]>
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> |
43b2e5d8 | 30 | #include "md.h" |
ef740c37 | 31 | #include "bitmap.h" |
32a7627c | 32 | |
ac2f40be | 33 | static inline char *bmname(struct bitmap *bitmap) |
32a7627c N |
34 | { |
35 | return bitmap->mddev ? mdname(bitmap->mddev) : "mdX"; | |
36 | } | |
37 | ||
32a7627c N |
38 | /* |
39 | * check a page and, if necessary, allocate it (or hijack it if the alloc fails) | |
40 | * | |
41 | * 1) check to see if this page is allocated, if it's not then try to alloc | |
42 | * 2) if the alloc fails, set the page's hijacked flag so we'll use the | |
43 | * page pointer directly as a counter | |
44 | * | |
45 | * if we find our page, we increment the page's refcount so that it stays | |
46 | * allocated while we're using it | |
47 | */ | |
40cffcc0 | 48 | static int bitmap_checkpage(struct bitmap_counts *bitmap, |
c9d65032 | 49 | unsigned long page, int create, int no_hijack) |
ee305ace N |
50 | __releases(bitmap->lock) |
51 | __acquires(bitmap->lock) | |
32a7627c N |
52 | { |
53 | unsigned char *mappage; | |
54 | ||
55 | if (page >= bitmap->pages) { | |
1187cf0a N |
56 | /* This can happen if bitmap_start_sync goes beyond |
57 | * End-of-device while looking for a whole page. | |
58 | * It is harmless. | |
59 | */ | |
32a7627c N |
60 | return -EINVAL; |
61 | } | |
62 | ||
32a7627c N |
63 | if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */ |
64 | return 0; | |
65 | ||
66 | if (bitmap->bp[page].map) /* page is already allocated, just return */ | |
67 | return 0; | |
68 | ||
69 | if (!create) | |
70 | return -ENOENT; | |
71 | ||
32a7627c N |
72 | /* this page has not been allocated yet */ |
73 | ||
ac2f40be | 74 | spin_unlock_irq(&bitmap->lock); |
d9590143 N |
75 | /* It is possible that this is being called inside a |
76 | * prepare_to_wait/finish_wait loop from raid5c:make_request(). | |
77 | * In general it is not permitted to sleep in that context as it | |
78 | * can cause the loop to spin freely. | |
79 | * That doesn't apply here as we can only reach this point | |
80 | * once with any loop. | |
81 | * When this function completes, either bp[page].map or | |
82 | * bp[page].hijacked. In either case, this function will | |
83 | * abort before getting to this point again. So there is | |
84 | * no risk of a free-spin, and so it is safe to assert | |
85 | * that sleeping here is allowed. | |
86 | */ | |
87 | sched_annotate_sleep(); | |
792a1d4b | 88 | mappage = kzalloc(PAGE_SIZE, GFP_NOIO); |
ac2f40be N |
89 | spin_lock_irq(&bitmap->lock); |
90 | ||
91 | if (mappage == NULL) { | |
40cffcc0 | 92 | pr_debug("md/bitmap: map page allocation failed, hijacking\n"); |
c9d65032 GJ |
93 | /* We don't support hijack for cluster raid */ |
94 | if (no_hijack) | |
95 | return -ENOMEM; | |
32a7627c N |
96 | /* failed - set the hijacked flag so that we can use the |
97 | * pointer as a counter */ | |
32a7627c N |
98 | if (!bitmap->bp[page].map) |
99 | bitmap->bp[page].hijacked = 1; | |
ac2f40be N |
100 | } else if (bitmap->bp[page].map || |
101 | bitmap->bp[page].hijacked) { | |
32a7627c | 102 | /* somebody beat us to getting the page */ |
792a1d4b | 103 | kfree(mappage); |
ac2f40be | 104 | } else { |
32a7627c | 105 | |
ac2f40be | 106 | /* no page was in place and we have one, so install it */ |
32a7627c | 107 | |
ac2f40be N |
108 | bitmap->bp[page].map = mappage; |
109 | bitmap->missing_pages--; | |
110 | } | |
32a7627c N |
111 | return 0; |
112 | } | |
113 | ||
32a7627c N |
114 | /* if page is completely empty, put it back on the free list, or dealloc it */ |
115 | /* if page was hijacked, unmark the flag so it might get alloced next time */ | |
116 | /* Note: lock should be held when calling this */ | |
40cffcc0 | 117 | static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page) |
32a7627c N |
118 | { |
119 | char *ptr; | |
120 | ||
121 | if (bitmap->bp[page].count) /* page is still busy */ | |
122 | return; | |
123 | ||
124 | /* page is no longer in use, it can be released */ | |
125 | ||
126 | if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */ | |
127 | bitmap->bp[page].hijacked = 0; | |
128 | bitmap->bp[page].map = NULL; | |
ac2f40be N |
129 | } else { |
130 | /* normal case, free the page */ | |
131 | ptr = bitmap->bp[page].map; | |
132 | bitmap->bp[page].map = NULL; | |
133 | bitmap->missing_pages++; | |
792a1d4b | 134 | kfree(ptr); |
32a7627c | 135 | } |
32a7627c N |
136 | } |
137 | ||
32a7627c N |
138 | /* |
139 | * bitmap file handling - read and write the bitmap file and its superblock | |
140 | */ | |
141 | ||
32a7627c N |
142 | /* |
143 | * basic page I/O operations | |
144 | */ | |
145 | ||
a654b9d8 | 146 | /* IO operations when bitmap is stored near all superblocks */ |
27581e5a N |
147 | static int read_sb_page(struct mddev *mddev, loff_t offset, |
148 | struct page *page, | |
149 | unsigned long index, int size) | |
a654b9d8 N |
150 | { |
151 | /* choose a good rdev and read the page from there */ | |
152 | ||
3cb03002 | 153 | struct md_rdev *rdev; |
a654b9d8 | 154 | sector_t target; |
a654b9d8 | 155 | |
dafb20fa | 156 | rdev_for_each(rdev, mddev) { |
b2d444d7 N |
157 | if (! test_bit(In_sync, &rdev->flags) |
158 | || test_bit(Faulty, &rdev->flags)) | |
ab904d63 N |
159 | continue; |
160 | ||
ccebd4c4 | 161 | target = offset + index * (PAGE_SIZE/512); |
a654b9d8 | 162 | |
2b193363 | 163 | if (sync_page_io(rdev, target, |
e1defc4f | 164 | roundup(size, bdev_logical_block_size(rdev->bdev)), |
796a5cf0 | 165 | page, REQ_OP_READ, 0, true)) { |
ab904d63 | 166 | page->index = index; |
27581e5a | 167 | return 0; |
ab904d63 N |
168 | } |
169 | } | |
27581e5a | 170 | return -EIO; |
a654b9d8 N |
171 | } |
172 | ||
fd01b88c | 173 | static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev) |
b2d2c4ce N |
174 | { |
175 | /* Iterate the disks of an mddev, using rcu to protect access to the | |
176 | * linked list, and raising the refcount of devices we return to ensure | |
177 | * they don't disappear while in use. | |
178 | * As devices are only added or removed when raid_disk is < 0 and | |
179 | * nr_pending is 0 and In_sync is clear, the entries we return will | |
180 | * still be in the same position on the list when we re-enter | |
fd177481 | 181 | * list_for_each_entry_continue_rcu. |
8532e343 N |
182 | * |
183 | * Note that if entered with 'rdev == NULL' to start at the | |
184 | * beginning, we temporarily assign 'rdev' to an address which | |
185 | * isn't really an rdev, but which can be used by | |
186 | * list_for_each_entry_continue_rcu() to find the first entry. | |
b2d2c4ce | 187 | */ |
b2d2c4ce N |
188 | rcu_read_lock(); |
189 | if (rdev == NULL) | |
190 | /* start at the beginning */ | |
8532e343 | 191 | rdev = list_entry(&mddev->disks, struct md_rdev, same_set); |
b2d2c4ce N |
192 | else { |
193 | /* release the previous rdev and start from there. */ | |
194 | rdev_dec_pending(rdev, mddev); | |
b2d2c4ce | 195 | } |
fd177481 | 196 | list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) { |
b2d2c4ce | 197 | if (rdev->raid_disk >= 0 && |
b2d2c4ce N |
198 | !test_bit(Faulty, &rdev->flags)) { |
199 | /* this is a usable devices */ | |
200 | atomic_inc(&rdev->nr_pending); | |
201 | rcu_read_unlock(); | |
202 | return rdev; | |
203 | } | |
204 | } | |
205 | rcu_read_unlock(); | |
206 | return NULL; | |
207 | } | |
208 | ||
ab6085c7 | 209 | static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait) |
a654b9d8 | 210 | { |
3cb03002 | 211 | struct md_rdev *rdev = NULL; |
a6ff7e08 | 212 | struct block_device *bdev; |
fd01b88c | 213 | struct mddev *mddev = bitmap->mddev; |
1ec885cd | 214 | struct bitmap_storage *store = &bitmap->storage; |
a654b9d8 | 215 | |
b2d2c4ce | 216 | while ((rdev = next_active_rdev(rdev, mddev)) != NULL) { |
ac2f40be N |
217 | int size = PAGE_SIZE; |
218 | loff_t offset = mddev->bitmap_info.offset; | |
a6ff7e08 JB |
219 | |
220 | bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev; | |
221 | ||
9b1215c1 N |
222 | if (page->index == store->file_pages-1) { |
223 | int last_page_size = store->bytes & (PAGE_SIZE-1); | |
224 | if (last_page_size == 0) | |
225 | last_page_size = PAGE_SIZE; | |
226 | size = roundup(last_page_size, | |
a6ff7e08 | 227 | bdev_logical_block_size(bdev)); |
9b1215c1 | 228 | } |
ac2f40be N |
229 | /* Just make sure we aren't corrupting data or |
230 | * metadata | |
231 | */ | |
232 | if (mddev->external) { | |
233 | /* Bitmap could be anywhere. */ | |
234 | if (rdev->sb_start + offset + (page->index | |
235 | * (PAGE_SIZE/512)) | |
236 | > rdev->data_offset | |
237 | && | |
238 | rdev->sb_start + offset | |
239 | < (rdev->data_offset + mddev->dev_sectors | |
240 | + (PAGE_SIZE/512))) | |
241 | goto bad_alignment; | |
242 | } else if (offset < 0) { | |
243 | /* DATA BITMAP METADATA */ | |
244 | if (offset | |
245 | + (long)(page->index * (PAGE_SIZE/512)) | |
246 | + size/512 > 0) | |
247 | /* bitmap runs in to metadata */ | |
248 | goto bad_alignment; | |
249 | if (rdev->data_offset + mddev->dev_sectors | |
250 | > rdev->sb_start + offset) | |
251 | /* data runs in to bitmap */ | |
252 | goto bad_alignment; | |
253 | } else if (rdev->sb_start < rdev->data_offset) { | |
254 | /* METADATA BITMAP DATA */ | |
255 | if (rdev->sb_start | |
256 | + offset | |
257 | + page->index*(PAGE_SIZE/512) + size/512 | |
258 | > rdev->data_offset) | |
259 | /* bitmap runs in to data */ | |
260 | goto bad_alignment; | |
261 | } else { | |
262 | /* DATA METADATA BITMAP - no problems */ | |
263 | } | |
264 | md_super_write(mddev, rdev, | |
265 | rdev->sb_start + offset | |
266 | + page->index * (PAGE_SIZE/512), | |
267 | size, | |
268 | page); | |
b2d2c4ce | 269 | } |
a654b9d8 N |
270 | |
271 | if (wait) | |
a9701a30 | 272 | md_super_wait(mddev); |
a654b9d8 | 273 | return 0; |
4b80991c N |
274 | |
275 | bad_alignment: | |
4b80991c | 276 | return -EINVAL; |
a654b9d8 N |
277 | } |
278 | ||
4ad13663 | 279 | static void bitmap_file_kick(struct bitmap *bitmap); |
32a7627c | 280 | /* |
a654b9d8 | 281 | * write out a page to a file |
32a7627c | 282 | */ |
4ad13663 | 283 | static void write_page(struct bitmap *bitmap, struct page *page, int wait) |
32a7627c | 284 | { |
d785a06a | 285 | struct buffer_head *bh; |
32a7627c | 286 | |
1ec885cd | 287 | if (bitmap->storage.file == NULL) { |
f0d76d70 N |
288 | switch (write_sb_page(bitmap, page, wait)) { |
289 | case -EINVAL: | |
b405fe91 | 290 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); |
f0d76d70 | 291 | } |
4ad13663 | 292 | } else { |
a654b9d8 | 293 | |
4ad13663 | 294 | bh = page_buffers(page); |
c708443c | 295 | |
4ad13663 N |
296 | while (bh && bh->b_blocknr) { |
297 | atomic_inc(&bitmap->pending_writes); | |
298 | set_buffer_locked(bh); | |
299 | set_buffer_mapped(bh); | |
2a222ca9 | 300 | submit_bh(REQ_OP_WRITE, REQ_SYNC, bh); |
4ad13663 N |
301 | bh = bh->b_this_page; |
302 | } | |
d785a06a | 303 | |
ac2f40be | 304 | if (wait) |
4ad13663 N |
305 | wait_event(bitmap->write_wait, |
306 | atomic_read(&bitmap->pending_writes)==0); | |
8a5e9cf1 | 307 | } |
b405fe91 | 308 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
4ad13663 | 309 | bitmap_file_kick(bitmap); |
d785a06a N |
310 | } |
311 | ||
312 | static void end_bitmap_write(struct buffer_head *bh, int uptodate) | |
313 | { | |
314 | struct bitmap *bitmap = bh->b_private; | |
32a7627c | 315 | |
b405fe91 N |
316 | if (!uptodate) |
317 | set_bit(BITMAP_WRITE_ERROR, &bitmap->flags); | |
d785a06a N |
318 | if (atomic_dec_and_test(&bitmap->pending_writes)) |
319 | wake_up(&bitmap->write_wait); | |
320 | } | |
32a7627c | 321 | |
d785a06a N |
322 | /* copied from buffer.c */ |
323 | static void | |
324 | __clear_page_buffers(struct page *page) | |
325 | { | |
326 | ClearPagePrivate(page); | |
327 | set_page_private(page, 0); | |
09cbfeaf | 328 | put_page(page); |
d785a06a N |
329 | } |
330 | static void free_buffers(struct page *page) | |
331 | { | |
27581e5a | 332 | struct buffer_head *bh; |
77ad4bc7 | 333 | |
27581e5a N |
334 | if (!PagePrivate(page)) |
335 | return; | |
336 | ||
337 | bh = page_buffers(page); | |
d785a06a N |
338 | while (bh) { |
339 | struct buffer_head *next = bh->b_this_page; | |
340 | free_buffer_head(bh); | |
341 | bh = next; | |
77ad4bc7 | 342 | } |
d785a06a N |
343 | __clear_page_buffers(page); |
344 | put_page(page); | |
32a7627c N |
345 | } |
346 | ||
d785a06a N |
347 | /* read a page from a file. |
348 | * We both read the page, and attach buffers to the page to record the | |
349 | * address of each block (using bmap). These addresses will be used | |
350 | * to write the block later, completely bypassing the filesystem. | |
351 | * This usage is similar to how swap files are handled, and allows us | |
352 | * to write to a file with no concerns of memory allocation failing. | |
353 | */ | |
27581e5a N |
354 | static int read_page(struct file *file, unsigned long index, |
355 | struct bitmap *bitmap, | |
356 | unsigned long count, | |
357 | struct page *page) | |
32a7627c | 358 | { |
27581e5a | 359 | int ret = 0; |
496ad9aa | 360 | struct inode *inode = file_inode(file); |
d785a06a N |
361 | struct buffer_head *bh; |
362 | sector_t block; | |
32a7627c | 363 | |
36a4e1fe N |
364 | pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE, |
365 | (unsigned long long)index << PAGE_SHIFT); | |
32a7627c | 366 | |
d785a06a N |
367 | bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0); |
368 | if (!bh) { | |
27581e5a | 369 | ret = -ENOMEM; |
32a7627c N |
370 | goto out; |
371 | } | |
d785a06a N |
372 | attach_page_buffers(page, bh); |
373 | block = index << (PAGE_SHIFT - inode->i_blkbits); | |
374 | while (bh) { | |
375 | if (count == 0) | |
376 | bh->b_blocknr = 0; | |
377 | else { | |
378 | bh->b_blocknr = bmap(inode, block); | |
379 | if (bh->b_blocknr == 0) { | |
380 | /* Cannot use this file! */ | |
27581e5a | 381 | ret = -EINVAL; |
d785a06a N |
382 | goto out; |
383 | } | |
384 | bh->b_bdev = inode->i_sb->s_bdev; | |
385 | if (count < (1<<inode->i_blkbits)) | |
386 | count = 0; | |
387 | else | |
388 | count -= (1<<inode->i_blkbits); | |
389 | ||
390 | bh->b_end_io = end_bitmap_write; | |
391 | bh->b_private = bitmap; | |
ce25c31b N |
392 | atomic_inc(&bitmap->pending_writes); |
393 | set_buffer_locked(bh); | |
394 | set_buffer_mapped(bh); | |
2a222ca9 | 395 | submit_bh(REQ_OP_READ, 0, bh); |
d785a06a N |
396 | } |
397 | block++; | |
398 | bh = bh->b_this_page; | |
399 | } | |
d785a06a | 400 | page->index = index; |
ce25c31b N |
401 | |
402 | wait_event(bitmap->write_wait, | |
403 | atomic_read(&bitmap->pending_writes)==0); | |
b405fe91 | 404 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
27581e5a | 405 | ret = -EIO; |
32a7627c | 406 | out: |
27581e5a | 407 | if (ret) |
ec0cc226 N |
408 | pr_err("md: bitmap read error: (%dB @ %llu): %d\n", |
409 | (int)PAGE_SIZE, | |
410 | (unsigned long long)index << PAGE_SHIFT, | |
411 | ret); | |
27581e5a | 412 | return ret; |
32a7627c N |
413 | } |
414 | ||
415 | /* | |
416 | * bitmap file superblock operations | |
417 | */ | |
418 | ||
85c9ccd4 N |
419 | /* |
420 | * bitmap_wait_writes() should be called before writing any bitmap | |
421 | * blocks, to ensure previous writes, particularly from | |
422 | * bitmap_daemon_work(), have completed. | |
423 | */ | |
424 | static void bitmap_wait_writes(struct bitmap *bitmap) | |
425 | { | |
426 | if (bitmap->storage.file) | |
427 | wait_event(bitmap->write_wait, | |
428 | atomic_read(&bitmap->pending_writes)==0); | |
429 | else | |
430 | md_super_wait(bitmap->mddev); | |
431 | } | |
432 | ||
433 | ||
32a7627c | 434 | /* update the event counter and sync the superblock to disk */ |
4ad13663 | 435 | void bitmap_update_sb(struct bitmap *bitmap) |
32a7627c N |
436 | { |
437 | bitmap_super_t *sb; | |
32a7627c N |
438 | |
439 | if (!bitmap || !bitmap->mddev) /* no bitmap for this array */ | |
4ad13663 | 440 | return; |
ece5cff0 N |
441 | if (bitmap->mddev->bitmap_info.external) |
442 | return; | |
1ec885cd | 443 | if (!bitmap->storage.sb_page) /* no superblock */ |
4ad13663 | 444 | return; |
1ec885cd | 445 | sb = kmap_atomic(bitmap->storage.sb_page); |
32a7627c | 446 | sb->events = cpu_to_le64(bitmap->mddev->events); |
8258c532 | 447 | if (bitmap->mddev->events < bitmap->events_cleared) |
a0da84f3 NB |
448 | /* rocking back to read-only */ |
449 | bitmap->events_cleared = bitmap->mddev->events; | |
8258c532 N |
450 | sb->events_cleared = cpu_to_le64(bitmap->events_cleared); |
451 | sb->state = cpu_to_le32(bitmap->flags); | |
43a70507 N |
452 | /* Just in case these have been changed via sysfs: */ |
453 | sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ); | |
454 | sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind); | |
b81a0404 N |
455 | /* This might have been changed by a reshape */ |
456 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
457 | sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize); | |
c4ce867f | 458 | sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes); |
1dff2b87 N |
459 | sb->sectors_reserved = cpu_to_le32(bitmap->mddev-> |
460 | bitmap_info.space); | |
b2f46e68 | 461 | kunmap_atomic(sb); |
1ec885cd | 462 | write_page(bitmap, bitmap->storage.sb_page, 1); |
32a7627c N |
463 | } |
464 | ||
465 | /* print out the bitmap file superblock */ | |
466 | void bitmap_print_sb(struct bitmap *bitmap) | |
467 | { | |
468 | bitmap_super_t *sb; | |
469 | ||
1ec885cd | 470 | if (!bitmap || !bitmap->storage.sb_page) |
32a7627c | 471 | return; |
1ec885cd | 472 | sb = kmap_atomic(bitmap->storage.sb_page); |
ec0cc226 N |
473 | pr_debug("%s: bitmap file superblock:\n", bmname(bitmap)); |
474 | pr_debug(" magic: %08x\n", le32_to_cpu(sb->magic)); | |
475 | pr_debug(" version: %d\n", le32_to_cpu(sb->version)); | |
476 | pr_debug(" uuid: %08x.%08x.%08x.%08x\n", | |
477 | *(__u32 *)(sb->uuid+0), | |
478 | *(__u32 *)(sb->uuid+4), | |
479 | *(__u32 *)(sb->uuid+8), | |
480 | *(__u32 *)(sb->uuid+12)); | |
481 | pr_debug(" events: %llu\n", | |
482 | (unsigned long long) le64_to_cpu(sb->events)); | |
483 | pr_debug("events cleared: %llu\n", | |
484 | (unsigned long long) le64_to_cpu(sb->events_cleared)); | |
485 | pr_debug(" state: %08x\n", le32_to_cpu(sb->state)); | |
486 | pr_debug(" chunksize: %d B\n", le32_to_cpu(sb->chunksize)); | |
487 | pr_debug(" daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep)); | |
488 | pr_debug(" sync size: %llu KB\n", | |
489 | (unsigned long long)le64_to_cpu(sb->sync_size)/2); | |
490 | pr_debug("max write behind: %d\n", le32_to_cpu(sb->write_behind)); | |
b2f46e68 | 491 | kunmap_atomic(sb); |
32a7627c N |
492 | } |
493 | ||
9c81075f JB |
494 | /* |
495 | * bitmap_new_disk_sb | |
496 | * @bitmap | |
497 | * | |
498 | * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb | |
499 | * reads and verifies the on-disk bitmap superblock and populates bitmap_info. | |
500 | * This function verifies 'bitmap_info' and populates the on-disk bitmap | |
501 | * structure, which is to be written to disk. | |
502 | * | |
503 | * Returns: 0 on success, -Exxx on error | |
504 | */ | |
505 | static int bitmap_new_disk_sb(struct bitmap *bitmap) | |
506 | { | |
507 | bitmap_super_t *sb; | |
508 | unsigned long chunksize, daemon_sleep, write_behind; | |
9c81075f | 509 | |
d3b178ad | 510 | bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO); |
582e2e05 JM |
511 | if (bitmap->storage.sb_page == NULL) |
512 | return -ENOMEM; | |
1ec885cd | 513 | bitmap->storage.sb_page->index = 0; |
9c81075f | 514 | |
1ec885cd | 515 | sb = kmap_atomic(bitmap->storage.sb_page); |
9c81075f JB |
516 | |
517 | sb->magic = cpu_to_le32(BITMAP_MAGIC); | |
518 | sb->version = cpu_to_le32(BITMAP_MAJOR_HI); | |
519 | ||
520 | chunksize = bitmap->mddev->bitmap_info.chunksize; | |
521 | BUG_ON(!chunksize); | |
522 | if (!is_power_of_2(chunksize)) { | |
b2f46e68 | 523 | kunmap_atomic(sb); |
ec0cc226 | 524 | pr_warn("bitmap chunksize not a power of 2\n"); |
9c81075f JB |
525 | return -EINVAL; |
526 | } | |
527 | sb->chunksize = cpu_to_le32(chunksize); | |
528 | ||
529 | daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep; | |
c97e0602 | 530 | if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) { |
ec0cc226 | 531 | pr_debug("Choosing daemon_sleep default (5 sec)\n"); |
9c81075f JB |
532 | daemon_sleep = 5 * HZ; |
533 | } | |
534 | sb->daemon_sleep = cpu_to_le32(daemon_sleep); | |
535 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
536 | ||
537 | /* | |
538 | * FIXME: write_behind for RAID1. If not specified, what | |
539 | * is a good choice? We choose COUNTER_MAX / 2 arbitrarily. | |
540 | */ | |
541 | write_behind = bitmap->mddev->bitmap_info.max_write_behind; | |
542 | if (write_behind > COUNTER_MAX) | |
543 | write_behind = COUNTER_MAX / 2; | |
544 | sb->write_behind = cpu_to_le32(write_behind); | |
545 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
546 | ||
547 | /* keep the array size field of the bitmap superblock up to date */ | |
548 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
549 | ||
550 | memcpy(sb->uuid, bitmap->mddev->uuid, 16); | |
551 | ||
b405fe91 | 552 | set_bit(BITMAP_STALE, &bitmap->flags); |
84e92345 | 553 | sb->state = cpu_to_le32(bitmap->flags); |
9c81075f JB |
554 | bitmap->events_cleared = bitmap->mddev->events; |
555 | sb->events_cleared = cpu_to_le64(bitmap->mddev->events); | |
d3b178ad | 556 | bitmap->mddev->bitmap_info.nodes = 0; |
9c81075f | 557 | |
b2f46e68 | 558 | kunmap_atomic(sb); |
9c81075f JB |
559 | |
560 | return 0; | |
561 | } | |
562 | ||
32a7627c N |
563 | /* read the superblock from the bitmap file and initialize some bitmap fields */ |
564 | static int bitmap_read_sb(struct bitmap *bitmap) | |
565 | { | |
566 | char *reason = NULL; | |
567 | bitmap_super_t *sb; | |
4b6d287f | 568 | unsigned long chunksize, daemon_sleep, write_behind; |
32a7627c | 569 | unsigned long long events; |
c4ce867f | 570 | int nodes = 0; |
1dff2b87 | 571 | unsigned long sectors_reserved = 0; |
32a7627c | 572 | int err = -EINVAL; |
27581e5a | 573 | struct page *sb_page; |
33e38ac6 | 574 | loff_t offset = bitmap->mddev->bitmap_info.offset; |
32a7627c | 575 | |
1ec885cd | 576 | if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) { |
ef99bf48 N |
577 | chunksize = 128 * 1024 * 1024; |
578 | daemon_sleep = 5 * HZ; | |
579 | write_behind = 0; | |
b405fe91 | 580 | set_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
581 | err = 0; |
582 | goto out_no_sb; | |
583 | } | |
32a7627c | 584 | /* page 0 is the superblock, read it... */ |
27581e5a N |
585 | sb_page = alloc_page(GFP_KERNEL); |
586 | if (!sb_page) | |
587 | return -ENOMEM; | |
1ec885cd | 588 | bitmap->storage.sb_page = sb_page; |
27581e5a | 589 | |
b97e9257 | 590 | re_read: |
f9209a32 GR |
591 | /* If cluster_slot is set, the cluster is setup */ |
592 | if (bitmap->cluster_slot >= 0) { | |
3b0e6aac | 593 | sector_t bm_blocks = bitmap->mddev->resync_max_sectors; |
f9209a32 | 594 | |
3b0e6aac SR |
595 | sector_div(bm_blocks, |
596 | bitmap->mddev->bitmap_info.chunksize >> 9); | |
124eb761 GR |
597 | /* bits to bytes */ |
598 | bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t); | |
599 | /* to 4k blocks */ | |
935f3d4f | 600 | bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096); |
33e38ac6 | 601 | offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3)); |
ec0cc226 | 602 | pr_debug("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__, |
33e38ac6 | 603 | bitmap->cluster_slot, offset); |
f9209a32 GR |
604 | } |
605 | ||
1ec885cd N |
606 | if (bitmap->storage.file) { |
607 | loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host); | |
f49d5e62 N |
608 | int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize; |
609 | ||
1ec885cd | 610 | err = read_page(bitmap->storage.file, 0, |
27581e5a | 611 | bitmap, bytes, sb_page); |
f49d5e62 | 612 | } else { |
27581e5a | 613 | err = read_sb_page(bitmap->mddev, |
33e38ac6 | 614 | offset, |
27581e5a N |
615 | sb_page, |
616 | 0, sizeof(bitmap_super_t)); | |
a654b9d8 | 617 | } |
27581e5a | 618 | if (err) |
32a7627c | 619 | return err; |
32a7627c | 620 | |
b97e9257 | 621 | err = -EINVAL; |
27581e5a | 622 | sb = kmap_atomic(sb_page); |
32a7627c | 623 | |
32a7627c | 624 | chunksize = le32_to_cpu(sb->chunksize); |
1b04be96 | 625 | daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ; |
4b6d287f | 626 | write_behind = le32_to_cpu(sb->write_behind); |
1dff2b87 | 627 | sectors_reserved = le32_to_cpu(sb->sectors_reserved); |
3c462c88 GR |
628 | /* Setup nodes/clustername only if bitmap version is |
629 | * cluster-compatible | |
d3b178ad | 630 | */ |
3c462c88 | 631 | if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) { |
d3b178ad GR |
632 | nodes = le32_to_cpu(sb->nodes); |
633 | strlcpy(bitmap->mddev->bitmap_info.cluster_name, | |
634 | sb->cluster_name, 64); | |
635 | } | |
32a7627c N |
636 | |
637 | /* verify that the bitmap-specific fields are valid */ | |
638 | if (sb->magic != cpu_to_le32(BITMAP_MAGIC)) | |
639 | reason = "bad magic"; | |
bd926c63 | 640 | else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO || |
3c462c88 | 641 | le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED) |
32a7627c | 642 | reason = "unrecognized superblock version"; |
1187cf0a | 643 | else if (chunksize < 512) |
7dd5d34c | 644 | reason = "bitmap chunksize too small"; |
d744540c | 645 | else if (!is_power_of_2(chunksize)) |
32a7627c | 646 | reason = "bitmap chunksize not a power of 2"; |
1b04be96 | 647 | else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT) |
7dd5d34c | 648 | reason = "daemon sleep period out of range"; |
4b6d287f N |
649 | else if (write_behind > COUNTER_MAX) |
650 | reason = "write-behind limit out of range (0 - 16383)"; | |
32a7627c | 651 | if (reason) { |
ec0cc226 | 652 | pr_warn("%s: invalid bitmap file superblock: %s\n", |
32a7627c N |
653 | bmname(bitmap), reason); |
654 | goto out; | |
655 | } | |
656 | ||
657 | /* keep the array size field of the bitmap superblock up to date */ | |
658 | sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors); | |
659 | ||
278c1ca2 N |
660 | if (bitmap->mddev->persistent) { |
661 | /* | |
662 | * We have a persistent array superblock, so compare the | |
663 | * bitmap's UUID and event counter to the mddev's | |
664 | */ | |
665 | if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) { | |
ec0cc226 N |
666 | pr_warn("%s: bitmap superblock UUID mismatch\n", |
667 | bmname(bitmap)); | |
278c1ca2 N |
668 | goto out; |
669 | } | |
670 | events = le64_to_cpu(sb->events); | |
b97e9257 | 671 | if (!nodes && (events < bitmap->mddev->events)) { |
ec0cc226 N |
672 | pr_warn("%s: bitmap file is out of date (%llu < %llu) -- forcing full recovery\n", |
673 | bmname(bitmap), events, | |
674 | (unsigned long long) bitmap->mddev->events); | |
b405fe91 | 675 | set_bit(BITMAP_STALE, &bitmap->flags); |
278c1ca2 | 676 | } |
32a7627c | 677 | } |
278c1ca2 | 678 | |
32a7627c | 679 | /* assign fields using values from superblock */ |
4f2e639a | 680 | bitmap->flags |= le32_to_cpu(sb->state); |
bd926c63 | 681 | if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN) |
b405fe91 | 682 | set_bit(BITMAP_HOSTENDIAN, &bitmap->flags); |
32a7627c | 683 | bitmap->events_cleared = le64_to_cpu(sb->events_cleared); |
cf921cc1 | 684 | strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64); |
32a7627c | 685 | err = 0; |
b97e9257 | 686 | |
32a7627c | 687 | out: |
b2f46e68 | 688 | kunmap_atomic(sb); |
f9209a32 GR |
689 | /* Assiging chunksize is required for "re_read" */ |
690 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
f7357273 | 691 | if (err == 0 && nodes && (bitmap->cluster_slot < 0)) { |
b97e9257 GR |
692 | err = md_setup_cluster(bitmap->mddev, nodes); |
693 | if (err) { | |
ec0cc226 N |
694 | pr_warn("%s: Could not setup cluster service (%d)\n", |
695 | bmname(bitmap), err); | |
b97e9257 GR |
696 | goto out_no_sb; |
697 | } | |
698 | bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev); | |
b97e9257 GR |
699 | goto re_read; |
700 | } | |
701 | ||
702 | ||
ef99bf48 | 703 | out_no_sb: |
b405fe91 | 704 | if (test_bit(BITMAP_STALE, &bitmap->flags)) |
ef99bf48 N |
705 | bitmap->events_cleared = bitmap->mddev->events; |
706 | bitmap->mddev->bitmap_info.chunksize = chunksize; | |
707 | bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep; | |
708 | bitmap->mddev->bitmap_info.max_write_behind = write_behind; | |
c4ce867f | 709 | bitmap->mddev->bitmap_info.nodes = nodes; |
1dff2b87 N |
710 | if (bitmap->mddev->bitmap_info.space == 0 || |
711 | bitmap->mddev->bitmap_info.space > sectors_reserved) | |
712 | bitmap->mddev->bitmap_info.space = sectors_reserved; | |
b97e9257 | 713 | if (err) { |
32a7627c | 714 | bitmap_print_sb(bitmap); |
f9209a32 | 715 | if (bitmap->cluster_slot < 0) |
b97e9257 GR |
716 | md_cluster_stop(bitmap->mddev); |
717 | } | |
32a7627c N |
718 | return err; |
719 | } | |
720 | ||
32a7627c N |
721 | /* |
722 | * general bitmap file operations | |
723 | */ | |
724 | ||
ece5cff0 N |
725 | /* |
726 | * on-disk bitmap: | |
727 | * | |
728 | * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap | |
729 | * file a page at a time. There's a superblock at the start of the file. | |
730 | */ | |
32a7627c | 731 | /* calculate the index of the page that contains this bit */ |
1ec885cd N |
732 | static inline unsigned long file_page_index(struct bitmap_storage *store, |
733 | unsigned long chunk) | |
32a7627c | 734 | { |
1ec885cd | 735 | if (store->sb_page) |
ece5cff0 N |
736 | chunk += sizeof(bitmap_super_t) << 3; |
737 | return chunk >> PAGE_BIT_SHIFT; | |
32a7627c N |
738 | } |
739 | ||
740 | /* calculate the (bit) offset of this bit within a page */ | |
1ec885cd N |
741 | static inline unsigned long file_page_offset(struct bitmap_storage *store, |
742 | unsigned long chunk) | |
32a7627c | 743 | { |
1ec885cd | 744 | if (store->sb_page) |
ece5cff0 N |
745 | chunk += sizeof(bitmap_super_t) << 3; |
746 | return chunk & (PAGE_BITS - 1); | |
32a7627c N |
747 | } |
748 | ||
749 | /* | |
750 | * return a pointer to the page in the filemap that contains the given bit | |
751 | * | |
32a7627c | 752 | */ |
1ec885cd | 753 | static inline struct page *filemap_get_page(struct bitmap_storage *store, |
3520fa4d | 754 | unsigned long chunk) |
32a7627c | 755 | { |
1ec885cd | 756 | if (file_page_index(store, chunk) >= store->file_pages) |
ac2f40be | 757 | return NULL; |
f2e06c58 | 758 | return store->filemap[file_page_index(store, chunk)]; |
32a7627c N |
759 | } |
760 | ||
d1244cb0 | 761 | static int bitmap_storage_alloc(struct bitmap_storage *store, |
b97e9257 GR |
762 | unsigned long chunks, int with_super, |
763 | int slot_number) | |
d1244cb0 | 764 | { |
b97e9257 | 765 | int pnum, offset = 0; |
d1244cb0 N |
766 | unsigned long num_pages; |
767 | unsigned long bytes; | |
768 | ||
769 | bytes = DIV_ROUND_UP(chunks, 8); | |
770 | if (with_super) | |
771 | bytes += sizeof(bitmap_super_t); | |
772 | ||
773 | num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE); | |
7f86ffed | 774 | offset = slot_number * num_pages; |
d1244cb0 N |
775 | |
776 | store->filemap = kmalloc(sizeof(struct page *) | |
777 | * num_pages, GFP_KERNEL); | |
778 | if (!store->filemap) | |
779 | return -ENOMEM; | |
780 | ||
781 | if (with_super && !store->sb_page) { | |
d60b479d | 782 | store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
783 | if (store->sb_page == NULL) |
784 | return -ENOMEM; | |
d1244cb0 | 785 | } |
b97e9257 | 786 | |
d1244cb0 N |
787 | pnum = 0; |
788 | if (store->sb_page) { | |
789 | store->filemap[0] = store->sb_page; | |
790 | pnum = 1; | |
b97e9257 | 791 | store->sb_page->index = offset; |
d1244cb0 | 792 | } |
b97e9257 | 793 | |
d1244cb0 | 794 | for ( ; pnum < num_pages; pnum++) { |
d60b479d | 795 | store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO); |
d1244cb0 N |
796 | if (!store->filemap[pnum]) { |
797 | store->file_pages = pnum; | |
798 | return -ENOMEM; | |
799 | } | |
b97e9257 | 800 | store->filemap[pnum]->index = pnum + offset; |
d1244cb0 N |
801 | } |
802 | store->file_pages = pnum; | |
803 | ||
804 | /* We need 4 bits per page, rounded up to a multiple | |
805 | * of sizeof(unsigned long) */ | |
806 | store->filemap_attr = kzalloc( | |
807 | roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)), | |
808 | GFP_KERNEL); | |
809 | if (!store->filemap_attr) | |
810 | return -ENOMEM; | |
811 | ||
812 | store->bytes = bytes; | |
813 | ||
814 | return 0; | |
815 | } | |
816 | ||
fae7d326 | 817 | static void bitmap_file_unmap(struct bitmap_storage *store) |
32a7627c N |
818 | { |
819 | struct page **map, *sb_page; | |
32a7627c | 820 | int pages; |
fae7d326 | 821 | struct file *file; |
32a7627c | 822 | |
fae7d326 | 823 | file = store->file; |
1ec885cd | 824 | map = store->filemap; |
1ec885cd | 825 | pages = store->file_pages; |
1ec885cd | 826 | sb_page = store->sb_page; |
32a7627c N |
827 | |
828 | while (pages--) | |
ece5cff0 | 829 | if (map[pages] != sb_page) /* 0 is sb_page, release it below */ |
d785a06a | 830 | free_buffers(map[pages]); |
32a7627c | 831 | kfree(map); |
fae7d326 | 832 | kfree(store->filemap_attr); |
32a7627c | 833 | |
d785a06a N |
834 | if (sb_page) |
835 | free_buffers(sb_page); | |
32a7627c | 836 | |
d785a06a | 837 | if (file) { |
496ad9aa | 838 | struct inode *inode = file_inode(file); |
fc0ecff6 | 839 | invalidate_mapping_pages(inode->i_mapping, 0, -1); |
32a7627c | 840 | fput(file); |
d785a06a | 841 | } |
32a7627c N |
842 | } |
843 | ||
32a7627c N |
844 | /* |
845 | * bitmap_file_kick - if an error occurs while manipulating the bitmap file | |
846 | * then it is no longer reliable, so we stop using it and we mark the file | |
847 | * as failed in the superblock | |
848 | */ | |
849 | static void bitmap_file_kick(struct bitmap *bitmap) | |
850 | { | |
851 | char *path, *ptr = NULL; | |
852 | ||
b405fe91 | 853 | if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) { |
4ad13663 | 854 | bitmap_update_sb(bitmap); |
32a7627c | 855 | |
1ec885cd | 856 | if (bitmap->storage.file) { |
4ad13663 N |
857 | path = kmalloc(PAGE_SIZE, GFP_KERNEL); |
858 | if (path) | |
9bf39ab2 | 859 | ptr = file_path(bitmap->storage.file, |
1ec885cd | 860 | path, PAGE_SIZE); |
6bcfd601 | 861 | |
ec0cc226 N |
862 | pr_warn("%s: kicking failed bitmap file %s from array!\n", |
863 | bmname(bitmap), IS_ERR(ptr) ? "" : ptr); | |
32a7627c | 864 | |
4ad13663 N |
865 | kfree(path); |
866 | } else | |
ec0cc226 N |
867 | pr_warn("%s: disabling internal bitmap due to errors\n", |
868 | bmname(bitmap)); | |
a654b9d8 | 869 | } |
32a7627c N |
870 | } |
871 | ||
872 | enum bitmap_page_attr { | |
ac2f40be | 873 | BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */ |
5a537df4 N |
874 | BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned. |
875 | * i.e. counter is 1 or 2. */ | |
ac2f40be | 876 | BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */ |
32a7627c N |
877 | }; |
878 | ||
d189122d N |
879 | static inline void set_page_attr(struct bitmap *bitmap, int pnum, |
880 | enum bitmap_page_attr attr) | |
32a7627c | 881 | { |
bdfd1140 | 882 | set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
883 | } |
884 | ||
d189122d N |
885 | static inline void clear_page_attr(struct bitmap *bitmap, int pnum, |
886 | enum bitmap_page_attr attr) | |
32a7627c | 887 | { |
bdfd1140 | 888 | clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
889 | } |
890 | ||
bdfd1140 N |
891 | static inline int test_page_attr(struct bitmap *bitmap, int pnum, |
892 | enum bitmap_page_attr attr) | |
32a7627c | 893 | { |
1ec885cd | 894 | return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr); |
32a7627c N |
895 | } |
896 | ||
bdfd1140 N |
897 | static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum, |
898 | enum bitmap_page_attr attr) | |
899 | { | |
900 | return test_and_clear_bit((pnum<<2) + attr, | |
901 | bitmap->storage.filemap_attr); | |
902 | } | |
32a7627c N |
903 | /* |
904 | * bitmap_file_set_bit -- called before performing a write to the md device | |
905 | * to set (and eventually sync) a particular bit in the bitmap file | |
906 | * | |
907 | * we set the bit immediately, then we record the page number so that | |
908 | * when an unplug occurs, we can flush the dirty pages out to disk | |
909 | */ | |
910 | static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block) | |
911 | { | |
912 | unsigned long bit; | |
3520fa4d | 913 | struct page *page; |
32a7627c | 914 | void *kaddr; |
40cffcc0 | 915 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
23cea66a GJ |
916 | struct bitmap_storage *store = &bitmap->storage; |
917 | unsigned long node_offset = 0; | |
918 | ||
919 | if (mddev_is_clustered(bitmap->mddev)) | |
920 | node_offset = bitmap->cluster_slot * store->file_pages; | |
32a7627c | 921 | |
1ec885cd | 922 | page = filemap_get_page(&bitmap->storage, chunk); |
3520fa4d JB |
923 | if (!page) |
924 | return; | |
1ec885cd | 925 | bit = file_page_offset(&bitmap->storage, chunk); |
32a7627c | 926 | |
3520fa4d | 927 | /* set the bit */ |
b2f46e68 | 928 | kaddr = kmap_atomic(page); |
b405fe91 | 929 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
3520fa4d JB |
930 | set_bit(bit, kaddr); |
931 | else | |
3f810b6c | 932 | set_bit_le(bit, kaddr); |
b2f46e68 | 933 | kunmap_atomic(kaddr); |
36a4e1fe | 934 | pr_debug("set file bit %lu page %lu\n", bit, page->index); |
32a7627c | 935 | /* record page number so it gets flushed to disk when unplug occurs */ |
23cea66a | 936 | set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_DIRTY); |
32a7627c N |
937 | } |
938 | ||
ef99bf48 N |
939 | static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block) |
940 | { | |
941 | unsigned long bit; | |
942 | struct page *page; | |
943 | void *paddr; | |
40cffcc0 | 944 | unsigned long chunk = block >> bitmap->counts.chunkshift; |
23cea66a GJ |
945 | struct bitmap_storage *store = &bitmap->storage; |
946 | unsigned long node_offset = 0; | |
947 | ||
948 | if (mddev_is_clustered(bitmap->mddev)) | |
949 | node_offset = bitmap->cluster_slot * store->file_pages; | |
ef99bf48 | 950 | |
1ec885cd | 951 | page = filemap_get_page(&bitmap->storage, chunk); |
ef99bf48 N |
952 | if (!page) |
953 | return; | |
1ec885cd | 954 | bit = file_page_offset(&bitmap->storage, chunk); |
ef99bf48 | 955 | paddr = kmap_atomic(page); |
b405fe91 | 956 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ef99bf48 N |
957 | clear_bit(bit, paddr); |
958 | else | |
3f810b6c | 959 | clear_bit_le(bit, paddr); |
ef99bf48 | 960 | kunmap_atomic(paddr); |
23cea66a GJ |
961 | if (!test_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_NEEDWRITE)) { |
962 | set_page_attr(bitmap, page->index - node_offset, BITMAP_PAGE_PENDING); | |
ef99bf48 N |
963 | bitmap->allclean = 0; |
964 | } | |
965 | } | |
966 | ||
11dd35da GR |
967 | static int bitmap_file_test_bit(struct bitmap *bitmap, sector_t block) |
968 | { | |
969 | unsigned long bit; | |
970 | struct page *page; | |
971 | void *paddr; | |
972 | unsigned long chunk = block >> bitmap->counts.chunkshift; | |
973 | int set = 0; | |
974 | ||
975 | page = filemap_get_page(&bitmap->storage, chunk); | |
976 | if (!page) | |
977 | return -EINVAL; | |
978 | bit = file_page_offset(&bitmap->storage, chunk); | |
979 | paddr = kmap_atomic(page); | |
980 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) | |
981 | set = test_bit(bit, paddr); | |
982 | else | |
983 | set = test_bit_le(bit, paddr); | |
984 | kunmap_atomic(paddr); | |
985 | return set; | |
986 | } | |
987 | ||
988 | ||
32a7627c N |
989 | /* this gets called when the md device is ready to unplug its underlying |
990 | * (slave) device queues -- before we let any writes go down, we need to | |
991 | * sync the dirty pages of the bitmap file to disk */ | |
4ad13663 | 992 | void bitmap_unplug(struct bitmap *bitmap) |
32a7627c | 993 | { |
74667123 | 994 | unsigned long i; |
ec7a3197 | 995 | int dirty, need_write; |
85c9ccd4 | 996 | int writing = 0; |
32a7627c | 997 | |
62f82faa N |
998 | if (!bitmap || !bitmap->storage.filemap || |
999 | test_bit(BITMAP_STALE, &bitmap->flags)) | |
4ad13663 | 1000 | return; |
32a7627c N |
1001 | |
1002 | /* look at each page to see if there are any set bits that need to be | |
1003 | * flushed out to disk */ | |
1ec885cd | 1004 | for (i = 0; i < bitmap->storage.file_pages; i++) { |
bdfd1140 | 1005 | if (!bitmap->storage.filemap) |
4ad13663 | 1006 | return; |
bdfd1140 N |
1007 | dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); |
1008 | need_write = test_and_clear_page_attr(bitmap, i, | |
1009 | BITMAP_PAGE_NEEDWRITE); | |
1010 | if (dirty || need_write) { | |
85c9ccd4 N |
1011 | if (!writing) |
1012 | bitmap_wait_writes(bitmap); | |
d189122d | 1013 | clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING); |
bdfd1140 | 1014 | write_page(bitmap, bitmap->storage.filemap[i], 0); |
85c9ccd4 | 1015 | writing = 1; |
bdfd1140 | 1016 | } |
32a7627c | 1017 | } |
85c9ccd4 N |
1018 | if (writing) |
1019 | bitmap_wait_writes(bitmap); | |
4b5060dd | 1020 | |
b405fe91 | 1021 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
d785a06a | 1022 | bitmap_file_kick(bitmap); |
32a7627c | 1023 | } |
ac2f40be | 1024 | EXPORT_SYMBOL(bitmap_unplug); |
32a7627c | 1025 | |
6a07997f | 1026 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed); |
32a7627c N |
1027 | /* * bitmap_init_from_disk -- called at bitmap_create time to initialize |
1028 | * the in-memory bitmap from the on-disk bitmap -- also, sets up the | |
1029 | * memory mapping of the bitmap file | |
1030 | * Special cases: | |
1031 | * if there's no bitmap file, or if the bitmap file had been | |
1032 | * previously kicked from the array, we mark all the bits as | |
1033 | * 1's in order to cause a full resync. | |
6a07997f N |
1034 | * |
1035 | * We ignore all bits for sectors that end earlier than 'start'. | |
1036 | * This is used when reading an out-of-date bitmap... | |
32a7627c | 1037 | */ |
6a07997f | 1038 | static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start) |
32a7627c | 1039 | { |
b97e9257 | 1040 | unsigned long i, chunks, index, oldindex, bit, node_offset = 0; |
27581e5a | 1041 | struct page *page = NULL; |
d1244cb0 | 1042 | unsigned long bit_cnt = 0; |
32a7627c | 1043 | struct file *file; |
d1244cb0 | 1044 | unsigned long offset; |
32a7627c N |
1045 | int outofdate; |
1046 | int ret = -ENOSPC; | |
ea03aff9 | 1047 | void *paddr; |
1ec885cd | 1048 | struct bitmap_storage *store = &bitmap->storage; |
32a7627c | 1049 | |
40cffcc0 | 1050 | chunks = bitmap->counts.chunks; |
1ec885cd | 1051 | file = store->file; |
32a7627c | 1052 | |
ef99bf48 N |
1053 | if (!file && !bitmap->mddev->bitmap_info.offset) { |
1054 | /* No permanent bitmap - fill with '1s'. */ | |
1ec885cd N |
1055 | store->filemap = NULL; |
1056 | store->file_pages = 0; | |
ef99bf48 N |
1057 | for (i = 0; i < chunks ; i++) { |
1058 | /* if the disk bit is set, set the memory bit */ | |
40cffcc0 | 1059 | int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift) |
ef99bf48 N |
1060 | >= start); |
1061 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 1062 | (sector_t)i << bitmap->counts.chunkshift, |
ef99bf48 N |
1063 | needed); |
1064 | } | |
1065 | return 0; | |
1066 | } | |
32a7627c | 1067 | |
b405fe91 | 1068 | outofdate = test_bit(BITMAP_STALE, &bitmap->flags); |
32a7627c | 1069 | if (outofdate) |
ec0cc226 | 1070 | pr_warn("%s: bitmap file is out of date, doing full recovery\n", bmname(bitmap)); |
32a7627c | 1071 | |
d1244cb0 | 1072 | if (file && i_size_read(file->f_mapping->host) < store->bytes) { |
ec0cc226 N |
1073 | pr_warn("%s: bitmap file too short %lu < %lu\n", |
1074 | bmname(bitmap), | |
1075 | (unsigned long) i_size_read(file->f_mapping->host), | |
1076 | store->bytes); | |
4ad13663 | 1077 | goto err; |
32a7627c | 1078 | } |
bc7f77de | 1079 | |
d1244cb0 | 1080 | oldindex = ~0L; |
27581e5a | 1081 | offset = 0; |
d1244cb0 | 1082 | if (!bitmap->mddev->bitmap_info.external) |
27581e5a | 1083 | offset = sizeof(bitmap_super_t); |
32a7627c | 1084 | |
b97e9257 GR |
1085 | if (mddev_is_clustered(bitmap->mddev)) |
1086 | node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE)); | |
1087 | ||
32a7627c | 1088 | for (i = 0; i < chunks; i++) { |
bd926c63 | 1089 | int b; |
1ec885cd N |
1090 | index = file_page_index(&bitmap->storage, i); |
1091 | bit = file_page_offset(&bitmap->storage, i); | |
32a7627c | 1092 | if (index != oldindex) { /* this is a new page, read it in */ |
d785a06a | 1093 | int count; |
32a7627c | 1094 | /* unmap the old page, we're done with it */ |
d1244cb0 N |
1095 | if (index == store->file_pages-1) |
1096 | count = store->bytes - index * PAGE_SIZE; | |
d785a06a N |
1097 | else |
1098 | count = PAGE_SIZE; | |
1ec885cd | 1099 | page = store->filemap[index]; |
27581e5a N |
1100 | if (file) |
1101 | ret = read_page(file, index, bitmap, | |
1102 | count, page); | |
1103 | else | |
1104 | ret = read_sb_page( | |
1105 | bitmap->mddev, | |
1106 | bitmap->mddev->bitmap_info.offset, | |
1107 | page, | |
b97e9257 | 1108 | index + node_offset, count); |
27581e5a N |
1109 | |
1110 | if (ret) | |
4ad13663 | 1111 | goto err; |
a654b9d8 | 1112 | |
32a7627c | 1113 | oldindex = index; |
32a7627c N |
1114 | |
1115 | if (outofdate) { | |
1116 | /* | |
1117 | * if bitmap is out of date, dirty the | |
ac2f40be | 1118 | * whole page and write it out |
32a7627c | 1119 | */ |
b2f46e68 | 1120 | paddr = kmap_atomic(page); |
ea03aff9 | 1121 | memset(paddr + offset, 0xff, |
6a07997f | 1122 | PAGE_SIZE - offset); |
b2f46e68 | 1123 | kunmap_atomic(paddr); |
4ad13663 N |
1124 | write_page(bitmap, page, 1); |
1125 | ||
1126 | ret = -EIO; | |
b405fe91 N |
1127 | if (test_bit(BITMAP_WRITE_ERROR, |
1128 | &bitmap->flags)) | |
4ad13663 | 1129 | goto err; |
32a7627c | 1130 | } |
32a7627c | 1131 | } |
b2f46e68 | 1132 | paddr = kmap_atomic(page); |
b405fe91 | 1133 | if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags)) |
ea03aff9 | 1134 | b = test_bit(bit, paddr); |
bd926c63 | 1135 | else |
6b33aff3 | 1136 | b = test_bit_le(bit, paddr); |
b2f46e68 | 1137 | kunmap_atomic(paddr); |
bd926c63 | 1138 | if (b) { |
32a7627c | 1139 | /* if the disk bit is set, set the memory bit */ |
40cffcc0 | 1140 | int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift |
db305e50 N |
1141 | >= start); |
1142 | bitmap_set_memory_bits(bitmap, | |
40cffcc0 | 1143 | (sector_t)i << bitmap->counts.chunkshift, |
db305e50 | 1144 | needed); |
32a7627c N |
1145 | bit_cnt++; |
1146 | } | |
27581e5a | 1147 | offset = 0; |
32a7627c N |
1148 | } |
1149 | ||
ec0cc226 N |
1150 | pr_debug("%s: bitmap initialized from disk: read %lu pages, set %lu of %lu bits\n", |
1151 | bmname(bitmap), store->file_pages, | |
1152 | bit_cnt, chunks); | |
4ad13663 N |
1153 | |
1154 | return 0; | |
32a7627c | 1155 | |
4ad13663 | 1156 | err: |
ec0cc226 N |
1157 | pr_warn("%s: bitmap initialisation failed: %d\n", |
1158 | bmname(bitmap), ret); | |
32a7627c N |
1159 | return ret; |
1160 | } | |
1161 | ||
a654b9d8 N |
1162 | void bitmap_write_all(struct bitmap *bitmap) |
1163 | { | |
1164 | /* We don't actually write all bitmap blocks here, | |
1165 | * just flag them as needing to be written | |
1166 | */ | |
ec7a3197 | 1167 | int i; |
a654b9d8 | 1168 | |
1ec885cd | 1169 | if (!bitmap || !bitmap->storage.filemap) |
ef99bf48 | 1170 | return; |
1ec885cd | 1171 | if (bitmap->storage.file) |
ef99bf48 N |
1172 | /* Only one copy, so nothing needed */ |
1173 | return; | |
1174 | ||
1ec885cd | 1175 | for (i = 0; i < bitmap->storage.file_pages; i++) |
d189122d | 1176 | set_page_attr(bitmap, i, |
ec7a3197 | 1177 | BITMAP_PAGE_NEEDWRITE); |
2585f3ef | 1178 | bitmap->allclean = 0; |
a654b9d8 N |
1179 | } |
1180 | ||
40cffcc0 N |
1181 | static void bitmap_count_page(struct bitmap_counts *bitmap, |
1182 | sector_t offset, int inc) | |
32a7627c | 1183 | { |
61a0d80c | 1184 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1185 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1186 | bitmap->bp[page].count += inc; | |
32a7627c N |
1187 | bitmap_checkfree(bitmap, page); |
1188 | } | |
bf07bb7d | 1189 | |
40cffcc0 | 1190 | static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset) |
bf07bb7d N |
1191 | { |
1192 | sector_t chunk = offset >> bitmap->chunkshift; | |
1193 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; | |
1194 | struct bitmap_page *bp = &bitmap->bp[page]; | |
1195 | ||
1196 | if (!bp->pending) | |
1197 | bp->pending = 1; | |
1198 | } | |
1199 | ||
40cffcc0 | 1200 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1201 | sector_t offset, sector_t *blocks, |
32a7627c N |
1202 | int create); |
1203 | ||
1204 | /* | |
1205 | * bitmap daemon -- periodically wakes up to clean bits and flush pages | |
1206 | * out to disk | |
1207 | */ | |
1208 | ||
fd01b88c | 1209 | void bitmap_daemon_work(struct mddev *mddev) |
32a7627c | 1210 | { |
aa5cbd10 | 1211 | struct bitmap *bitmap; |
aa3163f8 | 1212 | unsigned long j; |
bf07bb7d | 1213 | unsigned long nextpage; |
57dab0bd | 1214 | sector_t blocks; |
40cffcc0 | 1215 | struct bitmap_counts *counts; |
32a7627c | 1216 | |
aa5cbd10 N |
1217 | /* Use a mutex to guard daemon_work against |
1218 | * bitmap_destroy. | |
1219 | */ | |
c3d9714e | 1220 | mutex_lock(&mddev->bitmap_info.mutex); |
aa5cbd10 N |
1221 | bitmap = mddev->bitmap; |
1222 | if (bitmap == NULL) { | |
c3d9714e | 1223 | mutex_unlock(&mddev->bitmap_info.mutex); |
4ad13663 | 1224 | return; |
aa5cbd10 | 1225 | } |
42a04b50 | 1226 | if (time_before(jiffies, bitmap->daemon_lastrun |
2e61ebbc | 1227 | + mddev->bitmap_info.daemon_sleep)) |
7be3dfec N |
1228 | goto done; |
1229 | ||
32a7627c | 1230 | bitmap->daemon_lastrun = jiffies; |
8311c29d | 1231 | if (bitmap->allclean) { |
2e61ebbc | 1232 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; |
aa5cbd10 | 1233 | goto done; |
8311c29d N |
1234 | } |
1235 | bitmap->allclean = 1; | |
32a7627c | 1236 | |
bf07bb7d N |
1237 | /* Any file-page which is PENDING now needs to be written. |
1238 | * So set NEEDWRITE now, then after we make any last-minute changes | |
1239 | * we will write it. | |
1240 | */ | |
1ec885cd | 1241 | for (j = 0; j < bitmap->storage.file_pages; j++) |
bdfd1140 N |
1242 | if (test_and_clear_page_attr(bitmap, j, |
1243 | BITMAP_PAGE_PENDING)) | |
d189122d | 1244 | set_page_attr(bitmap, j, |
bf07bb7d | 1245 | BITMAP_PAGE_NEEDWRITE); |
bf07bb7d N |
1246 | |
1247 | if (bitmap->need_sync && | |
1248 | mddev->bitmap_info.external == 0) { | |
1249 | /* Arrange for superblock update as well as | |
1250 | * other changes */ | |
1251 | bitmap_super_t *sb; | |
1252 | bitmap->need_sync = 0; | |
1ec885cd N |
1253 | if (bitmap->storage.filemap) { |
1254 | sb = kmap_atomic(bitmap->storage.sb_page); | |
ef99bf48 N |
1255 | sb->events_cleared = |
1256 | cpu_to_le64(bitmap->events_cleared); | |
1257 | kunmap_atomic(sb); | |
d189122d | 1258 | set_page_attr(bitmap, 0, |
ef99bf48 N |
1259 | BITMAP_PAGE_NEEDWRITE); |
1260 | } | |
bf07bb7d N |
1261 | } |
1262 | /* Now look at the bitmap counters and if any are '2' or '1', | |
1263 | * decrement and handle accordingly. | |
1264 | */ | |
40cffcc0 N |
1265 | counts = &bitmap->counts; |
1266 | spin_lock_irq(&counts->lock); | |
bf07bb7d | 1267 | nextpage = 0; |
40cffcc0 | 1268 | for (j = 0; j < counts->chunks; j++) { |
32a7627c | 1269 | bitmap_counter_t *bmc; |
40cffcc0 | 1270 | sector_t block = (sector_t)j << counts->chunkshift; |
3520fa4d | 1271 | |
bf07bb7d N |
1272 | if (j == nextpage) { |
1273 | nextpage += PAGE_COUNTER_RATIO; | |
40cffcc0 | 1274 | if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) { |
bf07bb7d | 1275 | j |= PAGE_COUNTER_MASK; |
aa3163f8 N |
1276 | continue; |
1277 | } | |
40cffcc0 | 1278 | counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0; |
32a7627c | 1279 | } |
40cffcc0 | 1280 | bmc = bitmap_get_counter(counts, |
ef99bf48 | 1281 | block, |
db305e50 | 1282 | &blocks, 0); |
bf07bb7d N |
1283 | |
1284 | if (!bmc) { | |
5a537df4 | 1285 | j |= PAGE_COUNTER_MASK; |
bf07bb7d N |
1286 | continue; |
1287 | } | |
1288 | if (*bmc == 1 && !bitmap->need_sync) { | |
1289 | /* We can clear the bit */ | |
bf07bb7d | 1290 | *bmc = 0; |
40cffcc0 | 1291 | bitmap_count_page(counts, block, -1); |
ef99bf48 | 1292 | bitmap_file_clear_bit(bitmap, block); |
bf07bb7d N |
1293 | } else if (*bmc && *bmc <= 2) { |
1294 | *bmc = 1; | |
40cffcc0 | 1295 | bitmap_set_pending(counts, block); |
bf07bb7d | 1296 | bitmap->allclean = 0; |
5a537df4 | 1297 | } |
32a7627c | 1298 | } |
40cffcc0 | 1299 | spin_unlock_irq(&counts->lock); |
32a7627c | 1300 | |
85c9ccd4 | 1301 | bitmap_wait_writes(bitmap); |
bf07bb7d N |
1302 | /* Now start writeout on any page in NEEDWRITE that isn't DIRTY. |
1303 | * DIRTY pages need to be written by bitmap_unplug so it can wait | |
1304 | * for them. | |
1305 | * If we find any DIRTY page we stop there and let bitmap_unplug | |
1306 | * handle all the rest. This is important in the case where | |
1307 | * the first blocking holds the superblock and it has been updated. | |
1308 | * We mustn't write any other blocks before the superblock. | |
1309 | */ | |
62f82faa N |
1310 | for (j = 0; |
1311 | j < bitmap->storage.file_pages | |
1312 | && !test_bit(BITMAP_STALE, &bitmap->flags); | |
1313 | j++) { | |
d189122d | 1314 | if (test_page_attr(bitmap, j, |
bf07bb7d N |
1315 | BITMAP_PAGE_DIRTY)) |
1316 | /* bitmap_unplug will handle the rest */ | |
1317 | break; | |
bdfd1140 N |
1318 | if (test_and_clear_page_attr(bitmap, j, |
1319 | BITMAP_PAGE_NEEDWRITE)) { | |
1ec885cd | 1320 | write_page(bitmap, bitmap->storage.filemap[j], 0); |
32a7627c | 1321 | } |
32a7627c N |
1322 | } |
1323 | ||
7be3dfec | 1324 | done: |
8311c29d | 1325 | if (bitmap->allclean == 0) |
2e61ebbc N |
1326 | mddev->thread->timeout = |
1327 | mddev->bitmap_info.daemon_sleep; | |
c3d9714e | 1328 | mutex_unlock(&mddev->bitmap_info.mutex); |
32a7627c N |
1329 | } |
1330 | ||
40cffcc0 | 1331 | static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap, |
57dab0bd | 1332 | sector_t offset, sector_t *blocks, |
32a7627c | 1333 | int create) |
ee305ace N |
1334 | __releases(bitmap->lock) |
1335 | __acquires(bitmap->lock) | |
32a7627c N |
1336 | { |
1337 | /* If 'create', we might release the lock and reclaim it. | |
1338 | * The lock must have been taken with interrupts enabled. | |
1339 | * If !create, we don't release the lock. | |
1340 | */ | |
61a0d80c | 1341 | sector_t chunk = offset >> bitmap->chunkshift; |
32a7627c N |
1342 | unsigned long page = chunk >> PAGE_COUNTER_SHIFT; |
1343 | unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT; | |
1344 | sector_t csize; | |
ef425673 | 1345 | int err; |
32a7627c | 1346 | |
c9d65032 | 1347 | err = bitmap_checkpage(bitmap, page, create, 0); |
ef425673 N |
1348 | |
1349 | if (bitmap->bp[page].hijacked || | |
1350 | bitmap->bp[page].map == NULL) | |
61a0d80c | 1351 | csize = ((sector_t)1) << (bitmap->chunkshift + |
ef425673 N |
1352 | PAGE_COUNTER_SHIFT - 1); |
1353 | else | |
61a0d80c | 1354 | csize = ((sector_t)1) << bitmap->chunkshift; |
ef425673 N |
1355 | *blocks = csize - (offset & (csize - 1)); |
1356 | ||
1357 | if (err < 0) | |
32a7627c | 1358 | return NULL; |
ef425673 | 1359 | |
32a7627c N |
1360 | /* now locked ... */ |
1361 | ||
1362 | if (bitmap->bp[page].hijacked) { /* hijacked pointer */ | |
1363 | /* should we use the first or second counter field | |
1364 | * of the hijacked pointer? */ | |
1365 | int hi = (pageoff > PAGE_COUNTER_MASK); | |
32a7627c N |
1366 | return &((bitmap_counter_t *) |
1367 | &bitmap->bp[page].map)[hi]; | |
ef425673 | 1368 | } else /* page is allocated */ |
32a7627c N |
1369 | return (bitmap_counter_t *) |
1370 | &(bitmap->bp[page].map[pageoff]); | |
32a7627c N |
1371 | } |
1372 | ||
4b6d287f | 1373 | int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind) |
32a7627c | 1374 | { |
ac2f40be N |
1375 | if (!bitmap) |
1376 | return 0; | |
4b6d287f N |
1377 | |
1378 | if (behind) { | |
696fcd53 | 1379 | int bw; |
4b6d287f | 1380 | atomic_inc(&bitmap->behind_writes); |
696fcd53 PC |
1381 | bw = atomic_read(&bitmap->behind_writes); |
1382 | if (bw > bitmap->behind_writes_used) | |
1383 | bitmap->behind_writes_used = bw; | |
1384 | ||
36a4e1fe N |
1385 | pr_debug("inc write-behind count %d/%lu\n", |
1386 | bw, bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1387 | } |
1388 | ||
32a7627c | 1389 | while (sectors) { |
57dab0bd | 1390 | sector_t blocks; |
32a7627c N |
1391 | bitmap_counter_t *bmc; |
1392 | ||
40cffcc0 N |
1393 | spin_lock_irq(&bitmap->counts.lock); |
1394 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1); | |
32a7627c | 1395 | if (!bmc) { |
40cffcc0 | 1396 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1397 | return 0; |
1398 | } | |
1399 | ||
27d5ea04 | 1400 | if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) { |
da6e1a32 NB |
1401 | DEFINE_WAIT(__wait); |
1402 | /* note that it is safe to do the prepare_to_wait | |
1403 | * after the test as long as we do it before dropping | |
1404 | * the spinlock. | |
1405 | */ | |
1406 | prepare_to_wait(&bitmap->overflow_wait, &__wait, | |
1407 | TASK_UNINTERRUPTIBLE); | |
40cffcc0 | 1408 | spin_unlock_irq(&bitmap->counts.lock); |
f54a9d0e | 1409 | schedule(); |
da6e1a32 NB |
1410 | finish_wait(&bitmap->overflow_wait, &__wait); |
1411 | continue; | |
1412 | } | |
1413 | ||
ac2f40be | 1414 | switch (*bmc) { |
32a7627c N |
1415 | case 0: |
1416 | bitmap_file_set_bit(bitmap, offset); | |
40cffcc0 | 1417 | bitmap_count_page(&bitmap->counts, offset, 1); |
32a7627c N |
1418 | /* fall through */ |
1419 | case 1: | |
1420 | *bmc = 2; | |
1421 | } | |
da6e1a32 | 1422 | |
32a7627c N |
1423 | (*bmc)++; |
1424 | ||
40cffcc0 | 1425 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1426 | |
1427 | offset += blocks; | |
1428 | if (sectors > blocks) | |
1429 | sectors -= blocks; | |
ac2f40be N |
1430 | else |
1431 | sectors = 0; | |
32a7627c N |
1432 | } |
1433 | return 0; | |
1434 | } | |
ac2f40be | 1435 | EXPORT_SYMBOL(bitmap_startwrite); |
32a7627c N |
1436 | |
1437 | void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, | |
4b6d287f | 1438 | int success, int behind) |
32a7627c | 1439 | { |
ac2f40be N |
1440 | if (!bitmap) |
1441 | return; | |
4b6d287f | 1442 | if (behind) { |
e555190d N |
1443 | if (atomic_dec_and_test(&bitmap->behind_writes)) |
1444 | wake_up(&bitmap->behind_wait); | |
36a4e1fe N |
1445 | pr_debug("dec write-behind count %d/%lu\n", |
1446 | atomic_read(&bitmap->behind_writes), | |
1447 | bitmap->mddev->bitmap_info.max_write_behind); | |
4b6d287f N |
1448 | } |
1449 | ||
32a7627c | 1450 | while (sectors) { |
57dab0bd | 1451 | sector_t blocks; |
32a7627c N |
1452 | unsigned long flags; |
1453 | bitmap_counter_t *bmc; | |
1454 | ||
40cffcc0 N |
1455 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1456 | bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0); | |
32a7627c | 1457 | if (!bmc) { |
40cffcc0 | 1458 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1459 | return; |
1460 | } | |
1461 | ||
961902c0 | 1462 | if (success && !bitmap->mddev->degraded && |
a0da84f3 NB |
1463 | bitmap->events_cleared < bitmap->mddev->events) { |
1464 | bitmap->events_cleared = bitmap->mddev->events; | |
1465 | bitmap->need_sync = 1; | |
5ff5afff | 1466 | sysfs_notify_dirent_safe(bitmap->sysfs_can_clear); |
a0da84f3 NB |
1467 | } |
1468 | ||
27d5ea04 | 1469 | if (!success && !NEEDED(*bmc)) |
32a7627c N |
1470 | *bmc |= NEEDED_MASK; |
1471 | ||
27d5ea04 | 1472 | if (COUNTER(*bmc) == COUNTER_MAX) |
da6e1a32 NB |
1473 | wake_up(&bitmap->overflow_wait); |
1474 | ||
32a7627c | 1475 | (*bmc)--; |
2585f3ef | 1476 | if (*bmc <= 2) { |
40cffcc0 | 1477 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1478 | bitmap->allclean = 0; |
1479 | } | |
40cffcc0 | 1480 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c N |
1481 | offset += blocks; |
1482 | if (sectors > blocks) | |
1483 | sectors -= blocks; | |
ac2f40be N |
1484 | else |
1485 | sectors = 0; | |
32a7627c N |
1486 | } |
1487 | } | |
ac2f40be | 1488 | EXPORT_SYMBOL(bitmap_endwrite); |
32a7627c | 1489 | |
57dab0bd | 1490 | static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a | 1491 | int degraded) |
32a7627c N |
1492 | { |
1493 | bitmap_counter_t *bmc; | |
1494 | int rv; | |
1495 | if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */ | |
1496 | *blocks = 1024; | |
1497 | return 1; /* always resync if no bitmap */ | |
1498 | } | |
40cffcc0 N |
1499 | spin_lock_irq(&bitmap->counts.lock); |
1500 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1501 | rv = 0; |
1502 | if (bmc) { | |
1503 | /* locked */ | |
1504 | if (RESYNC(*bmc)) | |
1505 | rv = 1; | |
1506 | else if (NEEDED(*bmc)) { | |
1507 | rv = 1; | |
6a806c51 N |
1508 | if (!degraded) { /* don't set/clear bits if degraded */ |
1509 | *bmc |= RESYNC_MASK; | |
1510 | *bmc &= ~NEEDED_MASK; | |
1511 | } | |
32a7627c N |
1512 | } |
1513 | } | |
40cffcc0 | 1514 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1515 | return rv; |
1516 | } | |
1517 | ||
57dab0bd | 1518 | int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, |
1187cf0a N |
1519 | int degraded) |
1520 | { | |
1521 | /* bitmap_start_sync must always report on multiples of whole | |
1522 | * pages, otherwise resync (which is very PAGE_SIZE based) will | |
1523 | * get confused. | |
1524 | * So call __bitmap_start_sync repeatedly (if needed) until | |
1525 | * At least PAGE_SIZE>>9 blocks are covered. | |
1526 | * Return the 'or' of the result. | |
1527 | */ | |
1528 | int rv = 0; | |
57dab0bd | 1529 | sector_t blocks1; |
1187cf0a N |
1530 | |
1531 | *blocks = 0; | |
1532 | while (*blocks < (PAGE_SIZE>>9)) { | |
1533 | rv |= __bitmap_start_sync(bitmap, offset, | |
1534 | &blocks1, degraded); | |
1535 | offset += blocks1; | |
1536 | *blocks += blocks1; | |
1537 | } | |
1538 | return rv; | |
1539 | } | |
ac2f40be | 1540 | EXPORT_SYMBOL(bitmap_start_sync); |
1187cf0a | 1541 | |
57dab0bd | 1542 | void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted) |
32a7627c N |
1543 | { |
1544 | bitmap_counter_t *bmc; | |
1545 | unsigned long flags; | |
ac2f40be N |
1546 | |
1547 | if (bitmap == NULL) { | |
32a7627c N |
1548 | *blocks = 1024; |
1549 | return; | |
1550 | } | |
40cffcc0 N |
1551 | spin_lock_irqsave(&bitmap->counts.lock, flags); |
1552 | bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0); | |
32a7627c N |
1553 | if (bmc == NULL) |
1554 | goto unlock; | |
1555 | /* locked */ | |
32a7627c N |
1556 | if (RESYNC(*bmc)) { |
1557 | *bmc &= ~RESYNC_MASK; | |
1558 | ||
1559 | if (!NEEDED(*bmc) && aborted) | |
1560 | *bmc |= NEEDED_MASK; | |
1561 | else { | |
2585f3ef | 1562 | if (*bmc <= 2) { |
40cffcc0 | 1563 | bitmap_set_pending(&bitmap->counts, offset); |
2585f3ef N |
1564 | bitmap->allclean = 0; |
1565 | } | |
32a7627c N |
1566 | } |
1567 | } | |
1568 | unlock: | |
40cffcc0 | 1569 | spin_unlock_irqrestore(&bitmap->counts.lock, flags); |
32a7627c | 1570 | } |
ac2f40be | 1571 | EXPORT_SYMBOL(bitmap_end_sync); |
32a7627c N |
1572 | |
1573 | void bitmap_close_sync(struct bitmap *bitmap) | |
1574 | { | |
1575 | /* Sync has finished, and any bitmap chunks that weren't synced | |
1576 | * properly have been aborted. It remains to us to clear the | |
1577 | * RESYNC bit wherever it is still on | |
1578 | */ | |
1579 | sector_t sector = 0; | |
57dab0bd | 1580 | sector_t blocks; |
b47490c9 N |
1581 | if (!bitmap) |
1582 | return; | |
32a7627c N |
1583 | while (sector < bitmap->mddev->resync_max_sectors) { |
1584 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
b47490c9 N |
1585 | sector += blocks; |
1586 | } | |
1587 | } | |
ac2f40be | 1588 | EXPORT_SYMBOL(bitmap_close_sync); |
b47490c9 | 1589 | |
c40f341f | 1590 | void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force) |
b47490c9 N |
1591 | { |
1592 | sector_t s = 0; | |
57dab0bd | 1593 | sector_t blocks; |
b47490c9 N |
1594 | |
1595 | if (!bitmap) | |
1596 | return; | |
1597 | if (sector == 0) { | |
1598 | bitmap->last_end_sync = jiffies; | |
1599 | return; | |
1600 | } | |
c40f341f | 1601 | if (!force && time_before(jiffies, (bitmap->last_end_sync |
1b04be96 | 1602 | + bitmap->mddev->bitmap_info.daemon_sleep))) |
b47490c9 N |
1603 | return; |
1604 | wait_event(bitmap->mddev->recovery_wait, | |
1605 | atomic_read(&bitmap->mddev->recovery_active) == 0); | |
1606 | ||
75d3da43 | 1607 | bitmap->mddev->curr_resync_completed = sector; |
070dc6dd | 1608 | set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags); |
40cffcc0 | 1609 | sector &= ~((1ULL << bitmap->counts.chunkshift) - 1); |
b47490c9 N |
1610 | s = 0; |
1611 | while (s < sector && s < bitmap->mddev->resync_max_sectors) { | |
1612 | bitmap_end_sync(bitmap, s, &blocks, 0); | |
1613 | s += blocks; | |
32a7627c | 1614 | } |
b47490c9 | 1615 | bitmap->last_end_sync = jiffies; |
acb180b0 | 1616 | sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed"); |
32a7627c | 1617 | } |
ac2f40be | 1618 | EXPORT_SYMBOL(bitmap_cond_end_sync); |
32a7627c | 1619 | |
18c9ff7f GJ |
1620 | void bitmap_sync_with_cluster(struct mddev *mddev, |
1621 | sector_t old_lo, sector_t old_hi, | |
1622 | sector_t new_lo, sector_t new_hi) | |
1623 | { | |
1624 | struct bitmap *bitmap = mddev->bitmap; | |
1625 | sector_t sector, blocks = 0; | |
1626 | ||
1627 | for (sector = old_lo; sector < new_lo; ) { | |
1628 | bitmap_end_sync(bitmap, sector, &blocks, 0); | |
1629 | sector += blocks; | |
1630 | } | |
1631 | WARN((blocks > new_lo) && old_lo, "alignment is not correct for lo\n"); | |
1632 | ||
1633 | for (sector = old_hi; sector < new_hi; ) { | |
1634 | bitmap_start_sync(bitmap, sector, &blocks, 0); | |
1635 | sector += blocks; | |
1636 | } | |
1637 | WARN((blocks > new_hi) && old_hi, "alignment is not correct for hi\n"); | |
1638 | } | |
1639 | EXPORT_SYMBOL(bitmap_sync_with_cluster); | |
1640 | ||
6a07997f | 1641 | static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed) |
32a7627c N |
1642 | { |
1643 | /* For each chunk covered by any of these sectors, set the | |
ef99bf48 | 1644 | * counter to 2 and possibly set resync_needed. They should all |
32a7627c N |
1645 | * be 0 at this point |
1646 | */ | |
193f1c93 | 1647 | |
57dab0bd | 1648 | sector_t secs; |
193f1c93 | 1649 | bitmap_counter_t *bmc; |
40cffcc0 N |
1650 | spin_lock_irq(&bitmap->counts.lock); |
1651 | bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1); | |
193f1c93 | 1652 | if (!bmc) { |
40cffcc0 | 1653 | spin_unlock_irq(&bitmap->counts.lock); |
193f1c93 | 1654 | return; |
32a7627c | 1655 | } |
ac2f40be | 1656 | if (!*bmc) { |
11dd35da | 1657 | *bmc = 2; |
40cffcc0 N |
1658 | bitmap_count_page(&bitmap->counts, offset, 1); |
1659 | bitmap_set_pending(&bitmap->counts, offset); | |
2585f3ef | 1660 | bitmap->allclean = 0; |
193f1c93 | 1661 | } |
11dd35da GR |
1662 | if (needed) |
1663 | *bmc |= NEEDED_MASK; | |
40cffcc0 | 1664 | spin_unlock_irq(&bitmap->counts.lock); |
32a7627c N |
1665 | } |
1666 | ||
9b1d1dac PC |
1667 | /* dirty the memory and file bits for bitmap chunks "s" to "e" */ |
1668 | void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e) | |
1669 | { | |
1670 | unsigned long chunk; | |
1671 | ||
1672 | for (chunk = s; chunk <= e; chunk++) { | |
40cffcc0 | 1673 | sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift; |
9b1d1dac PC |
1674 | bitmap_set_memory_bits(bitmap, sec, 1); |
1675 | bitmap_file_set_bit(bitmap, sec); | |
ffa23322 N |
1676 | if (sec < bitmap->mddev->recovery_cp) |
1677 | /* We are asserting that the array is dirty, | |
1678 | * so move the recovery_cp address back so | |
1679 | * that it is obvious that it is dirty | |
1680 | */ | |
1681 | bitmap->mddev->recovery_cp = sec; | |
9b1d1dac PC |
1682 | } |
1683 | } | |
1684 | ||
6b8b3e8a N |
1685 | /* |
1686 | * flush out any pending updates | |
1687 | */ | |
fd01b88c | 1688 | void bitmap_flush(struct mddev *mddev) |
6b8b3e8a N |
1689 | { |
1690 | struct bitmap *bitmap = mddev->bitmap; | |
42a04b50 | 1691 | long sleep; |
6b8b3e8a N |
1692 | |
1693 | if (!bitmap) /* there was no bitmap */ | |
1694 | return; | |
1695 | ||
1696 | /* run the daemon_work three time to ensure everything is flushed | |
1697 | * that can be | |
1698 | */ | |
1b04be96 | 1699 | sleep = mddev->bitmap_info.daemon_sleep * 2; |
42a04b50 | 1700 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1701 | bitmap_daemon_work(mddev); |
42a04b50 | 1702 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1703 | bitmap_daemon_work(mddev); |
42a04b50 | 1704 | bitmap->daemon_lastrun -= sleep; |
aa5cbd10 | 1705 | bitmap_daemon_work(mddev); |
6b8b3e8a N |
1706 | bitmap_update_sb(bitmap); |
1707 | } | |
1708 | ||
32a7627c N |
1709 | /* |
1710 | * free memory that was allocated | |
1711 | */ | |
3178b0db | 1712 | static void bitmap_free(struct bitmap *bitmap) |
32a7627c N |
1713 | { |
1714 | unsigned long k, pages; | |
1715 | struct bitmap_page *bp; | |
32a7627c N |
1716 | |
1717 | if (!bitmap) /* there was no bitmap */ | |
1718 | return; | |
1719 | ||
f9a67b11 GJ |
1720 | if (bitmap->sysfs_can_clear) |
1721 | sysfs_put(bitmap->sysfs_can_clear); | |
1722 | ||
f9209a32 GR |
1723 | if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info && |
1724 | bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev)) | |
b97e9257 GR |
1725 | md_cluster_stop(bitmap->mddev); |
1726 | ||
fae7d326 N |
1727 | /* Shouldn't be needed - but just in case.... */ |
1728 | wait_event(bitmap->write_wait, | |
1729 | atomic_read(&bitmap->pending_writes) == 0); | |
1730 | ||
1731 | /* release the bitmap file */ | |
1732 | bitmap_file_unmap(&bitmap->storage); | |
32a7627c | 1733 | |
40cffcc0 N |
1734 | bp = bitmap->counts.bp; |
1735 | pages = bitmap->counts.pages; | |
32a7627c N |
1736 | |
1737 | /* free all allocated memory */ | |
1738 | ||
32a7627c N |
1739 | if (bp) /* deallocate the page memory */ |
1740 | for (k = 0; k < pages; k++) | |
1741 | if (bp[k].map && !bp[k].hijacked) | |
1742 | kfree(bp[k].map); | |
1743 | kfree(bp); | |
1744 | kfree(bitmap); | |
1745 | } | |
aa5cbd10 | 1746 | |
fd01b88c | 1747 | void bitmap_destroy(struct mddev *mddev) |
3178b0db N |
1748 | { |
1749 | struct bitmap *bitmap = mddev->bitmap; | |
1750 | ||
1751 | if (!bitmap) /* there was no bitmap */ | |
1752 | return; | |
1753 | ||
c3d9714e | 1754 | mutex_lock(&mddev->bitmap_info.mutex); |
978a7a47 | 1755 | spin_lock(&mddev->lock); |
3178b0db | 1756 | mddev->bitmap = NULL; /* disconnect from the md device */ |
978a7a47 | 1757 | spin_unlock(&mddev->lock); |
c3d9714e | 1758 | mutex_unlock(&mddev->bitmap_info.mutex); |
b15c2e57 N |
1759 | if (mddev->thread) |
1760 | mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT; | |
3178b0db N |
1761 | |
1762 | bitmap_free(bitmap); | |
1763 | } | |
32a7627c N |
1764 | |
1765 | /* | |
1766 | * initialize the bitmap structure | |
1767 | * if this returns an error, bitmap_destroy must be called to do clean up | |
f9a67b11 | 1768 | * once mddev->bitmap is set |
32a7627c | 1769 | */ |
f9209a32 | 1770 | struct bitmap *bitmap_create(struct mddev *mddev, int slot) |
32a7627c N |
1771 | { |
1772 | struct bitmap *bitmap; | |
1f593903 | 1773 | sector_t blocks = mddev->resync_max_sectors; |
c3d9714e | 1774 | struct file *file = mddev->bitmap_info.file; |
32a7627c | 1775 | int err; |
324a56e1 | 1776 | struct kernfs_node *bm = NULL; |
32a7627c | 1777 | |
5f6e3c83 | 1778 | BUILD_BUG_ON(sizeof(bitmap_super_t) != 256); |
32a7627c | 1779 | |
c3d9714e | 1780 | BUG_ON(file && mddev->bitmap_info.offset); |
a654b9d8 | 1781 | |
9ffae0cf | 1782 | bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL); |
32a7627c | 1783 | if (!bitmap) |
f9209a32 | 1784 | return ERR_PTR(-ENOMEM); |
32a7627c | 1785 | |
40cffcc0 | 1786 | spin_lock_init(&bitmap->counts.lock); |
ce25c31b N |
1787 | atomic_set(&bitmap->pending_writes, 0); |
1788 | init_waitqueue_head(&bitmap->write_wait); | |
da6e1a32 | 1789 | init_waitqueue_head(&bitmap->overflow_wait); |
e555190d | 1790 | init_waitqueue_head(&bitmap->behind_wait); |
ce25c31b | 1791 | |
32a7627c | 1792 | bitmap->mddev = mddev; |
f9209a32 | 1793 | bitmap->cluster_slot = slot; |
32a7627c | 1794 | |
5ff5afff | 1795 | if (mddev->kobj.sd) |
388975cc | 1796 | bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap"); |
ece5cff0 | 1797 | if (bm) { |
388975cc | 1798 | bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear"); |
ece5cff0 N |
1799 | sysfs_put(bm); |
1800 | } else | |
1801 | bitmap->sysfs_can_clear = NULL; | |
1802 | ||
1ec885cd | 1803 | bitmap->storage.file = file; |
ce25c31b N |
1804 | if (file) { |
1805 | get_file(file); | |
ae8fa283 N |
1806 | /* As future accesses to this file will use bmap, |
1807 | * and bypass the page cache, we must sync the file | |
1808 | * first. | |
1809 | */ | |
8018ab05 | 1810 | vfs_fsync(file, 1); |
ce25c31b | 1811 | } |
42a04b50 | 1812 | /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */ |
9c81075f JB |
1813 | if (!mddev->bitmap_info.external) { |
1814 | /* | |
1815 | * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is | |
1816 | * instructing us to create a new on-disk bitmap instance. | |
1817 | */ | |
1818 | if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags)) | |
1819 | err = bitmap_new_disk_sb(bitmap); | |
1820 | else | |
1821 | err = bitmap_read_sb(bitmap); | |
1822 | } else { | |
ece5cff0 N |
1823 | err = 0; |
1824 | if (mddev->bitmap_info.chunksize == 0 || | |
1825 | mddev->bitmap_info.daemon_sleep == 0) | |
1826 | /* chunksize and time_base need to be | |
1827 | * set first. */ | |
1828 | err = -EINVAL; | |
1829 | } | |
32a7627c | 1830 | if (err) |
3178b0db | 1831 | goto error; |
32a7627c | 1832 | |
624ce4f5 | 1833 | bitmap->daemon_lastrun = jiffies; |
d60b479d N |
1834 | err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1); |
1835 | if (err) | |
3178b0db | 1836 | goto error; |
32a7627c | 1837 | |
ec0cc226 N |
1838 | pr_debug("created bitmap (%lu pages) for device %s\n", |
1839 | bitmap->counts.pages, bmname(bitmap)); | |
69e51b44 | 1840 | |
f9209a32 GR |
1841 | err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0; |
1842 | if (err) | |
1843 | goto error; | |
69e51b44 | 1844 | |
f9209a32 | 1845 | return bitmap; |
69e51b44 N |
1846 | error: |
1847 | bitmap_free(bitmap); | |
f9209a32 | 1848 | return ERR_PTR(err); |
69e51b44 N |
1849 | } |
1850 | ||
fd01b88c | 1851 | int bitmap_load(struct mddev *mddev) |
69e51b44 N |
1852 | { |
1853 | int err = 0; | |
3520fa4d | 1854 | sector_t start = 0; |
69e51b44 N |
1855 | sector_t sector = 0; |
1856 | struct bitmap *bitmap = mddev->bitmap; | |
1857 | ||
1858 | if (!bitmap) | |
1859 | goto out; | |
1860 | ||
51e453ae GJ |
1861 | if (mddev_is_clustered(mddev)) |
1862 | md_cluster_ops->load_bitmaps(mddev, mddev->bitmap_info.nodes); | |
1863 | ||
69e51b44 N |
1864 | /* Clear out old bitmap info first: Either there is none, or we |
1865 | * are resuming after someone else has possibly changed things, | |
1866 | * so we should forget old cached info. | |
1867 | * All chunks should be clean, but some might need_sync. | |
1868 | */ | |
1869 | while (sector < mddev->resync_max_sectors) { | |
57dab0bd | 1870 | sector_t blocks; |
69e51b44 N |
1871 | bitmap_start_sync(bitmap, sector, &blocks, 0); |
1872 | sector += blocks; | |
1873 | } | |
1874 | bitmap_close_sync(bitmap); | |
1875 | ||
3520fa4d JB |
1876 | if (mddev->degraded == 0 |
1877 | || bitmap->events_cleared == mddev->events) | |
1878 | /* no need to keep dirty bits to optimise a | |
1879 | * re-add of a missing device */ | |
1880 | start = mddev->recovery_cp; | |
1881 | ||
afbaa90b | 1882 | mutex_lock(&mddev->bitmap_info.mutex); |
3520fa4d | 1883 | err = bitmap_init_from_disk(bitmap, start); |
afbaa90b | 1884 | mutex_unlock(&mddev->bitmap_info.mutex); |
3520fa4d | 1885 | |
32a7627c | 1886 | if (err) |
69e51b44 | 1887 | goto out; |
b405fe91 | 1888 | clear_bit(BITMAP_STALE, &bitmap->flags); |
ef99bf48 N |
1889 | |
1890 | /* Kick recovery in case any bits were set */ | |
1891 | set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery); | |
3178b0db | 1892 | |
1b04be96 | 1893 | mddev->thread->timeout = mddev->bitmap_info.daemon_sleep; |
9cd30fdc | 1894 | md_wakeup_thread(mddev->thread); |
b15c2e57 | 1895 | |
4ad13663 N |
1896 | bitmap_update_sb(bitmap); |
1897 | ||
b405fe91 | 1898 | if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags)) |
69e51b44 N |
1899 | err = -EIO; |
1900 | out: | |
3178b0db | 1901 | return err; |
32a7627c | 1902 | } |
69e51b44 | 1903 | EXPORT_SYMBOL_GPL(bitmap_load); |
32a7627c | 1904 | |
11dd35da GR |
1905 | /* Loads the bitmap associated with slot and copies the resync information |
1906 | * to our bitmap | |
1907 | */ | |
1908 | int bitmap_copy_from_slot(struct mddev *mddev, int slot, | |
97f6cd39 | 1909 | sector_t *low, sector_t *high, bool clear_bits) |
11dd35da GR |
1910 | { |
1911 | int rv = 0, i, j; | |
1912 | sector_t block, lo = 0, hi = 0; | |
1913 | struct bitmap_counts *counts; | |
1914 | struct bitmap *bitmap = bitmap_create(mddev, slot); | |
1915 | ||
f71f1cf9 | 1916 | if (IS_ERR(bitmap)) |
11dd35da GR |
1917 | return PTR_ERR(bitmap); |
1918 | ||
11dd35da GR |
1919 | rv = bitmap_init_from_disk(bitmap, 0); |
1920 | if (rv) | |
1921 | goto err; | |
1922 | ||
1923 | counts = &bitmap->counts; | |
1924 | for (j = 0; j < counts->chunks; j++) { | |
1925 | block = (sector_t)j << counts->chunkshift; | |
1926 | if (bitmap_file_test_bit(bitmap, block)) { | |
1927 | if (!lo) | |
1928 | lo = block; | |
1929 | hi = block; | |
1930 | bitmap_file_clear_bit(bitmap, block); | |
1931 | bitmap_set_memory_bits(mddev->bitmap, block, 1); | |
1932 | bitmap_file_set_bit(mddev->bitmap, block); | |
1933 | } | |
1934 | } | |
1935 | ||
97f6cd39 GR |
1936 | if (clear_bits) { |
1937 | bitmap_update_sb(bitmap); | |
c84400c8 GJ |
1938 | /* BITMAP_PAGE_PENDING is set, but bitmap_unplug needs |
1939 | * BITMAP_PAGE_DIRTY or _NEEDWRITE to write ... */ | |
97f6cd39 | 1940 | for (i = 0; i < bitmap->storage.file_pages; i++) |
c84400c8 GJ |
1941 | if (test_page_attr(bitmap, i, BITMAP_PAGE_PENDING)) |
1942 | set_page_attr(bitmap, i, BITMAP_PAGE_NEEDWRITE); | |
97f6cd39 GR |
1943 | bitmap_unplug(bitmap); |
1944 | } | |
c84400c8 | 1945 | bitmap_unplug(mddev->bitmap); |
11dd35da GR |
1946 | *low = lo; |
1947 | *high = hi; | |
1948 | err: | |
1949 | bitmap_free(bitmap); | |
1950 | return rv; | |
1951 | } | |
1952 | EXPORT_SYMBOL_GPL(bitmap_copy_from_slot); | |
1953 | ||
1954 | ||
57148964 N |
1955 | void bitmap_status(struct seq_file *seq, struct bitmap *bitmap) |
1956 | { | |
1957 | unsigned long chunk_kb; | |
40cffcc0 | 1958 | struct bitmap_counts *counts; |
57148964 N |
1959 | |
1960 | if (!bitmap) | |
1961 | return; | |
1962 | ||
40cffcc0 N |
1963 | counts = &bitmap->counts; |
1964 | ||
57148964 N |
1965 | chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10; |
1966 | seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], " | |
1967 | "%lu%s chunk", | |
40cffcc0 N |
1968 | counts->pages - counts->missing_pages, |
1969 | counts->pages, | |
1970 | (counts->pages - counts->missing_pages) | |
57148964 N |
1971 | << (PAGE_SHIFT - 10), |
1972 | chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize, | |
1973 | chunk_kb ? "KB" : "B"); | |
1ec885cd | 1974 | if (bitmap->storage.file) { |
57148964 | 1975 | seq_printf(seq, ", file: "); |
2726d566 | 1976 | seq_file_path(seq, bitmap->storage.file, " \t\n"); |
57148964 N |
1977 | } |
1978 | ||
1979 | seq_printf(seq, "\n"); | |
57148964 N |
1980 | } |
1981 | ||
d60b479d N |
1982 | int bitmap_resize(struct bitmap *bitmap, sector_t blocks, |
1983 | int chunksize, int init) | |
1984 | { | |
1985 | /* If chunk_size is 0, choose an appropriate chunk size. | |
1986 | * Then possibly allocate new storage space. | |
1987 | * Then quiesce, copy bits, replace bitmap, and re-start | |
1988 | * | |
1989 | * This function is called both to set up the initial bitmap | |
1990 | * and to resize the bitmap while the array is active. | |
1991 | * If this happens as a result of the array being resized, | |
1992 | * chunksize will be zero, and we need to choose a suitable | |
1993 | * chunksize, otherwise we use what we are given. | |
1994 | */ | |
1995 | struct bitmap_storage store; | |
1996 | struct bitmap_counts old_counts; | |
1997 | unsigned long chunks; | |
1998 | sector_t block; | |
1999 | sector_t old_blocks, new_blocks; | |
2000 | int chunkshift; | |
2001 | int ret = 0; | |
2002 | long pages; | |
2003 | struct bitmap_page *new_bp; | |
2004 | ||
2005 | if (chunksize == 0) { | |
2006 | /* If there is enough space, leave the chunk size unchanged, | |
2007 | * else increase by factor of two until there is enough space. | |
2008 | */ | |
2009 | long bytes; | |
2010 | long space = bitmap->mddev->bitmap_info.space; | |
2011 | ||
2012 | if (space == 0) { | |
2013 | /* We don't know how much space there is, so limit | |
2014 | * to current size - in sectors. | |
2015 | */ | |
2016 | bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8); | |
2017 | if (!bitmap->mddev->bitmap_info.external) | |
2018 | bytes += sizeof(bitmap_super_t); | |
2019 | space = DIV_ROUND_UP(bytes, 512); | |
2020 | bitmap->mddev->bitmap_info.space = space; | |
2021 | } | |
2022 | chunkshift = bitmap->counts.chunkshift; | |
2023 | chunkshift--; | |
2024 | do { | |
2025 | /* 'chunkshift' is shift from block size to chunk size */ | |
2026 | chunkshift++; | |
2027 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
2028 | bytes = DIV_ROUND_UP(chunks, 8); | |
2029 | if (!bitmap->mddev->bitmap_info.external) | |
2030 | bytes += sizeof(bitmap_super_t); | |
2031 | } while (bytes > (space << 9)); | |
2032 | } else | |
2033 | chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT; | |
2034 | ||
2035 | chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift); | |
2036 | memset(&store, 0, sizeof(store)); | |
2037 | if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file) | |
2038 | ret = bitmap_storage_alloc(&store, chunks, | |
b97e9257 | 2039 | !bitmap->mddev->bitmap_info.external, |
da6fb7a9 N |
2040 | mddev_is_clustered(bitmap->mddev) |
2041 | ? bitmap->cluster_slot : 0); | |
cbb38732 GJ |
2042 | if (ret) { |
2043 | bitmap_file_unmap(&store); | |
d60b479d | 2044 | goto err; |
cbb38732 | 2045 | } |
d60b479d N |
2046 | |
2047 | pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO); | |
2048 | ||
2049 | new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL); | |
2050 | ret = -ENOMEM; | |
2051 | if (!new_bp) { | |
2052 | bitmap_file_unmap(&store); | |
2053 | goto err; | |
2054 | } | |
2055 | ||
2056 | if (!init) | |
2057 | bitmap->mddev->pers->quiesce(bitmap->mddev, 1); | |
2058 | ||
2059 | store.file = bitmap->storage.file; | |
2060 | bitmap->storage.file = NULL; | |
2061 | ||
2062 | if (store.sb_page && bitmap->storage.sb_page) | |
2063 | memcpy(page_address(store.sb_page), | |
2064 | page_address(bitmap->storage.sb_page), | |
2065 | sizeof(bitmap_super_t)); | |
2066 | bitmap_file_unmap(&bitmap->storage); | |
2067 | bitmap->storage = store; | |
2068 | ||
2069 | old_counts = bitmap->counts; | |
2070 | bitmap->counts.bp = new_bp; | |
2071 | bitmap->counts.pages = pages; | |
2072 | bitmap->counts.missing_pages = pages; | |
2073 | bitmap->counts.chunkshift = chunkshift; | |
2074 | bitmap->counts.chunks = chunks; | |
2075 | bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift + | |
2076 | BITMAP_BLOCK_SHIFT); | |
2077 | ||
2078 | blocks = min(old_counts.chunks << old_counts.chunkshift, | |
2079 | chunks << chunkshift); | |
2080 | ||
2081 | spin_lock_irq(&bitmap->counts.lock); | |
c9d65032 GJ |
2082 | /* For cluster raid, need to pre-allocate bitmap */ |
2083 | if (mddev_is_clustered(bitmap->mddev)) { | |
2084 | unsigned long page; | |
2085 | for (page = 0; page < pages; page++) { | |
2086 | ret = bitmap_checkpage(&bitmap->counts, page, 1, 1); | |
2087 | if (ret) { | |
2088 | unsigned long k; | |
2089 | ||
2090 | /* deallocate the page memory */ | |
2091 | for (k = 0; k < page; k++) { | |
bc47e842 | 2092 | kfree(new_bp[k].map); |
c9d65032 GJ |
2093 | } |
2094 | ||
2095 | /* restore some fields from old_counts */ | |
2096 | bitmap->counts.bp = old_counts.bp; | |
2097 | bitmap->counts.pages = old_counts.pages; | |
2098 | bitmap->counts.missing_pages = old_counts.pages; | |
2099 | bitmap->counts.chunkshift = old_counts.chunkshift; | |
2100 | bitmap->counts.chunks = old_counts.chunks; | |
2101 | bitmap->mddev->bitmap_info.chunksize = 1 << (old_counts.chunkshift + | |
2102 | BITMAP_BLOCK_SHIFT); | |
2103 | blocks = old_counts.chunks << old_counts.chunkshift; | |
ec0cc226 | 2104 | pr_warn("Could not pre-allocate in-memory bitmap for cluster raid\n"); |
c9d65032 GJ |
2105 | break; |
2106 | } else | |
2107 | bitmap->counts.bp[page].count += 1; | |
2108 | } | |
2109 | } | |
2110 | ||
d60b479d N |
2111 | for (block = 0; block < blocks; ) { |
2112 | bitmap_counter_t *bmc_old, *bmc_new; | |
2113 | int set; | |
2114 | ||
2115 | bmc_old = bitmap_get_counter(&old_counts, block, | |
2116 | &old_blocks, 0); | |
2117 | set = bmc_old && NEEDED(*bmc_old); | |
2118 | ||
2119 | if (set) { | |
2120 | bmc_new = bitmap_get_counter(&bitmap->counts, block, | |
2121 | &new_blocks, 1); | |
2122 | if (*bmc_new == 0) { | |
2123 | /* need to set on-disk bits too. */ | |
2124 | sector_t end = block + new_blocks; | |
2125 | sector_t start = block >> chunkshift; | |
2126 | start <<= chunkshift; | |
2127 | while (start < end) { | |
2128 | bitmap_file_set_bit(bitmap, block); | |
2129 | start += 1 << chunkshift; | |
2130 | } | |
2131 | *bmc_new = 2; | |
2132 | bitmap_count_page(&bitmap->counts, | |
2133 | block, 1); | |
2134 | bitmap_set_pending(&bitmap->counts, | |
2135 | block); | |
2136 | } | |
2137 | *bmc_new |= NEEDED_MASK; | |
2138 | if (new_blocks < old_blocks) | |
2139 | old_blocks = new_blocks; | |
2140 | } | |
2141 | block += old_blocks; | |
2142 | } | |
2143 | ||
2144 | if (!init) { | |
2145 | int i; | |
2146 | while (block < (chunks << chunkshift)) { | |
2147 | bitmap_counter_t *bmc; | |
2148 | bmc = bitmap_get_counter(&bitmap->counts, block, | |
2149 | &new_blocks, 1); | |
2150 | if (bmc) { | |
2151 | /* new space. It needs to be resynced, so | |
2152 | * we set NEEDED_MASK. | |
2153 | */ | |
2154 | if (*bmc == 0) { | |
2155 | *bmc = NEEDED_MASK | 2; | |
2156 | bitmap_count_page(&bitmap->counts, | |
2157 | block, 1); | |
2158 | bitmap_set_pending(&bitmap->counts, | |
2159 | block); | |
2160 | } | |
2161 | } | |
2162 | block += new_blocks; | |
2163 | } | |
2164 | for (i = 0; i < bitmap->storage.file_pages; i++) | |
2165 | set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY); | |
2166 | } | |
2167 | spin_unlock_irq(&bitmap->counts.lock); | |
2168 | ||
2169 | if (!init) { | |
2170 | bitmap_unplug(bitmap); | |
2171 | bitmap->mddev->pers->quiesce(bitmap->mddev, 0); | |
2172 | } | |
2173 | ret = 0; | |
2174 | err: | |
2175 | return ret; | |
2176 | } | |
2177 | EXPORT_SYMBOL_GPL(bitmap_resize); | |
2178 | ||
43a70507 | 2179 | static ssize_t |
fd01b88c | 2180 | location_show(struct mddev *mddev, char *page) |
43a70507 N |
2181 | { |
2182 | ssize_t len; | |
ac2f40be | 2183 | if (mddev->bitmap_info.file) |
43a70507 | 2184 | len = sprintf(page, "file"); |
ac2f40be | 2185 | else if (mddev->bitmap_info.offset) |
43a70507 | 2186 | len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset); |
ac2f40be | 2187 | else |
43a70507 N |
2188 | len = sprintf(page, "none"); |
2189 | len += sprintf(page+len, "\n"); | |
2190 | return len; | |
2191 | } | |
2192 | ||
2193 | static ssize_t | |
fd01b88c | 2194 | location_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 | 2195 | { |
d9dd26b2 | 2196 | int rv; |
43a70507 | 2197 | |
d9dd26b2 SL |
2198 | rv = mddev_lock(mddev); |
2199 | if (rv) | |
2200 | return rv; | |
43a70507 | 2201 | if (mddev->pers) { |
d9dd26b2 SL |
2202 | if (!mddev->pers->quiesce) { |
2203 | rv = -EBUSY; | |
2204 | goto out; | |
2205 | } | |
2206 | if (mddev->recovery || mddev->sync_thread) { | |
2207 | rv = -EBUSY; | |
2208 | goto out; | |
2209 | } | |
43a70507 N |
2210 | } |
2211 | ||
2212 | if (mddev->bitmap || mddev->bitmap_info.file || | |
2213 | mddev->bitmap_info.offset) { | |
2214 | /* bitmap already configured. Only option is to clear it */ | |
d9dd26b2 SL |
2215 | if (strncmp(buf, "none", 4) != 0) { |
2216 | rv = -EBUSY; | |
2217 | goto out; | |
2218 | } | |
43a70507 N |
2219 | if (mddev->pers) { |
2220 | mddev->pers->quiesce(mddev, 1); | |
2221 | bitmap_destroy(mddev); | |
2222 | mddev->pers->quiesce(mddev, 0); | |
2223 | } | |
2224 | mddev->bitmap_info.offset = 0; | |
2225 | if (mddev->bitmap_info.file) { | |
2226 | struct file *f = mddev->bitmap_info.file; | |
2227 | mddev->bitmap_info.file = NULL; | |
43a70507 N |
2228 | fput(f); |
2229 | } | |
2230 | } else { | |
2231 | /* No bitmap, OK to set a location */ | |
2232 | long long offset; | |
2233 | if (strncmp(buf, "none", 4) == 0) | |
2234 | /* nothing to be done */; | |
2235 | else if (strncmp(buf, "file:", 5) == 0) { | |
2236 | /* Not supported yet */ | |
d9dd26b2 SL |
2237 | rv = -EINVAL; |
2238 | goto out; | |
43a70507 | 2239 | } else { |
43a70507 | 2240 | if (buf[0] == '+') |
b29bebd6 | 2241 | rv = kstrtoll(buf+1, 10, &offset); |
43a70507 | 2242 | else |
b29bebd6 | 2243 | rv = kstrtoll(buf, 10, &offset); |
43a70507 | 2244 | if (rv) |
d9dd26b2 SL |
2245 | goto out; |
2246 | if (offset == 0) { | |
2247 | rv = -EINVAL; | |
2248 | goto out; | |
2249 | } | |
ece5cff0 N |
2250 | if (mddev->bitmap_info.external == 0 && |
2251 | mddev->major_version == 0 && | |
d9dd26b2 SL |
2252 | offset != mddev->bitmap_info.default_offset) { |
2253 | rv = -EINVAL; | |
2254 | goto out; | |
2255 | } | |
43a70507 N |
2256 | mddev->bitmap_info.offset = offset; |
2257 | if (mddev->pers) { | |
f9209a32 | 2258 | struct bitmap *bitmap; |
43a70507 | 2259 | mddev->pers->quiesce(mddev, 1); |
f9209a32 GR |
2260 | bitmap = bitmap_create(mddev, -1); |
2261 | if (IS_ERR(bitmap)) | |
2262 | rv = PTR_ERR(bitmap); | |
2263 | else { | |
2264 | mddev->bitmap = bitmap; | |
4474ca42 | 2265 | rv = bitmap_load(mddev); |
f9a67b11 | 2266 | if (rv) |
f9209a32 | 2267 | mddev->bitmap_info.offset = 0; |
43a70507 N |
2268 | } |
2269 | mddev->pers->quiesce(mddev, 0); | |
f9a67b11 GJ |
2270 | if (rv) { |
2271 | bitmap_destroy(mddev); | |
d9dd26b2 | 2272 | goto out; |
f9a67b11 | 2273 | } |
43a70507 N |
2274 | } |
2275 | } | |
2276 | } | |
2277 | if (!mddev->external) { | |
2278 | /* Ensure new bitmap info is stored in | |
2279 | * metadata promptly. | |
2280 | */ | |
2281 | set_bit(MD_CHANGE_DEVS, &mddev->flags); | |
2282 | md_wakeup_thread(mddev->thread); | |
2283 | } | |
d9dd26b2 SL |
2284 | rv = 0; |
2285 | out: | |
2286 | mddev_unlock(mddev); | |
2287 | if (rv) | |
2288 | return rv; | |
43a70507 N |
2289 | return len; |
2290 | } | |
2291 | ||
2292 | static struct md_sysfs_entry bitmap_location = | |
2293 | __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store); | |
2294 | ||
6409bb05 N |
2295 | /* 'bitmap/space' is the space available at 'location' for the |
2296 | * bitmap. This allows the kernel to know when it is safe to | |
2297 | * resize the bitmap to match a resized array. | |
2298 | */ | |
2299 | static ssize_t | |
2300 | space_show(struct mddev *mddev, char *page) | |
2301 | { | |
2302 | return sprintf(page, "%lu\n", mddev->bitmap_info.space); | |
2303 | } | |
2304 | ||
2305 | static ssize_t | |
2306 | space_store(struct mddev *mddev, const char *buf, size_t len) | |
2307 | { | |
2308 | unsigned long sectors; | |
2309 | int rv; | |
2310 | ||
2311 | rv = kstrtoul(buf, 10, §ors); | |
2312 | if (rv) | |
2313 | return rv; | |
2314 | ||
2315 | if (sectors == 0) | |
2316 | return -EINVAL; | |
2317 | ||
2318 | if (mddev->bitmap && | |
9b1215c1 | 2319 | sectors < (mddev->bitmap->storage.bytes + 511) >> 9) |
6409bb05 N |
2320 | return -EFBIG; /* Bitmap is too big for this small space */ |
2321 | ||
2322 | /* could make sure it isn't too big, but that isn't really | |
2323 | * needed - user-space should be careful. | |
2324 | */ | |
2325 | mddev->bitmap_info.space = sectors; | |
2326 | return len; | |
2327 | } | |
2328 | ||
2329 | static struct md_sysfs_entry bitmap_space = | |
2330 | __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store); | |
2331 | ||
43a70507 | 2332 | static ssize_t |
fd01b88c | 2333 | timeout_show(struct mddev *mddev, char *page) |
43a70507 N |
2334 | { |
2335 | ssize_t len; | |
2336 | unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ; | |
2337 | unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ; | |
ac2f40be | 2338 | |
43a70507 N |
2339 | len = sprintf(page, "%lu", secs); |
2340 | if (jifs) | |
2341 | len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs)); | |
2342 | len += sprintf(page+len, "\n"); | |
2343 | return len; | |
2344 | } | |
2345 | ||
2346 | static ssize_t | |
fd01b88c | 2347 | timeout_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2348 | { |
2349 | /* timeout can be set at any time */ | |
2350 | unsigned long timeout; | |
2351 | int rv = strict_strtoul_scaled(buf, &timeout, 4); | |
2352 | if (rv) | |
2353 | return rv; | |
2354 | ||
2355 | /* just to make sure we don't overflow... */ | |
2356 | if (timeout >= LONG_MAX / HZ) | |
2357 | return -EINVAL; | |
2358 | ||
2359 | timeout = timeout * HZ / 10000; | |
2360 | ||
2361 | if (timeout >= MAX_SCHEDULE_TIMEOUT) | |
2362 | timeout = MAX_SCHEDULE_TIMEOUT-1; | |
2363 | if (timeout < 1) | |
2364 | timeout = 1; | |
2365 | mddev->bitmap_info.daemon_sleep = timeout; | |
2366 | if (mddev->thread) { | |
2367 | /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then | |
2368 | * the bitmap is all clean and we don't need to | |
2369 | * adjust the timeout right now | |
2370 | */ | |
2371 | if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) { | |
2372 | mddev->thread->timeout = timeout; | |
2373 | md_wakeup_thread(mddev->thread); | |
2374 | } | |
2375 | } | |
2376 | return len; | |
2377 | } | |
2378 | ||
2379 | static struct md_sysfs_entry bitmap_timeout = | |
2380 | __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store); | |
2381 | ||
2382 | static ssize_t | |
fd01b88c | 2383 | backlog_show(struct mddev *mddev, char *page) |
43a70507 N |
2384 | { |
2385 | return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind); | |
2386 | } | |
2387 | ||
2388 | static ssize_t | |
fd01b88c | 2389 | backlog_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2390 | { |
2391 | unsigned long backlog; | |
b29bebd6 | 2392 | int rv = kstrtoul(buf, 10, &backlog); |
43a70507 N |
2393 | if (rv) |
2394 | return rv; | |
2395 | if (backlog > COUNTER_MAX) | |
2396 | return -EINVAL; | |
2397 | mddev->bitmap_info.max_write_behind = backlog; | |
2398 | return len; | |
2399 | } | |
2400 | ||
2401 | static struct md_sysfs_entry bitmap_backlog = | |
2402 | __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store); | |
2403 | ||
2404 | static ssize_t | |
fd01b88c | 2405 | chunksize_show(struct mddev *mddev, char *page) |
43a70507 N |
2406 | { |
2407 | return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize); | |
2408 | } | |
2409 | ||
2410 | static ssize_t | |
fd01b88c | 2411 | chunksize_store(struct mddev *mddev, const char *buf, size_t len) |
43a70507 N |
2412 | { |
2413 | /* Can only be changed when no bitmap is active */ | |
2414 | int rv; | |
2415 | unsigned long csize; | |
2416 | if (mddev->bitmap) | |
2417 | return -EBUSY; | |
b29bebd6 | 2418 | rv = kstrtoul(buf, 10, &csize); |
43a70507 N |
2419 | if (rv) |
2420 | return rv; | |
2421 | if (csize < 512 || | |
2422 | !is_power_of_2(csize)) | |
2423 | return -EINVAL; | |
2424 | mddev->bitmap_info.chunksize = csize; | |
2425 | return len; | |
2426 | } | |
2427 | ||
2428 | static struct md_sysfs_entry bitmap_chunksize = | |
2429 | __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store); | |
2430 | ||
fd01b88c | 2431 | static ssize_t metadata_show(struct mddev *mddev, char *page) |
ece5cff0 | 2432 | { |
c4ce867f GR |
2433 | if (mddev_is_clustered(mddev)) |
2434 | return sprintf(page, "clustered\n"); | |
ece5cff0 N |
2435 | return sprintf(page, "%s\n", (mddev->bitmap_info.external |
2436 | ? "external" : "internal")); | |
2437 | } | |
2438 | ||
fd01b88c | 2439 | static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2440 | { |
2441 | if (mddev->bitmap || | |
2442 | mddev->bitmap_info.file || | |
2443 | mddev->bitmap_info.offset) | |
2444 | return -EBUSY; | |
2445 | if (strncmp(buf, "external", 8) == 0) | |
2446 | mddev->bitmap_info.external = 1; | |
c4ce867f GR |
2447 | else if ((strncmp(buf, "internal", 8) == 0) || |
2448 | (strncmp(buf, "clustered", 9) == 0)) | |
ece5cff0 N |
2449 | mddev->bitmap_info.external = 0; |
2450 | else | |
2451 | return -EINVAL; | |
2452 | return len; | |
2453 | } | |
2454 | ||
2455 | static struct md_sysfs_entry bitmap_metadata = | |
2456 | __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store); | |
2457 | ||
fd01b88c | 2458 | static ssize_t can_clear_show(struct mddev *mddev, char *page) |
ece5cff0 N |
2459 | { |
2460 | int len; | |
b7b17c9b | 2461 | spin_lock(&mddev->lock); |
ece5cff0 N |
2462 | if (mddev->bitmap) |
2463 | len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ? | |
2464 | "false" : "true")); | |
2465 | else | |
2466 | len = sprintf(page, "\n"); | |
b7b17c9b | 2467 | spin_unlock(&mddev->lock); |
ece5cff0 N |
2468 | return len; |
2469 | } | |
2470 | ||
fd01b88c | 2471 | static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len) |
ece5cff0 N |
2472 | { |
2473 | if (mddev->bitmap == NULL) | |
2474 | return -ENOENT; | |
2475 | if (strncmp(buf, "false", 5) == 0) | |
2476 | mddev->bitmap->need_sync = 1; | |
2477 | else if (strncmp(buf, "true", 4) == 0) { | |
2478 | if (mddev->degraded) | |
2479 | return -EBUSY; | |
2480 | mddev->bitmap->need_sync = 0; | |
2481 | } else | |
2482 | return -EINVAL; | |
2483 | return len; | |
2484 | } | |
2485 | ||
2486 | static struct md_sysfs_entry bitmap_can_clear = | |
2487 | __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store); | |
2488 | ||
696fcd53 | 2489 | static ssize_t |
fd01b88c | 2490 | behind_writes_used_show(struct mddev *mddev, char *page) |
696fcd53 | 2491 | { |
b7b17c9b N |
2492 | ssize_t ret; |
2493 | spin_lock(&mddev->lock); | |
696fcd53 | 2494 | if (mddev->bitmap == NULL) |
b7b17c9b N |
2495 | ret = sprintf(page, "0\n"); |
2496 | else | |
2497 | ret = sprintf(page, "%lu\n", | |
2498 | mddev->bitmap->behind_writes_used); | |
2499 | spin_unlock(&mddev->lock); | |
2500 | return ret; | |
696fcd53 PC |
2501 | } |
2502 | ||
2503 | static ssize_t | |
fd01b88c | 2504 | behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len) |
696fcd53 PC |
2505 | { |
2506 | if (mddev->bitmap) | |
2507 | mddev->bitmap->behind_writes_used = 0; | |
2508 | return len; | |
2509 | } | |
2510 | ||
2511 | static struct md_sysfs_entry max_backlog_used = | |
2512 | __ATTR(max_backlog_used, S_IRUGO | S_IWUSR, | |
2513 | behind_writes_used_show, behind_writes_used_reset); | |
2514 | ||
43a70507 N |
2515 | static struct attribute *md_bitmap_attrs[] = { |
2516 | &bitmap_location.attr, | |
6409bb05 | 2517 | &bitmap_space.attr, |
43a70507 N |
2518 | &bitmap_timeout.attr, |
2519 | &bitmap_backlog.attr, | |
2520 | &bitmap_chunksize.attr, | |
ece5cff0 N |
2521 | &bitmap_metadata.attr, |
2522 | &bitmap_can_clear.attr, | |
696fcd53 | 2523 | &max_backlog_used.attr, |
43a70507 N |
2524 | NULL |
2525 | }; | |
2526 | struct attribute_group md_bitmap_group = { | |
2527 | .name = "bitmap", | |
2528 | .attrs = md_bitmap_attrs, | |
2529 | }; | |
2530 |