]>
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). | |
32a7627c N |
16 | */ |
17 | ||
bff61975 | 18 | #include <linux/blkdev.h> |
32a7627c | 19 | #include <linux/module.h> |
32a7627c N |
20 | #include <linux/errno.h> |
21 | #include <linux/slab.h> | |
22 | #include <linux/init.h> | |
32a7627c N |
23 | #include <linux/timer.h> |
24 | #include <linux/sched.h> | |
25 | #include <linux/list.h> | |
26 | #include <linux/file.h> | |
27 | #include <linux/mount.h> | |
28 | #include <linux/buffer_head.h> | |
57148964 | 29 | #include <linux/seq_file.h> |
581dbd94 | 30 | #include <trace/events/block.h> |
43b2e5d8 | 31 | #include "md.h" |
935fe098 | 32 | #include "md-bitmap.h" |
32a7627c | 33 | |
ac2f40be | 34 | static inline char *bmname(struct bitmap *bitmap) |
32a7627c N |
35 | { |
36 | return bitmap->mddev ? mdname(bitmap->mddev) : "mdX"; | |
37 | } | |
38 | ||
32a7627c N |
39 | /* |
40 | * check a page and, if necessary, allocate it (or hijack it if the alloc fails) | |
41 | * | |
42 | * 1) check to see if this page is allocated, if it's not then try to alloc | |
43 | * 2) if the alloc fails, set the page's hijacked flag so we'll use the | |
44 | * page pointer directly as a counter | |
45 | * | |
46 | * if we find our page, we increment the page's refcount so that it stays | |
47 | * allocated while we're using it | |
48 | */ | |
e64e4018 AS |
49 | static int md_bitmap_checkpage(struct bitmap_counts *bitmap, |
50 | unsigned long page, int create, int no_hijack) | |
ee305ace N |
51 | __releases(bitmap->lock) |
52 | __acquires(bitmap->lock) | |
32a7627c N |
53 | { |
54 | unsigned char *mappage; | |
55 | ||
56 | if (page >= bitmap->pages) { | |
1187cf0a N |
57 | /* This can happen if bitmap_start_sync goes beyond |
58 | * End-of-device while looking for a whole page. | |
59 | * It is harmless. | |
60 | */ | |
32a7627c N |
61 | return -EINVAL; |
62 | } | |
63 | ||
32a7627c N |
64 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ |
65 | return 0; | |
66 | ||
67 | if (bitmap->bp[page].map) /* page is already allocated, just return */ | |
68 | return 0; | |
69 | ||
70 | if (!create) | |
71 | return -ENOENT; | |
72 | ||
32a7627c N |
73 | /* this page has not been allocated yet */ |
74 | ||
ac2f40be | 75 | spin_unlock_irq(&bitmap->lock); |
d9590143 N |
76 | /* It is possible that this is being called inside a |
77 | * prepare_to_wait/finish_wait loop from raid5c:make_request(). | |
78 | * In general it is not permitted to sleep in that context as it | |
79 | * can cause the loop to spin freely. | |
80 | * That doesn't apply here as we can only reach this point | |
81 | * once with any loop. | |
82 | * When this function completes, either bp[page].map or | |
83 | * bp[page].hijacked. In either case, this function will | |
84 | * abort before getting to this point again. So there is | |
85 | * no risk of a free-spin, and so it is safe to assert | |
86 | * that sleeping here is allowed. | |
87 | */ | |
88 | sched_annotate_sleep(); | |
792a1d4b | 89 | mappage = kzalloc(PAGE_SIZE, GFP_NOIO); |
ac2f40be N |
90 | spin_lock_irq(&bitmap->lock); |
91 | ||
92 | if (mappage == NULL) { | |
40cffcc0 | 93 | pr_debug("md/bitmap: map page allocation failed, hijacking\n"); |
c9d65032 GJ |
94 | /* We don't support hijack for cluster raid */ |
95 | if (no_hijack) | |
96 | return -ENOMEM; | |
32a7627c N |
97 | /* failed - set the hijacked flag so that we can use the |
98 | * pointer as a counter */ | |
32a7627c N |
99 | if (!bitmap->bp[page].map) |
100 | bitmap->bp[page].hijacked = 1; | |
ac2f40be N |
101 | } else if (bitmap->bp[page].map || |
102 | bitmap->bp[page].hijacked) { | |
32a7627c | 103 | /* somebody beat us to getting the page */ |
792a1d4b | 104 | kfree(mappage); |
ac2f40be | 105 | } else { |
32a7627c | 106 | |
ac2f40be | 107 | /* no page was in place and we have one, so install it */ |
32a7627c | 108 | |
ac2f40be N |
109 | bitmap->bp[page].map = mappage; |
110 | bitmap->missing_pages--; | |
111 | } | |
32a7627c N |
112 | return 0; |
113 | } | |
114 | ||
32a7627c N |
115 | /* if page is completely empty, put it back on the free list, or dealloc it */ |
116 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
117 | /* Note: lock should be held when calling this */ | |
e64e4018 | 118 | static void md_bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page) |
32a7627c N |
119 | { |
120 | char *ptr; | |
121 | ||
122 | if (bitmap->bp[page].count) /* page is still busy */ | |
123 | return; | |
124 | ||
125 | /* page is no longer in use, it can be released */ | |
126 | ||
127 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
128 | bitmap->bp[page].hijacked = 0; | |
129 | bitmap->bp[page].map = NULL; | |
ac2f40be N |
130 | } else { |
131 | /* normal case, free the page */ | |
132 | ptr = bitmap->bp[page].map; | |
133 | bitmap->bp[page].map = NULL; | |
134 | bitmap->missing_pages++; | |
792a1d4b | 135 | kfree(ptr); |
32a7627c | 136 | } |
32a7627c N |
137 | } |
138 | ||
32a7627c N |
139 | /* |
140 | * bitmap file handling - read and write the bitmap file and its superblock | |
141 | */ | |
142 | ||
32a7627c N |
143 | /* |
144 | * basic page I/O operations | |
145 | */ | |
146 | ||
a654b9d8 | 147 | /* IO operations when bitmap is stored near all superblocks */ |
27581e5a N |
148 | static int read_sb_page(struct mddev *mddev, loff_t offset, |
149 | struct page *page, | |
150 | unsigned long index, int size) | |
a654b9d8 N |
151 | { |
152 | /* choose a good rdev and read the page from there */ | |
153 | ||
3cb03002 | 154 | struct md_rdev *rdev; |
a654b9d8 | 155 | sector_t target; |
a654b9d8 | 156 | |
dafb20fa | 157 | rdev_for_each(rdev, mddev) { |
b2d444d7 | 158 | if (! test_bit(In_sync, &rdev->flags) |
4aaf7694 GJ |
159 | || test_bit(Faulty, &rdev->flags) |
160 | || test_bit(Bitmap_sync, &rdev->flags)) | |
ab904d63 N |
161 | continue; |
162 | ||
ccebd4c4 | 163 | target = offset + index * (PAGE_SIZE/512); |
a654b9d8 | 164 | |
2b193363 | 165 | if (sync_page_io(rdev, target, |
e1defc4f | 166 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
796a5cf0 | 167 | page, REQ_OP_READ, 0, true)) { |
ab904d63 | 168 | page->index = index; |
27581e5a | 169 | return 0; |
ab904d63 N |
170 | } |
171 | } | |
27581e5a | 172 | return -EIO; |
a654b9d8 N |
173 | } |
174 | ||
fd01b88c | 175 | static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) |
b2d2c4ce N |
176 | { |
177 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
178 | * linked list, and raising the refcount of devices we return to ensure | |
179 | * they don't disappear while in use. | |
180 | * As devices are only added or removed when raid_disk is < 0 and | |
181 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
182 | * still be in the same position on the list when we re-enter | |
fd177481 | 183 | * list_for_each_entry_continue_rcu. |
8532e343 N |
184 | * |
185 | * Note that if entered with 'rdev == NULL' to start at the | |
186 | * beginning, we temporarily assign 'rdev' to an address which | |
187 | * isn't really an rdev, but which can be used by | |
188 | * list_for_each_entry_continue_rcu() to find the first entry. | |
b2d2c4ce | 189 | */ |
b2d2c4ce N |
190 | rcu_read_lock(); |
191 | if (rdev == NULL) | |
192 | /* start at the beginning */ | |
8532e343 | 193 | rdev = list_entry(&mddev->disks, struct md_rdev, same_set); |
b2d2c4ce N |
194 | else { |
195 | /* release the previous rdev and start from there. */ | |
196 | rdev_dec_pending(rdev, mddev); | |
b2d2c4ce | 197 | } |
fd177481 | 198 | list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) { |
b2d2c4ce | 199 | if (rdev->raid_disk >= 0 && |
b2d2c4ce N |
200 | !test_bit(Faulty, &rdev->flags)) { |
201 | /* this is a usable devices */ | |
202 | atomic_inc(&rdev->nr_pending); | |
203 | rcu_read_unlock(); | |
204 | return rdev; | |
205 | } | |
206 | } | |
207 | rcu_read_unlock(); | |
208 | return NULL; | |
209 | } | |
210 | ||
ab6085c7 | 211 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 212 | { |
46533ff7 | 213 | struct md_rdev *rdev; |
a6ff7e08 | 214 | struct block_device *bdev; |
fd01b88c | 215 | struct mddev *mddev = bitmap->mddev; |
1ec885cd | 216 | struct bitmap_storage *store = &bitmap->storage; |
a654b9d8 | 217 | |
46533ff7 N |
218 | restart: |
219 | rdev = NULL; | |
b2d2c4ce | 220 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ac2f40be N |
221 | int size = PAGE_SIZE; |
222 | loff_t offset = mddev->bitmap_info.offset; | |
a6ff7e08 JB |
223 | |
224 | bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; | |
225 | ||
9b1215c1 N |
226 | if (page->index == store->file_pages-1) { |
227 | int last_page_size = store->bytes & (PAGE_SIZE-1); | |
228 | if (last_page_size == 0) | |
229 | last_page_size = PAGE_SIZE; | |
230 | size = roundup(last_page_size, | |
a6ff7e08 | 231 | bdev_logical_block_size(bdev)); |
9b1215c1 | 232 | } |
ac2f40be N |
233 | /* Just make sure we aren't corrupting data or |
234 | * metadata | |
235 | */ | |
236 | if (mddev->external) { | |
237 | /* Bitmap could be anywhere. */ | |
238 | if (rdev->sb_start + offset + (page->index | |
239 | * (PAGE_SIZE/512)) | |
240 | > rdev->data_offset | |
241 | && | |
242 | rdev->sb_start + offset | |
243 | < (rdev->data_offset + mddev->dev_sectors | |
244 | + (PAGE_SIZE/512))) | |
245 | goto bad_alignment; | |
246 | } else if (offset < 0) { | |
247 | /* DATA BITMAP METADATA */ | |
248 | if (offset | |
249 | + (long)(page->index * (PAGE_SIZE/512)) | |
250 | + size/512 > 0) | |
251 | /* bitmap runs in to metadata */ | |
252 | goto bad_alignment; | |
253 | if (rdev->data_offset + mddev->dev_sectors | |
254 | > rdev->sb_start + offset) | |
255 | /* data runs in to bitmap */ | |
256 | goto bad_alignment; | |
257 | } else if (rdev->sb_start < rdev->data_offset) { | |
258 | /* METADATA BITMAP DATA */ | |
259 | if (rdev->sb_start | |
260 | + offset | |
261 | + page->index*(PAGE_SIZE/512) + size/512 | |
262 | > rdev->data_offset) | |
263 | /* bitmap runs in to data */ | |
264 | goto bad_alignment; | |
265 | } else { | |
266 | /* DATA METADATA BITMAP - no problems */ | |
267 | } | |
268 | md_super_write(mddev, rdev, | |
269 | rdev->sb_start + offset | |
270 | + page->index * (PAGE_SIZE/512), | |
271 | size, | |
272 | page); | |
b2d2c4ce | 273 | } |
a654b9d8 | 274 | |
46533ff7 N |
275 | if (wait && md_super_wait(mddev) < 0) |
276 | goto restart; | |
a654b9d8 | 277 | return 0; |
4b80991c N |
278 | |
279 | bad_alignment: | |
4b80991c | 280 | return -EINVAL; |
a654b9d8 N |
281 | } |
282 | ||
e64e4018 | 283 | static void md_bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 284 | /* |
a654b9d8 | 285 | * write out a page to a file |
32a7627c | 286 | */ |
4ad13663 | 287 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 288 | { |
d785a06a | 289 | struct buffer_head *bh; |
32a7627c | 290 | |
1ec885cd | 291 | if (bitmap->storage.file == NULL) { |
f0d76d70 N |
292 | switch (write_sb_page(bitmap, page, wait)) { |
293 | case -EINVAL: | |
b405fe91 | 294 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); |
f0d76d70 | 295 | } |
4ad13663 | 296 | } else { |
a654b9d8 | 297 | |
4ad13663 | 298 | bh = page_buffers(page); |
c708443c | 299 | |
4ad13663 N |
300 | while (bh && bh->b_blocknr) { |
301 | atomic_inc(&bitmap->pending_writes); | |
302 | set_buffer_locked(bh); | |
303 | set_buffer_mapped(bh); | |
2a222ca9 | 304 | submit_bh(REQ_OP_WRITE, REQ_SYNC, bh); |
4ad13663 N |
305 | bh = bh->b_this_page; |
306 | } | |
d785a06a | 307 | |
ac2f40be | 308 | if (wait) |
4ad13663 N |
309 | wait_event(bitmap->write_wait, |
310 | atomic_read(&bitmap->pending_writes)==0); | |
8a5e9cf1 | 311 | } |
b405fe91 | 312 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
e64e4018 | 313 | md_bitmap_file_kick(bitmap); |
d785a06a N |
314 | } |
315 | ||
316 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
317 | { | |
318 | struct bitmap *bitmap = bh->b_private; | |
32a7627c | 319 | |
b405fe91 N |
320 | if (!uptodate) |
321 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); | |
d785a06a N |
322 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
323 | wake_up(&bitmap->write_wait); | |
324 | } | |
32a7627c | 325 | |
d785a06a N |
326 | /* copied from buffer.c */ |
327 | static void | |
328 | __clear_page_buffers(struct page *page) | |
329 | { | |
330 | ClearPagePrivate(page); | |
331 | set_page_private(page, 0); | |
09cbfeaf | 332 | put_page(page); |
d785a06a N |
333 | } |
334 | static void free_buffers(struct page *page) | |
335 | { | |
27581e5a | 336 | struct buffer_head *bh; |
77ad4bc7 | 337 | |
27581e5a N |
338 | if (!PagePrivate(page)) |
339 | return; | |
340 | ||
341 | bh = page_buffers(page); | |
d785a06a N |
342 | while (bh) { |
343 | struct buffer_head *next = bh->b_this_page; | |
344 | free_buffer_head(bh); | |
345 | bh = next; | |
77ad4bc7 | 346 | } |
d785a06a N |
347 | __clear_page_buffers(page); |
348 | put_page(page); | |
32a7627c N |
349 | } |
350 | ||
d785a06a N |
351 | /* read a page from a file. |
352 | * We both read the page, and attach buffers to the page to record the | |
353 | * address of each block (using bmap). These addresses will be used | |
354 | * to write the block later, completely bypassing the filesystem. | |
355 | * This usage is similar to how swap files are handled, and allows us | |
356 | * to write to a file with no concerns of memory allocation failing. | |
357 | */ | |
27581e5a N |
358 | static int read_page(struct file *file, unsigned long index, |
359 | struct bitmap *bitmap, | |
360 | unsigned long count, | |
361 | struct page *page) | |
32a7627c | 362 | { |
27581e5a | 363 | int ret = 0; |
496ad9aa | 364 | struct inode *inode = file_inode(file); |
d785a06a N |
365 | struct buffer_head *bh; |
366 | sector_t block; | |
32a7627c | 367 | |
36a4e1fe N |
368 | pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
369 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 370 | |
640ab98f | 371 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, false); |
d785a06a | 372 | if (!bh) { |
27581e5a | 373 | ret = -ENOMEM; |
32a7627c N |
374 | goto out; |
375 | } | |
d785a06a N |
376 | attach_page_buffers(page, bh); |
377 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
378 | while (bh) { | |
379 | if (count == 0) | |
380 | bh->b_blocknr = 0; | |
381 | else { | |
382 | bh->b_blocknr = bmap(inode, block); | |
383 | if (bh->b_blocknr == 0) { | |
384 | /* Cannot use this file! */ | |
27581e5a | 385 | ret = -EINVAL; |
d785a06a N |
386 | goto out; |
387 | } | |
388 | bh->b_bdev = inode->i_sb->s_bdev; | |
389 | if (count < (1<<inode->i_blkbits)) | |
390 | count = 0; | |
391 | else | |
392 | count -= (1<<inode->i_blkbits); | |
393 | ||
394 | bh->b_end_io = end_bitmap_write; | |
395 | bh->b_private = bitmap; | |
ce25c31b N |
396 | atomic_inc(&bitmap->pending_writes); |
397 | set_buffer_locked(bh); | |
398 | set_buffer_mapped(bh); | |
2a222ca9 | 399 | submit_bh(REQ_OP_READ, 0, bh); |
d785a06a N |
400 | } |
401 | block++; | |
402 | bh = bh->b_this_page; | |
403 | } | |
d785a06a | 404 | page->index = index; |
ce25c31b N |
405 | |
406 | wait_event(bitmap->write_wait, | |
407 | atomic_read(&bitmap->pending_writes)==0); | |
b405fe91 | 408 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
27581e5a | 409 | ret = -EIO; |
32a7627c | 410 | out: |
27581e5a | 411 | if (ret) |
ec0cc226 N |
412 | pr_err("md: bitmap read error: (%dB @ %llu): %d\n", |
413 | (int)PAGE_SIZE, | |
414 | (unsigned long long)index << PAGE_SHIFT, | |
415 | ret); | |
27581e5a | 416 | return ret; |
32a7627c N |
417 | } |
418 | ||
419 | /* | |
420 | * bitmap file superblock operations | |
421 | */ | |
422 | ||
85c9ccd4 | 423 | /* |
e64e4018 | 424 | * md_bitmap_wait_writes() should be called before writing any bitmap |
85c9ccd4 | 425 | * blocks, to ensure previous writes, particularly from |
e64e4018 | 426 | * md_bitmap_daemon_work(), have completed. |
85c9ccd4 | 427 | */ |
e64e4018 | 428 | static void md_bitmap_wait_writes(struct bitmap *bitmap) |
85c9ccd4 N |
429 | { |
430 | if (bitmap->storage.file) | |
431 | wait_event(bitmap->write_wait, | |
432 | atomic_read(&bitmap->pending_writes)==0); | |
433 | else | |
46533ff7 N |
434 | /* Note that we ignore the return value. The writes |
435 | * might have failed, but that would just mean that | |
436 | * some bits which should be cleared haven't been, | |
437 | * which is safe. The relevant bitmap blocks will | |
438 | * probably get written again, but there is no great | |
439 | * loss if they aren't. | |
440 | */ | |
85c9ccd4 N |
441 | md_super_wait(bitmap->mddev); |
442 | } | |
443 | ||
444 | ||
32a7627c | 445 | /* update the event counter and sync the superblock to disk */ |
e64e4018 | 446 | void md_bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
447 | { |
448 | bitmap_super_t *sb; | |
32a7627c N |
449 | |
450 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 451 | return; |
ece5cff0 N |
452 | if (bitmap->mddev->bitmap_info.external) |
453 | return; | |
1ec885cd | 454 | if (!bitmap->storage.sb_page) /* no superblock */ |
4ad13663 | 455 | return; |
1ec885cd | 456 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 457 | sb->events = cpu_to_le64(bitmap->mddev->events); |
8258c532 | 458 | if (bitmap->mddev->events < bitmap->events_cleared) |
a0da84f3 NB |
459 | /* rocking back to read-only */ |
460 | bitmap->events_cleared = bitmap->mddev->events; | |
8258c532 | 461 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); |
97f0eb9f HT |
462 | /* |
463 | * clear BITMAP_WRITE_ERROR bit to protect against the case that | |
464 | * a bitmap write error occurred but the later writes succeeded. | |
465 | */ | |
466 | sb->state = cpu_to_le32(bitmap->flags & ~BIT(BITMAP_WRITE_ERROR)); | |
43a70507 N |
467 | /* Just in case these have been changed via sysfs: */ |
468 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
469 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
b81a0404 N |
470 | /* This might have been changed by a reshape */ |
471 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
472 | sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize); | |
c4ce867f | 473 | sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes); |
1dff2b87 N |
474 | sb->sectors_reserved = cpu_to_le32(bitmap->mddev-> |
475 | bitmap_info.space); | |
b2f46e68 | 476 | kunmap_atomic(sb); |
1ec885cd | 477 | write_page(bitmap, bitmap->storage.sb_page, 1); |
32a7627c | 478 | } |
e64e4018 | 479 | EXPORT_SYMBOL(md_bitmap_update_sb); |
32a7627c N |
480 | |
481 | /* print out the bitmap file superblock */ | |
e64e4018 | 482 | void md_bitmap_print_sb(struct bitmap *bitmap) |
32a7627c N |
483 | { |
484 | bitmap_super_t *sb; | |
485 | ||
1ec885cd | 486 | if (!bitmap || !bitmap->storage.sb_page) |
32a7627c | 487 | return; |
1ec885cd | 488 | sb = kmap_atomic(bitmap->storage.sb_page); |
ec0cc226 N |
489 | pr_debug("%s: bitmap file superblock:\n", bmname(bitmap)); |
490 | pr_debug(" magic: %08x\n", le32_to_cpu(sb->magic)); | |
491 | pr_debug(" version: %d\n", le32_to_cpu(sb->version)); | |
492 | pr_debug(" uuid: %08x.%08x.%08x.%08x\n", | |
c35403f8 CH |
493 | le32_to_cpu(*(__le32 *)(sb->uuid+0)), |
494 | le32_to_cpu(*(__le32 *)(sb->uuid+4)), | |
495 | le32_to_cpu(*(__le32 *)(sb->uuid+8)), | |
496 | le32_to_cpu(*(__le32 *)(sb->uuid+12))); | |
ec0cc226 N |
497 | pr_debug(" events: %llu\n", |
498 | (unsigned long long) le64_to_cpu(sb->events)); | |
499 | pr_debug("events cleared: %llu\n", | |
500 | (unsigned long long) le64_to_cpu(sb->events_cleared)); | |
501 | pr_debug(" state: %08x\n", le32_to_cpu(sb->state)); | |
502 | pr_debug(" chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
503 | pr_debug(" daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
504 | pr_debug(" sync size: %llu KB\n", | |
505 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
506 | pr_debug("max write behind: %d\n", le32_to_cpu(sb->write_behind)); | |
b2f46e68 | 507 | kunmap_atomic(sb); |
32a7627c N |
508 | } |
509 | ||
9c81075f JB |
510 | /* |
511 | * bitmap_new_disk_sb | |
512 | * @bitmap | |
513 | * | |
514 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb | |
515 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. | |
516 | * This function verifies 'bitmap_info' and populates the on-disk bitmap | |
517 | * structure, which is to be written to disk. | |
518 | * | |
519 | * Returns: 0 on success, -Exxx on error | |
520 | */ | |
e64e4018 | 521 | static int md_bitmap_new_disk_sb(struct bitmap *bitmap) |
9c81075f JB |
522 | { |
523 | bitmap_super_t *sb; | |
524 | unsigned long chunksize, daemon_sleep, write_behind; | |
9c81075f | 525 | |
d3b178ad | 526 | bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
582e2e05 JM |
527 | if (bitmap->storage.sb_page == NULL) |
528 | return -ENOMEM; | |
1ec885cd | 529 | bitmap->storage.sb_page->index = 0; |
9c81075f | 530 | |
1ec885cd | 531 | sb = kmap_atomic(bitmap->storage.sb_page); |
9c81075f JB |
532 | |
533 | sb->magic = cpu_to_le32(BITMAP_MAGIC); | |
534 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); | |
535 | ||
536 | chunksize = bitmap->mddev->bitmap_info.chunksize; | |
537 | BUG_ON(!chunksize); | |
538 | if (!is_power_of_2(chunksize)) { | |
b2f46e68 | 539 | kunmap_atomic(sb); |
ec0cc226 | 540 | pr_warn("bitmap chunksize not a power of 2\n"); |
9c81075f JB |
541 | return -EINVAL; |
542 | } | |
543 | sb->chunksize = cpu_to_le32(chunksize); | |
544 | ||
545 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; | |
c97e0602 | 546 | if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { |
ec0cc226 | 547 | pr_debug("Choosing daemon_sleep default (5 sec)\n"); |
9c81075f JB |
548 | daemon_sleep = 5 * HZ; |
549 | } | |
550 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); | |
551 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
552 | ||
553 | /* | |
554 | * FIXME: write_behind for RAID1. If not specified, what | |
555 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. | |
556 | */ | |
557 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; | |
558 | if (write_behind > COUNTER_MAX) | |
559 | write_behind = COUNTER_MAX / 2; | |
560 | sb->write_behind = cpu_to_le32(write_behind); | |
561 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
562 | ||
563 | /* keep the array size field of the bitmap superblock up to date */ | |
564 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
565 | ||
566 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); | |
567 | ||
b405fe91 | 568 | set_bit(BITMAP_STALE, &bitmap->flags); |
84e92345 | 569 | sb->state = cpu_to_le32(bitmap->flags); |
9c81075f JB |
570 | bitmap->events_cleared = bitmap->mddev->events; |
571 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
d3b178ad | 572 | bitmap->mddev->bitmap_info.nodes = 0; |
9c81075f | 573 | |
b2f46e68 | 574 | kunmap_atomic(sb); |
9c81075f JB |
575 | |
576 | return 0; | |
577 | } | |
578 | ||
32a7627c | 579 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
e64e4018 | 580 | static int md_bitmap_read_sb(struct bitmap *bitmap) |
32a7627c N |
581 | { |
582 | char *reason = NULL; | |
583 | bitmap_super_t *sb; | |
4b6d287f | 584 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c | 585 | unsigned long long events; |
c4ce867f | 586 | int nodes = 0; |
1dff2b87 | 587 | unsigned long sectors_reserved = 0; |
32a7627c | 588 | int err = -EINVAL; |
27581e5a | 589 | struct page *sb_page; |
33e38ac6 | 590 | loff_t offset = bitmap->mddev->bitmap_info.offset; |
32a7627c | 591 | |
1ec885cd | 592 | if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { |
ef99bf48 N |
593 | chunksize = 128 * 1024 * 1024; |
594 | daemon_sleep = 5 * HZ; | |
595 | write_behind = 0; | |
b405fe91 | 596 | set_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
597 | err = 0; |
598 | goto out_no_sb; | |
599 | } | |
32a7627c | 600 | /* page 0 is the superblock, read it... */ |
27581e5a N |
601 | sb_page = alloc_page(GFP_KERNEL); |
602 | if (!sb_page) | |
603 | return -ENOMEM; | |
1ec885cd | 604 | bitmap->storage.sb_page = sb_page; |
27581e5a | 605 | |
b97e9257 | 606 | re_read: |
f9209a32 GR |
607 | /* If cluster_slot is set, the cluster is setup */ |
608 | if (bitmap->cluster_slot >= 0) { | |
3b0e6aac | 609 | sector_t bm_blocks = bitmap->mddev->resync_max_sectors; |
f9209a32 | 610 | |
3b0e6aac SR |
611 | sector_div(bm_blocks, |
612 | bitmap->mddev->bitmap_info.chunksize >> 9); | |
124eb761 GR |
613 | /* bits to bytes */ |
614 | bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t); | |
615 | /* to 4k blocks */ | |
935f3d4f | 616 | bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096); |
33e38ac6 | 617 | offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3)); |
ec0cc226 | 618 | pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__, |
33e38ac6 | 619 | bitmap->cluster_slot, offset); |
f9209a32 GR |
620 | } |
621 | ||
1ec885cd N |
622 | if (bitmap->storage.file) { |
623 | loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); | |
f49d5e62 N |
624 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; |
625 | ||
1ec885cd | 626 | err = read_page(bitmap->storage.file, 0, |
27581e5a | 627 | bitmap, bytes, sb_page); |
f49d5e62 | 628 | } else { |
27581e5a | 629 | err = read_sb_page(bitmap->mddev, |
33e38ac6 | 630 | offset, |
27581e5a | 631 | sb_page, |
938b533d | 632 | 0, sizeof(bitmap_super_t)); |
a654b9d8 | 633 | } |
27581e5a | 634 | if (err) |
32a7627c | 635 | return err; |
32a7627c | 636 | |
b97e9257 | 637 | err = -EINVAL; |
27581e5a | 638 | sb = kmap_atomic(sb_page); |
32a7627c | 639 | |
32a7627c | 640 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 641 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 642 | write_behind = le32_to_cpu(sb->write_behind); |
1dff2b87 | 643 | sectors_reserved = le32_to_cpu(sb->sectors_reserved); |
3c462c88 GR |
644 | /* Setup nodes/clustername only if bitmap version is |
645 | * cluster-compatible | |
d3b178ad | 646 | */ |
3c462c88 | 647 | if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) { |
d3b178ad GR |
648 | nodes = le32_to_cpu(sb->nodes); |
649 | strlcpy(bitmap->mddev->bitmap_info.cluster_name, | |
650 | sb->cluster_name, 64); | |
651 | } | |
32a7627c N |
652 | |
653 | /* verify that the bitmap-specific fields are valid */ | |
654 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
655 | reason = "bad magic"; | |
bd926c63 | 656 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
3c462c88 | 657 | le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED) |
32a7627c | 658 | reason = "unrecognized superblock version"; |
1187cf0a | 659 | else if (chunksize < 512) |
7dd5d34c | 660 | reason = "bitmap chunksize too small"; |
d744540c | 661 | else if (!is_power_of_2(chunksize)) |
32a7627c | 662 | reason = "bitmap chunksize not a power of 2"; |
1b04be96 | 663 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 664 | reason = "daemon sleep period out of range"; |
4b6d287f N |
665 | else if (write_behind > COUNTER_MAX) |
666 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c | 667 | if (reason) { |
ec0cc226 | 668 | pr_warn("%s: invalid bitmap file superblock: %s\n", |
32a7627c N |
669 | bmname(bitmap), reason); |
670 | goto out; | |
671 | } | |
672 | ||
673 | /* keep the array size field of the bitmap superblock up to date */ | |
674 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
675 | ||
278c1ca2 N |
676 | if (bitmap->mddev->persistent) { |
677 | /* | |
678 | * We have a persistent array superblock, so compare the | |
679 | * bitmap's UUID and event counter to the mddev's | |
680 | */ | |
681 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
ec0cc226 N |
682 | pr_warn("%s: bitmap superblock UUID mismatch\n", |
683 | bmname(bitmap)); | |
278c1ca2 N |
684 | goto out; |
685 | } | |
686 | events = le64_to_cpu(sb->events); | |
b97e9257 | 687 | if (!nodes && (events < bitmap->mddev->events)) { |
ec0cc226 N |
688 | pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n", |
689 | bmname(bitmap), events, | |
690 | (unsigned long long) bitmap->mddev->events); | |
b405fe91 | 691 | set_bit(BITMAP_STALE, &bitmap->flags); |
278c1ca2 | 692 | } |
32a7627c | 693 | } |
278c1ca2 | 694 | |
32a7627c | 695 | /* assign fields using values from superblock */ |
4f2e639a | 696 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 | 697 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
b405fe91 | 698 | set_bit(BITMAP_HOSTENDIAN, &bitmap->flags); |
32a7627c | 699 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
cf921cc1 | 700 | strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64); |
32a7627c | 701 | err = 0; |
b97e9257 | 702 | |
32a7627c | 703 | out: |
b2f46e68 | 704 | kunmap_atomic(sb); |
3560741e | 705 | /* Assigning chunksize is required for "re_read" */ |
f9209a32 | 706 | bitmap->mddev->bitmap_info.chunksize = chunksize; |
f7357273 | 707 | if (err == 0 && nodes && (bitmap->cluster_slot < 0)) { |
b97e9257 GR |
708 | err = md_setup_cluster(bitmap->mddev, nodes); |
709 | if (err) { | |
ec0cc226 N |
710 | pr_warn("%s: Could not setup cluster service (%d)\n", |
711 | bmname(bitmap), err); | |
b97e9257 GR |
712 | goto out_no_sb; |
713 | } | |
714 | bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev); | |
b97e9257 GR |
715 | goto re_read; |
716 | } | |
717 | ||
718 | ||
ef99bf48 | 719 | out_no_sb: |
b405fe91 | 720 | if (test_bit(BITMAP_STALE, &bitmap->flags)) |
ef99bf48 N |
721 | bitmap->events_cleared = bitmap->mddev->events; |
722 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
723 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
724 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
c4ce867f | 725 | bitmap->mddev->bitmap_info.nodes = nodes; |
1dff2b87 N |
726 | if (bitmap->mddev->bitmap_info.space == 0 || |
727 | bitmap->mddev->bitmap_info.space > sectors_reserved) | |
728 | bitmap->mddev->bitmap_info.space = sectors_reserved; | |
b97e9257 | 729 | if (err) { |
e64e4018 | 730 | md_bitmap_print_sb(bitmap); |
f9209a32 | 731 | if (bitmap->cluster_slot < 0) |
b97e9257 GR |
732 | md_cluster_stop(bitmap->mddev); |
733 | } | |
32a7627c N |
734 | return err; |
735 | } | |
736 | ||
32a7627c N |
737 | /* |
738 | * general bitmap file operations | |
739 | */ | |
740 | ||
ece5cff0 N |
741 | /* |
742 | * on-disk bitmap: | |
743 | * | |
744 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
745 | * file a page at a time. There's a superblock at the start of the file. | |
746 | */ | |
32a7627c | 747 | /* calculate the index of the page that contains this bit */ |
1ec885cd N |
748 | static inline unsigned long file_page_index(struct bitmap_storage *store, |
749 | unsigned long chunk) | |
32a7627c | 750 | { |
1ec885cd | 751 | if (store->sb_page) |
ece5cff0 N |
752 | chunk += sizeof(bitmap_super_t) << 3; |
753 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
754 | } |
755 | ||
756 | /* calculate the (bit) offset of this bit within a page */ | |
1ec885cd N |
757 | static inline unsigned long file_page_offset(struct bitmap_storage *store, |
758 | unsigned long chunk) | |
32a7627c | 759 | { |
1ec885cd | 760 | if (store->sb_page) |
ece5cff0 N |
761 | chunk += sizeof(bitmap_super_t) << 3; |
762 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
763 | } |
764 | ||
765 | /* | |
766 | * return a pointer to the page in the filemap that contains the given bit | |
767 | * | |
32a7627c | 768 | */ |
1ec885cd | 769 | static inline struct page *filemap_get_page(struct bitmap_storage *store, |
3520fa4d | 770 | unsigned long chunk) |
32a7627c | 771 | { |
1ec885cd | 772 | if (file_page_index(store, chunk) >= store->file_pages) |
ac2f40be | 773 | return NULL; |
f2e06c58 | 774 | return store->filemap[file_page_index(store, chunk)]; |
32a7627c N |
775 | } |
776 | ||
e64e4018 AS |
777 | static int md_bitmap_storage_alloc(struct bitmap_storage *store, |
778 | unsigned long chunks, int with_super, | |
779 | int slot_number) | |
d1244cb0 | 780 | { |
b97e9257 | 781 | int pnum, offset = 0; |
d1244cb0 N |
782 | unsigned long num_pages; |
783 | unsigned long bytes; | |
784 | ||
785 | bytes = DIV_ROUND_UP(chunks, 8); | |
786 | if (with_super) | |
787 | bytes += sizeof(bitmap_super_t); | |
788 | ||
789 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); | |
7f86ffed | 790 | offset = slot_number * num_pages; |
d1244cb0 | 791 | |
6da2ec56 KC |
792 | store->filemap = kmalloc_array(num_pages, sizeof(struct page *), |
793 | GFP_KERNEL); | |
d1244cb0 N |
794 | if (!store->filemap) |
795 | return -ENOMEM; | |
796 | ||
797 | if (with_super && !store->sb_page) { | |
d60b479d | 798 | store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
799 | if (store->sb_page == NULL) |
800 | return -ENOMEM; | |
d1244cb0 | 801 | } |
b97e9257 | 802 | |
d1244cb0 N |
803 | pnum = 0; |
804 | if (store->sb_page) { | |
805 | store->filemap[0] = store->sb_page; | |
806 | pnum = 1; | |
b97e9257 | 807 | store->sb_page->index = offset; |
d1244cb0 | 808 | } |
b97e9257 | 809 | |
d1244cb0 | 810 | for ( ; pnum < num_pages; pnum++) { |
d60b479d | 811 | store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
812 | if (!store->filemap[pnum]) { |
813 | store->file_pages = pnum; | |
814 | return -ENOMEM; | |
815 | } | |
b97e9257 | 816 | store->filemap[pnum]->index = pnum + offset; |
d1244cb0 N |
817 | } |
818 | store->file_pages = pnum; | |
819 | ||
820 | /* We need 4 bits per page, rounded up to a multiple | |
821 | * of sizeof(unsigned long) */ | |
822 | store->filemap_attr = kzalloc( | |
823 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), | |
824 | GFP_KERNEL); | |
825 | if (!store->filemap_attr) | |
826 | return -ENOMEM; | |
827 | ||
828 | store->bytes = bytes; | |
829 | ||
830 | return 0; | |
831 | } | |
832 | ||
e64e4018 | 833 | static void md_bitmap_file_unmap(struct bitmap_storage *store) |
32a7627c N |
834 | { |
835 | struct page **map, *sb_page; | |
32a7627c | 836 | int pages; |
fae7d326 | 837 | struct file *file; |
32a7627c | 838 | |
fae7d326 | 839 | file = store->file; |
1ec885cd | 840 | map = store->filemap; |
1ec885cd | 841 | pages = store->file_pages; |
1ec885cd | 842 | sb_page = store->sb_page; |
32a7627c N |
843 | |
844 | while (pages--) | |
ece5cff0 | 845 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 846 | free_buffers(map[pages]); |
32a7627c | 847 | kfree(map); |
fae7d326 | 848 | kfree(store->filemap_attr); |
32a7627c | 849 | |
d785a06a N |
850 | if (sb_page) |
851 | free_buffers(sb_page); | |
32a7627c | 852 | |
d785a06a | 853 | if (file) { |
496ad9aa | 854 | struct inode *inode = file_inode(file); |
fc0ecff6 | 855 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 856 | fput(file); |
d785a06a | 857 | } |
32a7627c N |
858 | } |
859 | ||
32a7627c N |
860 | /* |
861 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
862 | * then it is no longer reliable, so we stop using it and we mark the file | |
863 | * as failed in the superblock | |
864 | */ | |
e64e4018 | 865 | static void md_bitmap_file_kick(struct bitmap *bitmap) |
32a7627c N |
866 | { |
867 | char *path, *ptr = NULL; | |
868 | ||
b405fe91 | 869 | if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) { |
e64e4018 | 870 | md_bitmap_update_sb(bitmap); |
32a7627c | 871 | |
1ec885cd | 872 | if (bitmap->storage.file) { |
4ad13663 N |
873 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); |
874 | if (path) | |
9bf39ab2 | 875 | ptr = file_path(bitmap->storage.file, |
1ec885cd | 876 | path, PAGE_SIZE); |
6bcfd601 | 877 | |
ec0cc226 N |
878 | pr_warn("%s: kicking failed bitmap file %s from array!\n", |
879 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); | |
32a7627c | 880 | |
4ad13663 N |
881 | kfree(path); |
882 | } else | |
ec0cc226 N |
883 | pr_warn("%s: disabling internal bitmap due to errors\n", |
884 | bmname(bitmap)); | |
a654b9d8 | 885 | } |
32a7627c N |
886 | } |
887 | ||
888 | enum bitmap_page_attr { | |
ac2f40be | 889 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
5a537df4 N |
890 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
891 | * i.e. counter is 1 or 2. */ | |
ac2f40be | 892 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
32a7627c N |
893 | }; |
894 | ||
d189122d N |
895 | static inline void set_page_attr(struct bitmap *bitmap, int pnum, |
896 | enum bitmap_page_attr attr) | |
32a7627c | 897 | { |
bdfd1140 | 898 | set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
899 | } |
900 | ||
d189122d N |
901 | static inline void clear_page_attr(struct bitmap *bitmap, int pnum, |
902 | enum bitmap_page_attr attr) | |
32a7627c | 903 | { |
bdfd1140 | 904 | clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
905 | } |
906 | ||
bdfd1140 N |
907 | static inline int test_page_attr(struct bitmap *bitmap, int pnum, |
908 | enum bitmap_page_attr attr) | |
32a7627c | 909 | { |
1ec885cd | 910 | return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
911 | } |
912 | ||
bdfd1140 N |
913 | static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum, |
914 | enum bitmap_page_attr attr) | |
915 | { | |
916 | return test_and_clear_bit((pnum<<2) + attr, | |
917 | bitmap->storage.filemap_attr); | |
918 | } | |
32a7627c N |
919 | /* |
920 | * bitmap_file_set_bit -- called before performing a write to the md device | |
921 | * to set (and eventually sync) a particular bit in the bitmap file | |
922 | * | |
923 | * we set the bit immediately, then we record the page number so that | |
924 | * when an unplug occurs, we can flush the dirty pages out to disk | |
925 | */ | |
e64e4018 | 926 | static void md_bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) |
32a7627c N |
927 | { |
928 | unsigned long bit; | |
3520fa4d | 929 | struct page *page; |
32a7627c | 930 | void *kaddr; |
40cffcc0 | 931 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
23cea66a GJ |
932 | struct bitmap_storage *store = &bitmap->storage; |
933 | unsigned long node_offset = 0; | |
934 | ||
935 | if (mddev_is_clustered(bitmap->mddev)) | |
936 | node_offset = bitmap->cluster_slot * store->file_pages; | |
32a7627c | 937 | |
1ec885cd | 938 | page = filemap_get_page(&bitmap->storage, chunk); |
3520fa4d JB |
939 | if (!page) |
940 | return; | |
1ec885cd | 941 | bit = file_page_offset(&bitmap->storage, chunk); |
32a7627c | 942 | |
3520fa4d | 943 | /* set the bit */ |
b2f46e68 | 944 | kaddr = kmap_atomic(page); |
b405fe91 | 945 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
3520fa4d JB |
946 | set_bit(bit, kaddr); |
947 | else | |
3f810b6c | 948 | set_bit_le(bit, kaddr); |
b2f46e68 | 949 | kunmap_atomic(kaddr); |
36a4e1fe | 950 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
32a7627c | 951 | /* record page number so it gets flushed to disk when unplug occurs */ |
23cea66a | 952 | set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_DIRTY); |
32a7627c N |
953 | } |
954 | ||
e64e4018 | 955 | static void md_bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) |
ef99bf48 N |
956 | { |
957 | unsigned long bit; | |
958 | struct page *page; | |
959 | void *paddr; | |
40cffcc0 | 960 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
23cea66a GJ |
961 | struct bitmap_storage *store = &bitmap->storage; |
962 | unsigned long node_offset = 0; | |
963 | ||
964 | if (mddev_is_clustered(bitmap->mddev)) | |
965 | node_offset = bitmap->cluster_slot * store->file_pages; | |
ef99bf48 | 966 | |
1ec885cd | 967 | page = filemap_get_page(&bitmap->storage, chunk); |
ef99bf48 N |
968 | if (!page) |
969 | return; | |
1ec885cd | 970 | bit = file_page_offset(&bitmap->storage, chunk); |
ef99bf48 | 971 | paddr = kmap_atomic(page); |
b405fe91 | 972 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ef99bf48 N |
973 | clear_bit(bit, paddr); |
974 | else | |
3f810b6c | 975 | clear_bit_le(bit, paddr); |
ef99bf48 | 976 | kunmap_atomic(paddr); |
23cea66a GJ |
977 | if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) { |
978 | set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING); | |
ef99bf48 N |
979 | bitmap->allclean = 0; |
980 | } | |
981 | } | |
982 | ||
e64e4018 | 983 | static int md_bitmap_file_test_bit(struct bitmap *bitmap, sector_t block) |
11dd35da GR |
984 | { |
985 | unsigned long bit; | |
986 | struct page *page; | |
987 | void *paddr; | |
988 | unsigned long chunk = block >> bitmap->counts.chunkshift; | |
989 | int set = 0; | |
990 | ||
991 | page = filemap_get_page(&bitmap->storage, chunk); | |
992 | if (!page) | |
993 | return -EINVAL; | |
994 | bit = file_page_offset(&bitmap->storage, chunk); | |
995 | paddr = kmap_atomic(page); | |
996 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) | |
997 | set = test_bit(bit, paddr); | |
998 | else | |
999 | set = test_bit_le(bit, paddr); | |
1000 | kunmap_atomic(paddr); | |
1001 | return set; | |
1002 | } | |
1003 | ||
1004 | ||
32a7627c N |
1005 | /* this gets called when the md device is ready to unplug its underlying |
1006 | * (slave) device queues -- before we let any writes go down, we need to | |
1007 | * sync the dirty pages of the bitmap file to disk */ | |
e64e4018 | 1008 | void md_bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 1009 | { |
74667123 | 1010 | unsigned long i; |
ec7a3197 | 1011 | int dirty, need_write; |
85c9ccd4 | 1012 | int writing = 0; |
32a7627c | 1013 | |
62f82faa N |
1014 | if (!bitmap || !bitmap->storage.filemap || |
1015 | test_bit(BITMAP_STALE, &bitmap->flags)) | |
4ad13663 | 1016 | return; |
32a7627c N |
1017 | |
1018 | /* look at each page to see if there are any set bits that need to be | |
1019 | * flushed out to disk */ | |
1ec885cd | 1020 | for (i = 0; i < bitmap->storage.file_pages; i++) { |
bdfd1140 | 1021 | if (!bitmap->storage.filemap) |
4ad13663 | 1022 | return; |
bdfd1140 N |
1023 | dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
1024 | need_write = test_and_clear_page_attr(bitmap, i, | |
1025 | BITMAP_PAGE_NEEDWRITE); | |
1026 | if (dirty || need_write) { | |
581dbd94 | 1027 | if (!writing) { |
e64e4018 | 1028 | md_bitmap_wait_writes(bitmap); |
581dbd94 N |
1029 | if (bitmap->mddev->queue) |
1030 | blk_add_trace_msg(bitmap->mddev->queue, | |
1031 | "md bitmap_unplug"); | |
1032 | } | |
d189122d | 1033 | clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); |
bdfd1140 | 1034 | write_page(bitmap, bitmap->storage.filemap[i], 0); |
85c9ccd4 | 1035 | writing = 1; |
bdfd1140 | 1036 | } |
32a7627c | 1037 | } |
85c9ccd4 | 1038 | if (writing) |
e64e4018 | 1039 | md_bitmap_wait_writes(bitmap); |
4b5060dd | 1040 | |
b405fe91 | 1041 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
e64e4018 | 1042 | md_bitmap_file_kick(bitmap); |
32a7627c | 1043 | } |
e64e4018 | 1044 | EXPORT_SYMBOL(md_bitmap_unplug); |
32a7627c | 1045 | |
e64e4018 | 1046 | static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
1047 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
1048 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
1049 | * memory mapping of the bitmap file | |
1050 | * Special cases: | |
1051 | * if there's no bitmap file, or if the bitmap file had been | |
1052 | * previously kicked from the array, we mark all the bits as | |
1053 | * 1's in order to cause a full resync. | |
6a07997f N |
1054 | * |
1055 | * We ignore all bits for sectors that end earlier than 'start'. | |
1056 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 1057 | */ |
e64e4018 | 1058 | static int md_bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c | 1059 | { |
b97e9257 | 1060 | unsigned long i, chunks, index, oldindex, bit, node_offset = 0; |
27581e5a | 1061 | struct page *page = NULL; |
d1244cb0 | 1062 | unsigned long bit_cnt = 0; |
32a7627c | 1063 | struct file *file; |
d1244cb0 | 1064 | unsigned long offset; |
32a7627c N |
1065 | int outofdate; |
1066 | int ret = -ENOSPC; | |
ea03aff9 | 1067 | void *paddr; |
1ec885cd | 1068 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c | 1069 | |
40cffcc0 | 1070 | chunks = bitmap->counts.chunks; |
1ec885cd | 1071 | file = store->file; |
32a7627c | 1072 | |
ef99bf48 N |
1073 | if (!file && !bitmap->mddev->bitmap_info.offset) { |
1074 | /* No permanent bitmap - fill with '1s'. */ | |
1ec885cd N |
1075 | store->filemap = NULL; |
1076 | store->file_pages = 0; | |
ef99bf48 N |
1077 | for (i = 0; i < chunks ; i++) { |
1078 | /* if the disk bit is set, set the memory bit */ | |
40cffcc0 | 1079 | int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift) |
ef99bf48 | 1080 | >= start); |
e64e4018 AS |
1081 | md_bitmap_set_memory_bits(bitmap, |
1082 | (sector_t)i << bitmap->counts.chunkshift, | |
1083 | needed); | |
ef99bf48 N |
1084 | } |
1085 | return 0; | |
1086 | } | |
32a7627c | 1087 | |
b405fe91 | 1088 | outofdate = test_bit(BITMAP_STALE, &bitmap->flags); |
32a7627c | 1089 | if (outofdate) |
ec0cc226 | 1090 | pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap)); |
32a7627c | 1091 | |
d1244cb0 | 1092 | if (file && i_size_read(file->f_mapping->host) < store->bytes) { |
ec0cc226 N |
1093 | pr_warn("%s: bitmap file too short %lu < %lu\n", |
1094 | bmname(bitmap), | |
1095 | (unsigned long) i_size_read(file->f_mapping->host), | |
1096 | store->bytes); | |
4ad13663 | 1097 | goto err; |
32a7627c | 1098 | } |
bc7f77de | 1099 | |
d1244cb0 | 1100 | oldindex = ~0L; |
27581e5a | 1101 | offset = 0; |
d1244cb0 | 1102 | if (!bitmap->mddev->bitmap_info.external) |
27581e5a | 1103 | offset = sizeof(bitmap_super_t); |
32a7627c | 1104 | |
b97e9257 GR |
1105 | if (mddev_is_clustered(bitmap->mddev)) |
1106 | node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE)); | |
1107 | ||
32a7627c | 1108 | for (i = 0; i < chunks; i++) { |
bd926c63 | 1109 | int b; |
1ec885cd N |
1110 | index = file_page_index(&bitmap->storage, i); |
1111 | bit = file_page_offset(&bitmap->storage, i); | |
32a7627c | 1112 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 1113 | int count; |
32a7627c | 1114 | /* unmap the old page, we're done with it */ |
d1244cb0 N |
1115 | if (index == store->file_pages-1) |
1116 | count = store->bytes - index * PAGE_SIZE; | |
d785a06a N |
1117 | else |
1118 | count = PAGE_SIZE; | |
1ec885cd | 1119 | page = store->filemap[index]; |
27581e5a N |
1120 | if (file) |
1121 | ret = read_page(file, index, bitmap, | |
1122 | count, page); | |
1123 | else | |
1124 | ret = read_sb_page( | |
1125 | bitmap->mddev, | |
1126 | bitmap->mddev->bitmap_info.offset, | |
1127 | page, | |
b97e9257 | 1128 | index + node_offset, count); |
27581e5a N |
1129 | |
1130 | if (ret) | |
4ad13663 | 1131 | goto err; |
a654b9d8 | 1132 | |
32a7627c | 1133 | oldindex = index; |
32a7627c N |
1134 | |
1135 | if (outofdate) { | |
1136 | /* | |
1137 | * if bitmap is out of date, dirty the | |
ac2f40be | 1138 | * whole page and write it out |
32a7627c | 1139 | */ |
b2f46e68 | 1140 | paddr = kmap_atomic(page); |
ea03aff9 | 1141 | memset(paddr + offset, 0xff, |
6a07997f | 1142 | PAGE_SIZE - offset); |
b2f46e68 | 1143 | kunmap_atomic(paddr); |
4ad13663 N |
1144 | write_page(bitmap, page, 1); |
1145 | ||
1146 | ret = -EIO; | |
b405fe91 N |
1147 | if (test_bit(BITMAP_WRITE_ERROR, |
1148 | &bitmap->flags)) | |
4ad13663 | 1149 | goto err; |
32a7627c | 1150 | } |
32a7627c | 1151 | } |
b2f46e68 | 1152 | paddr = kmap_atomic(page); |
b405fe91 | 1153 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ea03aff9 | 1154 | b = test_bit(bit, paddr); |
bd926c63 | 1155 | else |
6b33aff3 | 1156 | b = test_bit_le(bit, paddr); |
b2f46e68 | 1157 | kunmap_atomic(paddr); |
bd926c63 | 1158 | if (b) { |
32a7627c | 1159 | /* if the disk bit is set, set the memory bit */ |
40cffcc0 | 1160 | int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift |
db305e50 | 1161 | >= start); |
e64e4018 AS |
1162 | md_bitmap_set_memory_bits(bitmap, |
1163 | (sector_t)i << bitmap->counts.chunkshift, | |
1164 | needed); | |
32a7627c N |
1165 | bit_cnt++; |
1166 | } | |
27581e5a | 1167 | offset = 0; |
32a7627c N |
1168 | } |
1169 | ||
ec0cc226 N |
1170 | pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n", |
1171 | bmname(bitmap), store->file_pages, | |
1172 | bit_cnt, chunks); | |
4ad13663 N |
1173 | |
1174 | return 0; | |
32a7627c | 1175 | |
4ad13663 | 1176 | err: |
ec0cc226 N |
1177 | pr_warn("%s: bitmap initialisation failed: %d\n", |
1178 | bmname(bitmap), ret); | |
32a7627c N |
1179 | return ret; |
1180 | } | |
1181 | ||
e64e4018 | 1182 | void md_bitmap_write_all(struct bitmap *bitmap) |
a654b9d8 N |
1183 | { |
1184 | /* We don't actually write all bitmap blocks here, | |
1185 | * just flag them as needing to be written | |
1186 | */ | |
ec7a3197 | 1187 | int i; |
a654b9d8 | 1188 | |
1ec885cd | 1189 | if (!bitmap || !bitmap->storage.filemap) |
ef99bf48 | 1190 | return; |
1ec885cd | 1191 | if (bitmap->storage.file) |
ef99bf48 N |
1192 | /* Only one copy, so nothing needed */ |
1193 | return; | |
1194 | ||
1ec885cd | 1195 | for (i = 0; i < bitmap->storage.file_pages; i++) |
d189122d | 1196 | set_page_attr(bitmap, i, |
ec7a3197 | 1197 | BITMAP_PAGE_NEEDWRITE); |
2585f3ef | 1198 | bitmap->allclean = 0; |
a654b9d8 N |
1199 | } |
1200 | ||
e64e4018 AS |
1201 | static void md_bitmap_count_page(struct bitmap_counts *bitmap, |
1202 | sector_t offset, int inc) | |
32a7627c | 1203 | { |
61a0d80c | 1204 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1205 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1206 | bitmap->bp[page].count += inc; | |
e64e4018 | 1207 | md_bitmap_checkfree(bitmap, page); |
32a7627c | 1208 | } |
bf07bb7d | 1209 | |
e64e4018 | 1210 | static void md_bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset) |
bf07bb7d N |
1211 | { |
1212 | sector_t chunk = offset >> bitmap->chunkshift; | |
1213 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1214 | struct bitmap_page *bp = &bitmap->bp[page]; | |
1215 | ||
1216 | if (!bp->pending) | |
1217 | bp->pending = 1; | |
1218 | } | |
1219 | ||
e64e4018 AS |
1220 | static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap, |
1221 | sector_t offset, sector_t *blocks, | |
1222 | int create); | |
32a7627c N |
1223 | |
1224 | /* | |
1225 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1226 | * out to disk | |
1227 | */ | |
1228 | ||
e64e4018 | 1229 | void md_bitmap_daemon_work(struct mddev *mddev) |
32a7627c | 1230 | { |
aa5cbd10 | 1231 | struct bitmap *bitmap; |
aa3163f8 | 1232 | unsigned long j; |
bf07bb7d | 1233 | unsigned long nextpage; |
57dab0bd | 1234 | sector_t blocks; |
40cffcc0 | 1235 | struct bitmap_counts *counts; |
32a7627c | 1236 | |
aa5cbd10 N |
1237 | /* Use a mutex to guard daemon_work against |
1238 | * bitmap_destroy. | |
1239 | */ | |
c3d9714e | 1240 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1241 | bitmap = mddev->bitmap; |
1242 | if (bitmap == NULL) { | |
c3d9714e | 1243 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1244 | return; |
aa5cbd10 | 1245 | } |
42a04b50 | 1246 | if (time_before(jiffies, bitmap->daemon_lastrun |
2e61ebbc | 1247 | + mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1248 | goto done; |
1249 | ||
32a7627c | 1250 | bitmap->daemon_lastrun = jiffies; |
8311c29d | 1251 | if (bitmap->allclean) { |
2e61ebbc | 1252 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
aa5cbd10 | 1253 | goto done; |
8311c29d N |
1254 | } |
1255 | bitmap->allclean = 1; | |
32a7627c | 1256 | |
581dbd94 N |
1257 | if (bitmap->mddev->queue) |
1258 | blk_add_trace_msg(bitmap->mddev->queue, | |
1259 | "md bitmap_daemon_work"); | |
1260 | ||
bf07bb7d N |
1261 | /* Any file-page which is PENDING now needs to be written. |
1262 | * So set NEEDWRITE now, then after we make any last-minute changes | |
1263 | * we will write it. | |
1264 | */ | |
1ec885cd | 1265 | for (j = 0; j < bitmap->storage.file_pages; j++) |
bdfd1140 N |
1266 | if (test_and_clear_page_attr(bitmap, j, |
1267 | BITMAP_PAGE_PENDING)) | |
d189122d | 1268 | set_page_attr(bitmap, j, |
bf07bb7d | 1269 | BITMAP_PAGE_NEEDWRITE); |
bf07bb7d N |
1270 | |
1271 | if (bitmap->need_sync && | |
1272 | mddev->bitmap_info.external == 0) { | |
1273 | /* Arrange for superblock update as well as | |
1274 | * other changes */ | |
1275 | bitmap_super_t *sb; | |
1276 | bitmap->need_sync = 0; | |
1ec885cd N |
1277 | if (bitmap->storage.filemap) { |
1278 | sb = kmap_atomic(bitmap->storage.sb_page); | |
ef99bf48 N |
1279 | sb->events_cleared = |
1280 | cpu_to_le64(bitmap->events_cleared); | |
1281 | kunmap_atomic(sb); | |
d189122d | 1282 | set_page_attr(bitmap, 0, |
ef99bf48 N |
1283 | BITMAP_PAGE_NEEDWRITE); |
1284 | } | |
bf07bb7d N |
1285 | } |
1286 | /* Now look at the bitmap counters and if any are '2' or '1', | |
1287 | * decrement and handle accordingly. | |
1288 | */ | |
40cffcc0 N |
1289 | counts = &bitmap->counts; |
1290 | spin_lock_irq(&counts->lock); | |
bf07bb7d | 1291 | nextpage = 0; |
40cffcc0 | 1292 | for (j = 0; j < counts->chunks; j++) { |
32a7627c | 1293 | bitmap_counter_t *bmc; |
40cffcc0 | 1294 | sector_t block = (sector_t)j << counts->chunkshift; |
3520fa4d | 1295 | |
bf07bb7d N |
1296 | if (j == nextpage) { |
1297 | nextpage += PAGE_COUNTER_RATIO; | |
40cffcc0 | 1298 | if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) { |
bf07bb7d | 1299 | j |= PAGE_COUNTER_MASK; |
aa3163f8 N |
1300 | continue; |
1301 | } | |
40cffcc0 | 1302 | counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; |
32a7627c | 1303 | } |
bf07bb7d | 1304 | |
e64e4018 | 1305 | bmc = md_bitmap_get_counter(counts, block, &blocks, 0); |
bf07bb7d | 1306 | if (!bmc) { |
5a537df4 | 1307 | j |= PAGE_COUNTER_MASK; |
bf07bb7d N |
1308 | continue; |
1309 | } | |
1310 | if (*bmc == 1 && !bitmap->need_sync) { | |
1311 | /* We can clear the bit */ | |
bf07bb7d | 1312 | *bmc = 0; |
e64e4018 AS |
1313 | md_bitmap_count_page(counts, block, -1); |
1314 | md_bitmap_file_clear_bit(bitmap, block); | |
bf07bb7d N |
1315 | } else if (*bmc && *bmc <= 2) { |
1316 | *bmc = 1; | |
e64e4018 | 1317 | md_bitmap_set_pending(counts, block); |
bf07bb7d | 1318 | bitmap->allclean = 0; |
5a537df4 | 1319 | } |
32a7627c | 1320 | } |
40cffcc0 | 1321 | spin_unlock_irq(&counts->lock); |
32a7627c | 1322 | |
e64e4018 | 1323 | md_bitmap_wait_writes(bitmap); |
bf07bb7d N |
1324 | /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. |
1325 | * DIRTY pages need to be written by bitmap_unplug so it can wait | |
1326 | * for them. | |
1327 | * If we find any DIRTY page we stop there and let bitmap_unplug | |
1328 | * handle all the rest. This is important in the case where | |
1329 | * the first blocking holds the superblock and it has been updated. | |
1330 | * We mustn't write any other blocks before the superblock. | |
1331 | */ | |
62f82faa N |
1332 | for (j = 0; |
1333 | j < bitmap->storage.file_pages | |
1334 | && !test_bit(BITMAP_STALE, &bitmap->flags); | |
1335 | j++) { | |
d189122d | 1336 | if (test_page_attr(bitmap, j, |
bf07bb7d N |
1337 | BITMAP_PAGE_DIRTY)) |
1338 | /* bitmap_unplug will handle the rest */ | |
1339 | break; | |
bdfd1140 N |
1340 | if (test_and_clear_page_attr(bitmap, j, |
1341 | BITMAP_PAGE_NEEDWRITE)) { | |
1ec885cd | 1342 | write_page(bitmap, bitmap->storage.filemap[j], 0); |
32a7627c | 1343 | } |
32a7627c N |
1344 | } |
1345 | ||
7be3dfec | 1346 | done: |
8311c29d | 1347 | if (bitmap->allclean == 0) |
2e61ebbc N |
1348 | mddev->thread->timeout = |
1349 | mddev->bitmap_info.daemon_sleep; | |
c3d9714e | 1350 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1351 | } |
1352 | ||
e64e4018 AS |
1353 | static bitmap_counter_t *md_bitmap_get_counter(struct bitmap_counts *bitmap, |
1354 | sector_t offset, sector_t *blocks, | |
1355 | int create) | |
ee305ace N |
1356 | __releases(bitmap->lock) |
1357 | __acquires(bitmap->lock) | |
32a7627c N |
1358 | { |
1359 | /* If 'create', we might release the lock and reclaim it. | |
1360 | * The lock must have been taken with interrupts enabled. | |
1361 | * If !create, we don't release the lock. | |
1362 | */ | |
61a0d80c | 1363 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1364 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1365 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1366 | sector_t csize; | |
ef425673 | 1367 | int err; |
32a7627c | 1368 | |
e64e4018 | 1369 | err = md_bitmap_checkpage(bitmap, page, create, 0); |
ef425673 N |
1370 | |
1371 | if (bitmap->bp[page].hijacked || | |
1372 | bitmap->bp[page].map == NULL) | |
61a0d80c | 1373 | csize = ((sector_t)1) << (bitmap->chunkshift + |
ef425673 N |
1374 | PAGE_COUNTER_SHIFT - 1); |
1375 | else | |
61a0d80c | 1376 | csize = ((sector_t)1) << bitmap->chunkshift; |
ef425673 N |
1377 | *blocks = csize - (offset & (csize - 1)); |
1378 | ||
1379 | if (err < 0) | |
32a7627c | 1380 | return NULL; |
ef425673 | 1381 | |
32a7627c N |
1382 | /* now locked ... */ |
1383 | ||
1384 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1385 | /* should we use the first or second counter field | |
1386 | * of the hijacked pointer? */ | |
1387 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1388 | return &((bitmap_counter_t *) |
1389 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1390 | } else /* page is allocated */ |
32a7627c N |
1391 | return (bitmap_counter_t *) |
1392 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1393 | } |
1394 | ||
e64e4018 | 1395 | int md_bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1396 | { |
ac2f40be N |
1397 | if (!bitmap) |
1398 | return 0; | |
4b6d287f N |
1399 | |
1400 | if (behind) { | |
696fcd53 | 1401 | int bw; |
4b6d287f | 1402 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1403 | bw = atomic_read(&bitmap->behind_writes); |
1404 | if (bw > bitmap->behind_writes_used) | |
1405 | bitmap->behind_writes_used = bw; | |
1406 | ||
36a4e1fe N |
1407 | pr_debug("inc write-behind count %d/%lu\n", |
1408 | bw, bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1409 | } |
1410 | ||
32a7627c | 1411 | while (sectors) { |
57dab0bd | 1412 | sector_t blocks; |
32a7627c N |
1413 | bitmap_counter_t *bmc; |
1414 | ||
40cffcc0 | 1415 | spin_lock_irq(&bitmap->counts.lock); |
e64e4018 | 1416 | bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 1); |
32a7627c | 1417 | if (!bmc) { |
40cffcc0 | 1418 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1419 | return 0; |
1420 | } | |
1421 | ||
27d5ea04 | 1422 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
da6e1a32 NB |
1423 | DEFINE_WAIT(__wait); |
1424 | /* note that it is safe to do the prepare_to_wait | |
1425 | * after the test as long as we do it before dropping | |
1426 | * the spinlock. | |
1427 | */ | |
1428 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1429 | TASK_UNINTERRUPTIBLE); | |
40cffcc0 | 1430 | spin_unlock_irq(&bitmap->counts.lock); |
f54a9d0e | 1431 | schedule(); |
da6e1a32 NB |
1432 | finish_wait(&bitmap->overflow_wait, &__wait); |
1433 | continue; | |
1434 | } | |
1435 | ||
ac2f40be | 1436 | switch (*bmc) { |
32a7627c | 1437 | case 0: |
e64e4018 AS |
1438 | md_bitmap_file_set_bit(bitmap, offset); |
1439 | md_bitmap_count_page(&bitmap->counts, offset, 1); | |
32a7627c N |
1440 | /* fall through */ |
1441 | case 1: | |
1442 | *bmc = 2; | |
1443 | } | |
da6e1a32 | 1444 | |
32a7627c N |
1445 | (*bmc)++; |
1446 | ||
40cffcc0 | 1447 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1448 | |
1449 | offset += blocks; | |
1450 | if (sectors > blocks) | |
1451 | sectors -= blocks; | |
ac2f40be N |
1452 | else |
1453 | sectors = 0; | |
32a7627c N |
1454 | } |
1455 | return 0; | |
1456 | } | |
e64e4018 | 1457 | EXPORT_SYMBOL(md_bitmap_startwrite); |
32a7627c | 1458 | |
e64e4018 AS |
1459 | void md_bitmap_endwrite(struct bitmap *bitmap, sector_t offset, |
1460 | unsigned long sectors, int success, int behind) | |
32a7627c | 1461 | { |
ac2f40be N |
1462 | if (!bitmap) |
1463 | return; | |
4b6d287f | 1464 | if (behind) { |
e555190d N |
1465 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1466 | wake_up(&bitmap->behind_wait); | |
36a4e1fe N |
1467 | pr_debug("dec write-behind count %d/%lu\n", |
1468 | atomic_read(&bitmap->behind_writes), | |
1469 | bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1470 | } |
1471 | ||
32a7627c | 1472 | while (sectors) { |
57dab0bd | 1473 | sector_t blocks; |
32a7627c N |
1474 | unsigned long flags; |
1475 | bitmap_counter_t *bmc; | |
1476 | ||
40cffcc0 | 1477 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
e64e4018 | 1478 | bmc = md_bitmap_get_counter(&bitmap->counts, offset, &blocks, 0); |
32a7627c | 1479 | if (!bmc) { |
40cffcc0 | 1480 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1481 | return; |
1482 | } | |
1483 | ||
961902c0 | 1484 | if (success && !bitmap->mddev->degraded && |
a0da84f3 NB |
1485 | bitmap->events_cleared < bitmap->mddev->events) { |
1486 | bitmap->events_cleared = bitmap->mddev->events; | |
1487 | bitmap->need_sync = 1; | |
5ff5afff | 1488 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1489 | } |
1490 | ||
27d5ea04 | 1491 | if (!success && !NEEDED(*bmc)) |
32a7627c N |
1492 | *bmc |= NEEDED_MASK; |
1493 | ||
27d5ea04 | 1494 | if (COUNTER(*bmc) == COUNTER_MAX) |
da6e1a32 NB |
1495 | wake_up(&bitmap->overflow_wait); |
1496 | ||
32a7627c | 1497 | (*bmc)--; |
2585f3ef | 1498 | if (*bmc <= 2) { |
e64e4018 | 1499 | md_bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1500 | bitmap->allclean = 0; |
1501 | } | |
40cffcc0 | 1502 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1503 | offset += blocks; |
1504 | if (sectors > blocks) | |
1505 | sectors -= blocks; | |
ac2f40be N |
1506 | else |
1507 | sectors = 0; | |
32a7627c N |
1508 | } |
1509 | } | |
e64e4018 | 1510 | EXPORT_SYMBOL(md_bitmap_endwrite); |
32a7627c | 1511 | |
57dab0bd | 1512 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1513 | int degraded) |
32a7627c N |
1514 | { |
1515 | bitmap_counter_t *bmc; | |
1516 | int rv; | |
1517 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1518 | *blocks = 1024; | |
1519 | return 1; /* always resync if no bitmap */ | |
1520 | } | |
40cffcc0 | 1521 | spin_lock_irq(&bitmap->counts.lock); |
e64e4018 | 1522 | bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0); |
32a7627c N |
1523 | rv = 0; |
1524 | if (bmc) { | |
1525 | /* locked */ | |
1526 | if (RESYNC(*bmc)) | |
1527 | rv = 1; | |
1528 | else if (NEEDED(*bmc)) { | |
1529 | rv = 1; | |
6a806c51 N |
1530 | if (!degraded) { /* don't set/clear bits if degraded */ |
1531 | *bmc |= RESYNC_MASK; | |
1532 | *bmc &= ~NEEDED_MASK; | |
1533 | } | |
32a7627c N |
1534 | } |
1535 | } | |
40cffcc0 | 1536 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1537 | return rv; |
1538 | } | |
1539 | ||
e64e4018 AS |
1540 | int md_bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1541 | int degraded) | |
1187cf0a N |
1542 | { |
1543 | /* bitmap_start_sync must always report on multiples of whole | |
1544 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1545 | * get confused. | |
1546 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1547 | * At least PAGE_SIZE>>9 blocks are covered. | |
1548 | * Return the 'or' of the result. | |
1549 | */ | |
1550 | int rv = 0; | |
57dab0bd | 1551 | sector_t blocks1; |
1187cf0a N |
1552 | |
1553 | *blocks = 0; | |
1554 | while (*blocks < (PAGE_SIZE>>9)) { | |
1555 | rv |= __bitmap_start_sync(bitmap, offset, | |
1556 | &blocks1, degraded); | |
1557 | offset += blocks1; | |
1558 | *blocks += blocks1; | |
1559 | } | |
1560 | return rv; | |
1561 | } | |
e64e4018 | 1562 | EXPORT_SYMBOL(md_bitmap_start_sync); |
1187cf0a | 1563 | |
e64e4018 | 1564 | void md_bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1565 | { |
1566 | bitmap_counter_t *bmc; | |
1567 | unsigned long flags; | |
ac2f40be N |
1568 | |
1569 | if (bitmap == NULL) { | |
32a7627c N |
1570 | *blocks = 1024; |
1571 | return; | |
1572 | } | |
40cffcc0 | 1573 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
e64e4018 | 1574 | bmc = md_bitmap_get_counter(&bitmap->counts, offset, blocks, 0); |
32a7627c N |
1575 | if (bmc == NULL) |
1576 | goto unlock; | |
1577 | /* locked */ | |
32a7627c N |
1578 | if (RESYNC(*bmc)) { |
1579 | *bmc &= ~RESYNC_MASK; | |
1580 | ||
1581 | if (!NEEDED(*bmc) && aborted) | |
1582 | *bmc |= NEEDED_MASK; | |
1583 | else { | |
2585f3ef | 1584 | if (*bmc <= 2) { |
e64e4018 | 1585 | md_bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1586 | bitmap->allclean = 0; |
1587 | } | |
32a7627c N |
1588 | } |
1589 | } | |
1590 | unlock: | |
40cffcc0 | 1591 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c | 1592 | } |
e64e4018 | 1593 | EXPORT_SYMBOL(md_bitmap_end_sync); |
32a7627c | 1594 | |
e64e4018 | 1595 | void md_bitmap_close_sync(struct bitmap *bitmap) |
32a7627c N |
1596 | { |
1597 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1598 | * properly have been aborted. It remains to us to clear the | |
1599 | * RESYNC bit wherever it is still on | |
1600 | */ | |
1601 | sector_t sector = 0; | |
57dab0bd | 1602 | sector_t blocks; |
b47490c9 N |
1603 | if (!bitmap) |
1604 | return; | |
32a7627c | 1605 | while (sector < bitmap->mddev->resync_max_sectors) { |
e64e4018 | 1606 | md_bitmap_end_sync(bitmap, sector, &blocks, 0); |
b47490c9 N |
1607 | sector += blocks; |
1608 | } | |
1609 | } | |
e64e4018 | 1610 | EXPORT_SYMBOL(md_bitmap_close_sync); |
b47490c9 | 1611 | |
e64e4018 | 1612 | void md_bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force) |
b47490c9 N |
1613 | { |
1614 | sector_t s = 0; | |
57dab0bd | 1615 | sector_t blocks; |
b47490c9 N |
1616 | |
1617 | if (!bitmap) | |
1618 | return; | |
1619 | if (sector == 0) { | |
1620 | bitmap->last_end_sync = jiffies; | |
1621 | return; | |
1622 | } | |
c40f341f | 1623 | if (!force && time_before(jiffies, (bitmap->last_end_sync |
1b04be96 | 1624 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1625 | return; |
1626 | wait_event(bitmap->mddev->recovery_wait, | |
1627 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1628 | ||
75d3da43 | 1629 | bitmap->mddev->curr_resync_completed = sector; |
2953079c | 1630 | set_bit(MD_SB_CHANGE_CLEAN, &bitmap->mddev->sb_flags); |
40cffcc0 | 1631 | sector &= ~((1ULL << bitmap->counts.chunkshift) - 1); |
b47490c9 N |
1632 | s = 0; |
1633 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
e64e4018 | 1634 | md_bitmap_end_sync(bitmap, s, &blocks, 0); |
b47490c9 | 1635 | s += blocks; |
32a7627c | 1636 | } |
b47490c9 | 1637 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1638 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1639 | } |
e64e4018 | 1640 | EXPORT_SYMBOL(md_bitmap_cond_end_sync); |
32a7627c | 1641 | |
e64e4018 | 1642 | void md_bitmap_sync_with_cluster(struct mddev *mddev, |
18c9ff7f GJ |
1643 | sector_t old_lo, sector_t old_hi, |
1644 | sector_t new_lo, sector_t new_hi) | |
1645 | { | |
1646 | struct bitmap *bitmap = mddev->bitmap; | |
1647 | sector_t sector, blocks = 0; | |
1648 | ||
1649 | for (sector = old_lo; sector < new_lo; ) { | |
e64e4018 | 1650 | md_bitmap_end_sync(bitmap, sector, &blocks, 0); |
18c9ff7f GJ |
1651 | sector += blocks; |
1652 | } | |
1653 | WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n"); | |
1654 | ||
1655 | for (sector = old_hi; sector < new_hi; ) { | |
e64e4018 | 1656 | md_bitmap_start_sync(bitmap, sector, &blocks, 0); |
18c9ff7f GJ |
1657 | sector += blocks; |
1658 | } | |
1659 | WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n"); | |
1660 | } | |
e64e4018 | 1661 | EXPORT_SYMBOL(md_bitmap_sync_with_cluster); |
18c9ff7f | 1662 | |
e64e4018 | 1663 | static void md_bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1664 | { |
1665 | /* For each chunk covered by any of these sectors, set the | |
ef99bf48 | 1666 | * counter to 2 and possibly set resync_needed. They should all |
32a7627c N |
1667 | * be 0 at this point |
1668 | */ | |
193f1c93 | 1669 | |
57dab0bd | 1670 | sector_t secs; |
193f1c93 | 1671 | bitmap_counter_t *bmc; |
40cffcc0 | 1672 | spin_lock_irq(&bitmap->counts.lock); |
e64e4018 | 1673 | bmc = md_bitmap_get_counter(&bitmap->counts, offset, &secs, 1); |
193f1c93 | 1674 | if (!bmc) { |
40cffcc0 | 1675 | spin_unlock_irq(&bitmap->counts.lock); |
193f1c93 | 1676 | return; |
32a7627c | 1677 | } |
ac2f40be | 1678 | if (!*bmc) { |
11dd35da | 1679 | *bmc = 2; |
e64e4018 AS |
1680 | md_bitmap_count_page(&bitmap->counts, offset, 1); |
1681 | md_bitmap_set_pending(&bitmap->counts, offset); | |
2585f3ef | 1682 | bitmap->allclean = 0; |
193f1c93 | 1683 | } |
11dd35da GR |
1684 | if (needed) |
1685 | *bmc |= NEEDED_MASK; | |
40cffcc0 | 1686 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1687 | } |
1688 | ||
9b1d1dac | 1689 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
e64e4018 | 1690 | void md_bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) |
9b1d1dac PC |
1691 | { |
1692 | unsigned long chunk; | |
1693 | ||
1694 | for (chunk = s; chunk <= e; chunk++) { | |
40cffcc0 | 1695 | sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift; |
e64e4018 AS |
1696 | md_bitmap_set_memory_bits(bitmap, sec, 1); |
1697 | md_bitmap_file_set_bit(bitmap, sec); | |
ffa23322 N |
1698 | if (sec < bitmap->mddev->recovery_cp) |
1699 | /* We are asserting that the array is dirty, | |
1700 | * so move the recovery_cp address back so | |
1701 | * that it is obvious that it is dirty | |
1702 | */ | |
1703 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1704 | } |
1705 | } | |
1706 | ||
6b8b3e8a N |
1707 | /* |
1708 | * flush out any pending updates | |
1709 | */ | |
e64e4018 | 1710 | void md_bitmap_flush(struct mddev *mddev) |
6b8b3e8a N |
1711 | { |
1712 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1713 | long sleep; |
6b8b3e8a N |
1714 | |
1715 | if (!bitmap) /* there was no bitmap */ | |
1716 | return; | |
1717 | ||
1718 | /* run the daemon_work three time to ensure everything is flushed | |
1719 | * that can be | |
1720 | */ | |
1b04be96 | 1721 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1722 | bitmap->daemon_lastrun -= sleep; |
e64e4018 | 1723 | md_bitmap_daemon_work(mddev); |
42a04b50 | 1724 | bitmap->daemon_lastrun -= sleep; |
e64e4018 | 1725 | md_bitmap_daemon_work(mddev); |
42a04b50 | 1726 | bitmap->daemon_lastrun -= sleep; |
e64e4018 AS |
1727 | md_bitmap_daemon_work(mddev); |
1728 | md_bitmap_update_sb(bitmap); | |
6b8b3e8a N |
1729 | } |
1730 | ||
32a7627c N |
1731 | /* |
1732 | * free memory that was allocated | |
1733 | */ | |
e64e4018 | 1734 | void md_bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1735 | { |
1736 | unsigned long k, pages; | |
1737 | struct bitmap_page *bp; | |
32a7627c N |
1738 | |
1739 | if (!bitmap) /* there was no bitmap */ | |
1740 | return; | |
1741 | ||
f9a67b11 GJ |
1742 | if (bitmap->sysfs_can_clear) |
1743 | sysfs_put(bitmap->sysfs_can_clear); | |
1744 | ||
f9209a32 GR |
1745 | if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info && |
1746 | bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev)) | |
b97e9257 GR |
1747 | md_cluster_stop(bitmap->mddev); |
1748 | ||
fae7d326 N |
1749 | /* Shouldn't be needed - but just in case.... */ |
1750 | wait_event(bitmap->write_wait, | |
1751 | atomic_read(&bitmap->pending_writes) == 0); | |
1752 | ||
1753 | /* release the bitmap file */ | |
e64e4018 | 1754 | md_bitmap_file_unmap(&bitmap->storage); |
32a7627c | 1755 | |
40cffcc0 N |
1756 | bp = bitmap->counts.bp; |
1757 | pages = bitmap->counts.pages; | |
32a7627c N |
1758 | |
1759 | /* free all allocated memory */ | |
1760 | ||
32a7627c N |
1761 | if (bp) /* deallocate the page memory */ |
1762 | for (k = 0; k < pages; k++) | |
1763 | if (bp[k].map && !bp[k].hijacked) | |
1764 | kfree(bp[k].map); | |
1765 | kfree(bp); | |
1766 | kfree(bitmap); | |
1767 | } | |
e64e4018 | 1768 | EXPORT_SYMBOL(md_bitmap_free); |
aa5cbd10 | 1769 | |
e64e4018 | 1770 | void md_bitmap_wait_behind_writes(struct mddev *mddev) |
48df498d GJ |
1771 | { |
1772 | struct bitmap *bitmap = mddev->bitmap; | |
1773 | ||
1774 | /* wait for behind writes to complete */ | |
1775 | if (bitmap && atomic_read(&bitmap->behind_writes) > 0) { | |
1776 | pr_debug("md:%s: behind writes in progress - waiting to stop.\n", | |
1777 | mdname(mddev)); | |
1778 | /* need to kick something here to make sure I/O goes? */ | |
1779 | wait_event(bitmap->behind_wait, | |
1780 | atomic_read(&bitmap->behind_writes) == 0); | |
1781 | } | |
1782 | } | |
1783 | ||
e64e4018 | 1784 | void md_bitmap_destroy(struct mddev *mddev) |
3178b0db N |
1785 | { |
1786 | struct bitmap *bitmap = mddev->bitmap; | |
1787 | ||
1788 | if (!bitmap) /* there was no bitmap */ | |
1789 | return; | |
1790 | ||
e64e4018 | 1791 | md_bitmap_wait_behind_writes(mddev); |
48df498d | 1792 | |
c3d9714e | 1793 | mutex_lock(&mddev->bitmap_info.mutex); |
978a7a47 | 1794 | spin_lock(&mddev->lock); |
3178b0db | 1795 | mddev->bitmap = NULL; /* disconnect from the md device */ |
978a7a47 | 1796 | spin_unlock(&mddev->lock); |
c3d9714e | 1797 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1798 | if (mddev->thread) |
1799 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db | 1800 | |
e64e4018 | 1801 | md_bitmap_free(bitmap); |
3178b0db | 1802 | } |
32a7627c N |
1803 | |
1804 | /* | |
1805 | * initialize the bitmap structure | |
1806 | * if this returns an error, bitmap_destroy must be called to do clean up | |
f9a67b11 | 1807 | * once mddev->bitmap is set |
32a7627c | 1808 | */ |
e64e4018 | 1809 | struct bitmap *md_bitmap_create(struct mddev *mddev, int slot) |
32a7627c N |
1810 | { |
1811 | struct bitmap *bitmap; | |
1f593903 | 1812 | sector_t blocks = mddev->resync_max_sectors; |
c3d9714e | 1813 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1814 | int err; |
324a56e1 | 1815 | struct kernfs_node *bm = NULL; |
32a7627c | 1816 | |
5f6e3c83 | 1817 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1818 | |
c3d9714e | 1819 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1820 | |
230b55fa N |
1821 | if (test_bit(MD_HAS_JOURNAL, &mddev->flags)) { |
1822 | pr_notice("md/raid:%s: array with journal cannot have bitmap\n", | |
1823 | mdname(mddev)); | |
1824 | return ERR_PTR(-EBUSY); | |
1825 | } | |
1826 | ||
9ffae0cf | 1827 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c | 1828 | if (!bitmap) |
f9209a32 | 1829 | return ERR_PTR(-ENOMEM); |
32a7627c | 1830 | |
40cffcc0 | 1831 | spin_lock_init(&bitmap->counts.lock); |
ce25c31b N |
1832 | atomic_set(&bitmap->pending_writes, 0); |
1833 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1834 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1835 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1836 | |
32a7627c | 1837 | bitmap->mddev = mddev; |
f9209a32 | 1838 | bitmap->cluster_slot = slot; |
32a7627c | 1839 | |
5ff5afff | 1840 | if (mddev->kobj.sd) |
388975cc | 1841 | bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap"); |
ece5cff0 | 1842 | if (bm) { |
388975cc | 1843 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear"); |
ece5cff0 N |
1844 | sysfs_put(bm); |
1845 | } else | |
1846 | bitmap->sysfs_can_clear = NULL; | |
1847 | ||
1ec885cd | 1848 | bitmap->storage.file = file; |
ce25c31b N |
1849 | if (file) { |
1850 | get_file(file); | |
ae8fa283 N |
1851 | /* As future accesses to this file will use bmap, |
1852 | * and bypass the page cache, we must sync the file | |
1853 | * first. | |
1854 | */ | |
8018ab05 | 1855 | vfs_fsync(file, 1); |
ce25c31b | 1856 | } |
42a04b50 | 1857 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
9c81075f JB |
1858 | if (!mddev->bitmap_info.external) { |
1859 | /* | |
1860 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is | |
1861 | * instructing us to create a new on-disk bitmap instance. | |
1862 | */ | |
1863 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) | |
e64e4018 | 1864 | err = md_bitmap_new_disk_sb(bitmap); |
9c81075f | 1865 | else |
e64e4018 | 1866 | err = md_bitmap_read_sb(bitmap); |
9c81075f | 1867 | } else { |
ece5cff0 N |
1868 | err = 0; |
1869 | if (mddev->bitmap_info.chunksize == 0 || | |
1870 | mddev->bitmap_info.daemon_sleep == 0) | |
1871 | /* chunksize and time_base need to be | |
1872 | * set first. */ | |
1873 | err = -EINVAL; | |
1874 | } | |
32a7627c | 1875 | if (err) |
3178b0db | 1876 | goto error; |
32a7627c | 1877 | |
624ce4f5 | 1878 | bitmap->daemon_lastrun = jiffies; |
e64e4018 | 1879 | err = md_bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1); |
d60b479d | 1880 | if (err) |
3178b0db | 1881 | goto error; |
32a7627c | 1882 | |
ec0cc226 N |
1883 | pr_debug("created bitmap (%lu pages) for device %s\n", |
1884 | bitmap->counts.pages, bmname(bitmap)); | |
69e51b44 | 1885 | |
f9209a32 GR |
1886 | err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0; |
1887 | if (err) | |
1888 | goto error; | |
69e51b44 | 1889 | |
f9209a32 | 1890 | return bitmap; |
69e51b44 | 1891 | error: |
e64e4018 | 1892 | md_bitmap_free(bitmap); |
f9209a32 | 1893 | return ERR_PTR(err); |
69e51b44 N |
1894 | } |
1895 | ||
e64e4018 | 1896 | int md_bitmap_load(struct mddev *mddev) |
69e51b44 N |
1897 | { |
1898 | int err = 0; | |
3520fa4d | 1899 | sector_t start = 0; |
69e51b44 N |
1900 | sector_t sector = 0; |
1901 | struct bitmap *bitmap = mddev->bitmap; | |
1902 | ||
1903 | if (!bitmap) | |
1904 | goto out; | |
1905 | ||
51e453ae GJ |
1906 | if (mddev_is_clustered(mddev)) |
1907 | md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes); | |
1908 | ||
69e51b44 N |
1909 | /* Clear out old bitmap info first: Either there is none, or we |
1910 | * are resuming after someone else has possibly changed things, | |
1911 | * so we should forget old cached info. | |
1912 | * All chunks should be clean, but some might need_sync. | |
1913 | */ | |
1914 | while (sector < mddev->resync_max_sectors) { | |
57dab0bd | 1915 | sector_t blocks; |
e64e4018 | 1916 | md_bitmap_start_sync(bitmap, sector, &blocks, 0); |
69e51b44 N |
1917 | sector += blocks; |
1918 | } | |
e64e4018 | 1919 | md_bitmap_close_sync(bitmap); |
69e51b44 | 1920 | |
3520fa4d JB |
1921 | if (mddev->degraded == 0 |
1922 | || bitmap->events_cleared == mddev->events) | |
1923 | /* no need to keep dirty bits to optimise a | |
1924 | * re-add of a missing device */ | |
1925 | start = mddev->recovery_cp; | |
1926 | ||
afbaa90b | 1927 | mutex_lock(&mddev->bitmap_info.mutex); |
e64e4018 | 1928 | err = md_bitmap_init_from_disk(bitmap, start); |
afbaa90b | 1929 | mutex_unlock(&mddev->bitmap_info.mutex); |
3520fa4d | 1930 | |
32a7627c | 1931 | if (err) |
69e51b44 | 1932 | goto out; |
b405fe91 | 1933 | clear_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
1934 | |
1935 | /* Kick recovery in case any bits were set */ | |
1936 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
3178b0db | 1937 | |
1b04be96 | 1938 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1939 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1940 | |
e64e4018 | 1941 | md_bitmap_update_sb(bitmap); |
4ad13663 | 1942 | |
b405fe91 | 1943 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
69e51b44 N |
1944 | err = -EIO; |
1945 | out: | |
3178b0db | 1946 | return err; |
32a7627c | 1947 | } |
e64e4018 | 1948 | EXPORT_SYMBOL_GPL(md_bitmap_load); |
32a7627c | 1949 | |
b98938d1 GJ |
1950 | struct bitmap *get_bitmap_from_slot(struct mddev *mddev, int slot) |
1951 | { | |
1952 | int rv = 0; | |
1953 | struct bitmap *bitmap; | |
1954 | ||
e64e4018 | 1955 | bitmap = md_bitmap_create(mddev, slot); |
b98938d1 GJ |
1956 | if (IS_ERR(bitmap)) { |
1957 | rv = PTR_ERR(bitmap); | |
1958 | return ERR_PTR(rv); | |
1959 | } | |
1960 | ||
e64e4018 | 1961 | rv = md_bitmap_init_from_disk(bitmap, 0); |
b98938d1 | 1962 | if (rv) { |
e64e4018 | 1963 | md_bitmap_free(bitmap); |
b98938d1 GJ |
1964 | return ERR_PTR(rv); |
1965 | } | |
1966 | ||
1967 | return bitmap; | |
1968 | } | |
1969 | EXPORT_SYMBOL(get_bitmap_from_slot); | |
1970 | ||
11dd35da GR |
1971 | /* Loads the bitmap associated with slot and copies the resync information |
1972 | * to our bitmap | |
1973 | */ | |
e64e4018 | 1974 | int md_bitmap_copy_from_slot(struct mddev *mddev, int slot, |
97f6cd39 | 1975 | sector_t *low, sector_t *high, bool clear_bits) |
11dd35da GR |
1976 | { |
1977 | int rv = 0, i, j; | |
1978 | sector_t block, lo = 0, hi = 0; | |
1979 | struct bitmap_counts *counts; | |
b98938d1 | 1980 | struct bitmap *bitmap; |
11dd35da | 1981 | |
b98938d1 GJ |
1982 | bitmap = get_bitmap_from_slot(mddev, slot); |
1983 | if (IS_ERR(bitmap)) { | |
1984 | pr_err("%s can't get bitmap from slot %d\n", __func__, slot); | |
1985 | return -1; | |
1986 | } | |
11dd35da GR |
1987 | |
1988 | counts = &bitmap->counts; | |
1989 | for (j = 0; j < counts->chunks; j++) { | |
1990 | block = (sector_t)j << counts->chunkshift; | |
e64e4018 | 1991 | if (md_bitmap_file_test_bit(bitmap, block)) { |
11dd35da GR |
1992 | if (!lo) |
1993 | lo = block; | |
1994 | hi = block; | |
e64e4018 AS |
1995 | md_bitmap_file_clear_bit(bitmap, block); |
1996 | md_bitmap_set_memory_bits(mddev->bitmap, block, 1); | |
1997 | md_bitmap_file_set_bit(mddev->bitmap, block); | |
11dd35da GR |
1998 | } |
1999 | } | |
2000 | ||
97f6cd39 | 2001 | if (clear_bits) { |
e64e4018 | 2002 | md_bitmap_update_sb(bitmap); |
c84400c8 GJ |
2003 | /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs |
2004 | * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */ | |
97f6cd39 | 2005 | for (i = 0; i < bitmap->storage.file_pages; i++) |
c84400c8 GJ |
2006 | if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING)) |
2007 | set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE); | |
e64e4018 | 2008 | md_bitmap_unplug(bitmap); |
97f6cd39 | 2009 | } |
e64e4018 | 2010 | md_bitmap_unplug(mddev->bitmap); |
11dd35da GR |
2011 | *low = lo; |
2012 | *high = hi; | |
b98938d1 | 2013 | |
11dd35da GR |
2014 | return rv; |
2015 | } | |
e64e4018 | 2016 | EXPORT_SYMBOL_GPL(md_bitmap_copy_from_slot); |
11dd35da GR |
2017 | |
2018 | ||
e64e4018 | 2019 | void md_bitmap_status(struct seq_file *seq, struct bitmap *bitmap) |
57148964 N |
2020 | { |
2021 | unsigned long chunk_kb; | |
40cffcc0 | 2022 | struct bitmap_counts *counts; |
57148964 N |
2023 | |
2024 | if (!bitmap) | |
2025 | return; | |
2026 | ||
40cffcc0 N |
2027 | counts = &bitmap->counts; |
2028 | ||
57148964 N |
2029 | chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; |
2030 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
2031 | "%lu%s chunk", | |
40cffcc0 N |
2032 | counts->pages - counts->missing_pages, |
2033 | counts->pages, | |
2034 | (counts->pages - counts->missing_pages) | |
57148964 N |
2035 | << (PAGE_SHIFT - 10), |
2036 | chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, | |
2037 | chunk_kb ? "KB" : "B"); | |
1ec885cd | 2038 | if (bitmap->storage.file) { |
57148964 | 2039 | seq_printf(seq, ", file: "); |
2726d566 | 2040 | seq_file_path(seq, bitmap->storage.file, " \t\n"); |
57148964 N |
2041 | } |
2042 | ||
2043 | seq_printf(seq, "\n"); | |
57148964 N |
2044 | } |
2045 | ||
e64e4018 | 2046 | int md_bitmap_resize(struct bitmap *bitmap, sector_t blocks, |
d60b479d N |
2047 | int chunksize, int init) |
2048 | { | |
2049 | /* If chunk_size is 0, choose an appropriate chunk size. | |
2050 | * Then possibly allocate new storage space. | |
2051 | * Then quiesce, copy bits, replace bitmap, and re-start | |
2052 | * | |
2053 | * This function is called both to set up the initial bitmap | |
2054 | * and to resize the bitmap while the array is active. | |
2055 | * If this happens as a result of the array being resized, | |
2056 | * chunksize will be zero, and we need to choose a suitable | |
2057 | * chunksize, otherwise we use what we are given. | |
2058 | */ | |
2059 | struct bitmap_storage store; | |
2060 | struct bitmap_counts old_counts; | |
2061 | unsigned long chunks; | |
2062 | sector_t block; | |
2063 | sector_t old_blocks, new_blocks; | |
2064 | int chunkshift; | |
2065 | int ret = 0; | |
2066 | long pages; | |
2067 | struct bitmap_page *new_bp; | |
2068 | ||
e8a27f83 N |
2069 | if (bitmap->storage.file && !init) { |
2070 | pr_info("md: cannot resize file-based bitmap\n"); | |
2071 | return -EINVAL; | |
2072 | } | |
2073 | ||
d60b479d N |
2074 | if (chunksize == 0) { |
2075 | /* If there is enough space, leave the chunk size unchanged, | |
2076 | * else increase by factor of two until there is enough space. | |
2077 | */ | |
2078 | long bytes; | |
2079 | long space = bitmap->mddev->bitmap_info.space; | |
2080 | ||
2081 | if (space == 0) { | |
2082 | /* We don't know how much space there is, so limit | |
2083 | * to current size - in sectors. | |
2084 | */ | |
2085 | bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8); | |
2086 | if (!bitmap->mddev->bitmap_info.external) | |
2087 | bytes += sizeof(bitmap_super_t); | |
2088 | space = DIV_ROUND_UP(bytes, 512); | |
2089 | bitmap->mddev->bitmap_info.space = space; | |
2090 | } | |
2091 | chunkshift = bitmap->counts.chunkshift; | |
2092 | chunkshift--; | |
2093 | do { | |
2094 | /* 'chunkshift' is shift from block size to chunk size */ | |
2095 | chunkshift++; | |
2096 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
2097 | bytes = DIV_ROUND_UP(chunks, 8); | |
2098 | if (!bitmap->mddev->bitmap_info.external) | |
2099 | bytes += sizeof(bitmap_super_t); | |
2100 | } while (bytes > (space << 9)); | |
2101 | } else | |
2102 | chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT; | |
2103 | ||
2104 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
2105 | memset(&store, 0, sizeof(store)); | |
2106 | if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file) | |
e64e4018 AS |
2107 | ret = md_bitmap_storage_alloc(&store, chunks, |
2108 | !bitmap->mddev->bitmap_info.external, | |
2109 | mddev_is_clustered(bitmap->mddev) | |
2110 | ? bitmap->cluster_slot : 0); | |
cbb38732 | 2111 | if (ret) { |
e64e4018 | 2112 | md_bitmap_file_unmap(&store); |
d60b479d | 2113 | goto err; |
cbb38732 | 2114 | } |
d60b479d N |
2115 | |
2116 | pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO); | |
2117 | ||
6396bb22 | 2118 | new_bp = kcalloc(pages, sizeof(*new_bp), GFP_KERNEL); |
d60b479d N |
2119 | ret = -ENOMEM; |
2120 | if (!new_bp) { | |
e64e4018 | 2121 | md_bitmap_file_unmap(&store); |
d60b479d N |
2122 | goto err; |
2123 | } | |
2124 | ||
2125 | if (!init) | |
2126 | bitmap->mddev->pers->quiesce(bitmap->mddev, 1); | |
2127 | ||
2128 | store.file = bitmap->storage.file; | |
2129 | bitmap->storage.file = NULL; | |
2130 | ||
2131 | if (store.sb_page && bitmap->storage.sb_page) | |
2132 | memcpy(page_address(store.sb_page), | |
2133 | page_address(bitmap->storage.sb_page), | |
938b533d | 2134 | sizeof(bitmap_super_t)); |
e64e4018 | 2135 | md_bitmap_file_unmap(&bitmap->storage); |
d60b479d N |
2136 | bitmap->storage = store; |
2137 | ||
2138 | old_counts = bitmap->counts; | |
2139 | bitmap->counts.bp = new_bp; | |
2140 | bitmap->counts.pages = pages; | |
2141 | bitmap->counts.missing_pages = pages; | |
2142 | bitmap->counts.chunkshift = chunkshift; | |
2143 | bitmap->counts.chunks = chunks; | |
2144 | bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift + | |
2145 | BITMAP_BLOCK_SHIFT); | |
2146 | ||
2147 | blocks = min(old_counts.chunks << old_counts.chunkshift, | |
2148 | chunks << chunkshift); | |
2149 | ||
2150 | spin_lock_irq(&bitmap->counts.lock); | |
c9d65032 GJ |
2151 | /* For cluster raid, need to pre-allocate bitmap */ |
2152 | if (mddev_is_clustered(bitmap->mddev)) { | |
2153 | unsigned long page; | |
2154 | for (page = 0; page < pages; page++) { | |
e64e4018 | 2155 | ret = md_bitmap_checkpage(&bitmap->counts, page, 1, 1); |
c9d65032 GJ |
2156 | if (ret) { |
2157 | unsigned long k; | |
2158 | ||
2159 | /* deallocate the page memory */ | |
2160 | for (k = 0; k < page; k++) { | |
bc47e842 | 2161 | kfree(new_bp[k].map); |
c9d65032 | 2162 | } |
0868b99c | 2163 | kfree(new_bp); |
c9d65032 GJ |
2164 | |
2165 | /* restore some fields from old_counts */ | |
2166 | bitmap->counts.bp = old_counts.bp; | |
2167 | bitmap->counts.pages = old_counts.pages; | |
2168 | bitmap->counts.missing_pages = old_counts.pages; | |
2169 | bitmap->counts.chunkshift = old_counts.chunkshift; | |
2170 | bitmap->counts.chunks = old_counts.chunks; | |
2171 | bitmap->mddev->bitmap_info.chunksize = 1 << (old_counts.chunkshift + | |
2172 | BITMAP_BLOCK_SHIFT); | |
2173 | blocks = old_counts.chunks << old_counts.chunkshift; | |
ec0cc226 | 2174 | pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n"); |
c9d65032 GJ |
2175 | break; |
2176 | } else | |
2177 | bitmap->counts.bp[page].count += 1; | |
2178 | } | |
2179 | } | |
2180 | ||
d60b479d N |
2181 | for (block = 0; block < blocks; ) { |
2182 | bitmap_counter_t *bmc_old, *bmc_new; | |
2183 | int set; | |
2184 | ||
e64e4018 | 2185 | bmc_old = md_bitmap_get_counter(&old_counts, block, &old_blocks, 0); |
d60b479d N |
2186 | set = bmc_old && NEEDED(*bmc_old); |
2187 | ||
2188 | if (set) { | |
e64e4018 | 2189 | bmc_new = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1); |
d60b479d N |
2190 | if (*bmc_new == 0) { |
2191 | /* need to set on-disk bits too. */ | |
2192 | sector_t end = block + new_blocks; | |
2193 | sector_t start = block >> chunkshift; | |
2194 | start <<= chunkshift; | |
2195 | while (start < end) { | |
e64e4018 | 2196 | md_bitmap_file_set_bit(bitmap, block); |
d60b479d N |
2197 | start += 1 << chunkshift; |
2198 | } | |
2199 | *bmc_new = 2; | |
e64e4018 AS |
2200 | md_bitmap_count_page(&bitmap->counts, block, 1); |
2201 | md_bitmap_set_pending(&bitmap->counts, block); | |
d60b479d N |
2202 | } |
2203 | *bmc_new |= NEEDED_MASK; | |
2204 | if (new_blocks < old_blocks) | |
2205 | old_blocks = new_blocks; | |
2206 | } | |
2207 | block += old_blocks; | |
2208 | } | |
2209 | ||
0868b99c ZK |
2210 | if (bitmap->counts.bp != old_counts.bp) { |
2211 | unsigned long k; | |
2212 | for (k = 0; k < old_counts.pages; k++) | |
2213 | if (!old_counts.bp[k].hijacked) | |
2214 | kfree(old_counts.bp[k].map); | |
2215 | kfree(old_counts.bp); | |
2216 | } | |
2217 | ||
d60b479d N |
2218 | if (!init) { |
2219 | int i; | |
2220 | while (block < (chunks << chunkshift)) { | |
2221 | bitmap_counter_t *bmc; | |
e64e4018 | 2222 | bmc = md_bitmap_get_counter(&bitmap->counts, block, &new_blocks, 1); |
d60b479d N |
2223 | if (bmc) { |
2224 | /* new space. It needs to be resynced, so | |
2225 | * we set NEEDED_MASK. | |
2226 | */ | |
2227 | if (*bmc == 0) { | |
2228 | *bmc = NEEDED_MASK | 2; | |
e64e4018 AS |
2229 | md_bitmap_count_page(&bitmap->counts, block, 1); |
2230 | md_bitmap_set_pending(&bitmap->counts, block); | |
d60b479d N |
2231 | } |
2232 | } | |
2233 | block += new_blocks; | |
2234 | } | |
2235 | for (i = 0; i < bitmap->storage.file_pages; i++) | |
2236 | set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); | |
2237 | } | |
2238 | spin_unlock_irq(&bitmap->counts.lock); | |
2239 | ||
2240 | if (!init) { | |
e64e4018 | 2241 | md_bitmap_unplug(bitmap); |
d60b479d N |
2242 | bitmap->mddev->pers->quiesce(bitmap->mddev, 0); |
2243 | } | |
2244 | ret = 0; | |
2245 | err: | |
2246 | return ret; | |
2247 | } | |
e64e4018 | 2248 | EXPORT_SYMBOL_GPL(md_bitmap_resize); |
d60b479d | 2249 | |
43a70507 | 2250 | static ssize_t |
fd01b88c | 2251 | location_show(struct mddev *mddev, char *page) |
43a70507 N |
2252 | { |
2253 | ssize_t len; | |
ac2f40be | 2254 | if (mddev->bitmap_info.file) |
43a70507 | 2255 | len = sprintf(page, "file"); |
ac2f40be | 2256 | else if (mddev->bitmap_info.offset) |
43a70507 | 2257 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 2258 | else |
43a70507 N |
2259 | len = sprintf(page, "none"); |
2260 | len += sprintf(page+len, "\n"); | |
2261 | return len; | |
2262 | } | |
2263 | ||
2264 | static ssize_t | |
fd01b88c | 2265 | location_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 | 2266 | { |
d9dd26b2 | 2267 | int rv; |
43a70507 | 2268 | |
d9dd26b2 SL |
2269 | rv = mddev_lock(mddev); |
2270 | if (rv) | |
2271 | return rv; | |
43a70507 | 2272 | if (mddev->pers) { |
d9dd26b2 SL |
2273 | if (!mddev->pers->quiesce) { |
2274 | rv = -EBUSY; | |
2275 | goto out; | |
2276 | } | |
2277 | if (mddev->recovery || mddev->sync_thread) { | |
2278 | rv = -EBUSY; | |
2279 | goto out; | |
2280 | } | |
43a70507 N |
2281 | } |
2282 | ||
2283 | if (mddev->bitmap || mddev->bitmap_info.file || | |
2284 | mddev->bitmap_info.offset) { | |
2285 | /* bitmap already configured. Only option is to clear it */ | |
d9dd26b2 SL |
2286 | if (strncmp(buf, "none", 4) != 0) { |
2287 | rv = -EBUSY; | |
2288 | goto out; | |
2289 | } | |
43a70507 | 2290 | if (mddev->pers) { |
f8f83d8f | 2291 | mddev_suspend(mddev); |
e64e4018 | 2292 | md_bitmap_destroy(mddev); |
f8f83d8f | 2293 | mddev_resume(mddev); |
43a70507 N |
2294 | } |
2295 | mddev->bitmap_info.offset = 0; | |
2296 | if (mddev->bitmap_info.file) { | |
2297 | struct file *f = mddev->bitmap_info.file; | |
2298 | mddev->bitmap_info.file = NULL; | |
43a70507 N |
2299 | fput(f); |
2300 | } | |
2301 | } else { | |
2302 | /* No bitmap, OK to set a location */ | |
2303 | long long offset; | |
2304 | if (strncmp(buf, "none", 4) == 0) | |
2305 | /* nothing to be done */; | |
2306 | else if (strncmp(buf, "file:", 5) == 0) { | |
2307 | /* Not supported yet */ | |
d9dd26b2 SL |
2308 | rv = -EINVAL; |
2309 | goto out; | |
43a70507 | 2310 | } else { |
43a70507 | 2311 | if (buf[0] == '+') |
b29bebd6 | 2312 | rv = kstrtoll(buf+1, 10, &offset); |
43a70507 | 2313 | else |
b29bebd6 | 2314 | rv = kstrtoll(buf, 10, &offset); |
43a70507 | 2315 | if (rv) |
d9dd26b2 SL |
2316 | goto out; |
2317 | if (offset == 0) { | |
2318 | rv = -EINVAL; | |
2319 | goto out; | |
2320 | } | |
ece5cff0 N |
2321 | if (mddev->bitmap_info.external == 0 && |
2322 | mddev->major_version == 0 && | |
d9dd26b2 SL |
2323 | offset != mddev->bitmap_info.default_offset) { |
2324 | rv = -EINVAL; | |
2325 | goto out; | |
2326 | } | |
43a70507 N |
2327 | mddev->bitmap_info.offset = offset; |
2328 | if (mddev->pers) { | |
f9209a32 | 2329 | struct bitmap *bitmap; |
e64e4018 | 2330 | bitmap = md_bitmap_create(mddev, -1); |
f8f83d8f | 2331 | mddev_suspend(mddev); |
f9209a32 GR |
2332 | if (IS_ERR(bitmap)) |
2333 | rv = PTR_ERR(bitmap); | |
2334 | else { | |
2335 | mddev->bitmap = bitmap; | |
e64e4018 | 2336 | rv = md_bitmap_load(mddev); |
f9a67b11 | 2337 | if (rv) |
f9209a32 | 2338 | mddev->bitmap_info.offset = 0; |
43a70507 | 2339 | } |
f9a67b11 | 2340 | if (rv) { |
e64e4018 | 2341 | md_bitmap_destroy(mddev); |
f8f83d8f | 2342 | mddev_resume(mddev); |
d9dd26b2 | 2343 | goto out; |
f9a67b11 | 2344 | } |
f8f83d8f | 2345 | mddev_resume(mddev); |
43a70507 N |
2346 | } |
2347 | } | |
2348 | } | |
2349 | if (!mddev->external) { | |
2350 | /* Ensure new bitmap info is stored in | |
2351 | * metadata promptly. | |
2352 | */ | |
2953079c | 2353 | set_bit(MD_SB_CHANGE_DEVS, &mddev->sb_flags); |
43a70507 N |
2354 | md_wakeup_thread(mddev->thread); |
2355 | } | |
d9dd26b2 SL |
2356 | rv = 0; |
2357 | out: | |
2358 | mddev_unlock(mddev); | |
2359 | if (rv) | |
2360 | return rv; | |
43a70507 N |
2361 | return len; |
2362 | } | |
2363 | ||
2364 | static struct md_sysfs_entry bitmap_location = | |
2365 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
2366 | ||
6409bb05 N |
2367 | /* 'bitmap/space' is the space available at 'location' for the |
2368 | * bitmap. This allows the kernel to know when it is safe to | |
2369 | * resize the bitmap to match a resized array. | |
2370 | */ | |
2371 | static ssize_t | |
2372 | space_show(struct mddev *mddev, char *page) | |
2373 | { | |
2374 | return sprintf(page, "%lu\n", mddev->bitmap_info.space); | |
2375 | } | |
2376 | ||
2377 | static ssize_t | |
2378 | space_store(struct mddev *mddev, const char *buf, size_t len) | |
2379 | { | |
2380 | unsigned long sectors; | |
2381 | int rv; | |
2382 | ||
2383 | rv = kstrtoul(buf, 10, §ors); | |
2384 | if (rv) | |
2385 | return rv; | |
2386 | ||
2387 | if (sectors == 0) | |
2388 | return -EINVAL; | |
2389 | ||
2390 | if (mddev->bitmap && | |
9b1215c1 | 2391 | sectors < (mddev->bitmap->storage.bytes + 511) >> 9) |
6409bb05 N |
2392 | return -EFBIG; /* Bitmap is too big for this small space */ |
2393 | ||
2394 | /* could make sure it isn't too big, but that isn't really | |
2395 | * needed - user-space should be careful. | |
2396 | */ | |
2397 | mddev->bitmap_info.space = sectors; | |
2398 | return len; | |
2399 | } | |
2400 | ||
2401 | static struct md_sysfs_entry bitmap_space = | |
2402 | __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); | |
2403 | ||
43a70507 | 2404 | static ssize_t |
fd01b88c | 2405 | timeout_show(struct mddev *mddev, char *page) |
43a70507 N |
2406 | { |
2407 | ssize_t len; | |
2408 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
2409 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 2410 | |
43a70507 N |
2411 | len = sprintf(page, "%lu", secs); |
2412 | if (jifs) | |
2413 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
2414 | len += sprintf(page+len, "\n"); | |
2415 | return len; | |
2416 | } | |
2417 | ||
2418 | static ssize_t | |
fd01b88c | 2419 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2420 | { |
2421 | /* timeout can be set at any time */ | |
2422 | unsigned long timeout; | |
2423 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
2424 | if (rv) | |
2425 | return rv; | |
2426 | ||
2427 | /* just to make sure we don't overflow... */ | |
2428 | if (timeout >= LONG_MAX / HZ) | |
2429 | return -EINVAL; | |
2430 | ||
2431 | timeout = timeout * HZ / 10000; | |
2432 | ||
2433 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
2434 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
2435 | if (timeout < 1) | |
2436 | timeout = 1; | |
2437 | mddev->bitmap_info.daemon_sleep = timeout; | |
2438 | if (mddev->thread) { | |
2439 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
2440 | * the bitmap is all clean and we don't need to | |
2441 | * adjust the timeout right now | |
2442 | */ | |
2443 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2444 | mddev->thread->timeout = timeout; | |
2445 | md_wakeup_thread(mddev->thread); | |
2446 | } | |
2447 | } | |
2448 | return len; | |
2449 | } | |
2450 | ||
2451 | static struct md_sysfs_entry bitmap_timeout = | |
2452 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2453 | ||
2454 | static ssize_t | |
fd01b88c | 2455 | backlog_show(struct mddev *mddev, char *page) |
43a70507 N |
2456 | { |
2457 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2458 | } | |
2459 | ||
2460 | static ssize_t | |
fd01b88c | 2461 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2462 | { |
2463 | unsigned long backlog; | |
b29bebd6 | 2464 | int rv = kstrtoul(buf, 10, &backlog); |
43a70507 N |
2465 | if (rv) |
2466 | return rv; | |
2467 | if (backlog > COUNTER_MAX) | |
2468 | return -EINVAL; | |
2469 | mddev->bitmap_info.max_write_behind = backlog; | |
2470 | return len; | |
2471 | } | |
2472 | ||
2473 | static struct md_sysfs_entry bitmap_backlog = | |
2474 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2475 | ||
2476 | static ssize_t | |
fd01b88c | 2477 | chunksize_show(struct mddev *mddev, char *page) |
43a70507 N |
2478 | { |
2479 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2480 | } | |
2481 | ||
2482 | static ssize_t | |
fd01b88c | 2483 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2484 | { |
2485 | /* Can only be changed when no bitmap is active */ | |
2486 | int rv; | |
2487 | unsigned long csize; | |
2488 | if (mddev->bitmap) | |
2489 | return -EBUSY; | |
b29bebd6 | 2490 | rv = kstrtoul(buf, 10, &csize); |
43a70507 N |
2491 | if (rv) |
2492 | return rv; | |
2493 | if (csize < 512 || | |
2494 | !is_power_of_2(csize)) | |
2495 | return -EINVAL; | |
2496 | mddev->bitmap_info.chunksize = csize; | |
2497 | return len; | |
2498 | } | |
2499 | ||
2500 | static struct md_sysfs_entry bitmap_chunksize = | |
2501 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2502 | ||
fd01b88c | 2503 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
ece5cff0 | 2504 | { |
c4ce867f GR |
2505 | if (mddev_is_clustered(mddev)) |
2506 | return sprintf(page, "clustered\n"); | |
ece5cff0 N |
2507 | return sprintf(page, "%s\n", (mddev->bitmap_info.external |
2508 | ? "external" : "internal")); | |
2509 | } | |
2510 | ||
fd01b88c | 2511 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2512 | { |
2513 | if (mddev->bitmap || | |
2514 | mddev->bitmap_info.file || | |
2515 | mddev->bitmap_info.offset) | |
2516 | return -EBUSY; | |
2517 | if (strncmp(buf, "external", 8) == 0) | |
2518 | mddev->bitmap_info.external = 1; | |
c4ce867f GR |
2519 | else if ((strncmp(buf, "internal", 8) == 0) || |
2520 | (strncmp(buf, "clustered", 9) == 0)) | |
ece5cff0 N |
2521 | mddev->bitmap_info.external = 0; |
2522 | else | |
2523 | return -EINVAL; | |
2524 | return len; | |
2525 | } | |
2526 | ||
2527 | static struct md_sysfs_entry bitmap_metadata = | |
2528 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2529 | ||
fd01b88c | 2530 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2531 | { |
2532 | int len; | |
b7b17c9b | 2533 | spin_lock(&mddev->lock); |
ece5cff0 N |
2534 | if (mddev->bitmap) |
2535 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2536 | "false" : "true")); | |
2537 | else | |
2538 | len = sprintf(page, "\n"); | |
b7b17c9b | 2539 | spin_unlock(&mddev->lock); |
ece5cff0 N |
2540 | return len; |
2541 | } | |
2542 | ||
fd01b88c | 2543 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2544 | { |
2545 | if (mddev->bitmap == NULL) | |
2546 | return -ENOENT; | |
2547 | if (strncmp(buf, "false", 5) == 0) | |
2548 | mddev->bitmap->need_sync = 1; | |
2549 | else if (strncmp(buf, "true", 4) == 0) { | |
2550 | if (mddev->degraded) | |
2551 | return -EBUSY; | |
2552 | mddev->bitmap->need_sync = 0; | |
2553 | } else | |
2554 | return -EINVAL; | |
2555 | return len; | |
2556 | } | |
2557 | ||
2558 | static struct md_sysfs_entry bitmap_can_clear = | |
2559 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2560 | ||
696fcd53 | 2561 | static ssize_t |
fd01b88c | 2562 | behind_writes_used_show(struct mddev *mddev, char *page) |
696fcd53 | 2563 | { |
b7b17c9b N |
2564 | ssize_t ret; |
2565 | spin_lock(&mddev->lock); | |
696fcd53 | 2566 | if (mddev->bitmap == NULL) |
b7b17c9b N |
2567 | ret = sprintf(page, "0\n"); |
2568 | else | |
2569 | ret = sprintf(page, "%lu\n", | |
2570 | mddev->bitmap->behind_writes_used); | |
2571 | spin_unlock(&mddev->lock); | |
2572 | return ret; | |
696fcd53 PC |
2573 | } |
2574 | ||
2575 | static ssize_t | |
fd01b88c | 2576 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
696fcd53 PC |
2577 | { |
2578 | if (mddev->bitmap) | |
2579 | mddev->bitmap->behind_writes_used = 0; | |
2580 | return len; | |
2581 | } | |
2582 | ||
2583 | static struct md_sysfs_entry max_backlog_used = | |
2584 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2585 | behind_writes_used_show, behind_writes_used_reset); | |
2586 | ||
43a70507 N |
2587 | static struct attribute *md_bitmap_attrs[] = { |
2588 | &bitmap_location.attr, | |
6409bb05 | 2589 | &bitmap_space.attr, |
43a70507 N |
2590 | &bitmap_timeout.attr, |
2591 | &bitmap_backlog.attr, | |
2592 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2593 | &bitmap_metadata.attr, |
2594 | &bitmap_can_clear.attr, | |
696fcd53 | 2595 | &max_backlog_used.attr, |
43a70507 N |
2596 | NULL |
2597 | }; | |
2598 | struct attribute_group md_bitmap_group = { | |
2599 | .name = "bitmap", | |
2600 | .attrs = md_bitmap_attrs, | |
2601 | }; | |
2602 |