1636 | bitmap->offset = mddev->bitmap_info.offset; |
ce25c31b N |
1637 | if (file) { |
1638 | get_file(file); | |
ae8fa283 N |
1639 | /* As future accesses to this file will use bmap, |
1640 | * and bypass the page cache, we must sync the file | |
1641 | * first. | |
1642 | */ | |
1643 | vfs_fsync(file, file->f_dentry, 1); | |
ce25c31b | 1644 | } |
32a7627c N |
1645 | /* read superblock from bitmap file (this sets bitmap->chunksize) */ |
1646 | err = bitmap_read_sb(bitmap); | |
1647 | if (err) | |
3178b0db | 1648 | goto error; |
32a7627c | 1649 | |
a638b2dc | 1650 | bitmap->chunkshift = ffz(~bitmap->chunksize); |
32a7627c N |
1651 | |
1652 | /* now that chunksize and chunkshift are set, we can use these macros */ | |
1f593903 N |
1653 | chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >> |
1654 | CHUNK_BLOCK_SHIFT(bitmap); | |
32a7627c N |
1655 | pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; |
1656 | ||
1657 | BUG_ON(!pages); | |
1658 | ||
1659 | bitmap->chunks = chunks; | |
1660 | bitmap->pages = pages; | |
1661 | bitmap->missing_pages = pages; | |
1662 | bitmap->counter_bits = COUNTER_BITS; | |
1663 | ||
1664 | bitmap->syncchunk = ~0UL; | |
1665 | ||
44456d37 | 1666 | #ifdef INJECT_FATAL_FAULT_1 |
32a7627c N |
1667 | bitmap->bp = NULL; |
1668 | #else | |
9ffae0cf | 1669 | bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); |
32a7627c | 1670 | #endif |
3178b0db | 1671 | err = -ENOMEM; |
32a7627c | 1672 | if (!bitmap->bp) |
3178b0db | 1673 | goto error; |
32a7627c | 1674 | |
32a7627c N |
1675 | /* now that we have some pages available, initialize the in-memory |
1676 | * bitmap from the on-disk bitmap */ | |
6a07997f N |
1677 | start = 0; |
1678 | if (mddev->degraded == 0 | |
1679 | || bitmap->events_cleared == mddev->events) | |
1680 | /* no need to keep dirty bits to optimise a re-add of a missing device */ | |
1681 | start = mddev->recovery_cp; | |
1682 | err = bitmap_init_from_disk(bitmap, start); | |
193f1c93 | 1683 | |
32a7627c | 1684 | if (err) |
3178b0db | 1685 | goto error; |
32a7627c N |
1686 | |
1687 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", | |
1688 | pages, bmname(bitmap)); | |
1689 | ||
3178b0db N |
1690 | mddev->bitmap = bitmap; |
1691 | ||
b15c2e57 N |
1692 | mddev->thread->timeout = bitmap->daemon_sleep * HZ; |
1693 | ||
4ad13663 N |
1694 | bitmap_update_sb(bitmap); |
1695 | ||
1696 | return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0; | |
3178b0db N |
1697 | |
1698 | error: | |
1699 | bitmap_free(bitmap); | |
1700 | return err; | |
32a7627c N |
1701 | } |
1702 | ||
1703 | /* the bitmap API -- for raid personalities */ | |
1704 | EXPORT_SYMBOL(bitmap_startwrite); | |
1705 | EXPORT_SYMBOL(bitmap_endwrite); | |
1706 | EXPORT_SYMBOL(bitmap_start_sync); | |
1707 | EXPORT_SYMBOL(bitmap_end_sync); | |
1708 | EXPORT_SYMBOL(bitmap_unplug); | |
1709 | EXPORT_SYMBOL(bitmap_close_sync); | |
b47490c9 | 1710 | EXPORT_SYMBOL(bitmap_cond_end_sync); |