]>
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 | ||
19 | #include <linux/module.h> | |
32a7627c N |
20 | #include <linux/errno.h> |
21 | #include <linux/slab.h> | |
22 | #include <linux/init.h> | |
32a7627c N |
23 | #include <linux/timer.h> |
24 | #include <linux/sched.h> | |
25 | #include <linux/list.h> | |
26 | #include <linux/file.h> | |
27 | #include <linux/mount.h> | |
28 | #include <linux/buffer_head.h> | |
29 | #include <linux/raid/md.h> | |
30 | #include <linux/raid/bitmap.h> | |
31 | ||
32 | /* debug macros */ | |
33 | ||
34 | #define DEBUG 0 | |
35 | ||
36 | #if DEBUG | |
37 | /* these are for debugging purposes only! */ | |
38 | ||
39 | /* define one and only one of these */ | |
40 | #define INJECT_FAULTS_1 0 /* cause bitmap_alloc_page to fail always */ | |
41 | #define INJECT_FAULTS_2 0 /* cause bitmap file to be kicked when first bit set*/ | |
42 | #define INJECT_FAULTS_3 0 /* treat bitmap file as kicked at init time */ | |
43 | #define INJECT_FAULTS_4 0 /* undef */ | |
44 | #define INJECT_FAULTS_5 0 /* undef */ | |
45 | #define INJECT_FAULTS_6 0 | |
46 | ||
47 | /* if these are defined, the driver will fail! debug only */ | |
48 | #define INJECT_FATAL_FAULT_1 0 /* fail kmalloc, causing bitmap_create to fail */ | |
49 | #define INJECT_FATAL_FAULT_2 0 /* undef */ | |
50 | #define INJECT_FATAL_FAULT_3 0 /* undef */ | |
51 | #endif | |
52 | ||
53 | //#define DPRINTK PRINTK /* set this NULL to avoid verbose debug output */ | |
54 | #define DPRINTK(x...) do { } while(0) | |
55 | ||
56 | #ifndef PRINTK | |
57 | # if DEBUG > 0 | |
58 | # define PRINTK(x...) printk(KERN_DEBUG x) | |
59 | # else | |
60 | # define PRINTK(x...) | |
61 | # endif | |
62 | #endif | |
63 | ||
64 | static inline char * bmname(struct bitmap *bitmap) | |
65 | { | |
66 | return bitmap->mddev ? mdname(bitmap->mddev) : "mdX"; | |
67 | } | |
68 | ||
32a7627c N |
69 | |
70 | /* | |
71 | * just a placeholder - calls kmalloc for bitmap pages | |
72 | */ | |
73 | static unsigned char *bitmap_alloc_page(struct bitmap *bitmap) | |
74 | { | |
75 | unsigned char *page; | |
76 | ||
44456d37 | 77 | #ifdef INJECT_FAULTS_1 |
32a7627c N |
78 | page = NULL; |
79 | #else | |
80 | page = kmalloc(PAGE_SIZE, GFP_NOIO); | |
81 | #endif | |
82 | if (!page) | |
83 | printk("%s: bitmap_alloc_page FAILED\n", bmname(bitmap)); | |
84 | else | |
a654b9d8 | 85 | PRINTK("%s: bitmap_alloc_page: allocated page at %p\n", |
32a7627c N |
86 | bmname(bitmap), page); |
87 | return page; | |
88 | } | |
89 | ||
90 | /* | |
91 | * for now just a placeholder -- just calls kfree for bitmap pages | |
92 | */ | |
93 | static void bitmap_free_page(struct bitmap *bitmap, unsigned char *page) | |
94 | { | |
95 | PRINTK("%s: bitmap_free_page: free page %p\n", bmname(bitmap), page); | |
96 | kfree(page); | |
97 | } | |
98 | ||
99 | /* | |
100 | * check a page and, if necessary, allocate it (or hijack it if the alloc fails) | |
101 | * | |
102 | * 1) check to see if this page is allocated, if it's not then try to alloc | |
103 | * 2) if the alloc fails, set the page's hijacked flag so we'll use the | |
104 | * page pointer directly as a counter | |
105 | * | |
106 | * if we find our page, we increment the page's refcount so that it stays | |
107 | * allocated while we're using it | |
108 | */ | |
109 | static int bitmap_checkpage(struct bitmap *bitmap, unsigned long page, int create) | |
110 | { | |
111 | unsigned char *mappage; | |
112 | ||
113 | if (page >= bitmap->pages) { | |
114 | printk(KERN_ALERT | |
115 | "%s: invalid bitmap page request: %lu (> %lu)\n", | |
116 | bmname(bitmap), page, bitmap->pages-1); | |
117 | return -EINVAL; | |
118 | } | |
119 | ||
120 | ||
121 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ | |
122 | return 0; | |
123 | ||
124 | if (bitmap->bp[page].map) /* page is already allocated, just return */ | |
125 | return 0; | |
126 | ||
127 | if (!create) | |
128 | return -ENOENT; | |
129 | ||
130 | spin_unlock_irq(&bitmap->lock); | |
131 | ||
132 | /* this page has not been allocated yet */ | |
133 | ||
134 | if ((mappage = bitmap_alloc_page(bitmap)) == NULL) { | |
135 | PRINTK("%s: bitmap map page allocation failed, hijacking\n", | |
136 | bmname(bitmap)); | |
137 | /* failed - set the hijacked flag so that we can use the | |
138 | * pointer as a counter */ | |
139 | spin_lock_irq(&bitmap->lock); | |
140 | if (!bitmap->bp[page].map) | |
141 | bitmap->bp[page].hijacked = 1; | |
142 | goto out; | |
143 | } | |
144 | ||
145 | /* got a page */ | |
146 | ||
147 | spin_lock_irq(&bitmap->lock); | |
148 | ||
149 | /* recheck the page */ | |
150 | ||
151 | if (bitmap->bp[page].map || bitmap->bp[page].hijacked) { | |
152 | /* somebody beat us to getting the page */ | |
153 | bitmap_free_page(bitmap, mappage); | |
154 | return 0; | |
155 | } | |
156 | ||
157 | /* no page was in place and we have one, so install it */ | |
158 | ||
159 | memset(mappage, 0, PAGE_SIZE); | |
160 | bitmap->bp[page].map = mappage; | |
161 | bitmap->missing_pages--; | |
162 | out: | |
163 | return 0; | |
164 | } | |
165 | ||
166 | ||
167 | /* if page is completely empty, put it back on the free list, or dealloc it */ | |
168 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
169 | /* Note: lock should be held when calling this */ | |
858119e1 | 170 | static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page) |
32a7627c N |
171 | { |
172 | char *ptr; | |
173 | ||
174 | if (bitmap->bp[page].count) /* page is still busy */ | |
175 | return; | |
176 | ||
177 | /* page is no longer in use, it can be released */ | |
178 | ||
179 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
180 | bitmap->bp[page].hijacked = 0; | |
181 | bitmap->bp[page].map = NULL; | |
182 | return; | |
183 | } | |
184 | ||
185 | /* normal case, free the page */ | |
186 | ||
187 | #if 0 | |
188 | /* actually ... let's not. We will probably need the page again exactly when | |
189 | * memory is tight and we are flusing to disk | |
190 | */ | |
191 | return; | |
192 | #else | |
193 | ptr = bitmap->bp[page].map; | |
194 | bitmap->bp[page].map = NULL; | |
195 | bitmap->missing_pages++; | |
196 | bitmap_free_page(bitmap, ptr); | |
197 | return; | |
198 | #endif | |
199 | } | |
200 | ||
201 | ||
202 | /* | |
203 | * bitmap file handling - read and write the bitmap file and its superblock | |
204 | */ | |
205 | ||
32a7627c N |
206 | /* |
207 | * basic page I/O operations | |
208 | */ | |
209 | ||
a654b9d8 | 210 | /* IO operations when bitmap is stored near all superblocks */ |
a2ed9615 N |
211 | static struct page *read_sb_page(mddev_t *mddev, long offset, |
212 | struct page *page, | |
213 | unsigned long index, int size) | |
a654b9d8 N |
214 | { |
215 | /* choose a good rdev and read the page from there */ | |
216 | ||
217 | mdk_rdev_t *rdev; | |
a654b9d8 N |
218 | sector_t target; |
219 | ||
a2ed9615 N |
220 | if (!page) |
221 | page = alloc_page(GFP_KERNEL); | |
a654b9d8 N |
222 | if (!page) |
223 | return ERR_PTR(-ENOMEM); | |
a654b9d8 | 224 | |
159ec1fc | 225 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 N |
226 | if (! test_bit(In_sync, &rdev->flags) |
227 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
228 | continue; |
229 | ||
0f420358 | 230 | target = rdev->sb_start + offset + index * (PAGE_SIZE/512); |
a654b9d8 | 231 | |
a2ed9615 N |
232 | if (sync_page_io(rdev->bdev, target, |
233 | roundup(size, bdev_hardsect_size(rdev->bdev)), | |
234 | page, READ)) { | |
ab904d63 | 235 | page->index = index; |
ce25c31b N |
236 | attach_page_buffers(page, NULL); /* so that free_buffer will |
237 | * quietly no-op */ | |
ab904d63 N |
238 | return page; |
239 | } | |
240 | } | |
241 | return ERR_PTR(-EIO); | |
a654b9d8 | 242 | |
a654b9d8 N |
243 | } |
244 | ||
b2d2c4ce N |
245 | static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
246 | { | |
247 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
248 | * linked list, and raising the refcount of devices we return to ensure | |
249 | * they don't disappear while in use. | |
250 | * As devices are only added or removed when raid_disk is < 0 and | |
251 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
252 | * still be in the same position on the list when we re-enter | |
253 | * list_for_each_continue_rcu. | |
254 | */ | |
255 | struct list_head *pos; | |
256 | rcu_read_lock(); | |
257 | if (rdev == NULL) | |
258 | /* start at the beginning */ | |
259 | pos = &mddev->disks; | |
260 | else { | |
261 | /* release the previous rdev and start from there. */ | |
262 | rdev_dec_pending(rdev, mddev); | |
263 | pos = &rdev->same_set; | |
264 | } | |
265 | list_for_each_continue_rcu(pos, &mddev->disks) { | |
266 | rdev = list_entry(pos, mdk_rdev_t, same_set); | |
267 | if (rdev->raid_disk >= 0 && | |
268 | test_bit(In_sync, &rdev->flags) && | |
269 | !test_bit(Faulty, &rdev->flags)) { | |
270 | /* this is a usable devices */ | |
271 | atomic_inc(&rdev->nr_pending); | |
272 | rcu_read_unlock(); | |
273 | return rdev; | |
274 | } | |
275 | } | |
276 | rcu_read_unlock(); | |
277 | return NULL; | |
278 | } | |
279 | ||
ab6085c7 | 280 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 281 | { |
b2d2c4ce | 282 | mdk_rdev_t *rdev = NULL; |
ab6085c7 | 283 | mddev_t *mddev = bitmap->mddev; |
a654b9d8 | 284 | |
b2d2c4ce | 285 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ab6085c7 N |
286 | int size = PAGE_SIZE; |
287 | if (page->index == bitmap->file_pages-1) | |
288 | size = roundup(bitmap->last_page_size, | |
289 | bdev_hardsect_size(rdev->bdev)); | |
f0d76d70 N |
290 | /* Just make sure we aren't corrupting data or |
291 | * metadata | |
292 | */ | |
293 | if (bitmap->offset < 0) { | |
294 | /* DATA BITMAP METADATA */ | |
295 | if (bitmap->offset | |
85bfb4da | 296 | + (long)(page->index * (PAGE_SIZE/512)) |
f0d76d70 N |
297 | + size/512 > 0) |
298 | /* bitmap runs in to metadata */ | |
4b80991c | 299 | goto bad_alignment; |
f0d76d70 | 300 | if (rdev->data_offset + mddev->size*2 |
0f420358 | 301 | > rdev->sb_start + bitmap->offset) |
f0d76d70 | 302 | /* data runs in to bitmap */ |
4b80991c | 303 | goto bad_alignment; |
0f420358 | 304 | } else if (rdev->sb_start < rdev->data_offset) { |
f0d76d70 | 305 | /* METADATA BITMAP DATA */ |
0f420358 | 306 | if (rdev->sb_start |
f0d76d70 N |
307 | + bitmap->offset |
308 | + page->index*(PAGE_SIZE/512) + size/512 | |
309 | > rdev->data_offset) | |
310 | /* bitmap runs in to data */ | |
4b80991c | 311 | goto bad_alignment; |
f0d76d70 N |
312 | } else { |
313 | /* DATA METADATA BITMAP - no problems */ | |
314 | } | |
a654b9d8 | 315 | md_super_write(mddev, rdev, |
0f420358 | 316 | rdev->sb_start + bitmap->offset |
a654b9d8 | 317 | + page->index * (PAGE_SIZE/512), |
ab6085c7 | 318 | size, |
a654b9d8 | 319 | page); |
b2d2c4ce | 320 | } |
a654b9d8 N |
321 | |
322 | if (wait) | |
a9701a30 | 323 | md_super_wait(mddev); |
a654b9d8 | 324 | return 0; |
4b80991c N |
325 | |
326 | bad_alignment: | |
327 | rcu_read_unlock(); | |
328 | return -EINVAL; | |
a654b9d8 N |
329 | } |
330 | ||
4ad13663 | 331 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 332 | /* |
a654b9d8 | 333 | * write out a page to a file |
32a7627c | 334 | */ |
4ad13663 | 335 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 336 | { |
d785a06a | 337 | struct buffer_head *bh; |
32a7627c | 338 | |
f0d76d70 N |
339 | if (bitmap->file == NULL) { |
340 | switch (write_sb_page(bitmap, page, wait)) { | |
341 | case -EINVAL: | |
342 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
f0d76d70 | 343 | } |
4ad13663 | 344 | } else { |
a654b9d8 | 345 | |
4ad13663 | 346 | bh = page_buffers(page); |
c708443c | 347 | |
4ad13663 N |
348 | while (bh && bh->b_blocknr) { |
349 | atomic_inc(&bitmap->pending_writes); | |
350 | set_buffer_locked(bh); | |
351 | set_buffer_mapped(bh); | |
352 | submit_bh(WRITE, bh); | |
353 | bh = bh->b_this_page; | |
354 | } | |
d785a06a | 355 | |
4ad13663 N |
356 | if (wait) { |
357 | wait_event(bitmap->write_wait, | |
358 | atomic_read(&bitmap->pending_writes)==0); | |
359 | } | |
8a5e9cf1 | 360 | } |
4ad13663 N |
361 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
362 | bitmap_file_kick(bitmap); | |
d785a06a N |
363 | } |
364 | ||
365 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
366 | { | |
367 | struct bitmap *bitmap = bh->b_private; | |
368 | unsigned long flags; | |
32a7627c | 369 | |
d785a06a N |
370 | if (!uptodate) { |
371 | spin_lock_irqsave(&bitmap->lock, flags); | |
372 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
373 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
32a7627c | 374 | } |
d785a06a N |
375 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
376 | wake_up(&bitmap->write_wait); | |
377 | } | |
32a7627c | 378 | |
d785a06a N |
379 | /* copied from buffer.c */ |
380 | static void | |
381 | __clear_page_buffers(struct page *page) | |
382 | { | |
383 | ClearPagePrivate(page); | |
384 | set_page_private(page, 0); | |
385 | page_cache_release(page); | |
386 | } | |
387 | static void free_buffers(struct page *page) | |
388 | { | |
389 | struct buffer_head *bh = page_buffers(page); | |
77ad4bc7 | 390 | |
d785a06a N |
391 | while (bh) { |
392 | struct buffer_head *next = bh->b_this_page; | |
393 | free_buffer_head(bh); | |
394 | bh = next; | |
77ad4bc7 | 395 | } |
d785a06a N |
396 | __clear_page_buffers(page); |
397 | put_page(page); | |
32a7627c N |
398 | } |
399 | ||
d785a06a N |
400 | /* read a page from a file. |
401 | * We both read the page, and attach buffers to the page to record the | |
402 | * address of each block (using bmap). These addresses will be used | |
403 | * to write the block later, completely bypassing the filesystem. | |
404 | * This usage is similar to how swap files are handled, and allows us | |
405 | * to write to a file with no concerns of memory allocation failing. | |
406 | */ | |
32a7627c | 407 | static struct page *read_page(struct file *file, unsigned long index, |
d785a06a N |
408 | struct bitmap *bitmap, |
409 | unsigned long count) | |
32a7627c | 410 | { |
32a7627c | 411 | struct page *page = NULL; |
c649bb9c | 412 | struct inode *inode = file->f_path.dentry->d_inode; |
d785a06a N |
413 | struct buffer_head *bh; |
414 | sector_t block; | |
32a7627c | 415 | |
2d1f3b5d N |
416 | PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE, |
417 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 418 | |
d785a06a N |
419 | page = alloc_page(GFP_KERNEL); |
420 | if (!page) | |
421 | page = ERR_PTR(-ENOMEM); | |
32a7627c N |
422 | if (IS_ERR(page)) |
423 | goto out; | |
d785a06a | 424 | |
d785a06a N |
425 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
426 | if (!bh) { | |
2d1f3b5d | 427 | put_page(page); |
d785a06a | 428 | page = ERR_PTR(-ENOMEM); |
32a7627c N |
429 | goto out; |
430 | } | |
d785a06a N |
431 | attach_page_buffers(page, bh); |
432 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
433 | while (bh) { | |
434 | if (count == 0) | |
435 | bh->b_blocknr = 0; | |
436 | else { | |
437 | bh->b_blocknr = bmap(inode, block); | |
438 | if (bh->b_blocknr == 0) { | |
439 | /* Cannot use this file! */ | |
440 | free_buffers(page); | |
441 | page = ERR_PTR(-EINVAL); | |
442 | goto out; | |
443 | } | |
444 | bh->b_bdev = inode->i_sb->s_bdev; | |
445 | if (count < (1<<inode->i_blkbits)) | |
446 | count = 0; | |
447 | else | |
448 | count -= (1<<inode->i_blkbits); | |
449 | ||
450 | bh->b_end_io = end_bitmap_write; | |
451 | bh->b_private = bitmap; | |
ce25c31b N |
452 | atomic_inc(&bitmap->pending_writes); |
453 | set_buffer_locked(bh); | |
454 | set_buffer_mapped(bh); | |
455 | submit_bh(READ, bh); | |
d785a06a N |
456 | } |
457 | block++; | |
458 | bh = bh->b_this_page; | |
459 | } | |
d785a06a | 460 | page->index = index; |
ce25c31b N |
461 | |
462 | wait_event(bitmap->write_wait, | |
463 | atomic_read(&bitmap->pending_writes)==0); | |
464 | if (bitmap->flags & BITMAP_WRITE_ERROR) { | |
465 | free_buffers(page); | |
466 | page = ERR_PTR(-EIO); | |
467 | } | |
32a7627c N |
468 | out: |
469 | if (IS_ERR(page)) | |
470 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n", | |
2d1f3b5d N |
471 | (int)PAGE_SIZE, |
472 | (unsigned long long)index << PAGE_SHIFT, | |
32a7627c N |
473 | PTR_ERR(page)); |
474 | return page; | |
475 | } | |
476 | ||
477 | /* | |
478 | * bitmap file superblock operations | |
479 | */ | |
480 | ||
481 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 482 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
483 | { |
484 | bitmap_super_t *sb; | |
485 | unsigned long flags; | |
486 | ||
487 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 488 | return; |
32a7627c N |
489 | spin_lock_irqsave(&bitmap->lock, flags); |
490 | if (!bitmap->sb_page) { /* no superblock */ | |
491 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 492 | return; |
32a7627c | 493 | } |
32a7627c | 494 | spin_unlock_irqrestore(&bitmap->lock, flags); |
ea03aff9 | 495 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 496 | sb->events = cpu_to_le64(bitmap->mddev->events); |
a0da84f3 NB |
497 | if (bitmap->mddev->events < bitmap->events_cleared) { |
498 | /* rocking back to read-only */ | |
499 | bitmap->events_cleared = bitmap->mddev->events; | |
500 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); | |
501 | } | |
ea03aff9 | 502 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 503 | write_page(bitmap, bitmap->sb_page, 1); |
32a7627c N |
504 | } |
505 | ||
506 | /* print out the bitmap file superblock */ | |
507 | void bitmap_print_sb(struct bitmap *bitmap) | |
508 | { | |
509 | bitmap_super_t *sb; | |
510 | ||
511 | if (!bitmap || !bitmap->sb_page) | |
512 | return; | |
ea03aff9 | 513 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 514 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
515 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
516 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
517 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
518 | *(__u32 *)(sb->uuid+0), |
519 | *(__u32 *)(sb->uuid+4), | |
520 | *(__u32 *)(sb->uuid+8), | |
521 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 522 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 523 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 524 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 525 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
526 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
527 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
528 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
529 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
530 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 531 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
ea03aff9 | 532 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
533 | } |
534 | ||
535 | /* read the superblock from the bitmap file and initialize some bitmap fields */ | |
536 | static int bitmap_read_sb(struct bitmap *bitmap) | |
537 | { | |
538 | char *reason = NULL; | |
539 | bitmap_super_t *sb; | |
4b6d287f | 540 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c N |
541 | unsigned long long events; |
542 | int err = -EINVAL; | |
543 | ||
544 | /* page 0 is the superblock, read it... */ | |
f49d5e62 N |
545 | if (bitmap->file) { |
546 | loff_t isize = i_size_read(bitmap->file->f_mapping->host); | |
547 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; | |
548 | ||
549 | bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes); | |
550 | } else { | |
a2ed9615 N |
551 | bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, |
552 | NULL, | |
553 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 554 | } |
32a7627c N |
555 | if (IS_ERR(bitmap->sb_page)) { |
556 | err = PTR_ERR(bitmap->sb_page); | |
557 | bitmap->sb_page = NULL; | |
558 | return err; | |
559 | } | |
560 | ||
ea03aff9 | 561 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 562 | |
32a7627c N |
563 | chunksize = le32_to_cpu(sb->chunksize); |
564 | daemon_sleep = le32_to_cpu(sb->daemon_sleep); | |
4b6d287f | 565 | write_behind = le32_to_cpu(sb->write_behind); |
32a7627c N |
566 | |
567 | /* verify that the bitmap-specific fields are valid */ | |
568 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
569 | reason = "bad magic"; | |
bd926c63 N |
570 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
571 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 572 | reason = "unrecognized superblock version"; |
7dd5d34c N |
573 | else if (chunksize < PAGE_SIZE) |
574 | reason = "bitmap chunksize too small"; | |
32a7627c N |
575 | else if ((1 << ffz(~chunksize)) != chunksize) |
576 | reason = "bitmap chunksize not a power of 2"; | |
7dd5d34c N |
577 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ) |
578 | reason = "daemon sleep period out of range"; | |
4b6d287f N |
579 | else if (write_behind > COUNTER_MAX) |
580 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
581 | if (reason) { |
582 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
583 | bmname(bitmap), reason); | |
584 | goto out; | |
585 | } | |
586 | ||
587 | /* keep the array size field of the bitmap superblock up to date */ | |
588 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
589 | ||
590 | if (!bitmap->mddev->persistent) | |
591 | goto success; | |
592 | ||
593 | /* | |
594 | * if we have a persistent array superblock, compare the | |
595 | * bitmap's UUID and event counter to the mddev's | |
596 | */ | |
597 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
598 | printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n", | |
599 | bmname(bitmap)); | |
600 | goto out; | |
601 | } | |
602 | events = le64_to_cpu(sb->events); | |
603 | if (events < bitmap->mddev->events) { | |
604 | printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) " | |
605 | "-- forcing full recovery\n", bmname(bitmap), events, | |
606 | (unsigned long long) bitmap->mddev->events); | |
4f2e639a | 607 | sb->state |= cpu_to_le32(BITMAP_STALE); |
32a7627c N |
608 | } |
609 | success: | |
610 | /* assign fields using values from superblock */ | |
611 | bitmap->chunksize = chunksize; | |
612 | bitmap->daemon_sleep = daemon_sleep; | |
585f0dd5 | 613 | bitmap->daemon_lastrun = jiffies; |
4b6d287f | 614 | bitmap->max_write_behind = write_behind; |
4f2e639a | 615 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 N |
616 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
617 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
32a7627c | 618 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
4f2e639a | 619 | if (sb->state & cpu_to_le32(BITMAP_STALE)) |
6a07997f | 620 | bitmap->events_cleared = bitmap->mddev->events; |
32a7627c N |
621 | err = 0; |
622 | out: | |
ea03aff9 | 623 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
624 | if (err) |
625 | bitmap_print_sb(bitmap); | |
626 | return err; | |
627 | } | |
628 | ||
629 | enum bitmap_mask_op { | |
630 | MASK_SET, | |
631 | MASK_UNSET | |
632 | }; | |
633 | ||
4ad13663 N |
634 | /* record the state of the bitmap in the superblock. Return the old value */ |
635 | static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits, | |
636 | enum bitmap_mask_op op) | |
32a7627c N |
637 | { |
638 | bitmap_super_t *sb; | |
639 | unsigned long flags; | |
4ad13663 | 640 | int old; |
32a7627c N |
641 | |
642 | spin_lock_irqsave(&bitmap->lock, flags); | |
7e317655 | 643 | if (!bitmap->sb_page) { /* can't set the state */ |
32a7627c | 644 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 645 | return 0; |
32a7627c | 646 | } |
32a7627c | 647 | spin_unlock_irqrestore(&bitmap->lock, flags); |
ea03aff9 | 648 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
4ad13663 | 649 | old = le32_to_cpu(sb->state) & bits; |
32a7627c | 650 | switch (op) { |
4f2e639a | 651 | case MASK_SET: sb->state |= cpu_to_le32(bits); |
32a7627c | 652 | break; |
4f2e639a | 653 | case MASK_UNSET: sb->state &= cpu_to_le32(~bits); |
32a7627c N |
654 | break; |
655 | default: BUG(); | |
656 | } | |
ea03aff9 | 657 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 658 | return old; |
32a7627c N |
659 | } |
660 | ||
661 | /* | |
662 | * general bitmap file operations | |
663 | */ | |
664 | ||
665 | /* calculate the index of the page that contains this bit */ | |
666 | static inline unsigned long file_page_index(unsigned long chunk) | |
667 | { | |
668 | return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT; | |
669 | } | |
670 | ||
671 | /* calculate the (bit) offset of this bit within a page */ | |
672 | static inline unsigned long file_page_offset(unsigned long chunk) | |
673 | { | |
674 | return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1); | |
675 | } | |
676 | ||
677 | /* | |
678 | * return a pointer to the page in the filemap that contains the given bit | |
679 | * | |
680 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
681 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
682 | * 0 or page 1 | |
683 | */ | |
684 | static inline struct page *filemap_get_page(struct bitmap *bitmap, | |
685 | unsigned long chunk) | |
686 | { | |
9b1d1dac | 687 | if (file_page_index(chunk) >= bitmap->file_pages) return NULL; |
32a7627c N |
688 | return bitmap->filemap[file_page_index(chunk) - file_page_index(0)]; |
689 | } | |
690 | ||
691 | ||
692 | static void bitmap_file_unmap(struct bitmap *bitmap) | |
693 | { | |
694 | struct page **map, *sb_page; | |
695 | unsigned long *attr; | |
696 | int pages; | |
697 | unsigned long flags; | |
698 | ||
699 | spin_lock_irqsave(&bitmap->lock, flags); | |
700 | map = bitmap->filemap; | |
701 | bitmap->filemap = NULL; | |
702 | attr = bitmap->filemap_attr; | |
703 | bitmap->filemap_attr = NULL; | |
704 | pages = bitmap->file_pages; | |
705 | bitmap->file_pages = 0; | |
706 | sb_page = bitmap->sb_page; | |
707 | bitmap->sb_page = NULL; | |
708 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
709 | ||
710 | while (pages--) | |
711 | if (map[pages]->index != 0) /* 0 is sb_page, release it below */ | |
d785a06a | 712 | free_buffers(map[pages]); |
32a7627c N |
713 | kfree(map); |
714 | kfree(attr); | |
715 | ||
d785a06a N |
716 | if (sb_page) |
717 | free_buffers(sb_page); | |
32a7627c N |
718 | } |
719 | ||
720 | static void bitmap_file_put(struct bitmap *bitmap) | |
721 | { | |
722 | struct file *file; | |
32a7627c N |
723 | unsigned long flags; |
724 | ||
725 | spin_lock_irqsave(&bitmap->lock, flags); | |
726 | file = bitmap->file; | |
727 | bitmap->file = NULL; | |
728 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
729 | ||
d785a06a N |
730 | if (file) |
731 | wait_event(bitmap->write_wait, | |
732 | atomic_read(&bitmap->pending_writes)==0); | |
32a7627c N |
733 | bitmap_file_unmap(bitmap); |
734 | ||
d785a06a | 735 | if (file) { |
c649bb9c | 736 | struct inode *inode = file->f_path.dentry->d_inode; |
fc0ecff6 | 737 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 738 | fput(file); |
d785a06a | 739 | } |
32a7627c N |
740 | } |
741 | ||
742 | ||
743 | /* | |
744 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
745 | * then it is no longer reliable, so we stop using it and we mark the file | |
746 | * as failed in the superblock | |
747 | */ | |
748 | static void bitmap_file_kick(struct bitmap *bitmap) | |
749 | { | |
750 | char *path, *ptr = NULL; | |
751 | ||
4ad13663 N |
752 | if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) { |
753 | bitmap_update_sb(bitmap); | |
32a7627c | 754 | |
4ad13663 N |
755 | if (bitmap->file) { |
756 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
757 | if (path) | |
6bcfd601 CH |
758 | ptr = d_path(&bitmap->file->f_path, path, |
759 | PAGE_SIZE); | |
760 | ||
32a7627c | 761 | |
4ad13663 N |
762 | printk(KERN_ALERT |
763 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 764 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 765 | |
4ad13663 N |
766 | kfree(path); |
767 | } else | |
768 | printk(KERN_ALERT | |
769 | "%s: disabling internal bitmap due to errors\n", | |
770 | bmname(bitmap)); | |
a654b9d8 | 771 | } |
32a7627c N |
772 | |
773 | bitmap_file_put(bitmap); | |
774 | ||
775 | return; | |
776 | } | |
777 | ||
778 | enum bitmap_page_attr { | |
e16b68b6 N |
779 | BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced |
780 | BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared | |
781 | BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced | |
32a7627c N |
782 | }; |
783 | ||
784 | static inline void set_page_attr(struct bitmap *bitmap, struct page *page, | |
785 | enum bitmap_page_attr attr) | |
786 | { | |
e16b68b6 | 787 | __set_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
788 | } |
789 | ||
790 | static inline void clear_page_attr(struct bitmap *bitmap, struct page *page, | |
791 | enum bitmap_page_attr attr) | |
792 | { | |
e16b68b6 | 793 | __clear_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
794 | } |
795 | ||
ec7a3197 N |
796 | static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page, |
797 | enum bitmap_page_attr attr) | |
32a7627c | 798 | { |
e16b68b6 | 799 | return test_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
800 | } |
801 | ||
802 | /* | |
803 | * bitmap_file_set_bit -- called before performing a write to the md device | |
804 | * to set (and eventually sync) a particular bit in the bitmap file | |
805 | * | |
806 | * we set the bit immediately, then we record the page number so that | |
807 | * when an unplug occurs, we can flush the dirty pages out to disk | |
808 | */ | |
809 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
810 | { | |
811 | unsigned long bit; | |
812 | struct page *page; | |
813 | void *kaddr; | |
814 | unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap); | |
815 | ||
a654b9d8 | 816 | if (!bitmap->filemap) { |
32a7627c N |
817 | return; |
818 | } | |
819 | ||
820 | page = filemap_get_page(bitmap, chunk); | |
9b1d1dac | 821 | if (!page) return; |
32a7627c N |
822 | bit = file_page_offset(chunk); |
823 | ||
32a7627c N |
824 | /* set the bit */ |
825 | kaddr = kmap_atomic(page, KM_USER0); | |
bd926c63 N |
826 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
827 | set_bit(bit, kaddr); | |
828 | else | |
829 | ext2_set_bit(bit, kaddr); | |
32a7627c N |
830 | kunmap_atomic(kaddr, KM_USER0); |
831 | PRINTK("set file bit %lu page %lu\n", bit, page->index); | |
832 | ||
833 | /* record page number so it gets flushed to disk when unplug occurs */ | |
834 | set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); | |
835 | ||
836 | } | |
837 | ||
838 | /* this gets called when the md device is ready to unplug its underlying | |
839 | * (slave) device queues -- before we let any writes go down, we need to | |
840 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 841 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 842 | { |
ec7a3197 N |
843 | unsigned long i, flags; |
844 | int dirty, need_write; | |
32a7627c N |
845 | struct page *page; |
846 | int wait = 0; | |
847 | ||
848 | if (!bitmap) | |
4ad13663 | 849 | return; |
32a7627c N |
850 | |
851 | /* look at each page to see if there are any set bits that need to be | |
852 | * flushed out to disk */ | |
853 | for (i = 0; i < bitmap->file_pages; i++) { | |
854 | spin_lock_irqsave(&bitmap->lock, flags); | |
a654b9d8 | 855 | if (!bitmap->filemap) { |
32a7627c | 856 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 857 | return; |
32a7627c N |
858 | } |
859 | page = bitmap->filemap[i]; | |
ec7a3197 N |
860 | dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
861 | need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
32a7627c N |
862 | clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
863 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
ec7a3197 | 864 | if (dirty) |
32a7627c N |
865 | wait = 1; |
866 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
867 | ||
d785a06a | 868 | if (dirty | need_write) |
4ad13663 | 869 | write_page(bitmap, page, 0); |
32a7627c N |
870 | } |
871 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
0b79ccf0 | 872 | if (bitmap->file) |
d785a06a N |
873 | wait_event(bitmap->write_wait, |
874 | atomic_read(&bitmap->pending_writes)==0); | |
0b79ccf0 | 875 | else |
a9701a30 | 876 | md_super_wait(bitmap->mddev); |
32a7627c | 877 | } |
d785a06a N |
878 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
879 | bitmap_file_kick(bitmap); | |
32a7627c N |
880 | } |
881 | ||
6a07997f | 882 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
883 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
884 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
885 | * memory mapping of the bitmap file | |
886 | * Special cases: | |
887 | * if there's no bitmap file, or if the bitmap file had been | |
888 | * previously kicked from the array, we mark all the bits as | |
889 | * 1's in order to cause a full resync. | |
6a07997f N |
890 | * |
891 | * We ignore all bits for sectors that end earlier than 'start'. | |
892 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 893 | */ |
6a07997f | 894 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
895 | { |
896 | unsigned long i, chunks, index, oldindex, bit; | |
897 | struct page *page = NULL, *oldpage = NULL; | |
898 | unsigned long num_pages, bit_cnt = 0; | |
899 | struct file *file; | |
d785a06a | 900 | unsigned long bytes, offset; |
32a7627c N |
901 | int outofdate; |
902 | int ret = -ENOSPC; | |
ea03aff9 | 903 | void *paddr; |
32a7627c N |
904 | |
905 | chunks = bitmap->chunks; | |
906 | file = bitmap->file; | |
907 | ||
a654b9d8 | 908 | BUG_ON(!file && !bitmap->offset); |
32a7627c | 909 | |
44456d37 | 910 | #ifdef INJECT_FAULTS_3 |
32a7627c N |
911 | outofdate = 1; |
912 | #else | |
913 | outofdate = bitmap->flags & BITMAP_STALE; | |
914 | #endif | |
915 | if (outofdate) | |
916 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
917 | "recovery\n", bmname(bitmap)); | |
918 | ||
919 | bytes = (chunks + 7) / 8; | |
bc7f77de | 920 | |
cdbb4cc2 | 921 | num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE; |
bc7f77de | 922 | |
a654b9d8 | 923 | if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) { |
32a7627c N |
924 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
925 | bmname(bitmap), | |
926 | (unsigned long) i_size_read(file->f_mapping->host), | |
927 | bytes + sizeof(bitmap_super_t)); | |
4ad13663 | 928 | goto err; |
32a7627c | 929 | } |
bc7f77de N |
930 | |
931 | ret = -ENOMEM; | |
932 | ||
32a7627c | 933 | bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL); |
bc7f77de | 934 | if (!bitmap->filemap) |
4ad13663 | 935 | goto err; |
32a7627c | 936 | |
e16b68b6 N |
937 | /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */ |
938 | bitmap->filemap_attr = kzalloc( | |
505fa2c4 | 939 | roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), |
e16b68b6 | 940 | GFP_KERNEL); |
bc7f77de | 941 | if (!bitmap->filemap_attr) |
4ad13663 | 942 | goto err; |
32a7627c | 943 | |
32a7627c N |
944 | oldindex = ~0L; |
945 | ||
946 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 947 | int b; |
32a7627c N |
948 | index = file_page_index(i); |
949 | bit = file_page_offset(i); | |
950 | if (index != oldindex) { /* this is a new page, read it in */ | |
d785a06a | 951 | int count; |
32a7627c | 952 | /* unmap the old page, we're done with it */ |
d785a06a | 953 | if (index == num_pages-1) |
f49d5e62 N |
954 | count = bytes + sizeof(bitmap_super_t) |
955 | - index * PAGE_SIZE; | |
d785a06a N |
956 | else |
957 | count = PAGE_SIZE; | |
32a7627c N |
958 | if (index == 0) { |
959 | /* | |
960 | * if we're here then the superblock page | |
961 | * contains some bits (PAGE_SIZE != sizeof sb) | |
962 | * we've already read it in, so just use it | |
963 | */ | |
964 | page = bitmap->sb_page; | |
965 | offset = sizeof(bitmap_super_t); | |
53845270 N |
966 | if (!file) |
967 | read_sb_page(bitmap->mddev, | |
968 | bitmap->offset, | |
969 | page, | |
970 | index, count); | |
a654b9d8 | 971 | } else if (file) { |
d785a06a | 972 | page = read_page(file, index, bitmap, count); |
a654b9d8 N |
973 | offset = 0; |
974 | } else { | |
a2ed9615 N |
975 | page = read_sb_page(bitmap->mddev, bitmap->offset, |
976 | NULL, | |
977 | index, count); | |
32a7627c N |
978 | offset = 0; |
979 | } | |
a654b9d8 N |
980 | if (IS_ERR(page)) { /* read error */ |
981 | ret = PTR_ERR(page); | |
4ad13663 | 982 | goto err; |
a654b9d8 N |
983 | } |
984 | ||
32a7627c N |
985 | oldindex = index; |
986 | oldpage = page; | |
32a7627c N |
987 | |
988 | if (outofdate) { | |
989 | /* | |
990 | * if bitmap is out of date, dirty the | |
991 | * whole page and write it out | |
992 | */ | |
ea03aff9 N |
993 | paddr = kmap_atomic(page, KM_USER0); |
994 | memset(paddr + offset, 0xff, | |
6a07997f | 995 | PAGE_SIZE - offset); |
ea03aff9 | 996 | kunmap_atomic(paddr, KM_USER0); |
4ad13663 N |
997 | write_page(bitmap, page, 1); |
998 | ||
999 | ret = -EIO; | |
1000 | if (bitmap->flags & BITMAP_WRITE_ERROR) { | |
32a7627c | 1001 | /* release, page not in filemap yet */ |
2d1f3b5d | 1002 | put_page(page); |
4ad13663 | 1003 | goto err; |
32a7627c N |
1004 | } |
1005 | } | |
1006 | ||
1007 | bitmap->filemap[bitmap->file_pages++] = page; | |
ab6085c7 | 1008 | bitmap->last_page_size = count; |
32a7627c | 1009 | } |
ea03aff9 | 1010 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1011 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1012 | b = test_bit(bit, paddr); |
bd926c63 | 1013 | else |
ea03aff9 N |
1014 | b = ext2_test_bit(bit, paddr); |
1015 | kunmap_atomic(paddr, KM_USER0); | |
bd926c63 | 1016 | if (b) { |
32a7627c | 1017 | /* if the disk bit is set, set the memory bit */ |
6a07997f N |
1018 | bitmap_set_memory_bits(bitmap, i << CHUNK_BLOCK_SHIFT(bitmap), |
1019 | ((i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) >= start) | |
1020 | ); | |
32a7627c | 1021 | bit_cnt++; |
6a07997f | 1022 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); |
32a7627c | 1023 | } |
32a7627c N |
1024 | } |
1025 | ||
1026 | /* everything went OK */ | |
1027 | ret = 0; | |
1028 | bitmap_mask_state(bitmap, BITMAP_STALE, MASK_UNSET); | |
1029 | ||
32a7627c N |
1030 | if (bit_cnt) { /* Kick recovery if any bits were set */ |
1031 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
1032 | md_wakeup_thread(bitmap->mddev->thread); | |
1033 | } | |
1034 | ||
32a7627c | 1035 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
4ad13663 N |
1036 | "read %lu/%lu pages, set %lu bits\n", |
1037 | bmname(bitmap), bitmap->file_pages, num_pages, bit_cnt); | |
1038 | ||
1039 | return 0; | |
32a7627c | 1040 | |
4ad13663 N |
1041 | err: |
1042 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", | |
1043 | bmname(bitmap), ret); | |
32a7627c N |
1044 | return ret; |
1045 | } | |
1046 | ||
a654b9d8 N |
1047 | void bitmap_write_all(struct bitmap *bitmap) |
1048 | { | |
1049 | /* We don't actually write all bitmap blocks here, | |
1050 | * just flag them as needing to be written | |
1051 | */ | |
ec7a3197 | 1052 | int i; |
a654b9d8 | 1053 | |
ec7a3197 N |
1054 | for (i=0; i < bitmap->file_pages; i++) |
1055 | set_page_attr(bitmap, bitmap->filemap[i], | |
1056 | BITMAP_PAGE_NEEDWRITE); | |
a654b9d8 N |
1057 | } |
1058 | ||
32a7627c N |
1059 | |
1060 | static void bitmap_count_page(struct bitmap *bitmap, sector_t offset, int inc) | |
1061 | { | |
1062 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1063 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1064 | bitmap->bp[page].count += inc; | |
1065 | /* | |
1066 | if (page == 0) printk("count page 0, offset %llu: %d gives %d\n", | |
1067 | (unsigned long long)offset, inc, bitmap->bp[page].count); | |
1068 | */ | |
1069 | bitmap_checkfree(bitmap, page); | |
1070 | } | |
1071 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, | |
1072 | sector_t offset, int *blocks, | |
1073 | int create); | |
1074 | ||
1075 | /* | |
1076 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1077 | * out to disk | |
1078 | */ | |
1079 | ||
4ad13663 | 1080 | void bitmap_daemon_work(struct bitmap *bitmap) |
32a7627c | 1081 | { |
aa3163f8 | 1082 | unsigned long j; |
32a7627c N |
1083 | unsigned long flags; |
1084 | struct page *page = NULL, *lastpage = NULL; | |
32a7627c | 1085 | int blocks; |
ea03aff9 | 1086 | void *paddr; |
32a7627c N |
1087 | |
1088 | if (bitmap == NULL) | |
4ad13663 | 1089 | return; |
32a7627c | 1090 | if (time_before(jiffies, bitmap->daemon_lastrun + bitmap->daemon_sleep*HZ)) |
7be3dfec N |
1091 | goto done; |
1092 | ||
32a7627c | 1093 | bitmap->daemon_lastrun = jiffies; |
8311c29d N |
1094 | if (bitmap->allclean) { |
1095 | bitmap->mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
1096 | return; | |
1097 | } | |
1098 | bitmap->allclean = 1; | |
32a7627c N |
1099 | |
1100 | for (j = 0; j < bitmap->chunks; j++) { | |
1101 | bitmap_counter_t *bmc; | |
1102 | spin_lock_irqsave(&bitmap->lock, flags); | |
a654b9d8 | 1103 | if (!bitmap->filemap) { |
32a7627c N |
1104 | /* error or shutdown */ |
1105 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1106 | break; | |
1107 | } | |
1108 | ||
1109 | page = filemap_get_page(bitmap, j); | |
32a7627c N |
1110 | |
1111 | if (page != lastpage) { | |
aa3163f8 | 1112 | /* skip this page unless it's marked as needing cleaning */ |
ec7a3197 N |
1113 | if (!test_page_attr(bitmap, page, BITMAP_PAGE_CLEAN)) { |
1114 | int need_write = test_page_attr(bitmap, page, | |
1115 | BITMAP_PAGE_NEEDWRITE); | |
a647e4bc | 1116 | if (need_write) |
aa3163f8 | 1117 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); |
a647e4bc | 1118 | |
aa3163f8 | 1119 | spin_unlock_irqrestore(&bitmap->lock, flags); |
8311c29d | 1120 | if (need_write) { |
4ad13663 | 1121 | write_page(bitmap, page, 0); |
8311c29d N |
1122 | bitmap->allclean = 0; |
1123 | } | |
aa3163f8 N |
1124 | continue; |
1125 | } | |
1126 | ||
32a7627c | 1127 | /* grab the new page, sync and release the old */ |
32a7627c | 1128 | if (lastpage != NULL) { |
ec7a3197 | 1129 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1130 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1131 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 1132 | write_page(bitmap, lastpage, 0); |
32a7627c N |
1133 | } else { |
1134 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1135 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1136 | } | |
32a7627c N |
1137 | } else |
1138 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1139 | lastpage = page; | |
a0da84f3 NB |
1140 | |
1141 | /* We are possibly going to clear some bits, so make | |
1142 | * sure that events_cleared is up-to-date. | |
1143 | */ | |
1144 | if (bitmap->need_sync) { | |
1145 | bitmap_super_t *sb; | |
1146 | bitmap->need_sync = 0; | |
1147 | sb = kmap_atomic(bitmap->sb_page, KM_USER0); | |
1148 | sb->events_cleared = | |
1149 | cpu_to_le64(bitmap->events_cleared); | |
1150 | kunmap_atomic(sb, KM_USER0); | |
1151 | write_page(bitmap, bitmap->sb_page, 1); | |
1152 | } | |
32a7627c N |
1153 | spin_lock_irqsave(&bitmap->lock, flags); |
1154 | clear_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1155 | } | |
1156 | bmc = bitmap_get_counter(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap), | |
1157 | &blocks, 0); | |
1158 | if (bmc) { | |
1159 | /* | |
1160 | if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc); | |
1161 | */ | |
8311c29d N |
1162 | if (*bmc) |
1163 | bitmap->allclean = 0; | |
1164 | ||
32a7627c N |
1165 | if (*bmc == 2) { |
1166 | *bmc=1; /* maybe clear the bit next time */ | |
1167 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1168 | } else if (*bmc == 1) { | |
1169 | /* we can clear the bit */ | |
1170 | *bmc = 0; | |
1171 | bitmap_count_page(bitmap, j << CHUNK_BLOCK_SHIFT(bitmap), | |
1172 | -1); | |
1173 | ||
1174 | /* clear the bit */ | |
ea03aff9 | 1175 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1176 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1177 | clear_bit(file_page_offset(j), paddr); |
bd926c63 | 1178 | else |
ea03aff9 N |
1179 | ext2_clear_bit(file_page_offset(j), paddr); |
1180 | kunmap_atomic(paddr, KM_USER0); | |
32a7627c N |
1181 | } |
1182 | } | |
1183 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1184 | } | |
1185 | ||
1186 | /* now sync the final page */ | |
1187 | if (lastpage != NULL) { | |
32a7627c | 1188 | spin_lock_irqsave(&bitmap->lock, flags); |
ec7a3197 | 1189 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1190 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1191 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 1192 | write_page(bitmap, lastpage, 0); |
32a7627c N |
1193 | } else { |
1194 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1195 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1196 | } | |
32a7627c N |
1197 | } |
1198 | ||
7be3dfec | 1199 | done: |
8311c29d N |
1200 | if (bitmap->allclean == 0) |
1201 | bitmap->mddev->thread->timeout = bitmap->daemon_sleep * HZ; | |
32a7627c N |
1202 | } |
1203 | ||
32a7627c N |
1204 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
1205 | sector_t offset, int *blocks, | |
1206 | int create) | |
1207 | { | |
1208 | /* If 'create', we might release the lock and reclaim it. | |
1209 | * The lock must have been taken with interrupts enabled. | |
1210 | * If !create, we don't release the lock. | |
1211 | */ | |
1212 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1213 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1214 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1215 | sector_t csize; | |
1216 | ||
1217 | if (bitmap_checkpage(bitmap, page, create) < 0) { | |
1218 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); | |
1219 | *blocks = csize - (offset & (csize- 1)); | |
1220 | return NULL; | |
1221 | } | |
1222 | /* now locked ... */ | |
1223 | ||
1224 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1225 | /* should we use the first or second counter field | |
1226 | * of the hijacked pointer? */ | |
1227 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
1228 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) + | |
1229 | PAGE_COUNTER_SHIFT - 1); | |
1230 | *blocks = csize - (offset & (csize- 1)); | |
1231 | return &((bitmap_counter_t *) | |
1232 | &bitmap->bp[page].map)[hi]; | |
1233 | } else { /* page is allocated */ | |
1234 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); | |
1235 | *blocks = csize - (offset & (csize- 1)); | |
1236 | return (bitmap_counter_t *) | |
1237 | &(bitmap->bp[page].map[pageoff]); | |
1238 | } | |
1239 | } | |
1240 | ||
4b6d287f | 1241 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c N |
1242 | { |
1243 | if (!bitmap) return 0; | |
4b6d287f N |
1244 | |
1245 | if (behind) { | |
1246 | atomic_inc(&bitmap->behind_writes); | |
1247 | PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n", | |
1248 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1249 | } | |
1250 | ||
32a7627c N |
1251 | while (sectors) { |
1252 | int blocks; | |
1253 | bitmap_counter_t *bmc; | |
1254 | ||
1255 | spin_lock_irq(&bitmap->lock); | |
1256 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); | |
1257 | if (!bmc) { | |
1258 | spin_unlock_irq(&bitmap->lock); | |
1259 | return 0; | |
1260 | } | |
1261 | ||
da6e1a32 NB |
1262 | if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) { |
1263 | DEFINE_WAIT(__wait); | |
1264 | /* note that it is safe to do the prepare_to_wait | |
1265 | * after the test as long as we do it before dropping | |
1266 | * the spinlock. | |
1267 | */ | |
1268 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1269 | TASK_UNINTERRUPTIBLE); | |
1270 | spin_unlock_irq(&bitmap->lock); | |
2ad8b1ef | 1271 | blk_unplug(bitmap->mddev->queue); |
da6e1a32 NB |
1272 | schedule(); |
1273 | finish_wait(&bitmap->overflow_wait, &__wait); | |
1274 | continue; | |
1275 | } | |
1276 | ||
32a7627c N |
1277 | switch(*bmc) { |
1278 | case 0: | |
1279 | bitmap_file_set_bit(bitmap, offset); | |
1280 | bitmap_count_page(bitmap,offset, 1); | |
93769f58 | 1281 | blk_plug_device_unlocked(bitmap->mddev->queue); |
32a7627c N |
1282 | /* fall through */ |
1283 | case 1: | |
1284 | *bmc = 2; | |
1285 | } | |
da6e1a32 | 1286 | |
32a7627c N |
1287 | (*bmc)++; |
1288 | ||
1289 | spin_unlock_irq(&bitmap->lock); | |
1290 | ||
1291 | offset += blocks; | |
1292 | if (sectors > blocks) | |
1293 | sectors -= blocks; | |
1294 | else sectors = 0; | |
1295 | } | |
8311c29d | 1296 | bitmap->allclean = 0; |
32a7627c N |
1297 | return 0; |
1298 | } | |
1299 | ||
1300 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1301 | int success, int behind) |
32a7627c N |
1302 | { |
1303 | if (!bitmap) return; | |
4b6d287f N |
1304 | if (behind) { |
1305 | atomic_dec(&bitmap->behind_writes); | |
1306 | PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n", | |
1307 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1308 | } | |
1309 | ||
32a7627c N |
1310 | while (sectors) { |
1311 | int blocks; | |
1312 | unsigned long flags; | |
1313 | bitmap_counter_t *bmc; | |
1314 | ||
1315 | spin_lock_irqsave(&bitmap->lock, flags); | |
1316 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); | |
1317 | if (!bmc) { | |
1318 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1319 | return; | |
1320 | } | |
1321 | ||
a0da84f3 NB |
1322 | if (success && |
1323 | bitmap->events_cleared < bitmap->mddev->events) { | |
1324 | bitmap->events_cleared = bitmap->mddev->events; | |
1325 | bitmap->need_sync = 1; | |
1326 | } | |
1327 | ||
32a7627c N |
1328 | if (!success && ! (*bmc & NEEDED_MASK)) |
1329 | *bmc |= NEEDED_MASK; | |
1330 | ||
da6e1a32 NB |
1331 | if ((*bmc & COUNTER_MAX) == COUNTER_MAX) |
1332 | wake_up(&bitmap->overflow_wait); | |
1333 | ||
32a7627c N |
1334 | (*bmc)--; |
1335 | if (*bmc <= 2) { | |
1336 | set_page_attr(bitmap, | |
1337 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1338 | BITMAP_PAGE_CLEAN); | |
1339 | } | |
1340 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1341 | offset += blocks; | |
1342 | if (sectors > blocks) | |
1343 | sectors -= blocks; | |
1344 | else sectors = 0; | |
1345 | } | |
1346 | } | |
1347 | ||
6a806c51 N |
1348 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, |
1349 | int degraded) | |
32a7627c N |
1350 | { |
1351 | bitmap_counter_t *bmc; | |
1352 | int rv; | |
1353 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1354 | *blocks = 1024; | |
1355 | return 1; /* always resync if no bitmap */ | |
1356 | } | |
1357 | spin_lock_irq(&bitmap->lock); | |
1358 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1359 | rv = 0; | |
1360 | if (bmc) { | |
1361 | /* locked */ | |
1362 | if (RESYNC(*bmc)) | |
1363 | rv = 1; | |
1364 | else if (NEEDED(*bmc)) { | |
1365 | rv = 1; | |
6a806c51 N |
1366 | if (!degraded) { /* don't set/clear bits if degraded */ |
1367 | *bmc |= RESYNC_MASK; | |
1368 | *bmc &= ~NEEDED_MASK; | |
1369 | } | |
32a7627c N |
1370 | } |
1371 | } | |
1372 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1373 | bitmap->allclean = 0; |
32a7627c N |
1374 | return rv; |
1375 | } | |
1376 | ||
1377 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted) | |
1378 | { | |
1379 | bitmap_counter_t *bmc; | |
1380 | unsigned long flags; | |
1381 | /* | |
1382 | if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted); | |
1383 | */ if (bitmap == NULL) { | |
1384 | *blocks = 1024; | |
1385 | return; | |
1386 | } | |
1387 | spin_lock_irqsave(&bitmap->lock, flags); | |
1388 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1389 | if (bmc == NULL) | |
1390 | goto unlock; | |
1391 | /* locked */ | |
1392 | /* | |
1393 | if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks); | |
1394 | */ | |
1395 | if (RESYNC(*bmc)) { | |
1396 | *bmc &= ~RESYNC_MASK; | |
1397 | ||
1398 | if (!NEEDED(*bmc) && aborted) | |
1399 | *bmc |= NEEDED_MASK; | |
1400 | else { | |
1401 | if (*bmc <= 2) { | |
1402 | set_page_attr(bitmap, | |
1403 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1404 | BITMAP_PAGE_CLEAN); | |
1405 | } | |
1406 | } | |
1407 | } | |
1408 | unlock: | |
1409 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
8311c29d | 1410 | bitmap->allclean = 0; |
32a7627c N |
1411 | } |
1412 | ||
1413 | void bitmap_close_sync(struct bitmap *bitmap) | |
1414 | { | |
1415 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1416 | * properly have been aborted. It remains to us to clear the | |
1417 | * RESYNC bit wherever it is still on | |
1418 | */ | |
1419 | sector_t sector = 0; | |
1420 | int blocks; | |
b47490c9 N |
1421 | if (!bitmap) |
1422 | return; | |
32a7627c N |
1423 | while (sector < bitmap->mddev->resync_max_sectors) { |
1424 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1425 | sector += blocks; |
1426 | } | |
1427 | } | |
1428 | ||
1429 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1430 | { | |
1431 | sector_t s = 0; | |
1432 | int blocks; | |
1433 | ||
1434 | if (!bitmap) | |
1435 | return; | |
1436 | if (sector == 0) { | |
1437 | bitmap->last_end_sync = jiffies; | |
1438 | return; | |
1439 | } | |
1440 | if (time_before(jiffies, (bitmap->last_end_sync | |
1441 | + bitmap->daemon_sleep * HZ))) | |
1442 | return; | |
1443 | wait_event(bitmap->mddev->recovery_wait, | |
1444 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1445 | ||
1446 | sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1); | |
1447 | s = 0; | |
1448 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1449 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1450 | s += blocks; | |
32a7627c | 1451 | } |
b47490c9 | 1452 | bitmap->last_end_sync = jiffies; |
32a7627c N |
1453 | } |
1454 | ||
6a07997f | 1455 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1456 | { |
1457 | /* For each chunk covered by any of these sectors, set the | |
193f1c93 | 1458 | * counter to 1 and set resync_needed. They should all |
32a7627c N |
1459 | * be 0 at this point |
1460 | */ | |
193f1c93 N |
1461 | |
1462 | int secs; | |
1463 | bitmap_counter_t *bmc; | |
1464 | spin_lock_irq(&bitmap->lock); | |
1465 | bmc = bitmap_get_counter(bitmap, offset, &secs, 1); | |
1466 | if (!bmc) { | |
32a7627c | 1467 | spin_unlock_irq(&bitmap->lock); |
193f1c93 | 1468 | return; |
32a7627c | 1469 | } |
193f1c93 N |
1470 | if (! *bmc) { |
1471 | struct page *page; | |
6a07997f | 1472 | *bmc = 1 | (needed?NEEDED_MASK:0); |
193f1c93 N |
1473 | bitmap_count_page(bitmap, offset, 1); |
1474 | page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)); | |
1475 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1476 | } | |
1477 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1478 | bitmap->allclean = 0; |
32a7627c N |
1479 | } |
1480 | ||
9b1d1dac PC |
1481 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1482 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1483 | { | |
1484 | unsigned long chunk; | |
1485 | ||
1486 | for (chunk = s; chunk <= e; chunk++) { | |
1487 | sector_t sec = chunk << CHUNK_BLOCK_SHIFT(bitmap); | |
1488 | bitmap_set_memory_bits(bitmap, sec, 1); | |
1489 | bitmap_file_set_bit(bitmap, sec); | |
1490 | } | |
1491 | } | |
1492 | ||
6b8b3e8a N |
1493 | /* |
1494 | * flush out any pending updates | |
1495 | */ | |
1496 | void bitmap_flush(mddev_t *mddev) | |
1497 | { | |
1498 | struct bitmap *bitmap = mddev->bitmap; | |
1499 | int sleep; | |
1500 | ||
1501 | if (!bitmap) /* there was no bitmap */ | |
1502 | return; | |
1503 | ||
1504 | /* run the daemon_work three time to ensure everything is flushed | |
1505 | * that can be | |
1506 | */ | |
1507 | sleep = bitmap->daemon_sleep; | |
1508 | bitmap->daemon_sleep = 0; | |
1509 | bitmap_daemon_work(bitmap); | |
1510 | bitmap_daemon_work(bitmap); | |
1511 | bitmap_daemon_work(bitmap); | |
1512 | bitmap->daemon_sleep = sleep; | |
1513 | bitmap_update_sb(bitmap); | |
1514 | } | |
1515 | ||
32a7627c N |
1516 | /* |
1517 | * free memory that was allocated | |
1518 | */ | |
3178b0db | 1519 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1520 | { |
1521 | unsigned long k, pages; | |
1522 | struct bitmap_page *bp; | |
32a7627c N |
1523 | |
1524 | if (!bitmap) /* there was no bitmap */ | |
1525 | return; | |
1526 | ||
32a7627c N |
1527 | /* release the bitmap file and kill the daemon */ |
1528 | bitmap_file_put(bitmap); | |
1529 | ||
1530 | bp = bitmap->bp; | |
1531 | pages = bitmap->pages; | |
1532 | ||
1533 | /* free all allocated memory */ | |
1534 | ||
32a7627c N |
1535 | if (bp) /* deallocate the page memory */ |
1536 | for (k = 0; k < pages; k++) | |
1537 | if (bp[k].map && !bp[k].hijacked) | |
1538 | kfree(bp[k].map); | |
1539 | kfree(bp); | |
1540 | kfree(bitmap); | |
1541 | } | |
3178b0db N |
1542 | void bitmap_destroy(mddev_t *mddev) |
1543 | { | |
1544 | struct bitmap *bitmap = mddev->bitmap; | |
1545 | ||
1546 | if (!bitmap) /* there was no bitmap */ | |
1547 | return; | |
1548 | ||
1549 | mddev->bitmap = NULL; /* disconnect from the md device */ | |
b15c2e57 N |
1550 | if (mddev->thread) |
1551 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db N |
1552 | |
1553 | bitmap_free(bitmap); | |
1554 | } | |
32a7627c N |
1555 | |
1556 | /* | |
1557 | * initialize the bitmap structure | |
1558 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1559 | */ | |
1560 | int bitmap_create(mddev_t *mddev) | |
1561 | { | |
1562 | struct bitmap *bitmap; | |
1563 | unsigned long blocks = mddev->resync_max_sectors; | |
1564 | unsigned long chunks; | |
1565 | unsigned long pages; | |
1566 | struct file *file = mddev->bitmap_file; | |
1567 | int err; | |
6a07997f | 1568 | sector_t start; |
32a7627c | 1569 | |
5f6e3c83 | 1570 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1571 | |
a654b9d8 | 1572 | if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */ |
32a7627c N |
1573 | return 0; |
1574 | ||
a654b9d8 N |
1575 | BUG_ON(file && mddev->bitmap_offset); |
1576 | ||
9ffae0cf | 1577 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1578 | if (!bitmap) |
1579 | return -ENOMEM; | |
1580 | ||
32a7627c | 1581 | spin_lock_init(&bitmap->lock); |
ce25c31b N |
1582 | atomic_set(&bitmap->pending_writes, 0); |
1583 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1584 | init_waitqueue_head(&bitmap->overflow_wait); |
ce25c31b | 1585 | |
32a7627c | 1586 | bitmap->mddev = mddev; |
32a7627c | 1587 | |
32a7627c | 1588 | bitmap->file = file; |
a654b9d8 | 1589 | bitmap->offset = mddev->bitmap_offset; |
ce25c31b N |
1590 | if (file) { |
1591 | get_file(file); | |
ef51c976 MF |
1592 | do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX, |
1593 | SYNC_FILE_RANGE_WAIT_BEFORE | | |
1594 | SYNC_FILE_RANGE_WRITE | | |
1595 | SYNC_FILE_RANGE_WAIT_AFTER); | |
ce25c31b | 1596 | } |
32a7627c N |
1597 | /* read superblock from bitmap file (this sets bitmap->chunksize) */ |
1598 | err = bitmap_read_sb(bitmap); | |
1599 | if (err) | |
3178b0db | 1600 | goto error; |
32a7627c | 1601 | |
a638b2dc | 1602 | bitmap->chunkshift = ffz(~bitmap->chunksize); |
32a7627c N |
1603 | |
1604 | /* now that chunksize and chunkshift are set, we can use these macros */ | |
1605 | chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) / | |
1606 | CHUNK_BLOCK_RATIO(bitmap); | |
1607 | pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; | |
1608 | ||
1609 | BUG_ON(!pages); | |
1610 | ||
1611 | bitmap->chunks = chunks; | |
1612 | bitmap->pages = pages; | |
1613 | bitmap->missing_pages = pages; | |
1614 | bitmap->counter_bits = COUNTER_BITS; | |
1615 | ||
1616 | bitmap->syncchunk = ~0UL; | |
1617 | ||
44456d37 | 1618 | #ifdef INJECT_FATAL_FAULT_1 |
32a7627c N |
1619 | bitmap->bp = NULL; |
1620 | #else | |
9ffae0cf | 1621 | bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); |
32a7627c | 1622 | #endif |
3178b0db | 1623 | err = -ENOMEM; |
32a7627c | 1624 | if (!bitmap->bp) |
3178b0db | 1625 | goto error; |
32a7627c | 1626 | |
32a7627c N |
1627 | /* now that we have some pages available, initialize the in-memory |
1628 | * bitmap from the on-disk bitmap */ | |
6a07997f N |
1629 | start = 0; |
1630 | if (mddev->degraded == 0 | |
1631 | || bitmap->events_cleared == mddev->events) | |
1632 | /* no need to keep dirty bits to optimise a re-add of a missing device */ | |
1633 | start = mddev->recovery_cp; | |
1634 | err = bitmap_init_from_disk(bitmap, start); | |
193f1c93 | 1635 | |
32a7627c | 1636 | if (err) |
3178b0db | 1637 | goto error; |
32a7627c N |
1638 | |
1639 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", | |
1640 | pages, bmname(bitmap)); | |
1641 | ||
3178b0db N |
1642 | mddev->bitmap = bitmap; |
1643 | ||
b15c2e57 N |
1644 | mddev->thread->timeout = bitmap->daemon_sleep * HZ; |
1645 | ||
4ad13663 N |
1646 | bitmap_update_sb(bitmap); |
1647 | ||
1648 | return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0; | |
3178b0db N |
1649 | |
1650 | error: | |
1651 | bitmap_free(bitmap); | |
1652 | return err; | |
32a7627c N |
1653 | } |
1654 | ||
1655 | /* the bitmap API -- for raid personalities */ | |
1656 | EXPORT_SYMBOL(bitmap_startwrite); | |
1657 | EXPORT_SYMBOL(bitmap_endwrite); | |
1658 | EXPORT_SYMBOL(bitmap_start_sync); | |
1659 | EXPORT_SYMBOL(bitmap_end_sync); | |
1660 | EXPORT_SYMBOL(bitmap_unplug); | |
1661 | EXPORT_SYMBOL(bitmap_close_sync); | |
b47490c9 | 1662 | EXPORT_SYMBOL(bitmap_cond_end_sync); |