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) | |
111 | { | |
112 | unsigned char *mappage; | |
113 | ||
114 | if (page >= bitmap->pages) { | |
1187cf0a N |
115 | /* This can happen if bitmap_start_sync goes beyond |
116 | * End-of-device while looking for a whole page. | |
117 | * It is harmless. | |
118 | */ | |
32a7627c N |
119 | return -EINVAL; |
120 | } | |
121 | ||
122 | ||
123 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ | |
124 | return 0; | |
125 | ||
126 | if (bitmap->bp[page].map) /* page is already allocated, just return */ | |
127 | return 0; | |
128 | ||
129 | if (!create) | |
130 | return -ENOENT; | |
131 | ||
132 | spin_unlock_irq(&bitmap->lock); | |
133 | ||
134 | /* this page has not been allocated yet */ | |
135 | ||
136 | if ((mappage = bitmap_alloc_page(bitmap)) == NULL) { | |
137 | PRINTK("%s: bitmap map page allocation failed, hijacking\n", | |
138 | bmname(bitmap)); | |
139 | /* failed - set the hijacked flag so that we can use the | |
140 | * pointer as a counter */ | |
141 | spin_lock_irq(&bitmap->lock); | |
142 | if (!bitmap->bp[page].map) | |
143 | bitmap->bp[page].hijacked = 1; | |
144 | goto out; | |
145 | } | |
146 | ||
147 | /* got a page */ | |
148 | ||
149 | spin_lock_irq(&bitmap->lock); | |
150 | ||
151 | /* recheck the page */ | |
152 | ||
153 | if (bitmap->bp[page].map || bitmap->bp[page].hijacked) { | |
154 | /* somebody beat us to getting the page */ | |
155 | bitmap_free_page(bitmap, mappage); | |
156 | return 0; | |
157 | } | |
158 | ||
159 | /* no page was in place and we have one, so install it */ | |
160 | ||
161 | memset(mappage, 0, PAGE_SIZE); | |
162 | bitmap->bp[page].map = mappage; | |
163 | bitmap->missing_pages--; | |
164 | out: | |
165 | return 0; | |
166 | } | |
167 | ||
168 | ||
169 | /* if page is completely empty, put it back on the free list, or dealloc it */ | |
170 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
171 | /* Note: lock should be held when calling this */ | |
858119e1 | 172 | static void bitmap_checkfree(struct bitmap *bitmap, unsigned long page) |
32a7627c N |
173 | { |
174 | char *ptr; | |
175 | ||
176 | if (bitmap->bp[page].count) /* page is still busy */ | |
177 | return; | |
178 | ||
179 | /* page is no longer in use, it can be released */ | |
180 | ||
181 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
182 | bitmap->bp[page].hijacked = 0; | |
183 | bitmap->bp[page].map = NULL; | |
184 | return; | |
185 | } | |
186 | ||
187 | /* normal case, free the page */ | |
188 | ||
189 | #if 0 | |
190 | /* actually ... let's not. We will probably need the page again exactly when | |
191 | * memory is tight and we are flusing to disk | |
192 | */ | |
193 | return; | |
194 | #else | |
195 | ptr = bitmap->bp[page].map; | |
196 | bitmap->bp[page].map = NULL; | |
197 | bitmap->missing_pages++; | |
198 | bitmap_free_page(bitmap, ptr); | |
199 | return; | |
200 | #endif | |
201 | } | |
202 | ||
203 | ||
204 | /* | |
205 | * bitmap file handling - read and write the bitmap file and its superblock | |
206 | */ | |
207 | ||
32a7627c N |
208 | /* |
209 | * basic page I/O operations | |
210 | */ | |
211 | ||
a654b9d8 | 212 | /* IO operations when bitmap is stored near all superblocks */ |
a2ed9615 N |
213 | static struct page *read_sb_page(mddev_t *mddev, long offset, |
214 | struct page *page, | |
215 | unsigned long index, int size) | |
a654b9d8 N |
216 | { |
217 | /* choose a good rdev and read the page from there */ | |
218 | ||
219 | mdk_rdev_t *rdev; | |
a654b9d8 N |
220 | sector_t target; |
221 | ||
a2ed9615 N |
222 | if (!page) |
223 | page = alloc_page(GFP_KERNEL); | |
a654b9d8 N |
224 | if (!page) |
225 | return ERR_PTR(-ENOMEM); | |
a654b9d8 | 226 | |
159ec1fc | 227 | list_for_each_entry(rdev, &mddev->disks, same_set) { |
b2d444d7 N |
228 | if (! test_bit(In_sync, &rdev->flags) |
229 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
230 | continue; |
231 | ||
0f420358 | 232 | target = rdev->sb_start + offset + index * (PAGE_SIZE/512); |
a654b9d8 | 233 | |
a2ed9615 N |
234 | if (sync_page_io(rdev->bdev, target, |
235 | roundup(size, bdev_hardsect_size(rdev->bdev)), | |
236 | page, READ)) { | |
ab904d63 | 237 | page->index = index; |
ce25c31b N |
238 | attach_page_buffers(page, NULL); /* so that free_buffer will |
239 | * quietly no-op */ | |
ab904d63 N |
240 | return page; |
241 | } | |
242 | } | |
243 | return ERR_PTR(-EIO); | |
a654b9d8 | 244 | |
a654b9d8 N |
245 | } |
246 | ||
b2d2c4ce N |
247 | static mdk_rdev_t *next_active_rdev(mdk_rdev_t *rdev, mddev_t *mddev) |
248 | { | |
249 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
250 | * linked list, and raising the refcount of devices we return to ensure | |
251 | * they don't disappear while in use. | |
252 | * As devices are only added or removed when raid_disk is < 0 and | |
253 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
254 | * still be in the same position on the list when we re-enter | |
255 | * list_for_each_continue_rcu. | |
256 | */ | |
257 | struct list_head *pos; | |
258 | rcu_read_lock(); | |
259 | if (rdev == NULL) | |
260 | /* start at the beginning */ | |
261 | pos = &mddev->disks; | |
262 | else { | |
263 | /* release the previous rdev and start from there. */ | |
264 | rdev_dec_pending(rdev, mddev); | |
265 | pos = &rdev->same_set; | |
266 | } | |
267 | list_for_each_continue_rcu(pos, &mddev->disks) { | |
268 | rdev = list_entry(pos, mdk_rdev_t, same_set); | |
269 | if (rdev->raid_disk >= 0 && | |
b2d2c4ce N |
270 | !test_bit(Faulty, &rdev->flags)) { |
271 | /* this is a usable devices */ | |
272 | atomic_inc(&rdev->nr_pending); | |
273 | rcu_read_unlock(); | |
274 | return rdev; | |
275 | } | |
276 | } | |
277 | rcu_read_unlock(); | |
278 | return NULL; | |
279 | } | |
280 | ||
ab6085c7 | 281 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 282 | { |
b2d2c4ce | 283 | mdk_rdev_t *rdev = NULL; |
ab6085c7 | 284 | mddev_t *mddev = bitmap->mddev; |
a654b9d8 | 285 | |
b2d2c4ce | 286 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ab6085c7 N |
287 | int size = PAGE_SIZE; |
288 | if (page->index == bitmap->file_pages-1) | |
289 | size = roundup(bitmap->last_page_size, | |
290 | bdev_hardsect_size(rdev->bdev)); | |
f0d76d70 N |
291 | /* Just make sure we aren't corrupting data or |
292 | * metadata | |
293 | */ | |
294 | if (bitmap->offset < 0) { | |
295 | /* DATA BITMAP METADATA */ | |
296 | if (bitmap->offset | |
85bfb4da | 297 | + (long)(page->index * (PAGE_SIZE/512)) |
f0d76d70 N |
298 | + size/512 > 0) |
299 | /* bitmap runs in to metadata */ | |
4b80991c | 300 | goto bad_alignment; |
58c0fed4 | 301 | if (rdev->data_offset + mddev->dev_sectors |
0f420358 | 302 | > rdev->sb_start + bitmap->offset) |
f0d76d70 | 303 | /* data runs in to bitmap */ |
4b80991c | 304 | goto bad_alignment; |
0f420358 | 305 | } else if (rdev->sb_start < rdev->data_offset) { |
f0d76d70 | 306 | /* METADATA BITMAP DATA */ |
0f420358 | 307 | if (rdev->sb_start |
f0d76d70 N |
308 | + bitmap->offset |
309 | + page->index*(PAGE_SIZE/512) + size/512 | |
310 | > rdev->data_offset) | |
311 | /* bitmap runs in to data */ | |
4b80991c | 312 | goto bad_alignment; |
f0d76d70 N |
313 | } else { |
314 | /* DATA METADATA BITMAP - no problems */ | |
315 | } | |
a654b9d8 | 316 | md_super_write(mddev, rdev, |
0f420358 | 317 | rdev->sb_start + bitmap->offset |
a654b9d8 | 318 | + page->index * (PAGE_SIZE/512), |
ab6085c7 | 319 | size, |
a654b9d8 | 320 | page); |
b2d2c4ce | 321 | } |
a654b9d8 N |
322 | |
323 | if (wait) | |
a9701a30 | 324 | md_super_wait(mddev); |
a654b9d8 | 325 | return 0; |
4b80991c N |
326 | |
327 | bad_alignment: | |
328 | rcu_read_unlock(); | |
329 | return -EINVAL; | |
a654b9d8 N |
330 | } |
331 | ||
4ad13663 | 332 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 333 | /* |
a654b9d8 | 334 | * write out a page to a file |
32a7627c | 335 | */ |
4ad13663 | 336 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 337 | { |
d785a06a | 338 | struct buffer_head *bh; |
32a7627c | 339 | |
f0d76d70 N |
340 | if (bitmap->file == NULL) { |
341 | switch (write_sb_page(bitmap, page, wait)) { | |
342 | case -EINVAL: | |
343 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
f0d76d70 | 344 | } |
4ad13663 | 345 | } else { |
a654b9d8 | 346 | |
4ad13663 | 347 | bh = page_buffers(page); |
c708443c | 348 | |
4ad13663 N |
349 | while (bh && bh->b_blocknr) { |
350 | atomic_inc(&bitmap->pending_writes); | |
351 | set_buffer_locked(bh); | |
352 | set_buffer_mapped(bh); | |
353 | submit_bh(WRITE, bh); | |
354 | bh = bh->b_this_page; | |
355 | } | |
d785a06a | 356 | |
4ad13663 N |
357 | if (wait) { |
358 | wait_event(bitmap->write_wait, | |
359 | atomic_read(&bitmap->pending_writes)==0); | |
360 | } | |
8a5e9cf1 | 361 | } |
4ad13663 N |
362 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
363 | bitmap_file_kick(bitmap); | |
d785a06a N |
364 | } |
365 | ||
366 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
367 | { | |
368 | struct bitmap *bitmap = bh->b_private; | |
369 | unsigned long flags; | |
32a7627c | 370 | |
d785a06a N |
371 | if (!uptodate) { |
372 | spin_lock_irqsave(&bitmap->lock, flags); | |
373 | bitmap->flags |= BITMAP_WRITE_ERROR; | |
374 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
32a7627c | 375 | } |
d785a06a N |
376 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
377 | wake_up(&bitmap->write_wait); | |
378 | } | |
32a7627c | 379 | |
d785a06a N |
380 | /* copied from buffer.c */ |
381 | static void | |
382 | __clear_page_buffers(struct page *page) | |
383 | { | |
384 | ClearPagePrivate(page); | |
385 | set_page_private(page, 0); | |
386 | page_cache_release(page); | |
387 | } | |
388 | static void free_buffers(struct page *page) | |
389 | { | |
390 | struct buffer_head *bh = page_buffers(page); | |
77ad4bc7 | 391 | |
d785a06a N |
392 | while (bh) { |
393 | struct buffer_head *next = bh->b_this_page; | |
394 | free_buffer_head(bh); | |
395 | bh = next; | |
77ad4bc7 | 396 | } |
d785a06a N |
397 | __clear_page_buffers(page); |
398 | put_page(page); | |
32a7627c N |
399 | } |
400 | ||
d785a06a N |
401 | /* read a page from a file. |
402 | * We both read the page, and attach buffers to the page to record the | |
403 | * address of each block (using bmap). These addresses will be used | |
404 | * to write the block later, completely bypassing the filesystem. | |
405 | * This usage is similar to how swap files are handled, and allows us | |
406 | * to write to a file with no concerns of memory allocation failing. | |
407 | */ | |
32a7627c | 408 | static struct page *read_page(struct file *file, unsigned long index, |
d785a06a N |
409 | struct bitmap *bitmap, |
410 | unsigned long count) | |
32a7627c | 411 | { |
32a7627c | 412 | struct page *page = NULL; |
c649bb9c | 413 | struct inode *inode = file->f_path.dentry->d_inode; |
d785a06a N |
414 | struct buffer_head *bh; |
415 | sector_t block; | |
32a7627c | 416 | |
2d1f3b5d N |
417 | PRINTK("read bitmap file (%dB @ %Lu)\n", (int)PAGE_SIZE, |
418 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 419 | |
d785a06a N |
420 | page = alloc_page(GFP_KERNEL); |
421 | if (!page) | |
422 | page = ERR_PTR(-ENOMEM); | |
32a7627c N |
423 | if (IS_ERR(page)) |
424 | goto out; | |
d785a06a | 425 | |
d785a06a N |
426 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
427 | if (!bh) { | |
2d1f3b5d | 428 | put_page(page); |
d785a06a | 429 | page = ERR_PTR(-ENOMEM); |
32a7627c N |
430 | goto out; |
431 | } | |
d785a06a N |
432 | attach_page_buffers(page, bh); |
433 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
434 | while (bh) { | |
435 | if (count == 0) | |
436 | bh->b_blocknr = 0; | |
437 | else { | |
438 | bh->b_blocknr = bmap(inode, block); | |
439 | if (bh->b_blocknr == 0) { | |
440 | /* Cannot use this file! */ | |
441 | free_buffers(page); | |
442 | page = ERR_PTR(-EINVAL); | |
443 | goto out; | |
444 | } | |
445 | bh->b_bdev = inode->i_sb->s_bdev; | |
446 | if (count < (1<<inode->i_blkbits)) | |
447 | count = 0; | |
448 | else | |
449 | count -= (1<<inode->i_blkbits); | |
450 | ||
451 | bh->b_end_io = end_bitmap_write; | |
452 | bh->b_private = bitmap; | |
ce25c31b N |
453 | atomic_inc(&bitmap->pending_writes); |
454 | set_buffer_locked(bh); | |
455 | set_buffer_mapped(bh); | |
456 | submit_bh(READ, bh); | |
d785a06a N |
457 | } |
458 | block++; | |
459 | bh = bh->b_this_page; | |
460 | } | |
d785a06a | 461 | page->index = index; |
ce25c31b N |
462 | |
463 | wait_event(bitmap->write_wait, | |
464 | atomic_read(&bitmap->pending_writes)==0); | |
465 | if (bitmap->flags & BITMAP_WRITE_ERROR) { | |
466 | free_buffers(page); | |
467 | page = ERR_PTR(-EIO); | |
468 | } | |
32a7627c N |
469 | out: |
470 | if (IS_ERR(page)) | |
471 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %Lu): %ld\n", | |
2d1f3b5d N |
472 | (int)PAGE_SIZE, |
473 | (unsigned long long)index << PAGE_SHIFT, | |
32a7627c N |
474 | PTR_ERR(page)); |
475 | return page; | |
476 | } | |
477 | ||
478 | /* | |
479 | * bitmap file superblock operations | |
480 | */ | |
481 | ||
482 | /* update the event counter and sync the superblock to disk */ | |
4ad13663 | 483 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
484 | { |
485 | bitmap_super_t *sb; | |
486 | unsigned long flags; | |
487 | ||
488 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 489 | return; |
32a7627c N |
490 | spin_lock_irqsave(&bitmap->lock, flags); |
491 | if (!bitmap->sb_page) { /* no superblock */ | |
492 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 493 | return; |
32a7627c | 494 | } |
32a7627c | 495 | spin_unlock_irqrestore(&bitmap->lock, flags); |
ea03aff9 | 496 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 497 | sb->events = cpu_to_le64(bitmap->mddev->events); |
a0da84f3 NB |
498 | if (bitmap->mddev->events < bitmap->events_cleared) { |
499 | /* rocking back to read-only */ | |
500 | bitmap->events_cleared = bitmap->mddev->events; | |
501 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); | |
502 | } | |
ea03aff9 | 503 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 504 | write_page(bitmap, bitmap->sb_page, 1); |
32a7627c N |
505 | } |
506 | ||
507 | /* print out the bitmap file superblock */ | |
508 | void bitmap_print_sb(struct bitmap *bitmap) | |
509 | { | |
510 | bitmap_super_t *sb; | |
511 | ||
512 | if (!bitmap || !bitmap->sb_page) | |
513 | return; | |
ea03aff9 | 514 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 515 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
a2cff26a N |
516 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
517 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); | |
518 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", | |
32a7627c N |
519 | *(__u32 *)(sb->uuid+0), |
520 | *(__u32 *)(sb->uuid+4), | |
521 | *(__u32 *)(sb->uuid+8), | |
522 | *(__u32 *)(sb->uuid+12)); | |
a2cff26a | 523 | printk(KERN_DEBUG " events: %llu\n", |
32a7627c | 524 | (unsigned long long) le64_to_cpu(sb->events)); |
a2cff26a | 525 | printk(KERN_DEBUG "events cleared: %llu\n", |
32a7627c | 526 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
a2cff26a N |
527 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
528 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
529 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
530 | printk(KERN_DEBUG " sync size: %llu KB\n", | |
531 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
4b6d287f | 532 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
ea03aff9 | 533 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
534 | } |
535 | ||
536 | /* read the superblock from the bitmap file and initialize some bitmap fields */ | |
537 | static int bitmap_read_sb(struct bitmap *bitmap) | |
538 | { | |
539 | char *reason = NULL; | |
540 | bitmap_super_t *sb; | |
4b6d287f | 541 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c N |
542 | unsigned long long events; |
543 | int err = -EINVAL; | |
544 | ||
545 | /* page 0 is the superblock, read it... */ | |
f49d5e62 N |
546 | if (bitmap->file) { |
547 | loff_t isize = i_size_read(bitmap->file->f_mapping->host); | |
548 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; | |
549 | ||
550 | bitmap->sb_page = read_page(bitmap->file, 0, bitmap, bytes); | |
551 | } else { | |
a2ed9615 N |
552 | bitmap->sb_page = read_sb_page(bitmap->mddev, bitmap->offset, |
553 | NULL, | |
554 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 555 | } |
32a7627c N |
556 | if (IS_ERR(bitmap->sb_page)) { |
557 | err = PTR_ERR(bitmap->sb_page); | |
558 | bitmap->sb_page = NULL; | |
559 | return err; | |
560 | } | |
561 | ||
ea03aff9 | 562 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
32a7627c | 563 | |
32a7627c N |
564 | chunksize = le32_to_cpu(sb->chunksize); |
565 | daemon_sleep = le32_to_cpu(sb->daemon_sleep); | |
4b6d287f | 566 | write_behind = le32_to_cpu(sb->write_behind); |
32a7627c N |
567 | |
568 | /* verify that the bitmap-specific fields are valid */ | |
569 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
570 | reason = "bad magic"; | |
bd926c63 N |
571 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
572 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) | |
32a7627c | 573 | reason = "unrecognized superblock version"; |
1187cf0a | 574 | else if (chunksize < 512) |
7dd5d34c | 575 | reason = "bitmap chunksize too small"; |
32a7627c N |
576 | else if ((1 << ffz(~chunksize)) != chunksize) |
577 | reason = "bitmap chunksize not a power of 2"; | |
7dd5d34c N |
578 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT / HZ) |
579 | reason = "daemon sleep period out of range"; | |
4b6d287f N |
580 | else if (write_behind > COUNTER_MAX) |
581 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c N |
582 | if (reason) { |
583 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", | |
584 | bmname(bitmap), reason); | |
585 | goto out; | |
586 | } | |
587 | ||
588 | /* keep the array size field of the bitmap superblock up to date */ | |
589 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
590 | ||
591 | if (!bitmap->mddev->persistent) | |
592 | goto success; | |
593 | ||
594 | /* | |
595 | * if we have a persistent array superblock, compare the | |
596 | * bitmap's UUID and event counter to the mddev's | |
597 | */ | |
598 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
599 | printk(KERN_INFO "%s: bitmap superblock UUID mismatch\n", | |
600 | bmname(bitmap)); | |
601 | goto out; | |
602 | } | |
603 | events = le64_to_cpu(sb->events); | |
604 | if (events < bitmap->mddev->events) { | |
605 | printk(KERN_INFO "%s: bitmap file is out of date (%llu < %llu) " | |
606 | "-- forcing full recovery\n", bmname(bitmap), events, | |
607 | (unsigned long long) bitmap->mddev->events); | |
4f2e639a | 608 | sb->state |= cpu_to_le32(BITMAP_STALE); |
32a7627c N |
609 | } |
610 | success: | |
611 | /* assign fields using values from superblock */ | |
612 | bitmap->chunksize = chunksize; | |
613 | bitmap->daemon_sleep = daemon_sleep; | |
585f0dd5 | 614 | bitmap->daemon_lastrun = jiffies; |
4b6d287f | 615 | bitmap->max_write_behind = write_behind; |
4f2e639a | 616 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 N |
617 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
618 | bitmap->flags |= BITMAP_HOSTENDIAN; | |
32a7627c | 619 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
4f2e639a | 620 | if (sb->state & cpu_to_le32(BITMAP_STALE)) |
6a07997f | 621 | bitmap->events_cleared = bitmap->mddev->events; |
32a7627c N |
622 | err = 0; |
623 | out: | |
ea03aff9 | 624 | kunmap_atomic(sb, KM_USER0); |
32a7627c N |
625 | if (err) |
626 | bitmap_print_sb(bitmap); | |
627 | return err; | |
628 | } | |
629 | ||
630 | enum bitmap_mask_op { | |
631 | MASK_SET, | |
632 | MASK_UNSET | |
633 | }; | |
634 | ||
4ad13663 N |
635 | /* record the state of the bitmap in the superblock. Return the old value */ |
636 | static int bitmap_mask_state(struct bitmap *bitmap, enum bitmap_state bits, | |
637 | enum bitmap_mask_op op) | |
32a7627c N |
638 | { |
639 | bitmap_super_t *sb; | |
640 | unsigned long flags; | |
4ad13663 | 641 | int old; |
32a7627c N |
642 | |
643 | spin_lock_irqsave(&bitmap->lock, flags); | |
7e317655 | 644 | if (!bitmap->sb_page) { /* can't set the state */ |
32a7627c | 645 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 646 | return 0; |
32a7627c | 647 | } |
32a7627c | 648 | spin_unlock_irqrestore(&bitmap->lock, flags); |
ea03aff9 | 649 | sb = (bitmap_super_t *)kmap_atomic(bitmap->sb_page, KM_USER0); |
4ad13663 | 650 | old = le32_to_cpu(sb->state) & bits; |
32a7627c | 651 | switch (op) { |
4f2e639a | 652 | case MASK_SET: sb->state |= cpu_to_le32(bits); |
32a7627c | 653 | break; |
4f2e639a | 654 | case MASK_UNSET: sb->state &= cpu_to_le32(~bits); |
32a7627c N |
655 | break; |
656 | default: BUG(); | |
657 | } | |
ea03aff9 | 658 | kunmap_atomic(sb, KM_USER0); |
4ad13663 | 659 | return old; |
32a7627c N |
660 | } |
661 | ||
662 | /* | |
663 | * general bitmap file operations | |
664 | */ | |
665 | ||
666 | /* calculate the index of the page that contains this bit */ | |
667 | static inline unsigned long file_page_index(unsigned long chunk) | |
668 | { | |
669 | return CHUNK_BIT_OFFSET(chunk) >> PAGE_BIT_SHIFT; | |
670 | } | |
671 | ||
672 | /* calculate the (bit) offset of this bit within a page */ | |
673 | static inline unsigned long file_page_offset(unsigned long chunk) | |
674 | { | |
675 | return CHUNK_BIT_OFFSET(chunk) & (PAGE_BITS - 1); | |
676 | } | |
677 | ||
678 | /* | |
679 | * return a pointer to the page in the filemap that contains the given bit | |
680 | * | |
681 | * this lookup is complicated by the fact that the bitmap sb might be exactly | |
682 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page | |
683 | * 0 or page 1 | |
684 | */ | |
685 | static inline struct page *filemap_get_page(struct bitmap *bitmap, | |
686 | unsigned long chunk) | |
687 | { | |
9b1d1dac | 688 | if (file_page_index(chunk) >= bitmap->file_pages) return NULL; |
32a7627c N |
689 | return bitmap->filemap[file_page_index(chunk) - file_page_index(0)]; |
690 | } | |
691 | ||
692 | ||
693 | static void bitmap_file_unmap(struct bitmap *bitmap) | |
694 | { | |
695 | struct page **map, *sb_page; | |
696 | unsigned long *attr; | |
697 | int pages; | |
698 | unsigned long flags; | |
699 | ||
700 | spin_lock_irqsave(&bitmap->lock, flags); | |
701 | map = bitmap->filemap; | |
702 | bitmap->filemap = NULL; | |
703 | attr = bitmap->filemap_attr; | |
704 | bitmap->filemap_attr = NULL; | |
705 | pages = bitmap->file_pages; | |
706 | bitmap->file_pages = 0; | |
707 | sb_page = bitmap->sb_page; | |
708 | bitmap->sb_page = NULL; | |
709 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
710 | ||
711 | while (pages--) | |
712 | if (map[pages]->index != 0) /* 0 is sb_page, release it below */ | |
d785a06a | 713 | free_buffers(map[pages]); |
32a7627c N |
714 | kfree(map); |
715 | kfree(attr); | |
716 | ||
d785a06a N |
717 | if (sb_page) |
718 | free_buffers(sb_page); | |
32a7627c N |
719 | } |
720 | ||
721 | static void bitmap_file_put(struct bitmap *bitmap) | |
722 | { | |
723 | struct file *file; | |
32a7627c N |
724 | unsigned long flags; |
725 | ||
726 | spin_lock_irqsave(&bitmap->lock, flags); | |
727 | file = bitmap->file; | |
728 | bitmap->file = NULL; | |
729 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
730 | ||
d785a06a N |
731 | if (file) |
732 | wait_event(bitmap->write_wait, | |
733 | atomic_read(&bitmap->pending_writes)==0); | |
32a7627c N |
734 | bitmap_file_unmap(bitmap); |
735 | ||
d785a06a | 736 | if (file) { |
c649bb9c | 737 | struct inode *inode = file->f_path.dentry->d_inode; |
fc0ecff6 | 738 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 739 | fput(file); |
d785a06a | 740 | } |
32a7627c N |
741 | } |
742 | ||
743 | ||
744 | /* | |
745 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
746 | * then it is no longer reliable, so we stop using it and we mark the file | |
747 | * as failed in the superblock | |
748 | */ | |
749 | static void bitmap_file_kick(struct bitmap *bitmap) | |
750 | { | |
751 | char *path, *ptr = NULL; | |
752 | ||
4ad13663 N |
753 | if (bitmap_mask_state(bitmap, BITMAP_STALE, MASK_SET) == 0) { |
754 | bitmap_update_sb(bitmap); | |
32a7627c | 755 | |
4ad13663 N |
756 | if (bitmap->file) { |
757 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); | |
758 | if (path) | |
6bcfd601 CH |
759 | ptr = d_path(&bitmap->file->f_path, path, |
760 | PAGE_SIZE); | |
761 | ||
32a7627c | 762 | |
4ad13663 N |
763 | printk(KERN_ALERT |
764 | "%s: kicking failed bitmap file %s from array!\n", | |
6bcfd601 | 765 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
32a7627c | 766 | |
4ad13663 N |
767 | kfree(path); |
768 | } else | |
769 | printk(KERN_ALERT | |
770 | "%s: disabling internal bitmap due to errors\n", | |
771 | bmname(bitmap)); | |
a654b9d8 | 772 | } |
32a7627c N |
773 | |
774 | bitmap_file_put(bitmap); | |
775 | ||
776 | return; | |
777 | } | |
778 | ||
779 | enum bitmap_page_attr { | |
e16b68b6 N |
780 | BITMAP_PAGE_DIRTY = 0, // there are set bits that need to be synced |
781 | BITMAP_PAGE_CLEAN = 1, // there are bits that might need to be cleared | |
782 | BITMAP_PAGE_NEEDWRITE=2, // there are cleared bits that need to be synced | |
32a7627c N |
783 | }; |
784 | ||
785 | static inline void set_page_attr(struct bitmap *bitmap, struct page *page, | |
786 | enum bitmap_page_attr attr) | |
787 | { | |
e16b68b6 | 788 | __set_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
789 | } |
790 | ||
791 | static inline void clear_page_attr(struct bitmap *bitmap, struct page *page, | |
792 | enum bitmap_page_attr attr) | |
793 | { | |
e16b68b6 | 794 | __clear_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
795 | } |
796 | ||
ec7a3197 N |
797 | static inline unsigned long test_page_attr(struct bitmap *bitmap, struct page *page, |
798 | enum bitmap_page_attr attr) | |
32a7627c | 799 | { |
e16b68b6 | 800 | return test_bit((page->index<<2) + attr, bitmap->filemap_attr); |
32a7627c N |
801 | } |
802 | ||
803 | /* | |
804 | * bitmap_file_set_bit -- called before performing a write to the md device | |
805 | * to set (and eventually sync) a particular bit in the bitmap file | |
806 | * | |
807 | * we set the bit immediately, then we record the page number so that | |
808 | * when an unplug occurs, we can flush the dirty pages out to disk | |
809 | */ | |
810 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
811 | { | |
812 | unsigned long bit; | |
813 | struct page *page; | |
814 | void *kaddr; | |
815 | unsigned long chunk = block >> CHUNK_BLOCK_SHIFT(bitmap); | |
816 | ||
a654b9d8 | 817 | if (!bitmap->filemap) { |
32a7627c N |
818 | return; |
819 | } | |
820 | ||
821 | page = filemap_get_page(bitmap, chunk); | |
9b1d1dac | 822 | if (!page) return; |
32a7627c N |
823 | bit = file_page_offset(chunk); |
824 | ||
32a7627c N |
825 | /* set the bit */ |
826 | kaddr = kmap_atomic(page, KM_USER0); | |
bd926c63 N |
827 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
828 | set_bit(bit, kaddr); | |
829 | else | |
830 | ext2_set_bit(bit, kaddr); | |
32a7627c N |
831 | kunmap_atomic(kaddr, KM_USER0); |
832 | PRINTK("set file bit %lu page %lu\n", bit, page->index); | |
833 | ||
834 | /* record page number so it gets flushed to disk when unplug occurs */ | |
835 | set_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); | |
836 | ||
837 | } | |
838 | ||
839 | /* this gets called when the md device is ready to unplug its underlying | |
840 | * (slave) device queues -- before we let any writes go down, we need to | |
841 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 842 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 843 | { |
ec7a3197 N |
844 | unsigned long i, flags; |
845 | int dirty, need_write; | |
32a7627c N |
846 | struct page *page; |
847 | int wait = 0; | |
848 | ||
849 | if (!bitmap) | |
4ad13663 | 850 | return; |
32a7627c N |
851 | |
852 | /* look at each page to see if there are any set bits that need to be | |
853 | * flushed out to disk */ | |
854 | for (i = 0; i < bitmap->file_pages; i++) { | |
855 | spin_lock_irqsave(&bitmap->lock, flags); | |
a654b9d8 | 856 | if (!bitmap->filemap) { |
32a7627c | 857 | spin_unlock_irqrestore(&bitmap->lock, flags); |
4ad13663 | 858 | return; |
32a7627c N |
859 | } |
860 | page = bitmap->filemap[i]; | |
ec7a3197 N |
861 | dirty = test_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
862 | need_write = test_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
32a7627c N |
863 | clear_page_attr(bitmap, page, BITMAP_PAGE_DIRTY); |
864 | clear_page_attr(bitmap, page, BITMAP_PAGE_NEEDWRITE); | |
ec7a3197 | 865 | if (dirty) |
32a7627c N |
866 | wait = 1; |
867 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
868 | ||
d785a06a | 869 | if (dirty | need_write) |
4ad13663 | 870 | write_page(bitmap, page, 0); |
32a7627c N |
871 | } |
872 | if (wait) { /* if any writes were performed, we need to wait on them */ | |
0b79ccf0 | 873 | if (bitmap->file) |
d785a06a N |
874 | wait_event(bitmap->write_wait, |
875 | atomic_read(&bitmap->pending_writes)==0); | |
0b79ccf0 | 876 | else |
a9701a30 | 877 | md_super_wait(bitmap->mddev); |
32a7627c | 878 | } |
d785a06a N |
879 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
880 | bitmap_file_kick(bitmap); | |
32a7627c N |
881 | } |
882 | ||
6a07997f | 883 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
884 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
885 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
886 | * memory mapping of the bitmap file | |
887 | * Special cases: | |
888 | * if there's no bitmap file, or if the bitmap file had been | |
889 | * previously kicked from the array, we mark all the bits as | |
890 | * 1's in order to cause a full resync. | |
6a07997f N |
891 | * |
892 | * We ignore all bits for sectors that end earlier than 'start'. | |
893 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 894 | */ |
6a07997f | 895 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c N |
896 | { |
897 | unsigned long i, chunks, index, oldindex, bit; | |
898 | struct page *page = NULL, *oldpage = NULL; | |
899 | unsigned long num_pages, bit_cnt = 0; | |
900 | struct file *file; | |
d785a06a | 901 | unsigned long bytes, offset; |
32a7627c N |
902 | int outofdate; |
903 | int ret = -ENOSPC; | |
ea03aff9 | 904 | void *paddr; |
32a7627c N |
905 | |
906 | chunks = bitmap->chunks; | |
907 | file = bitmap->file; | |
908 | ||
a654b9d8 | 909 | BUG_ON(!file && !bitmap->offset); |
32a7627c | 910 | |
44456d37 | 911 | #ifdef INJECT_FAULTS_3 |
32a7627c N |
912 | outofdate = 1; |
913 | #else | |
914 | outofdate = bitmap->flags & BITMAP_STALE; | |
915 | #endif | |
916 | if (outofdate) | |
917 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " | |
918 | "recovery\n", bmname(bitmap)); | |
919 | ||
920 | bytes = (chunks + 7) / 8; | |
bc7f77de | 921 | |
cdbb4cc2 | 922 | num_pages = (bytes + sizeof(bitmap_super_t) + PAGE_SIZE - 1) / PAGE_SIZE; |
bc7f77de | 923 | |
a654b9d8 | 924 | if (file && i_size_read(file->f_mapping->host) < bytes + sizeof(bitmap_super_t)) { |
32a7627c N |
925 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
926 | bmname(bitmap), | |
927 | (unsigned long) i_size_read(file->f_mapping->host), | |
928 | bytes + sizeof(bitmap_super_t)); | |
4ad13663 | 929 | goto err; |
32a7627c | 930 | } |
bc7f77de N |
931 | |
932 | ret = -ENOMEM; | |
933 | ||
32a7627c | 934 | bitmap->filemap = kmalloc(sizeof(struct page *) * num_pages, GFP_KERNEL); |
bc7f77de | 935 | if (!bitmap->filemap) |
4ad13663 | 936 | goto err; |
32a7627c | 937 | |
e16b68b6 N |
938 | /* We need 4 bits per page, rounded up to a multiple of sizeof(unsigned long) */ |
939 | bitmap->filemap_attr = kzalloc( | |
505fa2c4 | 940 | roundup( DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), |
e16b68b6 | 941 | GFP_KERNEL); |
bc7f77de | 942 | if (!bitmap->filemap_attr) |
4ad13663 | 943 | goto err; |
32a7627c | 944 | |
32a7627c N |
945 | oldindex = ~0L; |
946 | ||
947 | for (i = 0; i < chunks; i++) { | |
bd926c63 | 948 | int b; |
32a7627c N |
949 | index = file_page_index(i); |
950 | bit = file_page_offset(i); | |
951 | if (index != oldindex) { /* this is a new page, read it in */ | |
d785a06a | 952 | int count; |
32a7627c | 953 | /* unmap the old page, we're done with it */ |
d785a06a | 954 | if (index == num_pages-1) |
f49d5e62 N |
955 | count = bytes + sizeof(bitmap_super_t) |
956 | - index * PAGE_SIZE; | |
d785a06a N |
957 | else |
958 | count = PAGE_SIZE; | |
32a7627c N |
959 | if (index == 0) { |
960 | /* | |
961 | * if we're here then the superblock page | |
962 | * contains some bits (PAGE_SIZE != sizeof sb) | |
963 | * we've already read it in, so just use it | |
964 | */ | |
965 | page = bitmap->sb_page; | |
966 | offset = sizeof(bitmap_super_t); | |
53845270 N |
967 | if (!file) |
968 | read_sb_page(bitmap->mddev, | |
969 | bitmap->offset, | |
970 | page, | |
971 | index, count); | |
a654b9d8 | 972 | } else if (file) { |
d785a06a | 973 | page = read_page(file, index, bitmap, count); |
a654b9d8 N |
974 | offset = 0; |
975 | } else { | |
a2ed9615 N |
976 | page = read_sb_page(bitmap->mddev, bitmap->offset, |
977 | NULL, | |
978 | index, count); | |
32a7627c N |
979 | offset = 0; |
980 | } | |
a654b9d8 N |
981 | if (IS_ERR(page)) { /* read error */ |
982 | ret = PTR_ERR(page); | |
4ad13663 | 983 | goto err; |
a654b9d8 N |
984 | } |
985 | ||
32a7627c N |
986 | oldindex = index; |
987 | oldpage = page; | |
32a7627c | 988 | |
b74fd282 N |
989 | bitmap->filemap[bitmap->file_pages++] = page; |
990 | bitmap->last_page_size = count; | |
991 | ||
32a7627c N |
992 | if (outofdate) { |
993 | /* | |
994 | * if bitmap is out of date, dirty the | |
995 | * whole page and write it out | |
996 | */ | |
ea03aff9 N |
997 | paddr = kmap_atomic(page, KM_USER0); |
998 | memset(paddr + offset, 0xff, | |
6a07997f | 999 | PAGE_SIZE - offset); |
ea03aff9 | 1000 | kunmap_atomic(paddr, KM_USER0); |
4ad13663 N |
1001 | write_page(bitmap, page, 1); |
1002 | ||
1003 | ret = -EIO; | |
b74fd282 | 1004 | if (bitmap->flags & BITMAP_WRITE_ERROR) |
4ad13663 | 1005 | goto err; |
32a7627c | 1006 | } |
32a7627c | 1007 | } |
ea03aff9 | 1008 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1009 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1010 | b = test_bit(bit, paddr); |
bd926c63 | 1011 | else |
ea03aff9 N |
1012 | b = ext2_test_bit(bit, paddr); |
1013 | kunmap_atomic(paddr, KM_USER0); | |
bd926c63 | 1014 | if (b) { |
32a7627c | 1015 | /* if the disk bit is set, set the memory bit */ |
db305e50 N |
1016 | int needed = ((sector_t)(i+1) << (CHUNK_BLOCK_SHIFT(bitmap)) |
1017 | >= start); | |
1018 | bitmap_set_memory_bits(bitmap, | |
1019 | (sector_t)i << CHUNK_BLOCK_SHIFT(bitmap), | |
1020 | needed); | |
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 | } | |
db305e50 N |
1156 | bmc = bitmap_get_counter(bitmap, |
1157 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
1158 | &blocks, 0); | |
32a7627c N |
1159 | if (bmc) { |
1160 | /* | |
1161 | if (j < 100) printk("bitmap: j=%lu, *bmc = 0x%x\n", j, *bmc); | |
1162 | */ | |
8311c29d N |
1163 | if (*bmc) |
1164 | bitmap->allclean = 0; | |
1165 | ||
32a7627c N |
1166 | if (*bmc == 2) { |
1167 | *bmc=1; /* maybe clear the bit next time */ | |
1168 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1169 | } else if (*bmc == 1) { | |
1170 | /* we can clear the bit */ | |
1171 | *bmc = 0; | |
db305e50 N |
1172 | bitmap_count_page(bitmap, |
1173 | (sector_t)j << CHUNK_BLOCK_SHIFT(bitmap), | |
32a7627c N |
1174 | -1); |
1175 | ||
1176 | /* clear the bit */ | |
ea03aff9 | 1177 | paddr = kmap_atomic(page, KM_USER0); |
bd926c63 | 1178 | if (bitmap->flags & BITMAP_HOSTENDIAN) |
ea03aff9 | 1179 | clear_bit(file_page_offset(j), paddr); |
bd926c63 | 1180 | else |
ea03aff9 N |
1181 | ext2_clear_bit(file_page_offset(j), paddr); |
1182 | kunmap_atomic(paddr, KM_USER0); | |
32a7627c N |
1183 | } |
1184 | } | |
1185 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1186 | } | |
1187 | ||
1188 | /* now sync the final page */ | |
1189 | if (lastpage != NULL) { | |
32a7627c | 1190 | spin_lock_irqsave(&bitmap->lock, flags); |
ec7a3197 | 1191 | if (test_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE)) { |
32a7627c N |
1192 | clear_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); |
1193 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
4ad13663 | 1194 | write_page(bitmap, lastpage, 0); |
32a7627c N |
1195 | } else { |
1196 | set_page_attr(bitmap, lastpage, BITMAP_PAGE_NEEDWRITE); | |
1197 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1198 | } | |
32a7627c N |
1199 | } |
1200 | ||
7be3dfec | 1201 | done: |
8311c29d N |
1202 | if (bitmap->allclean == 0) |
1203 | bitmap->mddev->thread->timeout = bitmap->daemon_sleep * HZ; | |
32a7627c N |
1204 | } |
1205 | ||
32a7627c N |
1206 | static bitmap_counter_t *bitmap_get_counter(struct bitmap *bitmap, |
1207 | sector_t offset, int *blocks, | |
1208 | int create) | |
1209 | { | |
1210 | /* If 'create', we might release the lock and reclaim it. | |
1211 | * The lock must have been taken with interrupts enabled. | |
1212 | * If !create, we don't release the lock. | |
1213 | */ | |
1214 | sector_t chunk = offset >> CHUNK_BLOCK_SHIFT(bitmap); | |
1215 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1216 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1217 | sector_t csize; | |
1218 | ||
1219 | if (bitmap_checkpage(bitmap, page, create) < 0) { | |
1220 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); | |
1221 | *blocks = csize - (offset & (csize- 1)); | |
1222 | return NULL; | |
1223 | } | |
1224 | /* now locked ... */ | |
1225 | ||
1226 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1227 | /* should we use the first or second counter field | |
1228 | * of the hijacked pointer? */ | |
1229 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
1230 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap) + | |
1231 | PAGE_COUNTER_SHIFT - 1); | |
1232 | *blocks = csize - (offset & (csize- 1)); | |
1233 | return &((bitmap_counter_t *) | |
1234 | &bitmap->bp[page].map)[hi]; | |
1235 | } else { /* page is allocated */ | |
1236 | csize = ((sector_t)1) << (CHUNK_BLOCK_SHIFT(bitmap)); | |
1237 | *blocks = csize - (offset & (csize- 1)); | |
1238 | return (bitmap_counter_t *) | |
1239 | &(bitmap->bp[page].map[pageoff]); | |
1240 | } | |
1241 | } | |
1242 | ||
4b6d287f | 1243 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c N |
1244 | { |
1245 | if (!bitmap) return 0; | |
4b6d287f N |
1246 | |
1247 | if (behind) { | |
1248 | atomic_inc(&bitmap->behind_writes); | |
1249 | PRINTK(KERN_DEBUG "inc write-behind count %d/%d\n", | |
1250 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1251 | } | |
1252 | ||
32a7627c N |
1253 | while (sectors) { |
1254 | int blocks; | |
1255 | bitmap_counter_t *bmc; | |
1256 | ||
1257 | spin_lock_irq(&bitmap->lock); | |
1258 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 1); | |
1259 | if (!bmc) { | |
1260 | spin_unlock_irq(&bitmap->lock); | |
1261 | return 0; | |
1262 | } | |
1263 | ||
da6e1a32 NB |
1264 | if (unlikely((*bmc & COUNTER_MAX) == COUNTER_MAX)) { |
1265 | DEFINE_WAIT(__wait); | |
1266 | /* note that it is safe to do the prepare_to_wait | |
1267 | * after the test as long as we do it before dropping | |
1268 | * the spinlock. | |
1269 | */ | |
1270 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1271 | TASK_UNINTERRUPTIBLE); | |
1272 | spin_unlock_irq(&bitmap->lock); | |
2ad8b1ef | 1273 | blk_unplug(bitmap->mddev->queue); |
da6e1a32 NB |
1274 | schedule(); |
1275 | finish_wait(&bitmap->overflow_wait, &__wait); | |
1276 | continue; | |
1277 | } | |
1278 | ||
32a7627c N |
1279 | switch(*bmc) { |
1280 | case 0: | |
1281 | bitmap_file_set_bit(bitmap, offset); | |
1282 | bitmap_count_page(bitmap,offset, 1); | |
93769f58 | 1283 | blk_plug_device_unlocked(bitmap->mddev->queue); |
32a7627c N |
1284 | /* fall through */ |
1285 | case 1: | |
1286 | *bmc = 2; | |
1287 | } | |
da6e1a32 | 1288 | |
32a7627c N |
1289 | (*bmc)++; |
1290 | ||
1291 | spin_unlock_irq(&bitmap->lock); | |
1292 | ||
1293 | offset += blocks; | |
1294 | if (sectors > blocks) | |
1295 | sectors -= blocks; | |
1296 | else sectors = 0; | |
1297 | } | |
8311c29d | 1298 | bitmap->allclean = 0; |
32a7627c N |
1299 | return 0; |
1300 | } | |
1301 | ||
1302 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1303 | int success, int behind) |
32a7627c N |
1304 | { |
1305 | if (!bitmap) return; | |
4b6d287f N |
1306 | if (behind) { |
1307 | atomic_dec(&bitmap->behind_writes); | |
1308 | PRINTK(KERN_DEBUG "dec write-behind count %d/%d\n", | |
1309 | atomic_read(&bitmap->behind_writes), bitmap->max_write_behind); | |
1310 | } | |
d0a4bb49 N |
1311 | if (bitmap->mddev->degraded) |
1312 | /* Never clear bits or update events_cleared when degraded */ | |
1313 | success = 0; | |
4b6d287f | 1314 | |
32a7627c N |
1315 | while (sectors) { |
1316 | int blocks; | |
1317 | unsigned long flags; | |
1318 | bitmap_counter_t *bmc; | |
1319 | ||
1320 | spin_lock_irqsave(&bitmap->lock, flags); | |
1321 | bmc = bitmap_get_counter(bitmap, offset, &blocks, 0); | |
1322 | if (!bmc) { | |
1323 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1324 | return; | |
1325 | } | |
1326 | ||
a0da84f3 NB |
1327 | if (success && |
1328 | bitmap->events_cleared < bitmap->mddev->events) { | |
1329 | bitmap->events_cleared = bitmap->mddev->events; | |
1330 | bitmap->need_sync = 1; | |
1331 | } | |
1332 | ||
32a7627c N |
1333 | if (!success && ! (*bmc & NEEDED_MASK)) |
1334 | *bmc |= NEEDED_MASK; | |
1335 | ||
da6e1a32 NB |
1336 | if ((*bmc & COUNTER_MAX) == COUNTER_MAX) |
1337 | wake_up(&bitmap->overflow_wait); | |
1338 | ||
32a7627c N |
1339 | (*bmc)--; |
1340 | if (*bmc <= 2) { | |
1341 | set_page_attr(bitmap, | |
1342 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1343 | BITMAP_PAGE_CLEAN); | |
1344 | } | |
1345 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
1346 | offset += blocks; | |
1347 | if (sectors > blocks) | |
1348 | sectors -= blocks; | |
1349 | else sectors = 0; | |
1350 | } | |
1351 | } | |
1352 | ||
1187cf0a N |
1353 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, |
1354 | int degraded) | |
32a7627c N |
1355 | { |
1356 | bitmap_counter_t *bmc; | |
1357 | int rv; | |
1358 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1359 | *blocks = 1024; | |
1360 | return 1; /* always resync if no bitmap */ | |
1361 | } | |
1362 | spin_lock_irq(&bitmap->lock); | |
1363 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1364 | rv = 0; | |
1365 | if (bmc) { | |
1366 | /* locked */ | |
1367 | if (RESYNC(*bmc)) | |
1368 | rv = 1; | |
1369 | else if (NEEDED(*bmc)) { | |
1370 | rv = 1; | |
6a806c51 N |
1371 | if (!degraded) { /* don't set/clear bits if degraded */ |
1372 | *bmc |= RESYNC_MASK; | |
1373 | *bmc &= ~NEEDED_MASK; | |
1374 | } | |
32a7627c N |
1375 | } |
1376 | } | |
1377 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1378 | bitmap->allclean = 0; |
32a7627c N |
1379 | return rv; |
1380 | } | |
1381 | ||
1187cf0a N |
1382 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, int *blocks, |
1383 | int degraded) | |
1384 | { | |
1385 | /* bitmap_start_sync must always report on multiples of whole | |
1386 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1387 | * get confused. | |
1388 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1389 | * At least PAGE_SIZE>>9 blocks are covered. | |
1390 | * Return the 'or' of the result. | |
1391 | */ | |
1392 | int rv = 0; | |
1393 | int blocks1; | |
1394 | ||
1395 | *blocks = 0; | |
1396 | while (*blocks < (PAGE_SIZE>>9)) { | |
1397 | rv |= __bitmap_start_sync(bitmap, offset, | |
1398 | &blocks1, degraded); | |
1399 | offset += blocks1; | |
1400 | *blocks += blocks1; | |
1401 | } | |
1402 | return rv; | |
1403 | } | |
1404 | ||
32a7627c N |
1405 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, int *blocks, int aborted) |
1406 | { | |
1407 | bitmap_counter_t *bmc; | |
1408 | unsigned long flags; | |
1409 | /* | |
1410 | if (offset == 0) printk("bitmap_end_sync 0 (%d)\n", aborted); | |
1411 | */ if (bitmap == NULL) { | |
1412 | *blocks = 1024; | |
1413 | return; | |
1414 | } | |
1415 | spin_lock_irqsave(&bitmap->lock, flags); | |
1416 | bmc = bitmap_get_counter(bitmap, offset, blocks, 0); | |
1417 | if (bmc == NULL) | |
1418 | goto unlock; | |
1419 | /* locked */ | |
1420 | /* | |
1421 | if (offset == 0) printk("bitmap_end sync found 0x%x, blocks %d\n", *bmc, *blocks); | |
1422 | */ | |
1423 | if (RESYNC(*bmc)) { | |
1424 | *bmc &= ~RESYNC_MASK; | |
1425 | ||
1426 | if (!NEEDED(*bmc) && aborted) | |
1427 | *bmc |= NEEDED_MASK; | |
1428 | else { | |
1429 | if (*bmc <= 2) { | |
1430 | set_page_attr(bitmap, | |
1431 | filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)), | |
1432 | BITMAP_PAGE_CLEAN); | |
1433 | } | |
1434 | } | |
1435 | } | |
1436 | unlock: | |
1437 | spin_unlock_irqrestore(&bitmap->lock, flags); | |
8311c29d | 1438 | bitmap->allclean = 0; |
32a7627c N |
1439 | } |
1440 | ||
1441 | void bitmap_close_sync(struct bitmap *bitmap) | |
1442 | { | |
1443 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1444 | * properly have been aborted. It remains to us to clear the | |
1445 | * RESYNC bit wherever it is still on | |
1446 | */ | |
1447 | sector_t sector = 0; | |
1448 | int blocks; | |
b47490c9 N |
1449 | if (!bitmap) |
1450 | return; | |
32a7627c N |
1451 | while (sector < bitmap->mddev->resync_max_sectors) { |
1452 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1453 | sector += blocks; |
1454 | } | |
1455 | } | |
1456 | ||
1457 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) | |
1458 | { | |
1459 | sector_t s = 0; | |
1460 | int blocks; | |
1461 | ||
1462 | if (!bitmap) | |
1463 | return; | |
1464 | if (sector == 0) { | |
1465 | bitmap->last_end_sync = jiffies; | |
1466 | return; | |
1467 | } | |
1468 | if (time_before(jiffies, (bitmap->last_end_sync | |
1469 | + bitmap->daemon_sleep * HZ))) | |
1470 | return; | |
1471 | wait_event(bitmap->mddev->recovery_wait, | |
1472 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1473 | ||
97e4f42d N |
1474 | bitmap->mddev->curr_resync_completed = bitmap->mddev->curr_resync; |
1475 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); | |
b47490c9 N |
1476 | sector &= ~((1ULL << CHUNK_BLOCK_SHIFT(bitmap)) - 1); |
1477 | s = 0; | |
1478 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1479 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1480 | s += blocks; | |
32a7627c | 1481 | } |
b47490c9 | 1482 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1483 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c N |
1484 | } |
1485 | ||
6a07997f | 1486 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1487 | { |
1488 | /* For each chunk covered by any of these sectors, set the | |
193f1c93 | 1489 | * counter to 1 and set resync_needed. They should all |
32a7627c N |
1490 | * be 0 at this point |
1491 | */ | |
193f1c93 N |
1492 | |
1493 | int secs; | |
1494 | bitmap_counter_t *bmc; | |
1495 | spin_lock_irq(&bitmap->lock); | |
1496 | bmc = bitmap_get_counter(bitmap, offset, &secs, 1); | |
1497 | if (!bmc) { | |
32a7627c | 1498 | spin_unlock_irq(&bitmap->lock); |
193f1c93 | 1499 | return; |
32a7627c | 1500 | } |
193f1c93 N |
1501 | if (! *bmc) { |
1502 | struct page *page; | |
6a07997f | 1503 | *bmc = 1 | (needed?NEEDED_MASK:0); |
193f1c93 N |
1504 | bitmap_count_page(bitmap, offset, 1); |
1505 | page = filemap_get_page(bitmap, offset >> CHUNK_BLOCK_SHIFT(bitmap)); | |
1506 | set_page_attr(bitmap, page, BITMAP_PAGE_CLEAN); | |
1507 | } | |
1508 | spin_unlock_irq(&bitmap->lock); | |
8311c29d | 1509 | bitmap->allclean = 0; |
32a7627c N |
1510 | } |
1511 | ||
9b1d1dac PC |
1512 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1513 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1514 | { | |
1515 | unsigned long chunk; | |
1516 | ||
1517 | for (chunk = s; chunk <= e; chunk++) { | |
db305e50 | 1518 | sector_t sec = (sector_t)chunk << CHUNK_BLOCK_SHIFT(bitmap); |
9b1d1dac PC |
1519 | bitmap_set_memory_bits(bitmap, sec, 1); |
1520 | bitmap_file_set_bit(bitmap, sec); | |
1521 | } | |
1522 | } | |
1523 | ||
6b8b3e8a N |
1524 | /* |
1525 | * flush out any pending updates | |
1526 | */ | |
1527 | void bitmap_flush(mddev_t *mddev) | |
1528 | { | |
1529 | struct bitmap *bitmap = mddev->bitmap; | |
1530 | int sleep; | |
1531 | ||
1532 | if (!bitmap) /* there was no bitmap */ | |
1533 | return; | |
1534 | ||
1535 | /* run the daemon_work three time to ensure everything is flushed | |
1536 | * that can be | |
1537 | */ | |
1538 | sleep = bitmap->daemon_sleep; | |
1539 | bitmap->daemon_sleep = 0; | |
1540 | bitmap_daemon_work(bitmap); | |
1541 | bitmap_daemon_work(bitmap); | |
1542 | bitmap_daemon_work(bitmap); | |
1543 | bitmap->daemon_sleep = sleep; | |
1544 | bitmap_update_sb(bitmap); | |
1545 | } | |
1546 | ||
32a7627c N |
1547 | /* |
1548 | * free memory that was allocated | |
1549 | */ | |
3178b0db | 1550 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1551 | { |
1552 | unsigned long k, pages; | |
1553 | struct bitmap_page *bp; | |
32a7627c N |
1554 | |
1555 | if (!bitmap) /* there was no bitmap */ | |
1556 | return; | |
1557 | ||
32a7627c N |
1558 | /* release the bitmap file and kill the daemon */ |
1559 | bitmap_file_put(bitmap); | |
1560 | ||
1561 | bp = bitmap->bp; | |
1562 | pages = bitmap->pages; | |
1563 | ||
1564 | /* free all allocated memory */ | |
1565 | ||
32a7627c N |
1566 | if (bp) /* deallocate the page memory */ |
1567 | for (k = 0; k < pages; k++) | |
1568 | if (bp[k].map && !bp[k].hijacked) | |
1569 | kfree(bp[k].map); | |
1570 | kfree(bp); | |
1571 | kfree(bitmap); | |
1572 | } | |
3178b0db N |
1573 | void bitmap_destroy(mddev_t *mddev) |
1574 | { | |
1575 | struct bitmap *bitmap = mddev->bitmap; | |
1576 | ||
1577 | if (!bitmap) /* there was no bitmap */ | |
1578 | return; | |
1579 | ||
1580 | mddev->bitmap = NULL; /* disconnect from the md device */ | |
b15c2e57 N |
1581 | if (mddev->thread) |
1582 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db N |
1583 | |
1584 | bitmap_free(bitmap); | |
1585 | } | |
32a7627c N |
1586 | |
1587 | /* | |
1588 | * initialize the bitmap structure | |
1589 | * if this returns an error, bitmap_destroy must be called to do clean up | |
1590 | */ | |
1591 | int bitmap_create(mddev_t *mddev) | |
1592 | { | |
1593 | struct bitmap *bitmap; | |
1f593903 | 1594 | sector_t blocks = mddev->resync_max_sectors; |
32a7627c N |
1595 | unsigned long chunks; |
1596 | unsigned long pages; | |
1597 | struct file *file = mddev->bitmap_file; | |
1598 | int err; | |
6a07997f | 1599 | sector_t start; |
32a7627c | 1600 | |
5f6e3c83 | 1601 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1602 | |
a654b9d8 | 1603 | if (!file && !mddev->bitmap_offset) /* bitmap disabled, nothing to do */ |
32a7627c N |
1604 | return 0; |
1605 | ||
a654b9d8 N |
1606 | BUG_ON(file && mddev->bitmap_offset); |
1607 | ||
9ffae0cf | 1608 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c N |
1609 | if (!bitmap) |
1610 | return -ENOMEM; | |
1611 | ||
32a7627c | 1612 | spin_lock_init(&bitmap->lock); |
ce25c31b N |
1613 | atomic_set(&bitmap->pending_writes, 0); |
1614 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1615 | init_waitqueue_head(&bitmap->overflow_wait); |
ce25c31b | 1616 | |
32a7627c | 1617 | bitmap->mddev = mddev; |
32a7627c | 1618 | |
32a7627c | 1619 | bitmap->file = file; |
a654b9d8 | 1620 | bitmap->offset = mddev->bitmap_offset; |
ce25c31b N |
1621 | if (file) { |
1622 | get_file(file); | |
ef51c976 MF |
1623 | do_sync_mapping_range(file->f_mapping, 0, LLONG_MAX, |
1624 | SYNC_FILE_RANGE_WAIT_BEFORE | | |
1625 | SYNC_FILE_RANGE_WRITE | | |
1626 | SYNC_FILE_RANGE_WAIT_AFTER); | |
ce25c31b | 1627 | } |
32a7627c N |
1628 | /* read superblock from bitmap file (this sets bitmap->chunksize) */ |
1629 | err = bitmap_read_sb(bitmap); | |
1630 | if (err) | |
3178b0db | 1631 | goto error; |
32a7627c | 1632 | |
a638b2dc | 1633 | bitmap->chunkshift = ffz(~bitmap->chunksize); |
32a7627c N |
1634 | |
1635 | /* now that chunksize and chunkshift are set, we can use these macros */ | |
1f593903 N |
1636 | chunks = (blocks + CHUNK_BLOCK_RATIO(bitmap) - 1) >> |
1637 | CHUNK_BLOCK_SHIFT(bitmap); | |
32a7627c N |
1638 | pages = (chunks + PAGE_COUNTER_RATIO - 1) / PAGE_COUNTER_RATIO; |
1639 | ||
1640 | BUG_ON(!pages); | |
1641 | ||
1642 | bitmap->chunks = chunks; | |
1643 | bitmap->pages = pages; | |
1644 | bitmap->missing_pages = pages; | |
1645 | bitmap->counter_bits = COUNTER_BITS; | |
1646 | ||
1647 | bitmap->syncchunk = ~0UL; | |
1648 | ||
44456d37 | 1649 | #ifdef INJECT_FATAL_FAULT_1 |
32a7627c N |
1650 | bitmap->bp = NULL; |
1651 | #else | |
9ffae0cf | 1652 | bitmap->bp = kzalloc(pages * sizeof(*bitmap->bp), GFP_KERNEL); |
32a7627c | 1653 | #endif |
3178b0db | 1654 | err = -ENOMEM; |
32a7627c | 1655 | if (!bitmap->bp) |
3178b0db | 1656 | goto error; |
32a7627c | 1657 | |
32a7627c N |
1658 | /* now that we have some pages available, initialize the in-memory |
1659 | * bitmap from the on-disk bitmap */ | |
6a07997f N |
1660 | start = 0; |
1661 | if (mddev->degraded == 0 | |
1662 | || bitmap->events_cleared == mddev->events) | |
1663 | /* no need to keep dirty bits to optimise a re-add of a missing device */ | |
1664 | start = mddev->recovery_cp; | |
1665 | err = bitmap_init_from_disk(bitmap, start); | |
193f1c93 | 1666 | |
32a7627c | 1667 | if (err) |
3178b0db | 1668 | goto error; |
32a7627c N |
1669 | |
1670 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", | |
1671 | pages, bmname(bitmap)); | |
1672 | ||
3178b0db N |
1673 | mddev->bitmap = bitmap; |
1674 | ||
b15c2e57 N |
1675 | mddev->thread->timeout = bitmap->daemon_sleep * HZ; |
1676 | ||
4ad13663 N |
1677 | bitmap_update_sb(bitmap); |
1678 | ||
1679 | return (bitmap->flags & BITMAP_WRITE_ERROR) ? -EIO : 0; | |
3178b0db N |
1680 | |
1681 | error: | |
1682 | bitmap_free(bitmap); | |
1683 | return err; | |
32a7627c N |
1684 | } |
1685 | ||
1686 | /* the bitmap API -- for raid personalities */ | |
1687 | EXPORT_SYMBOL(bitmap_startwrite); | |
1688 | EXPORT_SYMBOL(bitmap_endwrite); | |
1689 | EXPORT_SYMBOL(bitmap_start_sync); | |
1690 | EXPORT_SYMBOL(bitmap_end_sync); | |
1691 | EXPORT_SYMBOL(bitmap_unplug); | |
1692 | EXPORT_SYMBOL(bitmap_close_sync); | |
b47490c9 | 1693 | EXPORT_SYMBOL(bitmap_cond_end_sync); |