Root/
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 |
10 | */ |
11 | |
12 | /* |
13 | * Still to do: |
14 | * |
15 | * flush after percent set rather than just time based. (maybe both). |
16 | */ |
17 | |
18 | #include <linux/blkdev.h> |
19 | #include <linux/module.h> |
20 | #include <linux/errno.h> |
21 | #include <linux/slab.h> |
22 | #include <linux/init.h> |
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/seq_file.h> |
30 | #include "md.h" |
31 | #include "bitmap.h" |
32 | |
33 | static inline char *bmname(struct bitmap *bitmap) |
34 | { |
35 | return bitmap->mddev ? mdname(bitmap->mddev) : "mdX"; |
36 | } |
37 | |
38 | /* |
39 | * check a page and, if necessary, allocate it (or hijack it if the alloc fails) |
40 | * |
41 | * 1) check to see if this page is allocated, if it's not then try to alloc |
42 | * 2) if the alloc fails, set the page's hijacked flag so we'll use the |
43 | * page pointer directly as a counter |
44 | * |
45 | * if we find our page, we increment the page's refcount so that it stays |
46 | * allocated while we're using it |
47 | */ |
48 | static int bitmap_checkpage(struct bitmap_counts *bitmap, |
49 | unsigned long page, int create) |
50 | __releases(bitmap->lock) |
51 | __acquires(bitmap->lock) |
52 | { |
53 | unsigned char *mappage; |
54 | |
55 | if (page >= bitmap->pages) { |
56 | /* This can happen if bitmap_start_sync goes beyond |
57 | * End-of-device while looking for a whole page. |
58 | * It is harmless. |
59 | */ |
60 | return -EINVAL; |
61 | } |
62 | |
63 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ |
64 | return 0; |
65 | |
66 | if (bitmap->bp[page].map) /* page is already allocated, just return */ |
67 | return 0; |
68 | |
69 | if (!create) |
70 | return -ENOENT; |
71 | |
72 | /* this page has not been allocated yet */ |
73 | |
74 | spin_unlock_irq(&bitmap->lock); |
75 | mappage = kzalloc(PAGE_SIZE, GFP_NOIO); |
76 | spin_lock_irq(&bitmap->lock); |
77 | |
78 | if (mappage == NULL) { |
79 | pr_debug("md/bitmap: map page allocation failed, hijacking\n"); |
80 | /* failed - set the hijacked flag so that we can use the |
81 | * pointer as a counter */ |
82 | if (!bitmap->bp[page].map) |
83 | bitmap->bp[page].hijacked = 1; |
84 | } else if (bitmap->bp[page].map || |
85 | bitmap->bp[page].hijacked) { |
86 | /* somebody beat us to getting the page */ |
87 | kfree(mappage); |
88 | return 0; |
89 | } else { |
90 | |
91 | /* no page was in place and we have one, so install it */ |
92 | |
93 | bitmap->bp[page].map = mappage; |
94 | bitmap->missing_pages--; |
95 | } |
96 | return 0; |
97 | } |
98 | |
99 | /* if page is completely empty, put it back on the free list, or dealloc it */ |
100 | /* if page was hijacked, unmark the flag so it might get alloced next time */ |
101 | /* Note: lock should be held when calling this */ |
102 | static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page) |
103 | { |
104 | char *ptr; |
105 | |
106 | if (bitmap->bp[page].count) /* page is still busy */ |
107 | return; |
108 | |
109 | /* page is no longer in use, it can be released */ |
110 | |
111 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ |
112 | bitmap->bp[page].hijacked = 0; |
113 | bitmap->bp[page].map = NULL; |
114 | } else { |
115 | /* normal case, free the page */ |
116 | ptr = bitmap->bp[page].map; |
117 | bitmap->bp[page].map = NULL; |
118 | bitmap->missing_pages++; |
119 | kfree(ptr); |
120 | } |
121 | } |
122 | |
123 | /* |
124 | * bitmap file handling - read and write the bitmap file and its superblock |
125 | */ |
126 | |
127 | /* |
128 | * basic page I/O operations |
129 | */ |
130 | |
131 | /* IO operations when bitmap is stored near all superblocks */ |
132 | static int read_sb_page(struct mddev *mddev, loff_t offset, |
133 | struct page *page, |
134 | unsigned long index, int size) |
135 | { |
136 | /* choose a good rdev and read the page from there */ |
137 | |
138 | struct md_rdev *rdev; |
139 | sector_t target; |
140 | |
141 | rdev_for_each(rdev, mddev) { |
142 | if (! test_bit(In_sync, &rdev->flags) |
143 | || test_bit(Faulty, &rdev->flags)) |
144 | continue; |
145 | |
146 | target = offset + index * (PAGE_SIZE/512); |
147 | |
148 | if (sync_page_io(rdev, target, |
149 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
150 | page, READ, true)) { |
151 | page->index = index; |
152 | return 0; |
153 | } |
154 | } |
155 | return -EIO; |
156 | } |
157 | |
158 | static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) |
159 | { |
160 | /* Iterate the disks of an mddev, using rcu to protect access to the |
161 | * linked list, and raising the refcount of devices we return to ensure |
162 | * they don't disappear while in use. |
163 | * As devices are only added or removed when raid_disk is < 0 and |
164 | * nr_pending is 0 and In_sync is clear, the entries we return will |
165 | * still be in the same position on the list when we re-enter |
166 | * list_for_each_entry_continue_rcu. |
167 | */ |
168 | rcu_read_lock(); |
169 | if (rdev == NULL) |
170 | /* start at the beginning */ |
171 | rdev = list_entry_rcu(&mddev->disks, struct md_rdev, same_set); |
172 | else { |
173 | /* release the previous rdev and start from there. */ |
174 | rdev_dec_pending(rdev, mddev); |
175 | } |
176 | list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) { |
177 | if (rdev->raid_disk >= 0 && |
178 | !test_bit(Faulty, &rdev->flags)) { |
179 | /* this is a usable devices */ |
180 | atomic_inc(&rdev->nr_pending); |
181 | rcu_read_unlock(); |
182 | return rdev; |
183 | } |
184 | } |
185 | rcu_read_unlock(); |
186 | return NULL; |
187 | } |
188 | |
189 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
190 | { |
191 | struct md_rdev *rdev = NULL; |
192 | struct block_device *bdev; |
193 | struct mddev *mddev = bitmap->mddev; |
194 | struct bitmap_storage *store = &bitmap->storage; |
195 | |
196 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
197 | int size = PAGE_SIZE; |
198 | loff_t offset = mddev->bitmap_info.offset; |
199 | |
200 | bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; |
201 | |
202 | if (page->index == store->file_pages-1) { |
203 | int last_page_size = store->bytes & (PAGE_SIZE-1); |
204 | if (last_page_size == 0) |
205 | last_page_size = PAGE_SIZE; |
206 | size = roundup(last_page_size, |
207 | bdev_logical_block_size(bdev)); |
208 | } |
209 | /* Just make sure we aren't corrupting data or |
210 | * metadata |
211 | */ |
212 | if (mddev->external) { |
213 | /* Bitmap could be anywhere. */ |
214 | if (rdev->sb_start + offset + (page->index |
215 | * (PAGE_SIZE/512)) |
216 | > rdev->data_offset |
217 | && |
218 | rdev->sb_start + offset |
219 | < (rdev->data_offset + mddev->dev_sectors |
220 | + (PAGE_SIZE/512))) |
221 | goto bad_alignment; |
222 | } else if (offset < 0) { |
223 | /* DATA BITMAP METADATA */ |
224 | if (offset |
225 | + (long)(page->index * (PAGE_SIZE/512)) |
226 | + size/512 > 0) |
227 | /* bitmap runs in to metadata */ |
228 | goto bad_alignment; |
229 | if (rdev->data_offset + mddev->dev_sectors |
230 | > rdev->sb_start + offset) |
231 | /* data runs in to bitmap */ |
232 | goto bad_alignment; |
233 | } else if (rdev->sb_start < rdev->data_offset) { |
234 | /* METADATA BITMAP DATA */ |
235 | if (rdev->sb_start |
236 | + offset |
237 | + page->index*(PAGE_SIZE/512) + size/512 |
238 | > rdev->data_offset) |
239 | /* bitmap runs in to data */ |
240 | goto bad_alignment; |
241 | } else { |
242 | /* DATA METADATA BITMAP - no problems */ |
243 | } |
244 | md_super_write(mddev, rdev, |
245 | rdev->sb_start + offset |
246 | + page->index * (PAGE_SIZE/512), |
247 | size, |
248 | page); |
249 | } |
250 | |
251 | if (wait) |
252 | md_super_wait(mddev); |
253 | return 0; |
254 | |
255 | bad_alignment: |
256 | return -EINVAL; |
257 | } |
258 | |
259 | static void bitmap_file_kick(struct bitmap *bitmap); |
260 | /* |
261 | * write out a page to a file |
262 | */ |
263 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
264 | { |
265 | struct buffer_head *bh; |
266 | |
267 | if (bitmap->storage.file == NULL) { |
268 | switch (write_sb_page(bitmap, page, wait)) { |
269 | case -EINVAL: |
270 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); |
271 | } |
272 | } else { |
273 | |
274 | bh = page_buffers(page); |
275 | |
276 | while (bh && bh->b_blocknr) { |
277 | atomic_inc(&bitmap->pending_writes); |
278 | set_buffer_locked(bh); |
279 | set_buffer_mapped(bh); |
280 | submit_bh(WRITE | REQ_SYNC, bh); |
281 | bh = bh->b_this_page; |
282 | } |
283 | |
284 | if (wait) |
285 | wait_event(bitmap->write_wait, |
286 | atomic_read(&bitmap->pending_writes)==0); |
287 | } |
288 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
289 | bitmap_file_kick(bitmap); |
290 | } |
291 | |
292 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) |
293 | { |
294 | struct bitmap *bitmap = bh->b_private; |
295 | |
296 | if (!uptodate) |
297 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); |
298 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
299 | wake_up(&bitmap->write_wait); |
300 | } |
301 | |
302 | /* copied from buffer.c */ |
303 | static void |
304 | __clear_page_buffers(struct page *page) |
305 | { |
306 | ClearPagePrivate(page); |
307 | set_page_private(page, 0); |
308 | page_cache_release(page); |
309 | } |
310 | static void free_buffers(struct page *page) |
311 | { |
312 | struct buffer_head *bh; |
313 | |
314 | if (!PagePrivate(page)) |
315 | return; |
316 | |
317 | bh = page_buffers(page); |
318 | while (bh) { |
319 | struct buffer_head *next = bh->b_this_page; |
320 | free_buffer_head(bh); |
321 | bh = next; |
322 | } |
323 | __clear_page_buffers(page); |
324 | put_page(page); |
325 | } |
326 | |
327 | /* read a page from a file. |
328 | * We both read the page, and attach buffers to the page to record the |
329 | * address of each block (using bmap). These addresses will be used |
330 | * to write the block later, completely bypassing the filesystem. |
331 | * This usage is similar to how swap files are handled, and allows us |
332 | * to write to a file with no concerns of memory allocation failing. |
333 | */ |
334 | static int read_page(struct file *file, unsigned long index, |
335 | struct bitmap *bitmap, |
336 | unsigned long count, |
337 | struct page *page) |
338 | { |
339 | int ret = 0; |
340 | struct inode *inode = file_inode(file); |
341 | struct buffer_head *bh; |
342 | sector_t block; |
343 | |
344 | pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
345 | (unsigned long long)index << PAGE_SHIFT); |
346 | |
347 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
348 | if (!bh) { |
349 | ret = -ENOMEM; |
350 | goto out; |
351 | } |
352 | attach_page_buffers(page, bh); |
353 | block = index << (PAGE_SHIFT - inode->i_blkbits); |
354 | while (bh) { |
355 | if (count == 0) |
356 | bh->b_blocknr = 0; |
357 | else { |
358 | bh->b_blocknr = bmap(inode, block); |
359 | if (bh->b_blocknr == 0) { |
360 | /* Cannot use this file! */ |
361 | ret = -EINVAL; |
362 | goto out; |
363 | } |
364 | bh->b_bdev = inode->i_sb->s_bdev; |
365 | if (count < (1<<inode->i_blkbits)) |
366 | count = 0; |
367 | else |
368 | count -= (1<<inode->i_blkbits); |
369 | |
370 | bh->b_end_io = end_bitmap_write; |
371 | bh->b_private = bitmap; |
372 | atomic_inc(&bitmap->pending_writes); |
373 | set_buffer_locked(bh); |
374 | set_buffer_mapped(bh); |
375 | submit_bh(READ, bh); |
376 | } |
377 | block++; |
378 | bh = bh->b_this_page; |
379 | } |
380 | page->index = index; |
381 | |
382 | wait_event(bitmap->write_wait, |
383 | atomic_read(&bitmap->pending_writes)==0); |
384 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
385 | ret = -EIO; |
386 | out: |
387 | if (ret) |
388 | printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n", |
389 | (int)PAGE_SIZE, |
390 | (unsigned long long)index << PAGE_SHIFT, |
391 | ret); |
392 | return ret; |
393 | } |
394 | |
395 | /* |
396 | * bitmap file superblock operations |
397 | */ |
398 | |
399 | /* update the event counter and sync the superblock to disk */ |
400 | void bitmap_update_sb(struct bitmap *bitmap) |
401 | { |
402 | bitmap_super_t *sb; |
403 | |
404 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ |
405 | return; |
406 | if (bitmap->mddev->bitmap_info.external) |
407 | return; |
408 | if (!bitmap->storage.sb_page) /* no superblock */ |
409 | return; |
410 | sb = kmap_atomic(bitmap->storage.sb_page); |
411 | sb->events = cpu_to_le64(bitmap->mddev->events); |
412 | if (bitmap->mddev->events < bitmap->events_cleared) |
413 | /* rocking back to read-only */ |
414 | bitmap->events_cleared = bitmap->mddev->events; |
415 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); |
416 | sb->state = cpu_to_le32(bitmap->flags); |
417 | /* Just in case these have been changed via sysfs: */ |
418 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); |
419 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); |
420 | /* This might have been changed by a reshape */ |
421 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); |
422 | sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize); |
423 | sb->sectors_reserved = cpu_to_le32(bitmap->mddev-> |
424 | bitmap_info.space); |
425 | kunmap_atomic(sb); |
426 | write_page(bitmap, bitmap->storage.sb_page, 1); |
427 | } |
428 | |
429 | /* print out the bitmap file superblock */ |
430 | void bitmap_print_sb(struct bitmap *bitmap) |
431 | { |
432 | bitmap_super_t *sb; |
433 | |
434 | if (!bitmap || !bitmap->storage.sb_page) |
435 | return; |
436 | sb = kmap_atomic(bitmap->storage.sb_page); |
437 | printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap)); |
438 | printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic)); |
439 | printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version)); |
440 | printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n", |
441 | *(__u32 *)(sb->uuid+0), |
442 | *(__u32 *)(sb->uuid+4), |
443 | *(__u32 *)(sb->uuid+8), |
444 | *(__u32 *)(sb->uuid+12)); |
445 | printk(KERN_DEBUG " events: %llu\n", |
446 | (unsigned long long) le64_to_cpu(sb->events)); |
447 | printk(KERN_DEBUG "events cleared: %llu\n", |
448 | (unsigned long long) le64_to_cpu(sb->events_cleared)); |
449 | printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state)); |
450 | printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize)); |
451 | printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); |
452 | printk(KERN_DEBUG " sync size: %llu KB\n", |
453 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); |
454 | printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind)); |
455 | kunmap_atomic(sb); |
456 | } |
457 | |
458 | /* |
459 | * bitmap_new_disk_sb |
460 | * @bitmap |
461 | * |
462 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb |
463 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. |
464 | * This function verifies 'bitmap_info' and populates the on-disk bitmap |
465 | * structure, which is to be written to disk. |
466 | * |
467 | * Returns: 0 on success, -Exxx on error |
468 | */ |
469 | static int bitmap_new_disk_sb(struct bitmap *bitmap) |
470 | { |
471 | bitmap_super_t *sb; |
472 | unsigned long chunksize, daemon_sleep, write_behind; |
473 | |
474 | bitmap->storage.sb_page = alloc_page(GFP_KERNEL); |
475 | if (bitmap->storage.sb_page == NULL) |
476 | return -ENOMEM; |
477 | bitmap->storage.sb_page->index = 0; |
478 | |
479 | sb = kmap_atomic(bitmap->storage.sb_page); |
480 | |
481 | sb->magic = cpu_to_le32(BITMAP_MAGIC); |
482 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); |
483 | |
484 | chunksize = bitmap->mddev->bitmap_info.chunksize; |
485 | BUG_ON(!chunksize); |
486 | if (!is_power_of_2(chunksize)) { |
487 | kunmap_atomic(sb); |
488 | printk(KERN_ERR "bitmap chunksize not a power of 2\n"); |
489 | return -EINVAL; |
490 | } |
491 | sb->chunksize = cpu_to_le32(chunksize); |
492 | |
493 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; |
494 | if (!daemon_sleep || |
495 | (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { |
496 | printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n"); |
497 | daemon_sleep = 5 * HZ; |
498 | } |
499 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); |
500 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; |
501 | |
502 | /* |
503 | * FIXME: write_behind for RAID1. If not specified, what |
504 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. |
505 | */ |
506 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; |
507 | if (write_behind > COUNTER_MAX) |
508 | write_behind = COUNTER_MAX / 2; |
509 | sb->write_behind = cpu_to_le32(write_behind); |
510 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; |
511 | |
512 | /* keep the array size field of the bitmap superblock up to date */ |
513 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); |
514 | |
515 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); |
516 | |
517 | set_bit(BITMAP_STALE, &bitmap->flags); |
518 | sb->state = cpu_to_le32(bitmap->flags); |
519 | bitmap->events_cleared = bitmap->mddev->events; |
520 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); |
521 | |
522 | kunmap_atomic(sb); |
523 | |
524 | return 0; |
525 | } |
526 | |
527 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
528 | static int bitmap_read_sb(struct bitmap *bitmap) |
529 | { |
530 | char *reason = NULL; |
531 | bitmap_super_t *sb; |
532 | unsigned long chunksize, daemon_sleep, write_behind; |
533 | unsigned long long events; |
534 | unsigned long sectors_reserved = 0; |
535 | int err = -EINVAL; |
536 | struct page *sb_page; |
537 | |
538 | if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { |
539 | chunksize = 128 * 1024 * 1024; |
540 | daemon_sleep = 5 * HZ; |
541 | write_behind = 0; |
542 | set_bit(BITMAP_STALE, &bitmap->flags); |
543 | err = 0; |
544 | goto out_no_sb; |
545 | } |
546 | /* page 0 is the superblock, read it... */ |
547 | sb_page = alloc_page(GFP_KERNEL); |
548 | if (!sb_page) |
549 | return -ENOMEM; |
550 | bitmap->storage.sb_page = sb_page; |
551 | |
552 | if (bitmap->storage.file) { |
553 | loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); |
554 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; |
555 | |
556 | err = read_page(bitmap->storage.file, 0, |
557 | bitmap, bytes, sb_page); |
558 | } else { |
559 | err = read_sb_page(bitmap->mddev, |
560 | bitmap->mddev->bitmap_info.offset, |
561 | sb_page, |
562 | 0, sizeof(bitmap_super_t)); |
563 | } |
564 | if (err) |
565 | return err; |
566 | |
567 | sb = kmap_atomic(sb_page); |
568 | |
569 | chunksize = le32_to_cpu(sb->chunksize); |
570 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
571 | write_behind = le32_to_cpu(sb->write_behind); |
572 | sectors_reserved = le32_to_cpu(sb->sectors_reserved); |
573 | |
574 | /* verify that the bitmap-specific fields are valid */ |
575 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) |
576 | reason = "bad magic"; |
577 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
578 | le32_to_cpu(sb->version) > BITMAP_MAJOR_HI) |
579 | reason = "unrecognized superblock version"; |
580 | else if (chunksize < 512) |
581 | reason = "bitmap chunksize too small"; |
582 | else if (!is_power_of_2(chunksize)) |
583 | reason = "bitmap chunksize not a power of 2"; |
584 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
585 | reason = "daemon sleep period out of range"; |
586 | else if (write_behind > COUNTER_MAX) |
587 | reason = "write-behind limit out of range (0 - 16383)"; |
588 | if (reason) { |
589 | printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n", |
590 | bmname(bitmap), reason); |
591 | goto out; |
592 | } |
593 | |
594 | /* keep the array size field of the bitmap superblock up to date */ |
595 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); |
596 | |
597 | if (bitmap->mddev->persistent) { |
598 | /* |
599 | * We have a persistent array superblock, so compare the |
600 | * bitmap's UUID and event counter to the mddev's |
601 | */ |
602 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { |
603 | printk(KERN_INFO |
604 | "%s: bitmap superblock UUID mismatch\n", |
605 | bmname(bitmap)); |
606 | goto out; |
607 | } |
608 | events = le64_to_cpu(sb->events); |
609 | if (events < bitmap->mddev->events) { |
610 | printk(KERN_INFO |
611 | "%s: bitmap file is out of date (%llu < %llu) " |
612 | "-- forcing full recovery\n", |
613 | bmname(bitmap), events, |
614 | (unsigned long long) bitmap->mddev->events); |
615 | set_bit(BITMAP_STALE, &bitmap->flags); |
616 | } |
617 | } |
618 | |
619 | /* assign fields using values from superblock */ |
620 | bitmap->flags |= le32_to_cpu(sb->state); |
621 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
622 | set_bit(BITMAP_HOSTENDIAN, &bitmap->flags); |
623 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
624 | err = 0; |
625 | out: |
626 | kunmap_atomic(sb); |
627 | out_no_sb: |
628 | if (test_bit(BITMAP_STALE, &bitmap->flags)) |
629 | bitmap->events_cleared = bitmap->mddev->events; |
630 | bitmap->mddev->bitmap_info.chunksize = chunksize; |
631 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; |
632 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; |
633 | if (bitmap->mddev->bitmap_info.space == 0 || |
634 | bitmap->mddev->bitmap_info.space > sectors_reserved) |
635 | bitmap->mddev->bitmap_info.space = sectors_reserved; |
636 | if (err) |
637 | bitmap_print_sb(bitmap); |
638 | return err; |
639 | } |
640 | |
641 | /* |
642 | * general bitmap file operations |
643 | */ |
644 | |
645 | /* |
646 | * on-disk bitmap: |
647 | * |
648 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap |
649 | * file a page at a time. There's a superblock at the start of the file. |
650 | */ |
651 | /* calculate the index of the page that contains this bit */ |
652 | static inline unsigned long file_page_index(struct bitmap_storage *store, |
653 | unsigned long chunk) |
654 | { |
655 | if (store->sb_page) |
656 | chunk += sizeof(bitmap_super_t) << 3; |
657 | return chunk >> PAGE_BIT_SHIFT; |
658 | } |
659 | |
660 | /* calculate the (bit) offset of this bit within a page */ |
661 | static inline unsigned long file_page_offset(struct bitmap_storage *store, |
662 | unsigned long chunk) |
663 | { |
664 | if (store->sb_page) |
665 | chunk += sizeof(bitmap_super_t) << 3; |
666 | return chunk & (PAGE_BITS - 1); |
667 | } |
668 | |
669 | /* |
670 | * return a pointer to the page in the filemap that contains the given bit |
671 | * |
672 | * this lookup is complicated by the fact that the bitmap sb might be exactly |
673 | * 1 page (e.g., x86) or less than 1 page -- so the bitmap might start on page |
674 | * 0 or page 1 |
675 | */ |
676 | static inline struct page *filemap_get_page(struct bitmap_storage *store, |
677 | unsigned long chunk) |
678 | { |
679 | if (file_page_index(store, chunk) >= store->file_pages) |
680 | return NULL; |
681 | return store->filemap[file_page_index(store, chunk) |
682 | - file_page_index(store, 0)]; |
683 | } |
684 | |
685 | static int bitmap_storage_alloc(struct bitmap_storage *store, |
686 | unsigned long chunks, int with_super) |
687 | { |
688 | int pnum; |
689 | unsigned long num_pages; |
690 | unsigned long bytes; |
691 | |
692 | bytes = DIV_ROUND_UP(chunks, 8); |
693 | if (with_super) |
694 | bytes += sizeof(bitmap_super_t); |
695 | |
696 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); |
697 | |
698 | store->filemap = kmalloc(sizeof(struct page *) |
699 | * num_pages, GFP_KERNEL); |
700 | if (!store->filemap) |
701 | return -ENOMEM; |
702 | |
703 | if (with_super && !store->sb_page) { |
704 | store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO); |
705 | if (store->sb_page == NULL) |
706 | return -ENOMEM; |
707 | store->sb_page->index = 0; |
708 | } |
709 | pnum = 0; |
710 | if (store->sb_page) { |
711 | store->filemap[0] = store->sb_page; |
712 | pnum = 1; |
713 | } |
714 | for ( ; pnum < num_pages; pnum++) { |
715 | store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
716 | if (!store->filemap[pnum]) { |
717 | store->file_pages = pnum; |
718 | return -ENOMEM; |
719 | } |
720 | store->filemap[pnum]->index = pnum; |
721 | } |
722 | store->file_pages = pnum; |
723 | |
724 | /* We need 4 bits per page, rounded up to a multiple |
725 | * of sizeof(unsigned long) */ |
726 | store->filemap_attr = kzalloc( |
727 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), |
728 | GFP_KERNEL); |
729 | if (!store->filemap_attr) |
730 | return -ENOMEM; |
731 | |
732 | store->bytes = bytes; |
733 | |
734 | return 0; |
735 | } |
736 | |
737 | static void bitmap_file_unmap(struct bitmap_storage *store) |
738 | { |
739 | struct page **map, *sb_page; |
740 | int pages; |
741 | struct file *file; |
742 | |
743 | file = store->file; |
744 | map = store->filemap; |
745 | pages = store->file_pages; |
746 | sb_page = store->sb_page; |
747 | |
748 | while (pages--) |
749 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
750 | free_buffers(map[pages]); |
751 | kfree(map); |
752 | kfree(store->filemap_attr); |
753 | |
754 | if (sb_page) |
755 | free_buffers(sb_page); |
756 | |
757 | if (file) { |
758 | struct inode *inode = file_inode(file); |
759 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
760 | fput(file); |
761 | } |
762 | } |
763 | |
764 | /* |
765 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file |
766 | * then it is no longer reliable, so we stop using it and we mark the file |
767 | * as failed in the superblock |
768 | */ |
769 | static void bitmap_file_kick(struct bitmap *bitmap) |
770 | { |
771 | char *path, *ptr = NULL; |
772 | |
773 | if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) { |
774 | bitmap_update_sb(bitmap); |
775 | |
776 | if (bitmap->storage.file) { |
777 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); |
778 | if (path) |
779 | ptr = d_path(&bitmap->storage.file->f_path, |
780 | path, PAGE_SIZE); |
781 | |
782 | printk(KERN_ALERT |
783 | "%s: kicking failed bitmap file %s from array!\n", |
784 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); |
785 | |
786 | kfree(path); |
787 | } else |
788 | printk(KERN_ALERT |
789 | "%s: disabling internal bitmap due to errors\n", |
790 | bmname(bitmap)); |
791 | } |
792 | } |
793 | |
794 | enum bitmap_page_attr { |
795 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
796 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
797 | * i.e. counter is 1 or 2. */ |
798 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
799 | }; |
800 | |
801 | static inline void set_page_attr(struct bitmap *bitmap, int pnum, |
802 | enum bitmap_page_attr attr) |
803 | { |
804 | set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
805 | } |
806 | |
807 | static inline void clear_page_attr(struct bitmap *bitmap, int pnum, |
808 | enum bitmap_page_attr attr) |
809 | { |
810 | clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
811 | } |
812 | |
813 | static inline int test_page_attr(struct bitmap *bitmap, int pnum, |
814 | enum bitmap_page_attr attr) |
815 | { |
816 | return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
817 | } |
818 | |
819 | static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum, |
820 | enum bitmap_page_attr attr) |
821 | { |
822 | return test_and_clear_bit((pnum<<2) + attr, |
823 | bitmap->storage.filemap_attr); |
824 | } |
825 | /* |
826 | * bitmap_file_set_bit -- called before performing a write to the md device |
827 | * to set (and eventually sync) a particular bit in the bitmap file |
828 | * |
829 | * we set the bit immediately, then we record the page number so that |
830 | * when an unplug occurs, we can flush the dirty pages out to disk |
831 | */ |
832 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) |
833 | { |
834 | unsigned long bit; |
835 | struct page *page; |
836 | void *kaddr; |
837 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
838 | |
839 | page = filemap_get_page(&bitmap->storage, chunk); |
840 | if (!page) |
841 | return; |
842 | bit = file_page_offset(&bitmap->storage, chunk); |
843 | |
844 | /* set the bit */ |
845 | kaddr = kmap_atomic(page); |
846 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
847 | set_bit(bit, kaddr); |
848 | else |
849 | test_and_set_bit_le(bit, kaddr); |
850 | kunmap_atomic(kaddr); |
851 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
852 | /* record page number so it gets flushed to disk when unplug occurs */ |
853 | set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY); |
854 | } |
855 | |
856 | static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) |
857 | { |
858 | unsigned long bit; |
859 | struct page *page; |
860 | void *paddr; |
861 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
862 | |
863 | page = filemap_get_page(&bitmap->storage, chunk); |
864 | if (!page) |
865 | return; |
866 | bit = file_page_offset(&bitmap->storage, chunk); |
867 | paddr = kmap_atomic(page); |
868 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
869 | clear_bit(bit, paddr); |
870 | else |
871 | test_and_clear_bit_le(bit, paddr); |
872 | kunmap_atomic(paddr); |
873 | if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) { |
874 | set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING); |
875 | bitmap->allclean = 0; |
876 | } |
877 | } |
878 | |
879 | /* this gets called when the md device is ready to unplug its underlying |
880 | * (slave) device queues -- before we let any writes go down, we need to |
881 | * sync the dirty pages of the bitmap file to disk */ |
882 | void bitmap_unplug(struct bitmap *bitmap) |
883 | { |
884 | unsigned long i; |
885 | int dirty, need_write; |
886 | int wait = 0; |
887 | |
888 | if (!bitmap || !bitmap->storage.filemap || |
889 | test_bit(BITMAP_STALE, &bitmap->flags)) |
890 | return; |
891 | |
892 | /* look at each page to see if there are any set bits that need to be |
893 | * flushed out to disk */ |
894 | for (i = 0; i < bitmap->storage.file_pages; i++) { |
895 | if (!bitmap->storage.filemap) |
896 | return; |
897 | dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
898 | need_write = test_and_clear_page_attr(bitmap, i, |
899 | BITMAP_PAGE_NEEDWRITE); |
900 | if (dirty || need_write) { |
901 | clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); |
902 | write_page(bitmap, bitmap->storage.filemap[i], 0); |
903 | } |
904 | if (dirty) |
905 | wait = 1; |
906 | } |
907 | if (wait) { /* if any writes were performed, we need to wait on them */ |
908 | if (bitmap->storage.file) |
909 | wait_event(bitmap->write_wait, |
910 | atomic_read(&bitmap->pending_writes)==0); |
911 | else |
912 | md_super_wait(bitmap->mddev); |
913 | } |
914 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
915 | bitmap_file_kick(bitmap); |
916 | } |
917 | EXPORT_SYMBOL(bitmap_unplug); |
918 | |
919 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
920 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
921 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the |
922 | * memory mapping of the bitmap file |
923 | * Special cases: |
924 | * if there's no bitmap file, or if the bitmap file had been |
925 | * previously kicked from the array, we mark all the bits as |
926 | * 1's in order to cause a full resync. |
927 | * |
928 | * We ignore all bits for sectors that end earlier than 'start'. |
929 | * This is used when reading an out-of-date bitmap... |
930 | */ |
931 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
932 | { |
933 | unsigned long i, chunks, index, oldindex, bit; |
934 | struct page *page = NULL; |
935 | unsigned long bit_cnt = 0; |
936 | struct file *file; |
937 | unsigned long offset; |
938 | int outofdate; |
939 | int ret = -ENOSPC; |
940 | void *paddr; |
941 | struct bitmap_storage *store = &bitmap->storage; |
942 | |
943 | chunks = bitmap->counts.chunks; |
944 | file = store->file; |
945 | |
946 | if (!file && !bitmap->mddev->bitmap_info.offset) { |
947 | /* No permanent bitmap - fill with '1s'. */ |
948 | store->filemap = NULL; |
949 | store->file_pages = 0; |
950 | for (i = 0; i < chunks ; i++) { |
951 | /* if the disk bit is set, set the memory bit */ |
952 | int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift) |
953 | >= start); |
954 | bitmap_set_memory_bits(bitmap, |
955 | (sector_t)i << bitmap->counts.chunkshift, |
956 | needed); |
957 | } |
958 | return 0; |
959 | } |
960 | |
961 | outofdate = test_bit(BITMAP_STALE, &bitmap->flags); |
962 | if (outofdate) |
963 | printk(KERN_INFO "%s: bitmap file is out of date, doing full " |
964 | "recovery\n", bmname(bitmap)); |
965 | |
966 | if (file && i_size_read(file->f_mapping->host) < store->bytes) { |
967 | printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n", |
968 | bmname(bitmap), |
969 | (unsigned long) i_size_read(file->f_mapping->host), |
970 | store->bytes); |
971 | goto err; |
972 | } |
973 | |
974 | oldindex = ~0L; |
975 | offset = 0; |
976 | if (!bitmap->mddev->bitmap_info.external) |
977 | offset = sizeof(bitmap_super_t); |
978 | |
979 | for (i = 0; i < chunks; i++) { |
980 | int b; |
981 | index = file_page_index(&bitmap->storage, i); |
982 | bit = file_page_offset(&bitmap->storage, i); |
983 | if (index != oldindex) { /* this is a new page, read it in */ |
984 | int count; |
985 | /* unmap the old page, we're done with it */ |
986 | if (index == store->file_pages-1) |
987 | count = store->bytes - index * PAGE_SIZE; |
988 | else |
989 | count = PAGE_SIZE; |
990 | page = store->filemap[index]; |
991 | if (file) |
992 | ret = read_page(file, index, bitmap, |
993 | count, page); |
994 | else |
995 | ret = read_sb_page( |
996 | bitmap->mddev, |
997 | bitmap->mddev->bitmap_info.offset, |
998 | page, |
999 | index, count); |
1000 | |
1001 | if (ret) |
1002 | goto err; |
1003 | |
1004 | oldindex = index; |
1005 | |
1006 | if (outofdate) { |
1007 | /* |
1008 | * if bitmap is out of date, dirty the |
1009 | * whole page and write it out |
1010 | */ |
1011 | paddr = kmap_atomic(page); |
1012 | memset(paddr + offset, 0xff, |
1013 | PAGE_SIZE - offset); |
1014 | kunmap_atomic(paddr); |
1015 | write_page(bitmap, page, 1); |
1016 | |
1017 | ret = -EIO; |
1018 | if (test_bit(BITMAP_WRITE_ERROR, |
1019 | &bitmap->flags)) |
1020 | goto err; |
1021 | } |
1022 | } |
1023 | paddr = kmap_atomic(page); |
1024 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
1025 | b = test_bit(bit, paddr); |
1026 | else |
1027 | b = test_bit_le(bit, paddr); |
1028 | kunmap_atomic(paddr); |
1029 | if (b) { |
1030 | /* if the disk bit is set, set the memory bit */ |
1031 | int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift |
1032 | >= start); |
1033 | bitmap_set_memory_bits(bitmap, |
1034 | (sector_t)i << bitmap->counts.chunkshift, |
1035 | needed); |
1036 | bit_cnt++; |
1037 | } |
1038 | offset = 0; |
1039 | } |
1040 | |
1041 | printk(KERN_INFO "%s: bitmap initialized from disk: " |
1042 | "read %lu pages, set %lu of %lu bits\n", |
1043 | bmname(bitmap), store->file_pages, |
1044 | bit_cnt, chunks); |
1045 | |
1046 | return 0; |
1047 | |
1048 | err: |
1049 | printk(KERN_INFO "%s: bitmap initialisation failed: %d\n", |
1050 | bmname(bitmap), ret); |
1051 | return ret; |
1052 | } |
1053 | |
1054 | void bitmap_write_all(struct bitmap *bitmap) |
1055 | { |
1056 | /* We don't actually write all bitmap blocks here, |
1057 | * just flag them as needing to be written |
1058 | */ |
1059 | int i; |
1060 | |
1061 | if (!bitmap || !bitmap->storage.filemap) |
1062 | return; |
1063 | if (bitmap->storage.file) |
1064 | /* Only one copy, so nothing needed */ |
1065 | return; |
1066 | |
1067 | for (i = 0; i < bitmap->storage.file_pages; i++) |
1068 | set_page_attr(bitmap, i, |
1069 | BITMAP_PAGE_NEEDWRITE); |
1070 | bitmap->allclean = 0; |
1071 | } |
1072 | |
1073 | static void bitmap_count_page(struct bitmap_counts *bitmap, |
1074 | sector_t offset, int inc) |
1075 | { |
1076 | sector_t chunk = offset >> bitmap->chunkshift; |
1077 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1078 | bitmap->bp[page].count += inc; |
1079 | bitmap_checkfree(bitmap, page); |
1080 | } |
1081 | |
1082 | static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset) |
1083 | { |
1084 | sector_t chunk = offset >> bitmap->chunkshift; |
1085 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1086 | struct bitmap_page *bp = &bitmap->bp[page]; |
1087 | |
1088 | if (!bp->pending) |
1089 | bp->pending = 1; |
1090 | } |
1091 | |
1092 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
1093 | sector_t offset, sector_t *blocks, |
1094 | int create); |
1095 | |
1096 | /* |
1097 | * bitmap daemon -- periodically wakes up to clean bits and flush pages |
1098 | * out to disk |
1099 | */ |
1100 | |
1101 | void bitmap_daemon_work(struct mddev *mddev) |
1102 | { |
1103 | struct bitmap *bitmap; |
1104 | unsigned long j; |
1105 | unsigned long nextpage; |
1106 | sector_t blocks; |
1107 | struct bitmap_counts *counts; |
1108 | |
1109 | /* Use a mutex to guard daemon_work against |
1110 | * bitmap_destroy. |
1111 | */ |
1112 | mutex_lock(&mddev->bitmap_info.mutex); |
1113 | bitmap = mddev->bitmap; |
1114 | if (bitmap == NULL) { |
1115 | mutex_unlock(&mddev->bitmap_info.mutex); |
1116 | return; |
1117 | } |
1118 | if (time_before(jiffies, bitmap->daemon_lastrun |
1119 | + mddev->bitmap_info.daemon_sleep)) |
1120 | goto done; |
1121 | |
1122 | bitmap->daemon_lastrun = jiffies; |
1123 | if (bitmap->allclean) { |
1124 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
1125 | goto done; |
1126 | } |
1127 | bitmap->allclean = 1; |
1128 | |
1129 | /* Any file-page which is PENDING now needs to be written. |
1130 | * So set NEEDWRITE now, then after we make any last-minute changes |
1131 | * we will write it. |
1132 | */ |
1133 | for (j = 0; j < bitmap->storage.file_pages; j++) |
1134 | if (test_and_clear_page_attr(bitmap, j, |
1135 | BITMAP_PAGE_PENDING)) |
1136 | set_page_attr(bitmap, j, |
1137 | BITMAP_PAGE_NEEDWRITE); |
1138 | |
1139 | if (bitmap->need_sync && |
1140 | mddev->bitmap_info.external == 0) { |
1141 | /* Arrange for superblock update as well as |
1142 | * other changes */ |
1143 | bitmap_super_t *sb; |
1144 | bitmap->need_sync = 0; |
1145 | if (bitmap->storage.filemap) { |
1146 | sb = kmap_atomic(bitmap->storage.sb_page); |
1147 | sb->events_cleared = |
1148 | cpu_to_le64(bitmap->events_cleared); |
1149 | kunmap_atomic(sb); |
1150 | set_page_attr(bitmap, 0, |
1151 | BITMAP_PAGE_NEEDWRITE); |
1152 | } |
1153 | } |
1154 | /* Now look at the bitmap counters and if any are '2' or '1', |
1155 | * decrement and handle accordingly. |
1156 | */ |
1157 | counts = &bitmap->counts; |
1158 | spin_lock_irq(&counts->lock); |
1159 | nextpage = 0; |
1160 | for (j = 0; j < counts->chunks; j++) { |
1161 | bitmap_counter_t *bmc; |
1162 | sector_t block = (sector_t)j << counts->chunkshift; |
1163 | |
1164 | if (j == nextpage) { |
1165 | nextpage += PAGE_COUNTER_RATIO; |
1166 | if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) { |
1167 | j |= PAGE_COUNTER_MASK; |
1168 | continue; |
1169 | } |
1170 | counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; |
1171 | } |
1172 | bmc = bitmap_get_counter(counts, |
1173 | block, |
1174 | &blocks, 0); |
1175 | |
1176 | if (!bmc) { |
1177 | j |= PAGE_COUNTER_MASK; |
1178 | continue; |
1179 | } |
1180 | if (*bmc == 1 && !bitmap->need_sync) { |
1181 | /* We can clear the bit */ |
1182 | *bmc = 0; |
1183 | bitmap_count_page(counts, block, -1); |
1184 | bitmap_file_clear_bit(bitmap, block); |
1185 | } else if (*bmc && *bmc <= 2) { |
1186 | *bmc = 1; |
1187 | bitmap_set_pending(counts, block); |
1188 | bitmap->allclean = 0; |
1189 | } |
1190 | } |
1191 | spin_unlock_irq(&counts->lock); |
1192 | |
1193 | /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. |
1194 | * DIRTY pages need to be written by bitmap_unplug so it can wait |
1195 | * for them. |
1196 | * If we find any DIRTY page we stop there and let bitmap_unplug |
1197 | * handle all the rest. This is important in the case where |
1198 | * the first blocking holds the superblock and it has been updated. |
1199 | * We mustn't write any other blocks before the superblock. |
1200 | */ |
1201 | for (j = 0; |
1202 | j < bitmap->storage.file_pages |
1203 | && !test_bit(BITMAP_STALE, &bitmap->flags); |
1204 | j++) { |
1205 | |
1206 | if (test_page_attr(bitmap, j, |
1207 | BITMAP_PAGE_DIRTY)) |
1208 | /* bitmap_unplug will handle the rest */ |
1209 | break; |
1210 | if (test_and_clear_page_attr(bitmap, j, |
1211 | BITMAP_PAGE_NEEDWRITE)) { |
1212 | write_page(bitmap, bitmap->storage.filemap[j], 0); |
1213 | } |
1214 | } |
1215 | |
1216 | done: |
1217 | if (bitmap->allclean == 0) |
1218 | mddev->thread->timeout = |
1219 | mddev->bitmap_info.daemon_sleep; |
1220 | mutex_unlock(&mddev->bitmap_info.mutex); |
1221 | } |
1222 | |
1223 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
1224 | sector_t offset, sector_t *blocks, |
1225 | int create) |
1226 | __releases(bitmap->lock) |
1227 | __acquires(bitmap->lock) |
1228 | { |
1229 | /* If 'create', we might release the lock and reclaim it. |
1230 | * The lock must have been taken with interrupts enabled. |
1231 | * If !create, we don't release the lock. |
1232 | */ |
1233 | sector_t chunk = offset >> bitmap->chunkshift; |
1234 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1235 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; |
1236 | sector_t csize; |
1237 | int err; |
1238 | |
1239 | err = bitmap_checkpage(bitmap, page, create); |
1240 | |
1241 | if (bitmap->bp[page].hijacked || |
1242 | bitmap->bp[page].map == NULL) |
1243 | csize = ((sector_t)1) << (bitmap->chunkshift + |
1244 | PAGE_COUNTER_SHIFT - 1); |
1245 | else |
1246 | csize = ((sector_t)1) << bitmap->chunkshift; |
1247 | *blocks = csize - (offset & (csize - 1)); |
1248 | |
1249 | if (err < 0) |
1250 | return NULL; |
1251 | |
1252 | /* now locked ... */ |
1253 | |
1254 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ |
1255 | /* should we use the first or second counter field |
1256 | * of the hijacked pointer? */ |
1257 | int hi = (pageoff > PAGE_COUNTER_MASK); |
1258 | return &((bitmap_counter_t *) |
1259 | &bitmap->bp[page].map)[hi]; |
1260 | } else /* page is allocated */ |
1261 | return (bitmap_counter_t *) |
1262 | &(bitmap->bp[page].map[pageoff]); |
1263 | } |
1264 | |
1265 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
1266 | { |
1267 | if (!bitmap) |
1268 | return 0; |
1269 | |
1270 | if (behind) { |
1271 | int bw; |
1272 | atomic_inc(&bitmap->behind_writes); |
1273 | bw = atomic_read(&bitmap->behind_writes); |
1274 | if (bw > bitmap->behind_writes_used) |
1275 | bitmap->behind_writes_used = bw; |
1276 | |
1277 | pr_debug("inc write-behind count %d/%lu\n", |
1278 | bw, bitmap->mddev->bitmap_info.max_write_behind); |
1279 | } |
1280 | |
1281 | while (sectors) { |
1282 | sector_t blocks; |
1283 | bitmap_counter_t *bmc; |
1284 | |
1285 | spin_lock_irq(&bitmap->counts.lock); |
1286 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1); |
1287 | if (!bmc) { |
1288 | spin_unlock_irq(&bitmap->counts.lock); |
1289 | return 0; |
1290 | } |
1291 | |
1292 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
1293 | DEFINE_WAIT(__wait); |
1294 | /* note that it is safe to do the prepare_to_wait |
1295 | * after the test as long as we do it before dropping |
1296 | * the spinlock. |
1297 | */ |
1298 | prepare_to_wait(&bitmap->overflow_wait, &__wait, |
1299 | TASK_UNINTERRUPTIBLE); |
1300 | spin_unlock_irq(&bitmap->counts.lock); |
1301 | schedule(); |
1302 | finish_wait(&bitmap->overflow_wait, &__wait); |
1303 | continue; |
1304 | } |
1305 | |
1306 | switch (*bmc) { |
1307 | case 0: |
1308 | bitmap_file_set_bit(bitmap, offset); |
1309 | bitmap_count_page(&bitmap->counts, offset, 1); |
1310 | /* fall through */ |
1311 | case 1: |
1312 | *bmc = 2; |
1313 | } |
1314 | |
1315 | (*bmc)++; |
1316 | |
1317 | spin_unlock_irq(&bitmap->counts.lock); |
1318 | |
1319 | offset += blocks; |
1320 | if (sectors > blocks) |
1321 | sectors -= blocks; |
1322 | else |
1323 | sectors = 0; |
1324 | } |
1325 | return 0; |
1326 | } |
1327 | EXPORT_SYMBOL(bitmap_startwrite); |
1328 | |
1329 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, |
1330 | int success, int behind) |
1331 | { |
1332 | if (!bitmap) |
1333 | return; |
1334 | if (behind) { |
1335 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1336 | wake_up(&bitmap->behind_wait); |
1337 | pr_debug("dec write-behind count %d/%lu\n", |
1338 | atomic_read(&bitmap->behind_writes), |
1339 | bitmap->mddev->bitmap_info.max_write_behind); |
1340 | } |
1341 | |
1342 | while (sectors) { |
1343 | sector_t blocks; |
1344 | unsigned long flags; |
1345 | bitmap_counter_t *bmc; |
1346 | |
1347 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1348 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0); |
1349 | if (!bmc) { |
1350 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
1351 | return; |
1352 | } |
1353 | |
1354 | if (success && !bitmap->mddev->degraded && |
1355 | bitmap->events_cleared < bitmap->mddev->events) { |
1356 | bitmap->events_cleared = bitmap->mddev->events; |
1357 | bitmap->need_sync = 1; |
1358 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
1359 | } |
1360 | |
1361 | if (!success && !NEEDED(*bmc)) |
1362 | *bmc |= NEEDED_MASK; |
1363 | |
1364 | if (COUNTER(*bmc) == COUNTER_MAX) |
1365 | wake_up(&bitmap->overflow_wait); |
1366 | |
1367 | (*bmc)--; |
1368 | if (*bmc <= 2) { |
1369 | bitmap_set_pending(&bitmap->counts, offset); |
1370 | bitmap->allclean = 0; |
1371 | } |
1372 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
1373 | offset += blocks; |
1374 | if (sectors > blocks) |
1375 | sectors -= blocks; |
1376 | else |
1377 | sectors = 0; |
1378 | } |
1379 | } |
1380 | EXPORT_SYMBOL(bitmap_endwrite); |
1381 | |
1382 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1383 | int degraded) |
1384 | { |
1385 | bitmap_counter_t *bmc; |
1386 | int rv; |
1387 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ |
1388 | *blocks = 1024; |
1389 | return 1; /* always resync if no bitmap */ |
1390 | } |
1391 | spin_lock_irq(&bitmap->counts.lock); |
1392 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); |
1393 | rv = 0; |
1394 | if (bmc) { |
1395 | /* locked */ |
1396 | if (RESYNC(*bmc)) |
1397 | rv = 1; |
1398 | else if (NEEDED(*bmc)) { |
1399 | rv = 1; |
1400 | if (!degraded) { /* don't set/clear bits if degraded */ |
1401 | *bmc |= RESYNC_MASK; |
1402 | *bmc &= ~NEEDED_MASK; |
1403 | } |
1404 | } |
1405 | } |
1406 | spin_unlock_irq(&bitmap->counts.lock); |
1407 | return rv; |
1408 | } |
1409 | |
1410 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1411 | int degraded) |
1412 | { |
1413 | /* bitmap_start_sync must always report on multiples of whole |
1414 | * pages, otherwise resync (which is very PAGE_SIZE based) will |
1415 | * get confused. |
1416 | * So call __bitmap_start_sync repeatedly (if needed) until |
1417 | * At least PAGE_SIZE>>9 blocks are covered. |
1418 | * Return the 'or' of the result. |
1419 | */ |
1420 | int rv = 0; |
1421 | sector_t blocks1; |
1422 | |
1423 | *blocks = 0; |
1424 | while (*blocks < (PAGE_SIZE>>9)) { |
1425 | rv |= __bitmap_start_sync(bitmap, offset, |
1426 | &blocks1, degraded); |
1427 | offset += blocks1; |
1428 | *blocks += blocks1; |
1429 | } |
1430 | return rv; |
1431 | } |
1432 | EXPORT_SYMBOL(bitmap_start_sync); |
1433 | |
1434 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
1435 | { |
1436 | bitmap_counter_t *bmc; |
1437 | unsigned long flags; |
1438 | |
1439 | if (bitmap == NULL) { |
1440 | *blocks = 1024; |
1441 | return; |
1442 | } |
1443 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1444 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); |
1445 | if (bmc == NULL) |
1446 | goto unlock; |
1447 | /* locked */ |
1448 | if (RESYNC(*bmc)) { |
1449 | *bmc &= ~RESYNC_MASK; |
1450 | |
1451 | if (!NEEDED(*bmc) && aborted) |
1452 | *bmc |= NEEDED_MASK; |
1453 | else { |
1454 | if (*bmc <= 2) { |
1455 | bitmap_set_pending(&bitmap->counts, offset); |
1456 | bitmap->allclean = 0; |
1457 | } |
1458 | } |
1459 | } |
1460 | unlock: |
1461 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
1462 | } |
1463 | EXPORT_SYMBOL(bitmap_end_sync); |
1464 | |
1465 | void bitmap_close_sync(struct bitmap *bitmap) |
1466 | { |
1467 | /* Sync has finished, and any bitmap chunks that weren't synced |
1468 | * properly have been aborted. It remains to us to clear the |
1469 | * RESYNC bit wherever it is still on |
1470 | */ |
1471 | sector_t sector = 0; |
1472 | sector_t blocks; |
1473 | if (!bitmap) |
1474 | return; |
1475 | while (sector < bitmap->mddev->resync_max_sectors) { |
1476 | bitmap_end_sync(bitmap, sector, &blocks, 0); |
1477 | sector += blocks; |
1478 | } |
1479 | } |
1480 | EXPORT_SYMBOL(bitmap_close_sync); |
1481 | |
1482 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector) |
1483 | { |
1484 | sector_t s = 0; |
1485 | sector_t blocks; |
1486 | |
1487 | if (!bitmap) |
1488 | return; |
1489 | if (sector == 0) { |
1490 | bitmap->last_end_sync = jiffies; |
1491 | return; |
1492 | } |
1493 | if (time_before(jiffies, (bitmap->last_end_sync |
1494 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
1495 | return; |
1496 | wait_event(bitmap->mddev->recovery_wait, |
1497 | atomic_read(&bitmap->mddev->recovery_active) == 0); |
1498 | |
1499 | bitmap->mddev->curr_resync_completed = sector; |
1500 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
1501 | sector &= ~((1ULL << bitmap->counts.chunkshift) - 1); |
1502 | s = 0; |
1503 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { |
1504 | bitmap_end_sync(bitmap, s, &blocks, 0); |
1505 | s += blocks; |
1506 | } |
1507 | bitmap->last_end_sync = jiffies; |
1508 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
1509 | } |
1510 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
1511 | |
1512 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
1513 | { |
1514 | /* For each chunk covered by any of these sectors, set the |
1515 | * counter to 2 and possibly set resync_needed. They should all |
1516 | * be 0 at this point |
1517 | */ |
1518 | |
1519 | sector_t secs; |
1520 | bitmap_counter_t *bmc; |
1521 | spin_lock_irq(&bitmap->counts.lock); |
1522 | bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1); |
1523 | if (!bmc) { |
1524 | spin_unlock_irq(&bitmap->counts.lock); |
1525 | return; |
1526 | } |
1527 | if (!*bmc) { |
1528 | *bmc = 2 | (needed ? NEEDED_MASK : 0); |
1529 | bitmap_count_page(&bitmap->counts, offset, 1); |
1530 | bitmap_set_pending(&bitmap->counts, offset); |
1531 | bitmap->allclean = 0; |
1532 | } |
1533 | spin_unlock_irq(&bitmap->counts.lock); |
1534 | } |
1535 | |
1536 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1537 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) |
1538 | { |
1539 | unsigned long chunk; |
1540 | |
1541 | for (chunk = s; chunk <= e; chunk++) { |
1542 | sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift; |
1543 | bitmap_set_memory_bits(bitmap, sec, 1); |
1544 | bitmap_file_set_bit(bitmap, sec); |
1545 | if (sec < bitmap->mddev->recovery_cp) |
1546 | /* We are asserting that the array is dirty, |
1547 | * so move the recovery_cp address back so |
1548 | * that it is obvious that it is dirty |
1549 | */ |
1550 | bitmap->mddev->recovery_cp = sec; |
1551 | } |
1552 | } |
1553 | |
1554 | /* |
1555 | * flush out any pending updates |
1556 | */ |
1557 | void bitmap_flush(struct mddev *mddev) |
1558 | { |
1559 | struct bitmap *bitmap = mddev->bitmap; |
1560 | long sleep; |
1561 | |
1562 | if (!bitmap) /* there was no bitmap */ |
1563 | return; |
1564 | |
1565 | /* run the daemon_work three time to ensure everything is flushed |
1566 | * that can be |
1567 | */ |
1568 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
1569 | bitmap->daemon_lastrun -= sleep; |
1570 | bitmap_daemon_work(mddev); |
1571 | bitmap->daemon_lastrun -= sleep; |
1572 | bitmap_daemon_work(mddev); |
1573 | bitmap->daemon_lastrun -= sleep; |
1574 | bitmap_daemon_work(mddev); |
1575 | bitmap_update_sb(bitmap); |
1576 | } |
1577 | |
1578 | /* |
1579 | * free memory that was allocated |
1580 | */ |
1581 | static void bitmap_free(struct bitmap *bitmap) |
1582 | { |
1583 | unsigned long k, pages; |
1584 | struct bitmap_page *bp; |
1585 | |
1586 | if (!bitmap) /* there was no bitmap */ |
1587 | return; |
1588 | |
1589 | /* Shouldn't be needed - but just in case.... */ |
1590 | wait_event(bitmap->write_wait, |
1591 | atomic_read(&bitmap->pending_writes) == 0); |
1592 | |
1593 | /* release the bitmap file */ |
1594 | bitmap_file_unmap(&bitmap->storage); |
1595 | |
1596 | bp = bitmap->counts.bp; |
1597 | pages = bitmap->counts.pages; |
1598 | |
1599 | /* free all allocated memory */ |
1600 | |
1601 | if (bp) /* deallocate the page memory */ |
1602 | for (k = 0; k < pages; k++) |
1603 | if (bp[k].map && !bp[k].hijacked) |
1604 | kfree(bp[k].map); |
1605 | kfree(bp); |
1606 | kfree(bitmap); |
1607 | } |
1608 | |
1609 | void bitmap_destroy(struct mddev *mddev) |
1610 | { |
1611 | struct bitmap *bitmap = mddev->bitmap; |
1612 | |
1613 | if (!bitmap) /* there was no bitmap */ |
1614 | return; |
1615 | |
1616 | mutex_lock(&mddev->bitmap_info.mutex); |
1617 | mddev->bitmap = NULL; /* disconnect from the md device */ |
1618 | mutex_unlock(&mddev->bitmap_info.mutex); |
1619 | if (mddev->thread) |
1620 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
1621 | |
1622 | if (bitmap->sysfs_can_clear) |
1623 | sysfs_put(bitmap->sysfs_can_clear); |
1624 | |
1625 | bitmap_free(bitmap); |
1626 | } |
1627 | |
1628 | /* |
1629 | * initialize the bitmap structure |
1630 | * if this returns an error, bitmap_destroy must be called to do clean up |
1631 | */ |
1632 | int bitmap_create(struct mddev *mddev) |
1633 | { |
1634 | struct bitmap *bitmap; |
1635 | sector_t blocks = mddev->resync_max_sectors; |
1636 | struct file *file = mddev->bitmap_info.file; |
1637 | int err; |
1638 | struct sysfs_dirent *bm = NULL; |
1639 | |
1640 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
1641 | |
1642 | BUG_ON(file && mddev->bitmap_info.offset); |
1643 | |
1644 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
1645 | if (!bitmap) |
1646 | return -ENOMEM; |
1647 | |
1648 | spin_lock_init(&bitmap->counts.lock); |
1649 | atomic_set(&bitmap->pending_writes, 0); |
1650 | init_waitqueue_head(&bitmap->write_wait); |
1651 | init_waitqueue_head(&bitmap->overflow_wait); |
1652 | init_waitqueue_head(&bitmap->behind_wait); |
1653 | |
1654 | bitmap->mddev = mddev; |
1655 | |
1656 | if (mddev->kobj.sd) |
1657 | bm = sysfs_get_dirent(mddev->kobj.sd, NULL, "bitmap"); |
1658 | if (bm) { |
1659 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, NULL, "can_clear"); |
1660 | sysfs_put(bm); |
1661 | } else |
1662 | bitmap->sysfs_can_clear = NULL; |
1663 | |
1664 | bitmap->storage.file = file; |
1665 | if (file) { |
1666 | get_file(file); |
1667 | /* As future accesses to this file will use bmap, |
1668 | * and bypass the page cache, we must sync the file |
1669 | * first. |
1670 | */ |
1671 | vfs_fsync(file, 1); |
1672 | } |
1673 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
1674 | if (!mddev->bitmap_info.external) { |
1675 | /* |
1676 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is |
1677 | * instructing us to create a new on-disk bitmap instance. |
1678 | */ |
1679 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) |
1680 | err = bitmap_new_disk_sb(bitmap); |
1681 | else |
1682 | err = bitmap_read_sb(bitmap); |
1683 | } else { |
1684 | err = 0; |
1685 | if (mddev->bitmap_info.chunksize == 0 || |
1686 | mddev->bitmap_info.daemon_sleep == 0) |
1687 | /* chunksize and time_base need to be |
1688 | * set first. */ |
1689 | err = -EINVAL; |
1690 | } |
1691 | if (err) |
1692 | goto error; |
1693 | |
1694 | bitmap->daemon_lastrun = jiffies; |
1695 | err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1); |
1696 | if (err) |
1697 | goto error; |
1698 | |
1699 | printk(KERN_INFO "created bitmap (%lu pages) for device %s\n", |
1700 | bitmap->counts.pages, bmname(bitmap)); |
1701 | |
1702 | mddev->bitmap = bitmap; |
1703 | return test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0; |
1704 | |
1705 | error: |
1706 | bitmap_free(bitmap); |
1707 | return err; |
1708 | } |
1709 | |
1710 | int bitmap_load(struct mddev *mddev) |
1711 | { |
1712 | int err = 0; |
1713 | sector_t start = 0; |
1714 | sector_t sector = 0; |
1715 | struct bitmap *bitmap = mddev->bitmap; |
1716 | |
1717 | if (!bitmap) |
1718 | goto out; |
1719 | |
1720 | /* Clear out old bitmap info first: Either there is none, or we |
1721 | * are resuming after someone else has possibly changed things, |
1722 | * so we should forget old cached info. |
1723 | * All chunks should be clean, but some might need_sync. |
1724 | */ |
1725 | while (sector < mddev->resync_max_sectors) { |
1726 | sector_t blocks; |
1727 | bitmap_start_sync(bitmap, sector, &blocks, 0); |
1728 | sector += blocks; |
1729 | } |
1730 | bitmap_close_sync(bitmap); |
1731 | |
1732 | if (mddev->degraded == 0 |
1733 | || bitmap->events_cleared == mddev->events) |
1734 | /* no need to keep dirty bits to optimise a |
1735 | * re-add of a missing device */ |
1736 | start = mddev->recovery_cp; |
1737 | |
1738 | mutex_lock(&mddev->bitmap_info.mutex); |
1739 | err = bitmap_init_from_disk(bitmap, start); |
1740 | mutex_unlock(&mddev->bitmap_info.mutex); |
1741 | |
1742 | if (err) |
1743 | goto out; |
1744 | clear_bit(BITMAP_STALE, &bitmap->flags); |
1745 | |
1746 | /* Kick recovery in case any bits were set */ |
1747 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); |
1748 | |
1749 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
1750 | md_wakeup_thread(mddev->thread); |
1751 | |
1752 | bitmap_update_sb(bitmap); |
1753 | |
1754 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
1755 | err = -EIO; |
1756 | out: |
1757 | return err; |
1758 | } |
1759 | EXPORT_SYMBOL_GPL(bitmap_load); |
1760 | |
1761 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap) |
1762 | { |
1763 | unsigned long chunk_kb; |
1764 | struct bitmap_counts *counts; |
1765 | |
1766 | if (!bitmap) |
1767 | return; |
1768 | |
1769 | counts = &bitmap->counts; |
1770 | |
1771 | chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; |
1772 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " |
1773 | "%lu%s chunk", |
1774 | counts->pages - counts->missing_pages, |
1775 | counts->pages, |
1776 | (counts->pages - counts->missing_pages) |
1777 | << (PAGE_SHIFT - 10), |
1778 | chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, |
1779 | chunk_kb ? "KB" : "B"); |
1780 | if (bitmap->storage.file) { |
1781 | seq_printf(seq, ", file: "); |
1782 | seq_path(seq, &bitmap->storage.file->f_path, " \t\n"); |
1783 | } |
1784 | |
1785 | seq_printf(seq, "\n"); |
1786 | } |
1787 | |
1788 | int bitmap_resize(struct bitmap *bitmap, sector_t blocks, |
1789 | int chunksize, int init) |
1790 | { |
1791 | /* If chunk_size is 0, choose an appropriate chunk size. |
1792 | * Then possibly allocate new storage space. |
1793 | * Then quiesce, copy bits, replace bitmap, and re-start |
1794 | * |
1795 | * This function is called both to set up the initial bitmap |
1796 | * and to resize the bitmap while the array is active. |
1797 | * If this happens as a result of the array being resized, |
1798 | * chunksize will be zero, and we need to choose a suitable |
1799 | * chunksize, otherwise we use what we are given. |
1800 | */ |
1801 | struct bitmap_storage store; |
1802 | struct bitmap_counts old_counts; |
1803 | unsigned long chunks; |
1804 | sector_t block; |
1805 | sector_t old_blocks, new_blocks; |
1806 | int chunkshift; |
1807 | int ret = 0; |
1808 | long pages; |
1809 | struct bitmap_page *new_bp; |
1810 | |
1811 | if (chunksize == 0) { |
1812 | /* If there is enough space, leave the chunk size unchanged, |
1813 | * else increase by factor of two until there is enough space. |
1814 | */ |
1815 | long bytes; |
1816 | long space = bitmap->mddev->bitmap_info.space; |
1817 | |
1818 | if (space == 0) { |
1819 | /* We don't know how much space there is, so limit |
1820 | * to current size - in sectors. |
1821 | */ |
1822 | bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8); |
1823 | if (!bitmap->mddev->bitmap_info.external) |
1824 | bytes += sizeof(bitmap_super_t); |
1825 | space = DIV_ROUND_UP(bytes, 512); |
1826 | bitmap->mddev->bitmap_info.space = space; |
1827 | } |
1828 | chunkshift = bitmap->counts.chunkshift; |
1829 | chunkshift--; |
1830 | do { |
1831 | /* 'chunkshift' is shift from block size to chunk size */ |
1832 | chunkshift++; |
1833 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); |
1834 | bytes = DIV_ROUND_UP(chunks, 8); |
1835 | if (!bitmap->mddev->bitmap_info.external) |
1836 | bytes += sizeof(bitmap_super_t); |
1837 | } while (bytes > (space << 9)); |
1838 | } else |
1839 | chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT; |
1840 | |
1841 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); |
1842 | memset(&store, 0, sizeof(store)); |
1843 | if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file) |
1844 | ret = bitmap_storage_alloc(&store, chunks, |
1845 | !bitmap->mddev->bitmap_info.external); |
1846 | if (ret) |
1847 | goto err; |
1848 | |
1849 | pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO); |
1850 | |
1851 | new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL); |
1852 | ret = -ENOMEM; |
1853 | if (!new_bp) { |
1854 | bitmap_file_unmap(&store); |
1855 | goto err; |
1856 | } |
1857 | |
1858 | if (!init) |
1859 | bitmap->mddev->pers->quiesce(bitmap->mddev, 1); |
1860 | |
1861 | store.file = bitmap->storage.file; |
1862 | bitmap->storage.file = NULL; |
1863 | |
1864 | if (store.sb_page && bitmap->storage.sb_page) |
1865 | memcpy(page_address(store.sb_page), |
1866 | page_address(bitmap->storage.sb_page), |
1867 | sizeof(bitmap_super_t)); |
1868 | bitmap_file_unmap(&bitmap->storage); |
1869 | bitmap->storage = store; |
1870 | |
1871 | old_counts = bitmap->counts; |
1872 | bitmap->counts.bp = new_bp; |
1873 | bitmap->counts.pages = pages; |
1874 | bitmap->counts.missing_pages = pages; |
1875 | bitmap->counts.chunkshift = chunkshift; |
1876 | bitmap->counts.chunks = chunks; |
1877 | bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift + |
1878 | BITMAP_BLOCK_SHIFT); |
1879 | |
1880 | blocks = min(old_counts.chunks << old_counts.chunkshift, |
1881 | chunks << chunkshift); |
1882 | |
1883 | spin_lock_irq(&bitmap->counts.lock); |
1884 | for (block = 0; block < blocks; ) { |
1885 | bitmap_counter_t *bmc_old, *bmc_new; |
1886 | int set; |
1887 | |
1888 | bmc_old = bitmap_get_counter(&old_counts, block, |
1889 | &old_blocks, 0); |
1890 | set = bmc_old && NEEDED(*bmc_old); |
1891 | |
1892 | if (set) { |
1893 | bmc_new = bitmap_get_counter(&bitmap->counts, block, |
1894 | &new_blocks, 1); |
1895 | if (*bmc_new == 0) { |
1896 | /* need to set on-disk bits too. */ |
1897 | sector_t end = block + new_blocks; |
1898 | sector_t start = block >> chunkshift; |
1899 | start <<= chunkshift; |
1900 | while (start < end) { |
1901 | bitmap_file_set_bit(bitmap, block); |
1902 | start += 1 << chunkshift; |
1903 | } |
1904 | *bmc_new = 2; |
1905 | bitmap_count_page(&bitmap->counts, |
1906 | block, 1); |
1907 | bitmap_set_pending(&bitmap->counts, |
1908 | block); |
1909 | } |
1910 | *bmc_new |= NEEDED_MASK; |
1911 | if (new_blocks < old_blocks) |
1912 | old_blocks = new_blocks; |
1913 | } |
1914 | block += old_blocks; |
1915 | } |
1916 | |
1917 | if (!init) { |
1918 | int i; |
1919 | while (block < (chunks << chunkshift)) { |
1920 | bitmap_counter_t *bmc; |
1921 | bmc = bitmap_get_counter(&bitmap->counts, block, |
1922 | &new_blocks, 1); |
1923 | if (bmc) { |
1924 | /* new space. It needs to be resynced, so |
1925 | * we set NEEDED_MASK. |
1926 | */ |
1927 | if (*bmc == 0) { |
1928 | *bmc = NEEDED_MASK | 2; |
1929 | bitmap_count_page(&bitmap->counts, |
1930 | block, 1); |
1931 | bitmap_set_pending(&bitmap->counts, |
1932 | block); |
1933 | } |
1934 | } |
1935 | block += new_blocks; |
1936 | } |
1937 | for (i = 0; i < bitmap->storage.file_pages; i++) |
1938 | set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
1939 | } |
1940 | spin_unlock_irq(&bitmap->counts.lock); |
1941 | |
1942 | if (!init) { |
1943 | bitmap_unplug(bitmap); |
1944 | bitmap->mddev->pers->quiesce(bitmap->mddev, 0); |
1945 | } |
1946 | ret = 0; |
1947 | err: |
1948 | return ret; |
1949 | } |
1950 | EXPORT_SYMBOL_GPL(bitmap_resize); |
1951 | |
1952 | static ssize_t |
1953 | location_show(struct mddev *mddev, char *page) |
1954 | { |
1955 | ssize_t len; |
1956 | if (mddev->bitmap_info.file) |
1957 | len = sprintf(page, "file"); |
1958 | else if (mddev->bitmap_info.offset) |
1959 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
1960 | else |
1961 | len = sprintf(page, "none"); |
1962 | len += sprintf(page+len, "\n"); |
1963 | return len; |
1964 | } |
1965 | |
1966 | static ssize_t |
1967 | location_store(struct mddev *mddev, const char *buf, size_t len) |
1968 | { |
1969 | |
1970 | if (mddev->pers) { |
1971 | if (!mddev->pers->quiesce) |
1972 | return -EBUSY; |
1973 | if (mddev->recovery || mddev->sync_thread) |
1974 | return -EBUSY; |
1975 | } |
1976 | |
1977 | if (mddev->bitmap || mddev->bitmap_info.file || |
1978 | mddev->bitmap_info.offset) { |
1979 | /* bitmap already configured. Only option is to clear it */ |
1980 | if (strncmp(buf, "none", 4) != 0) |
1981 | return -EBUSY; |
1982 | if (mddev->pers) { |
1983 | mddev->pers->quiesce(mddev, 1); |
1984 | bitmap_destroy(mddev); |
1985 | mddev->pers->quiesce(mddev, 0); |
1986 | } |
1987 | mddev->bitmap_info.offset = 0; |
1988 | if (mddev->bitmap_info.file) { |
1989 | struct file *f = mddev->bitmap_info.file; |
1990 | mddev->bitmap_info.file = NULL; |
1991 | restore_bitmap_write_access(f); |
1992 | fput(f); |
1993 | } |
1994 | } else { |
1995 | /* No bitmap, OK to set a location */ |
1996 | long long offset; |
1997 | if (strncmp(buf, "none", 4) == 0) |
1998 | /* nothing to be done */; |
1999 | else if (strncmp(buf, "file:", 5) == 0) { |
2000 | /* Not supported yet */ |
2001 | return -EINVAL; |
2002 | } else { |
2003 | int rv; |
2004 | if (buf[0] == '+') |
2005 | rv = strict_strtoll(buf+1, 10, &offset); |
2006 | else |
2007 | rv = strict_strtoll(buf, 10, &offset); |
2008 | if (rv) |
2009 | return rv; |
2010 | if (offset == 0) |
2011 | return -EINVAL; |
2012 | if (mddev->bitmap_info.external == 0 && |
2013 | mddev->major_version == 0 && |
2014 | offset != mddev->bitmap_info.default_offset) |
2015 | return -EINVAL; |
2016 | mddev->bitmap_info.offset = offset; |
2017 | if (mddev->pers) { |
2018 | mddev->pers->quiesce(mddev, 1); |
2019 | rv = bitmap_create(mddev); |
2020 | if (!rv) |
2021 | rv = bitmap_load(mddev); |
2022 | if (rv) { |
2023 | bitmap_destroy(mddev); |
2024 | mddev->bitmap_info.offset = 0; |
2025 | } |
2026 | mddev->pers->quiesce(mddev, 0); |
2027 | if (rv) |
2028 | return rv; |
2029 | } |
2030 | } |
2031 | } |
2032 | if (!mddev->external) { |
2033 | /* Ensure new bitmap info is stored in |
2034 | * metadata promptly. |
2035 | */ |
2036 | set_bit(MD_CHANGE_DEVS, &mddev->flags); |
2037 | md_wakeup_thread(mddev->thread); |
2038 | } |
2039 | return len; |
2040 | } |
2041 | |
2042 | static struct md_sysfs_entry bitmap_location = |
2043 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); |
2044 | |
2045 | /* 'bitmap/space' is the space available at 'location' for the |
2046 | * bitmap. This allows the kernel to know when it is safe to |
2047 | * resize the bitmap to match a resized array. |
2048 | */ |
2049 | static ssize_t |
2050 | space_show(struct mddev *mddev, char *page) |
2051 | { |
2052 | return sprintf(page, "%lu\n", mddev->bitmap_info.space); |
2053 | } |
2054 | |
2055 | static ssize_t |
2056 | space_store(struct mddev *mddev, const char *buf, size_t len) |
2057 | { |
2058 | unsigned long sectors; |
2059 | int rv; |
2060 | |
2061 | rv = kstrtoul(buf, 10, §ors); |
2062 | if (rv) |
2063 | return rv; |
2064 | |
2065 | if (sectors == 0) |
2066 | return -EINVAL; |
2067 | |
2068 | if (mddev->bitmap && |
2069 | sectors < (mddev->bitmap->storage.bytes + 511) >> 9) |
2070 | return -EFBIG; /* Bitmap is too big for this small space */ |
2071 | |
2072 | /* could make sure it isn't too big, but that isn't really |
2073 | * needed - user-space should be careful. |
2074 | */ |
2075 | mddev->bitmap_info.space = sectors; |
2076 | return len; |
2077 | } |
2078 | |
2079 | static struct md_sysfs_entry bitmap_space = |
2080 | __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); |
2081 | |
2082 | static ssize_t |
2083 | timeout_show(struct mddev *mddev, char *page) |
2084 | { |
2085 | ssize_t len; |
2086 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; |
2087 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; |
2088 | |
2089 | len = sprintf(page, "%lu", secs); |
2090 | if (jifs) |
2091 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); |
2092 | len += sprintf(page+len, "\n"); |
2093 | return len; |
2094 | } |
2095 | |
2096 | static ssize_t |
2097 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
2098 | { |
2099 | /* timeout can be set at any time */ |
2100 | unsigned long timeout; |
2101 | int rv = strict_strtoul_scaled(buf, &timeout, 4); |
2102 | if (rv) |
2103 | return rv; |
2104 | |
2105 | /* just to make sure we don't overflow... */ |
2106 | if (timeout >= LONG_MAX / HZ) |
2107 | return -EINVAL; |
2108 | |
2109 | timeout = timeout * HZ / 10000; |
2110 | |
2111 | if (timeout >= MAX_SCHEDULE_TIMEOUT) |
2112 | timeout = MAX_SCHEDULE_TIMEOUT-1; |
2113 | if (timeout < 1) |
2114 | timeout = 1; |
2115 | mddev->bitmap_info.daemon_sleep = timeout; |
2116 | if (mddev->thread) { |
2117 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then |
2118 | * the bitmap is all clean and we don't need to |
2119 | * adjust the timeout right now |
2120 | */ |
2121 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { |
2122 | mddev->thread->timeout = timeout; |
2123 | md_wakeup_thread(mddev->thread); |
2124 | } |
2125 | } |
2126 | return len; |
2127 | } |
2128 | |
2129 | static struct md_sysfs_entry bitmap_timeout = |
2130 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); |
2131 | |
2132 | static ssize_t |
2133 | backlog_show(struct mddev *mddev, char *page) |
2134 | { |
2135 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); |
2136 | } |
2137 | |
2138 | static ssize_t |
2139 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
2140 | { |
2141 | unsigned long backlog; |
2142 | int rv = strict_strtoul(buf, 10, &backlog); |
2143 | if (rv) |
2144 | return rv; |
2145 | if (backlog > COUNTER_MAX) |
2146 | return -EINVAL; |
2147 | mddev->bitmap_info.max_write_behind = backlog; |
2148 | return len; |
2149 | } |
2150 | |
2151 | static struct md_sysfs_entry bitmap_backlog = |
2152 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); |
2153 | |
2154 | static ssize_t |
2155 | chunksize_show(struct mddev *mddev, char *page) |
2156 | { |
2157 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); |
2158 | } |
2159 | |
2160 | static ssize_t |
2161 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
2162 | { |
2163 | /* Can only be changed when no bitmap is active */ |
2164 | int rv; |
2165 | unsigned long csize; |
2166 | if (mddev->bitmap) |
2167 | return -EBUSY; |
2168 | rv = strict_strtoul(buf, 10, &csize); |
2169 | if (rv) |
2170 | return rv; |
2171 | if (csize < 512 || |
2172 | !is_power_of_2(csize)) |
2173 | return -EINVAL; |
2174 | mddev->bitmap_info.chunksize = csize; |
2175 | return len; |
2176 | } |
2177 | |
2178 | static struct md_sysfs_entry bitmap_chunksize = |
2179 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); |
2180 | |
2181 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
2182 | { |
2183 | return sprintf(page, "%s\n", (mddev->bitmap_info.external |
2184 | ? "external" : "internal")); |
2185 | } |
2186 | |
2187 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
2188 | { |
2189 | if (mddev->bitmap || |
2190 | mddev->bitmap_info.file || |
2191 | mddev->bitmap_info.offset) |
2192 | return -EBUSY; |
2193 | if (strncmp(buf, "external", 8) == 0) |
2194 | mddev->bitmap_info.external = 1; |
2195 | else if (strncmp(buf, "internal", 8) == 0) |
2196 | mddev->bitmap_info.external = 0; |
2197 | else |
2198 | return -EINVAL; |
2199 | return len; |
2200 | } |
2201 | |
2202 | static struct md_sysfs_entry bitmap_metadata = |
2203 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); |
2204 | |
2205 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
2206 | { |
2207 | int len; |
2208 | if (mddev->bitmap) |
2209 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? |
2210 | "false" : "true")); |
2211 | else |
2212 | len = sprintf(page, "\n"); |
2213 | return len; |
2214 | } |
2215 | |
2216 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
2217 | { |
2218 | if (mddev->bitmap == NULL) |
2219 | return -ENOENT; |
2220 | if (strncmp(buf, "false", 5) == 0) |
2221 | mddev->bitmap->need_sync = 1; |
2222 | else if (strncmp(buf, "true", 4) == 0) { |
2223 | if (mddev->degraded) |
2224 | return -EBUSY; |
2225 | mddev->bitmap->need_sync = 0; |
2226 | } else |
2227 | return -EINVAL; |
2228 | return len; |
2229 | } |
2230 | |
2231 | static struct md_sysfs_entry bitmap_can_clear = |
2232 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); |
2233 | |
2234 | static ssize_t |
2235 | behind_writes_used_show(struct mddev *mddev, char *page) |
2236 | { |
2237 | if (mddev->bitmap == NULL) |
2238 | return sprintf(page, "0\n"); |
2239 | return sprintf(page, "%lu\n", |
2240 | mddev->bitmap->behind_writes_used); |
2241 | } |
2242 | |
2243 | static ssize_t |
2244 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
2245 | { |
2246 | if (mddev->bitmap) |
2247 | mddev->bitmap->behind_writes_used = 0; |
2248 | return len; |
2249 | } |
2250 | |
2251 | static struct md_sysfs_entry max_backlog_used = |
2252 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, |
2253 | behind_writes_used_show, behind_writes_used_reset); |
2254 | |
2255 | static struct attribute *md_bitmap_attrs[] = { |
2256 | &bitmap_location.attr, |
2257 | &bitmap_space.attr, |
2258 | &bitmap_timeout.attr, |
2259 | &bitmap_backlog.attr, |
2260 | &bitmap_chunksize.attr, |
2261 | &bitmap_metadata.attr, |
2262 | &bitmap_can_clear.attr, |
2263 | &max_backlog_used.attr, |
2264 | NULL |
2265 | }; |
2266 | struct attribute_group md_bitmap_group = { |
2267 | .name = "bitmap", |
2268 | .attrs = md_bitmap_attrs, |
2269 | }; |
2270 | |
2271 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9