]> git.ipfire.org Git - thirdparty/linux.git/blob - drivers/md/bitmap.c
md-cluster: always setup in-memory bitmap
[thirdparty/linux.git] / drivers / md / bitmap.c
1 /*
2 * bitmap.c two-level bitmap (C) Peter T. Breuer (ptb@ot.uc3m.es) 2003
3 *
4 * bitmap_create - sets up the bitmap structure
5 * bitmap_destroy - destroys the bitmap structure
6 *
7 * additions, Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.:
8 * - added disk storage for bitmap
9 * - changes to allow various bitmap chunk sizes
10 */
11
12 /*
13 * Still to do:
14 *
15 * flush after percent set rather than just time based. (maybe both).
16 */
17
18 #include <linux/blkdev.h>
19 #include <linux/module.h>
20 #include <linux/errno.h>
21 #include <linux/slab.h>
22 #include <linux/init.h>
23 #include <linux/timer.h>
24 #include <linux/sched.h>
25 #include <linux/list.h>
26 #include <linux/file.h>
27 #include <linux/mount.h>
28 #include <linux/buffer_head.h>
29 #include <linux/seq_file.h>
30 #include "md.h"
31 #include "bitmap.h"
32
33 static inline char *bmname(struct bitmap *bitmap)
34 {
35 return bitmap->mddev ? mdname(bitmap->mddev) : "mdX";
36 }
37
38 /*
39 * check a page and, if necessary, allocate it (or hijack it if the alloc fails)
40 *
41 * 1) check to see if this page is allocated, if it's not then try to alloc
42 * 2) if the alloc fails, set the page's hijacked flag so we'll use the
43 * page pointer directly as a counter
44 *
45 * if we find our page, we increment the page's refcount so that it stays
46 * allocated while we're using it
47 */
48 static int bitmap_checkpage(struct bitmap_counts *bitmap,
49 unsigned long page, int create, int no_hijack)
50 __releases(bitmap->lock)
51 __acquires(bitmap->lock)
52 {
53 unsigned char *mappage;
54
55 if (page >= bitmap->pages) {
56 /* This can happen if bitmap_start_sync goes beyond
57 * End-of-device while looking for a whole page.
58 * It is harmless.
59 */
60 return -EINVAL;
61 }
62
63 if (bitmap->bp[page].hijacked) /* it's hijacked, don't try to alloc */
64 return 0;
65
66 if (bitmap->bp[page].map) /* page is already allocated, just return */
67 return 0;
68
69 if (!create)
70 return -ENOENT;
71
72 /* this page has not been allocated yet */
73
74 spin_unlock_irq(&bitmap->lock);
75 /* 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();
88 mappage = kzalloc(PAGE_SIZE, GFP_NOIO);
89 spin_lock_irq(&bitmap->lock);
90
91 if (mappage == NULL) {
92 pr_debug("md/bitmap: map page allocation failed, hijacking\n");
93 /* We don't support hijack for cluster raid */
94 if (no_hijack)
95 return -ENOMEM;
96 /* failed - set the hijacked flag so that we can use the
97 * pointer as a counter */
98 if (!bitmap->bp[page].map)
99 bitmap->bp[page].hijacked = 1;
100 } else if (bitmap->bp[page].map ||
101 bitmap->bp[page].hijacked) {
102 /* somebody beat us to getting the page */
103 kfree(mappage);
104 } else {
105
106 /* no page was in place and we have one, so install it */
107
108 bitmap->bp[page].map = mappage;
109 bitmap->missing_pages--;
110 }
111 return 0;
112 }
113
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 */
117 static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
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;
129 } else {
130 /* normal case, free the page */
131 ptr = bitmap->bp[page].map;
132 bitmap->bp[page].map = NULL;
133 bitmap->missing_pages++;
134 kfree(ptr);
135 }
136 }
137
138 /*
139 * bitmap file handling - read and write the bitmap file and its superblock
140 */
141
142 /*
143 * basic page I/O operations
144 */
145
146 /* IO operations when bitmap is stored near all superblocks */
147 static int read_sb_page(struct mddev *mddev, loff_t offset,
148 struct page *page,
149 unsigned long index, int size)
150 {
151 /* choose a good rdev and read the page from there */
152
153 struct md_rdev *rdev;
154 sector_t target;
155
156 rdev_for_each(rdev, mddev) {
157 if (! test_bit(In_sync, &rdev->flags)
158 || test_bit(Faulty, &rdev->flags))
159 continue;
160
161 target = offset + index * (PAGE_SIZE/512);
162
163 if (sync_page_io(rdev, target,
164 roundup(size, bdev_logical_block_size(rdev->bdev)),
165 page, READ, true)) {
166 page->index = index;
167 return 0;
168 }
169 }
170 return -EIO;
171 }
172
173 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
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
181 * list_for_each_entry_continue_rcu.
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.
187 */
188 rcu_read_lock();
189 if (rdev == NULL)
190 /* start at the beginning */
191 rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
192 else {
193 /* release the previous rdev and start from there. */
194 rdev_dec_pending(rdev, mddev);
195 }
196 list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
197 if (rdev->raid_disk >= 0 &&
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
209 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
210 {
211 struct md_rdev *rdev = NULL;
212 struct block_device *bdev;
213 struct mddev *mddev = bitmap->mddev;
214 struct bitmap_storage *store = &bitmap->storage;
215
216 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
217 int size = PAGE_SIZE;
218 loff_t offset = mddev->bitmap_info.offset;
219
220 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
221
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,
227 bdev_logical_block_size(bdev));
228 }
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);
269 }
270
271 if (wait)
272 md_super_wait(mddev);
273 return 0;
274
275 bad_alignment:
276 return -EINVAL;
277 }
278
279 static void bitmap_file_kick(struct bitmap *bitmap);
280 /*
281 * write out a page to a file
282 */
283 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
284 {
285 struct buffer_head *bh;
286
287 if (bitmap->storage.file == NULL) {
288 switch (write_sb_page(bitmap, page, wait)) {
289 case -EINVAL:
290 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
291 }
292 } else {
293
294 bh = page_buffers(page);
295
296 while (bh && bh->b_blocknr) {
297 atomic_inc(&bitmap->pending_writes);
298 set_buffer_locked(bh);
299 set_buffer_mapped(bh);
300 submit_bh(WRITE | REQ_SYNC, bh);
301 bh = bh->b_this_page;
302 }
303
304 if (wait)
305 wait_event(bitmap->write_wait,
306 atomic_read(&bitmap->pending_writes)==0);
307 }
308 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
309 bitmap_file_kick(bitmap);
310 }
311
312 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
313 {
314 struct bitmap *bitmap = bh->b_private;
315
316 if (!uptodate)
317 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
318 if (atomic_dec_and_test(&bitmap->pending_writes))
319 wake_up(&bitmap->write_wait);
320 }
321
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);
328 put_page(page);
329 }
330 static void free_buffers(struct page *page)
331 {
332 struct buffer_head *bh;
333
334 if (!PagePrivate(page))
335 return;
336
337 bh = page_buffers(page);
338 while (bh) {
339 struct buffer_head *next = bh->b_this_page;
340 free_buffer_head(bh);
341 bh = next;
342 }
343 __clear_page_buffers(page);
344 put_page(page);
345 }
346
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 */
354 static int read_page(struct file *file, unsigned long index,
355 struct bitmap *bitmap,
356 unsigned long count,
357 struct page *page)
358 {
359 int ret = 0;
360 struct inode *inode = file_inode(file);
361 struct buffer_head *bh;
362 sector_t block;
363
364 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
365 (unsigned long long)index << PAGE_SHIFT);
366
367 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
368 if (!bh) {
369 ret = -ENOMEM;
370 goto out;
371 }
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! */
381 ret = -EINVAL;
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;
392 atomic_inc(&bitmap->pending_writes);
393 set_buffer_locked(bh);
394 set_buffer_mapped(bh);
395 submit_bh(READ, bh);
396 }
397 block++;
398 bh = bh->b_this_page;
399 }
400 page->index = index;
401
402 wait_event(bitmap->write_wait,
403 atomic_read(&bitmap->pending_writes)==0);
404 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
405 ret = -EIO;
406 out:
407 if (ret)
408 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
409 (int)PAGE_SIZE,
410 (unsigned long long)index << PAGE_SHIFT,
411 ret);
412 return ret;
413 }
414
415 /*
416 * bitmap file superblock operations
417 */
418
419 /* update the event counter and sync the superblock to disk */
420 void bitmap_update_sb(struct bitmap *bitmap)
421 {
422 bitmap_super_t *sb;
423
424 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
425 return;
426 if (bitmap->mddev->bitmap_info.external)
427 return;
428 if (!bitmap->storage.sb_page) /* no superblock */
429 return;
430 sb = kmap_atomic(bitmap->storage.sb_page);
431 sb->events = cpu_to_le64(bitmap->mddev->events);
432 if (bitmap->mddev->events < bitmap->events_cleared)
433 /* rocking back to read-only */
434 bitmap->events_cleared = bitmap->mddev->events;
435 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
436 sb->state = cpu_to_le32(bitmap->flags);
437 /* Just in case these have been changed via sysfs: */
438 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
439 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
440 /* This might have been changed by a reshape */
441 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
442 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
443 sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
444 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
445 bitmap_info.space);
446 kunmap_atomic(sb);
447 write_page(bitmap, bitmap->storage.sb_page, 1);
448 }
449
450 /* print out the bitmap file superblock */
451 void bitmap_print_sb(struct bitmap *bitmap)
452 {
453 bitmap_super_t *sb;
454
455 if (!bitmap || !bitmap->storage.sb_page)
456 return;
457 sb = kmap_atomic(bitmap->storage.sb_page);
458 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
459 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
460 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
461 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
462 *(__u32 *)(sb->uuid+0),
463 *(__u32 *)(sb->uuid+4),
464 *(__u32 *)(sb->uuid+8),
465 *(__u32 *)(sb->uuid+12));
466 printk(KERN_DEBUG " events: %llu\n",
467 (unsigned long long) le64_to_cpu(sb->events));
468 printk(KERN_DEBUG "events cleared: %llu\n",
469 (unsigned long long) le64_to_cpu(sb->events_cleared));
470 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
471 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
472 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
473 printk(KERN_DEBUG " sync size: %llu KB\n",
474 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
475 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
476 kunmap_atomic(sb);
477 }
478
479 /*
480 * bitmap_new_disk_sb
481 * @bitmap
482 *
483 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
484 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
485 * This function verifies 'bitmap_info' and populates the on-disk bitmap
486 * structure, which is to be written to disk.
487 *
488 * Returns: 0 on success, -Exxx on error
489 */
490 static int bitmap_new_disk_sb(struct bitmap *bitmap)
491 {
492 bitmap_super_t *sb;
493 unsigned long chunksize, daemon_sleep, write_behind;
494
495 bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
496 if (bitmap->storage.sb_page == NULL)
497 return -ENOMEM;
498 bitmap->storage.sb_page->index = 0;
499
500 sb = kmap_atomic(bitmap->storage.sb_page);
501
502 sb->magic = cpu_to_le32(BITMAP_MAGIC);
503 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
504
505 chunksize = bitmap->mddev->bitmap_info.chunksize;
506 BUG_ON(!chunksize);
507 if (!is_power_of_2(chunksize)) {
508 kunmap_atomic(sb);
509 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
510 return -EINVAL;
511 }
512 sb->chunksize = cpu_to_le32(chunksize);
513
514 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
515 if (!daemon_sleep || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
516 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
517 daemon_sleep = 5 * HZ;
518 }
519 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
520 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
521
522 /*
523 * FIXME: write_behind for RAID1. If not specified, what
524 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
525 */
526 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
527 if (write_behind > COUNTER_MAX)
528 write_behind = COUNTER_MAX / 2;
529 sb->write_behind = cpu_to_le32(write_behind);
530 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
531
532 /* keep the array size field of the bitmap superblock up to date */
533 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
534
535 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
536
537 set_bit(BITMAP_STALE, &bitmap->flags);
538 sb->state = cpu_to_le32(bitmap->flags);
539 bitmap->events_cleared = bitmap->mddev->events;
540 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
541 bitmap->mddev->bitmap_info.nodes = 0;
542
543 kunmap_atomic(sb);
544
545 return 0;
546 }
547
548 /* read the superblock from the bitmap file and initialize some bitmap fields */
549 static int bitmap_read_sb(struct bitmap *bitmap)
550 {
551 char *reason = NULL;
552 bitmap_super_t *sb;
553 unsigned long chunksize, daemon_sleep, write_behind;
554 unsigned long long events;
555 int nodes = 0;
556 unsigned long sectors_reserved = 0;
557 int err = -EINVAL;
558 struct page *sb_page;
559 loff_t offset = bitmap->mddev->bitmap_info.offset;
560
561 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
562 chunksize = 128 * 1024 * 1024;
563 daemon_sleep = 5 * HZ;
564 write_behind = 0;
565 set_bit(BITMAP_STALE, &bitmap->flags);
566 err = 0;
567 goto out_no_sb;
568 }
569 /* page 0 is the superblock, read it... */
570 sb_page = alloc_page(GFP_KERNEL);
571 if (!sb_page)
572 return -ENOMEM;
573 bitmap->storage.sb_page = sb_page;
574
575 re_read:
576 /* If cluster_slot is set, the cluster is setup */
577 if (bitmap->cluster_slot >= 0) {
578 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
579
580 sector_div(bm_blocks,
581 bitmap->mddev->bitmap_info.chunksize >> 9);
582 /* bits to bytes */
583 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
584 /* to 4k blocks */
585 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
586 offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
587 pr_info("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
588 bitmap->cluster_slot, offset);
589 }
590
591 if (bitmap->storage.file) {
592 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
593 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
594
595 err = read_page(bitmap->storage.file, 0,
596 bitmap, bytes, sb_page);
597 } else {
598 err = read_sb_page(bitmap->mddev,
599 offset,
600 sb_page,
601 0, sizeof(bitmap_super_t));
602 }
603 if (err)
604 return err;
605
606 err = -EINVAL;
607 sb = kmap_atomic(sb_page);
608
609 chunksize = le32_to_cpu(sb->chunksize);
610 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
611 write_behind = le32_to_cpu(sb->write_behind);
612 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
613 /* Setup nodes/clustername only if bitmap version is
614 * cluster-compatible
615 */
616 if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
617 nodes = le32_to_cpu(sb->nodes);
618 strlcpy(bitmap->mddev->bitmap_info.cluster_name,
619 sb->cluster_name, 64);
620 }
621
622 /* verify that the bitmap-specific fields are valid */
623 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
624 reason = "bad magic";
625 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
626 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
627 reason = "unrecognized superblock version";
628 else if (chunksize < 512)
629 reason = "bitmap chunksize too small";
630 else if (!is_power_of_2(chunksize))
631 reason = "bitmap chunksize not a power of 2";
632 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
633 reason = "daemon sleep period out of range";
634 else if (write_behind > COUNTER_MAX)
635 reason = "write-behind limit out of range (0 - 16383)";
636 if (reason) {
637 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
638 bmname(bitmap), reason);
639 goto out;
640 }
641
642 /* keep the array size field of the bitmap superblock up to date */
643 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
644
645 if (bitmap->mddev->persistent) {
646 /*
647 * We have a persistent array superblock, so compare the
648 * bitmap's UUID and event counter to the mddev's
649 */
650 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
651 printk(KERN_INFO
652 "%s: bitmap superblock UUID mismatch\n",
653 bmname(bitmap));
654 goto out;
655 }
656 events = le64_to_cpu(sb->events);
657 if (!nodes && (events < bitmap->mddev->events)) {
658 printk(KERN_INFO
659 "%s: bitmap file is out of date (%llu < %llu) "
660 "-- forcing full recovery\n",
661 bmname(bitmap), events,
662 (unsigned long long) bitmap->mddev->events);
663 set_bit(BITMAP_STALE, &bitmap->flags);
664 }
665 }
666
667 /* assign fields using values from superblock */
668 bitmap->flags |= le32_to_cpu(sb->state);
669 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
670 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
671 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
672 strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
673 err = 0;
674
675 out:
676 kunmap_atomic(sb);
677 /* Assiging chunksize is required for "re_read" */
678 bitmap->mddev->bitmap_info.chunksize = chunksize;
679 if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
680 err = md_setup_cluster(bitmap->mddev, nodes);
681 if (err) {
682 pr_err("%s: Could not setup cluster service (%d)\n",
683 bmname(bitmap), err);
684 goto out_no_sb;
685 }
686 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
687 goto re_read;
688 }
689
690
691 out_no_sb:
692 if (test_bit(BITMAP_STALE, &bitmap->flags))
693 bitmap->events_cleared = bitmap->mddev->events;
694 bitmap->mddev->bitmap_info.chunksize = chunksize;
695 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
696 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
697 bitmap->mddev->bitmap_info.nodes = nodes;
698 if (bitmap->mddev->bitmap_info.space == 0 ||
699 bitmap->mddev->bitmap_info.space > sectors_reserved)
700 bitmap->mddev->bitmap_info.space = sectors_reserved;
701 if (err) {
702 bitmap_print_sb(bitmap);
703 if (bitmap->cluster_slot < 0)
704 md_cluster_stop(bitmap->mddev);
705 }
706 return err;
707 }
708
709 /*
710 * general bitmap file operations
711 */
712
713 /*
714 * on-disk bitmap:
715 *
716 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
717 * file a page at a time. There's a superblock at the start of the file.
718 */
719 /* calculate the index of the page that contains this bit */
720 static inline unsigned long file_page_index(struct bitmap_storage *store,
721 unsigned long chunk)
722 {
723 if (store->sb_page)
724 chunk += sizeof(bitmap_super_t) << 3;
725 return chunk >> PAGE_BIT_SHIFT;
726 }
727
728 /* calculate the (bit) offset of this bit within a page */
729 static inline unsigned long file_page_offset(struct bitmap_storage *store,
730 unsigned long chunk)
731 {
732 if (store->sb_page)
733 chunk += sizeof(bitmap_super_t) << 3;
734 return chunk & (PAGE_BITS - 1);
735 }
736
737 /*
738 * return a pointer to the page in the filemap that contains the given bit
739 *
740 */
741 static inline struct page *filemap_get_page(struct bitmap_storage *store,
742 unsigned long chunk)
743 {
744 if (file_page_index(store, chunk) >= store->file_pages)
745 return NULL;
746 return store->filemap[file_page_index(store, chunk)];
747 }
748
749 static int bitmap_storage_alloc(struct bitmap_storage *store,
750 unsigned long chunks, int with_super,
751 int slot_number)
752 {
753 int pnum, offset = 0;
754 unsigned long num_pages;
755 unsigned long bytes;
756
757 bytes = DIV_ROUND_UP(chunks, 8);
758 if (with_super)
759 bytes += sizeof(bitmap_super_t);
760
761 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
762 offset = slot_number * (num_pages - 1);
763
764 store->filemap = kmalloc(sizeof(struct page *)
765 * num_pages, GFP_KERNEL);
766 if (!store->filemap)
767 return -ENOMEM;
768
769 if (with_super && !store->sb_page) {
770 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
771 if (store->sb_page == NULL)
772 return -ENOMEM;
773 }
774
775 pnum = 0;
776 if (store->sb_page) {
777 store->filemap[0] = store->sb_page;
778 pnum = 1;
779 store->sb_page->index = offset;
780 }
781
782 for ( ; pnum < num_pages; pnum++) {
783 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
784 if (!store->filemap[pnum]) {
785 store->file_pages = pnum;
786 return -ENOMEM;
787 }
788 store->filemap[pnum]->index = pnum + offset;
789 }
790 store->file_pages = pnum;
791
792 /* We need 4 bits per page, rounded up to a multiple
793 * of sizeof(unsigned long) */
794 store->filemap_attr = kzalloc(
795 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
796 GFP_KERNEL);
797 if (!store->filemap_attr)
798 return -ENOMEM;
799
800 store->bytes = bytes;
801
802 return 0;
803 }
804
805 static void bitmap_file_unmap(struct bitmap_storage *store)
806 {
807 struct page **map, *sb_page;
808 int pages;
809 struct file *file;
810
811 file = store->file;
812 map = store->filemap;
813 pages = store->file_pages;
814 sb_page = store->sb_page;
815
816 while (pages--)
817 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
818 free_buffers(map[pages]);
819 kfree(map);
820 kfree(store->filemap_attr);
821
822 if (sb_page)
823 free_buffers(sb_page);
824
825 if (file) {
826 struct inode *inode = file_inode(file);
827 invalidate_mapping_pages(inode->i_mapping, 0, -1);
828 fput(file);
829 }
830 }
831
832 /*
833 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
834 * then it is no longer reliable, so we stop using it and we mark the file
835 * as failed in the superblock
836 */
837 static void bitmap_file_kick(struct bitmap *bitmap)
838 {
839 char *path, *ptr = NULL;
840
841 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
842 bitmap_update_sb(bitmap);
843
844 if (bitmap->storage.file) {
845 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
846 if (path)
847 ptr = file_path(bitmap->storage.file,
848 path, PAGE_SIZE);
849
850 printk(KERN_ALERT
851 "%s: kicking failed bitmap file %s from array!\n",
852 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
853
854 kfree(path);
855 } else
856 printk(KERN_ALERT
857 "%s: disabling internal bitmap due to errors\n",
858 bmname(bitmap));
859 }
860 }
861
862 enum bitmap_page_attr {
863 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
864 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
865 * i.e. counter is 1 or 2. */
866 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
867 };
868
869 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
870 enum bitmap_page_attr attr)
871 {
872 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
873 }
874
875 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
876 enum bitmap_page_attr attr)
877 {
878 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
879 }
880
881 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
882 enum bitmap_page_attr attr)
883 {
884 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
885 }
886
887 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
888 enum bitmap_page_attr attr)
889 {
890 return test_and_clear_bit((pnum<<2) + attr,
891 bitmap->storage.filemap_attr);
892 }
893 /*
894 * bitmap_file_set_bit -- called before performing a write to the md device
895 * to set (and eventually sync) a particular bit in the bitmap file
896 *
897 * we set the bit immediately, then we record the page number so that
898 * when an unplug occurs, we can flush the dirty pages out to disk
899 */
900 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
901 {
902 unsigned long bit;
903 struct page *page;
904 void *kaddr;
905 unsigned long chunk = block >> bitmap->counts.chunkshift;
906
907 page = filemap_get_page(&bitmap->storage, chunk);
908 if (!page)
909 return;
910 bit = file_page_offset(&bitmap->storage, chunk);
911
912 /* set the bit */
913 kaddr = kmap_atomic(page);
914 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
915 set_bit(bit, kaddr);
916 else
917 set_bit_le(bit, kaddr);
918 kunmap_atomic(kaddr);
919 pr_debug("set file bit %lu page %lu\n", bit, page->index);
920 /* record page number so it gets flushed to disk when unplug occurs */
921 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
922 }
923
924 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
925 {
926 unsigned long bit;
927 struct page *page;
928 void *paddr;
929 unsigned long chunk = block >> bitmap->counts.chunkshift;
930
931 page = filemap_get_page(&bitmap->storage, chunk);
932 if (!page)
933 return;
934 bit = file_page_offset(&bitmap->storage, chunk);
935 paddr = kmap_atomic(page);
936 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
937 clear_bit(bit, paddr);
938 else
939 clear_bit_le(bit, paddr);
940 kunmap_atomic(paddr);
941 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
942 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
943 bitmap->allclean = 0;
944 }
945 }
946
947 static int bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
948 {
949 unsigned long bit;
950 struct page *page;
951 void *paddr;
952 unsigned long chunk = block >> bitmap->counts.chunkshift;
953 int set = 0;
954
955 page = filemap_get_page(&bitmap->storage, chunk);
956 if (!page)
957 return -EINVAL;
958 bit = file_page_offset(&bitmap->storage, chunk);
959 paddr = kmap_atomic(page);
960 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
961 set = test_bit(bit, paddr);
962 else
963 set = test_bit_le(bit, paddr);
964 kunmap_atomic(paddr);
965 return set;
966 }
967
968
969 /* this gets called when the md device is ready to unplug its underlying
970 * (slave) device queues -- before we let any writes go down, we need to
971 * sync the dirty pages of the bitmap file to disk */
972 void bitmap_unplug(struct bitmap *bitmap)
973 {
974 unsigned long i;
975 int dirty, need_write;
976
977 if (!bitmap || !bitmap->storage.filemap ||
978 test_bit(BITMAP_STALE, &bitmap->flags))
979 return;
980
981 /* look at each page to see if there are any set bits that need to be
982 * flushed out to disk */
983 for (i = 0; i < bitmap->storage.file_pages; i++) {
984 if (!bitmap->storage.filemap)
985 return;
986 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
987 need_write = test_and_clear_page_attr(bitmap, i,
988 BITMAP_PAGE_NEEDWRITE);
989 if (dirty || need_write) {
990 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
991 write_page(bitmap, bitmap->storage.filemap[i], 0);
992 }
993 }
994 if (bitmap->storage.file)
995 wait_event(bitmap->write_wait,
996 atomic_read(&bitmap->pending_writes)==0);
997 else
998 md_super_wait(bitmap->mddev);
999
1000 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1001 bitmap_file_kick(bitmap);
1002 }
1003 EXPORT_SYMBOL(bitmap_unplug);
1004
1005 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1006 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1007 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1008 * memory mapping of the bitmap file
1009 * Special cases:
1010 * if there's no bitmap file, or if the bitmap file had been
1011 * previously kicked from the array, we mark all the bits as
1012 * 1's in order to cause a full resync.
1013 *
1014 * We ignore all bits for sectors that end earlier than 'start'.
1015 * This is used when reading an out-of-date bitmap...
1016 */
1017 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1018 {
1019 unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1020 struct page *page = NULL;
1021 unsigned long bit_cnt = 0;
1022 struct file *file;
1023 unsigned long offset;
1024 int outofdate;
1025 int ret = -ENOSPC;
1026 void *paddr;
1027 struct bitmap_storage *store = &bitmap->storage;
1028
1029 chunks = bitmap->counts.chunks;
1030 file = store->file;
1031
1032 if (!file && !bitmap->mddev->bitmap_info.offset) {
1033 /* No permanent bitmap - fill with '1s'. */
1034 store->filemap = NULL;
1035 store->file_pages = 0;
1036 for (i = 0; i < chunks ; i++) {
1037 /* if the disk bit is set, set the memory bit */
1038 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1039 >= start);
1040 bitmap_set_memory_bits(bitmap,
1041 (sector_t)i << bitmap->counts.chunkshift,
1042 needed);
1043 }
1044 return 0;
1045 }
1046
1047 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1048 if (outofdate)
1049 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
1050 "recovery\n", bmname(bitmap));
1051
1052 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1053 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
1054 bmname(bitmap),
1055 (unsigned long) i_size_read(file->f_mapping->host),
1056 store->bytes);
1057 goto err;
1058 }
1059
1060 oldindex = ~0L;
1061 offset = 0;
1062 if (!bitmap->mddev->bitmap_info.external)
1063 offset = sizeof(bitmap_super_t);
1064
1065 if (mddev_is_clustered(bitmap->mddev))
1066 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1067
1068 for (i = 0; i < chunks; i++) {
1069 int b;
1070 index = file_page_index(&bitmap->storage, i);
1071 bit = file_page_offset(&bitmap->storage, i);
1072 if (index != oldindex) { /* this is a new page, read it in */
1073 int count;
1074 /* unmap the old page, we're done with it */
1075 if (index == store->file_pages-1)
1076 count = store->bytes - index * PAGE_SIZE;
1077 else
1078 count = PAGE_SIZE;
1079 page = store->filemap[index];
1080 if (file)
1081 ret = read_page(file, index, bitmap,
1082 count, page);
1083 else
1084 ret = read_sb_page(
1085 bitmap->mddev,
1086 bitmap->mddev->bitmap_info.offset,
1087 page,
1088 index + node_offset, count);
1089
1090 if (ret)
1091 goto err;
1092
1093 oldindex = index;
1094
1095 if (outofdate) {
1096 /*
1097 * if bitmap is out of date, dirty the
1098 * whole page and write it out
1099 */
1100 paddr = kmap_atomic(page);
1101 memset(paddr + offset, 0xff,
1102 PAGE_SIZE - offset);
1103 kunmap_atomic(paddr);
1104 write_page(bitmap, page, 1);
1105
1106 ret = -EIO;
1107 if (test_bit(BITMAP_WRITE_ERROR,
1108 &bitmap->flags))
1109 goto err;
1110 }
1111 }
1112 paddr = kmap_atomic(page);
1113 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1114 b = test_bit(bit, paddr);
1115 else
1116 b = test_bit_le(bit, paddr);
1117 kunmap_atomic(paddr);
1118 if (b) {
1119 /* if the disk bit is set, set the memory bit */
1120 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1121 >= start);
1122 bitmap_set_memory_bits(bitmap,
1123 (sector_t)i << bitmap->counts.chunkshift,
1124 needed);
1125 bit_cnt++;
1126 }
1127 offset = 0;
1128 }
1129
1130 printk(KERN_INFO "%s: bitmap initialized from disk: "
1131 "read %lu pages, set %lu of %lu bits\n",
1132 bmname(bitmap), store->file_pages,
1133 bit_cnt, chunks);
1134
1135 return 0;
1136
1137 err:
1138 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1139 bmname(bitmap), ret);
1140 return ret;
1141 }
1142
1143 void bitmap_write_all(struct bitmap *bitmap)
1144 {
1145 /* We don't actually write all bitmap blocks here,
1146 * just flag them as needing to be written
1147 */
1148 int i;
1149
1150 if (!bitmap || !bitmap->storage.filemap)
1151 return;
1152 if (bitmap->storage.file)
1153 /* Only one copy, so nothing needed */
1154 return;
1155
1156 for (i = 0; i < bitmap->storage.file_pages; i++)
1157 set_page_attr(bitmap, i,
1158 BITMAP_PAGE_NEEDWRITE);
1159 bitmap->allclean = 0;
1160 }
1161
1162 static void bitmap_count_page(struct bitmap_counts *bitmap,
1163 sector_t offset, int inc)
1164 {
1165 sector_t chunk = offset >> bitmap->chunkshift;
1166 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1167 bitmap->bp[page].count += inc;
1168 bitmap_checkfree(bitmap, page);
1169 }
1170
1171 static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1172 {
1173 sector_t chunk = offset >> bitmap->chunkshift;
1174 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1175 struct bitmap_page *bp = &bitmap->bp[page];
1176
1177 if (!bp->pending)
1178 bp->pending = 1;
1179 }
1180
1181 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1182 sector_t offset, sector_t *blocks,
1183 int create);
1184
1185 /*
1186 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1187 * out to disk
1188 */
1189
1190 void bitmap_daemon_work(struct mddev *mddev)
1191 {
1192 struct bitmap *bitmap;
1193 unsigned long j;
1194 unsigned long nextpage;
1195 sector_t blocks;
1196 struct bitmap_counts *counts;
1197
1198 /* Use a mutex to guard daemon_work against
1199 * bitmap_destroy.
1200 */
1201 mutex_lock(&mddev->bitmap_info.mutex);
1202 bitmap = mddev->bitmap;
1203 if (bitmap == NULL) {
1204 mutex_unlock(&mddev->bitmap_info.mutex);
1205 return;
1206 }
1207 if (time_before(jiffies, bitmap->daemon_lastrun
1208 + mddev->bitmap_info.daemon_sleep))
1209 goto done;
1210
1211 bitmap->daemon_lastrun = jiffies;
1212 if (bitmap->allclean) {
1213 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1214 goto done;
1215 }
1216 bitmap->allclean = 1;
1217
1218 /* Any file-page which is PENDING now needs to be written.
1219 * So set NEEDWRITE now, then after we make any last-minute changes
1220 * we will write it.
1221 */
1222 for (j = 0; j < bitmap->storage.file_pages; j++)
1223 if (test_and_clear_page_attr(bitmap, j,
1224 BITMAP_PAGE_PENDING))
1225 set_page_attr(bitmap, j,
1226 BITMAP_PAGE_NEEDWRITE);
1227
1228 if (bitmap->need_sync &&
1229 mddev->bitmap_info.external == 0) {
1230 /* Arrange for superblock update as well as
1231 * other changes */
1232 bitmap_super_t *sb;
1233 bitmap->need_sync = 0;
1234 if (bitmap->storage.filemap) {
1235 sb = kmap_atomic(bitmap->storage.sb_page);
1236 sb->events_cleared =
1237 cpu_to_le64(bitmap->events_cleared);
1238 kunmap_atomic(sb);
1239 set_page_attr(bitmap, 0,
1240 BITMAP_PAGE_NEEDWRITE);
1241 }
1242 }
1243 /* Now look at the bitmap counters and if any are '2' or '1',
1244 * decrement and handle accordingly.
1245 */
1246 counts = &bitmap->counts;
1247 spin_lock_irq(&counts->lock);
1248 nextpage = 0;
1249 for (j = 0; j < counts->chunks; j++) {
1250 bitmap_counter_t *bmc;
1251 sector_t block = (sector_t)j << counts->chunkshift;
1252
1253 if (j == nextpage) {
1254 nextpage += PAGE_COUNTER_RATIO;
1255 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1256 j |= PAGE_COUNTER_MASK;
1257 continue;
1258 }
1259 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1260 }
1261 bmc = bitmap_get_counter(counts,
1262 block,
1263 &blocks, 0);
1264
1265 if (!bmc) {
1266 j |= PAGE_COUNTER_MASK;
1267 continue;
1268 }
1269 if (*bmc == 1 && !bitmap->need_sync) {
1270 /* We can clear the bit */
1271 *bmc = 0;
1272 bitmap_count_page(counts, block, -1);
1273 bitmap_file_clear_bit(bitmap, block);
1274 } else if (*bmc && *bmc <= 2) {
1275 *bmc = 1;
1276 bitmap_set_pending(counts, block);
1277 bitmap->allclean = 0;
1278 }
1279 }
1280 spin_unlock_irq(&counts->lock);
1281
1282 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1283 * DIRTY pages need to be written by bitmap_unplug so it can wait
1284 * for them.
1285 * If we find any DIRTY page we stop there and let bitmap_unplug
1286 * handle all the rest. This is important in the case where
1287 * the first blocking holds the superblock and it has been updated.
1288 * We mustn't write any other blocks before the superblock.
1289 */
1290 for (j = 0;
1291 j < bitmap->storage.file_pages
1292 && !test_bit(BITMAP_STALE, &bitmap->flags);
1293 j++) {
1294 if (test_page_attr(bitmap, j,
1295 BITMAP_PAGE_DIRTY))
1296 /* bitmap_unplug will handle the rest */
1297 break;
1298 if (test_and_clear_page_attr(bitmap, j,
1299 BITMAP_PAGE_NEEDWRITE)) {
1300 write_page(bitmap, bitmap->storage.filemap[j], 0);
1301 }
1302 }
1303
1304 done:
1305 if (bitmap->allclean == 0)
1306 mddev->thread->timeout =
1307 mddev->bitmap_info.daemon_sleep;
1308 mutex_unlock(&mddev->bitmap_info.mutex);
1309 }
1310
1311 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1312 sector_t offset, sector_t *blocks,
1313 int create)
1314 __releases(bitmap->lock)
1315 __acquires(bitmap->lock)
1316 {
1317 /* If 'create', we might release the lock and reclaim it.
1318 * The lock must have been taken with interrupts enabled.
1319 * If !create, we don't release the lock.
1320 */
1321 sector_t chunk = offset >> bitmap->chunkshift;
1322 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1323 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1324 sector_t csize;
1325 int err;
1326
1327 err = bitmap_checkpage(bitmap, page, create, 0);
1328
1329 if (bitmap->bp[page].hijacked ||
1330 bitmap->bp[page].map == NULL)
1331 csize = ((sector_t)1) << (bitmap->chunkshift +
1332 PAGE_COUNTER_SHIFT - 1);
1333 else
1334 csize = ((sector_t)1) << bitmap->chunkshift;
1335 *blocks = csize - (offset & (csize - 1));
1336
1337 if (err < 0)
1338 return NULL;
1339
1340 /* now locked ... */
1341
1342 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1343 /* should we use the first or second counter field
1344 * of the hijacked pointer? */
1345 int hi = (pageoff > PAGE_COUNTER_MASK);
1346 return &((bitmap_counter_t *)
1347 &bitmap->bp[page].map)[hi];
1348 } else /* page is allocated */
1349 return (bitmap_counter_t *)
1350 &(bitmap->bp[page].map[pageoff]);
1351 }
1352
1353 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1354 {
1355 if (!bitmap)
1356 return 0;
1357
1358 if (behind) {
1359 int bw;
1360 atomic_inc(&bitmap->behind_writes);
1361 bw = atomic_read(&bitmap->behind_writes);
1362 if (bw > bitmap->behind_writes_used)
1363 bitmap->behind_writes_used = bw;
1364
1365 pr_debug("inc write-behind count %d/%lu\n",
1366 bw, bitmap->mddev->bitmap_info.max_write_behind);
1367 }
1368
1369 while (sectors) {
1370 sector_t blocks;
1371 bitmap_counter_t *bmc;
1372
1373 spin_lock_irq(&bitmap->counts.lock);
1374 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1375 if (!bmc) {
1376 spin_unlock_irq(&bitmap->counts.lock);
1377 return 0;
1378 }
1379
1380 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1381 DEFINE_WAIT(__wait);
1382 /* note that it is safe to do the prepare_to_wait
1383 * after the test as long as we do it before dropping
1384 * the spinlock.
1385 */
1386 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1387 TASK_UNINTERRUPTIBLE);
1388 spin_unlock_irq(&bitmap->counts.lock);
1389 schedule();
1390 finish_wait(&bitmap->overflow_wait, &__wait);
1391 continue;
1392 }
1393
1394 switch (*bmc) {
1395 case 0:
1396 bitmap_file_set_bit(bitmap, offset);
1397 bitmap_count_page(&bitmap->counts, offset, 1);
1398 /* fall through */
1399 case 1:
1400 *bmc = 2;
1401 }
1402
1403 (*bmc)++;
1404
1405 spin_unlock_irq(&bitmap->counts.lock);
1406
1407 offset += blocks;
1408 if (sectors > blocks)
1409 sectors -= blocks;
1410 else
1411 sectors = 0;
1412 }
1413 return 0;
1414 }
1415 EXPORT_SYMBOL(bitmap_startwrite);
1416
1417 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1418 int success, int behind)
1419 {
1420 if (!bitmap)
1421 return;
1422 if (behind) {
1423 if (atomic_dec_and_test(&bitmap->behind_writes))
1424 wake_up(&bitmap->behind_wait);
1425 pr_debug("dec write-behind count %d/%lu\n",
1426 atomic_read(&bitmap->behind_writes),
1427 bitmap->mddev->bitmap_info.max_write_behind);
1428 }
1429
1430 while (sectors) {
1431 sector_t blocks;
1432 unsigned long flags;
1433 bitmap_counter_t *bmc;
1434
1435 spin_lock_irqsave(&bitmap->counts.lock, flags);
1436 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1437 if (!bmc) {
1438 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1439 return;
1440 }
1441
1442 if (success && !bitmap->mddev->degraded &&
1443 bitmap->events_cleared < bitmap->mddev->events) {
1444 bitmap->events_cleared = bitmap->mddev->events;
1445 bitmap->need_sync = 1;
1446 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1447 }
1448
1449 if (!success && !NEEDED(*bmc))
1450 *bmc |= NEEDED_MASK;
1451
1452 if (COUNTER(*bmc) == COUNTER_MAX)
1453 wake_up(&bitmap->overflow_wait);
1454
1455 (*bmc)--;
1456 if (*bmc <= 2) {
1457 bitmap_set_pending(&bitmap->counts, offset);
1458 bitmap->allclean = 0;
1459 }
1460 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1461 offset += blocks;
1462 if (sectors > blocks)
1463 sectors -= blocks;
1464 else
1465 sectors = 0;
1466 }
1467 }
1468 EXPORT_SYMBOL(bitmap_endwrite);
1469
1470 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1471 int degraded)
1472 {
1473 bitmap_counter_t *bmc;
1474 int rv;
1475 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1476 *blocks = 1024;
1477 return 1; /* always resync if no bitmap */
1478 }
1479 spin_lock_irq(&bitmap->counts.lock);
1480 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1481 rv = 0;
1482 if (bmc) {
1483 /* locked */
1484 if (RESYNC(*bmc))
1485 rv = 1;
1486 else if (NEEDED(*bmc)) {
1487 rv = 1;
1488 if (!degraded) { /* don't set/clear bits if degraded */
1489 *bmc |= RESYNC_MASK;
1490 *bmc &= ~NEEDED_MASK;
1491 }
1492 }
1493 }
1494 spin_unlock_irq(&bitmap->counts.lock);
1495 return rv;
1496 }
1497
1498 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1499 int degraded)
1500 {
1501 /* bitmap_start_sync must always report on multiples of whole
1502 * pages, otherwise resync (which is very PAGE_SIZE based) will
1503 * get confused.
1504 * So call __bitmap_start_sync repeatedly (if needed) until
1505 * At least PAGE_SIZE>>9 blocks are covered.
1506 * Return the 'or' of the result.
1507 */
1508 int rv = 0;
1509 sector_t blocks1;
1510
1511 *blocks = 0;
1512 while (*blocks < (PAGE_SIZE>>9)) {
1513 rv |= __bitmap_start_sync(bitmap, offset,
1514 &blocks1, degraded);
1515 offset += blocks1;
1516 *blocks += blocks1;
1517 }
1518 return rv;
1519 }
1520 EXPORT_SYMBOL(bitmap_start_sync);
1521
1522 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1523 {
1524 bitmap_counter_t *bmc;
1525 unsigned long flags;
1526
1527 if (bitmap == NULL) {
1528 *blocks = 1024;
1529 return;
1530 }
1531 spin_lock_irqsave(&bitmap->counts.lock, flags);
1532 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1533 if (bmc == NULL)
1534 goto unlock;
1535 /* locked */
1536 if (RESYNC(*bmc)) {
1537 *bmc &= ~RESYNC_MASK;
1538
1539 if (!NEEDED(*bmc) && aborted)
1540 *bmc |= NEEDED_MASK;
1541 else {
1542 if (*bmc <= 2) {
1543 bitmap_set_pending(&bitmap->counts, offset);
1544 bitmap->allclean = 0;
1545 }
1546 }
1547 }
1548 unlock:
1549 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1550 }
1551 EXPORT_SYMBOL(bitmap_end_sync);
1552
1553 void bitmap_close_sync(struct bitmap *bitmap)
1554 {
1555 /* Sync has finished, and any bitmap chunks that weren't synced
1556 * properly have been aborted. It remains to us to clear the
1557 * RESYNC bit wherever it is still on
1558 */
1559 sector_t sector = 0;
1560 sector_t blocks;
1561 if (!bitmap)
1562 return;
1563 while (sector < bitmap->mddev->resync_max_sectors) {
1564 bitmap_end_sync(bitmap, sector, &blocks, 0);
1565 sector += blocks;
1566 }
1567 }
1568 EXPORT_SYMBOL(bitmap_close_sync);
1569
1570 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
1571 {
1572 sector_t s = 0;
1573 sector_t blocks;
1574
1575 if (!bitmap)
1576 return;
1577 if (sector == 0) {
1578 bitmap->last_end_sync = jiffies;
1579 return;
1580 }
1581 if (!force && time_before(jiffies, (bitmap->last_end_sync
1582 + bitmap->mddev->bitmap_info.daemon_sleep)))
1583 return;
1584 wait_event(bitmap->mddev->recovery_wait,
1585 atomic_read(&bitmap->mddev->recovery_active) == 0);
1586
1587 bitmap->mddev->curr_resync_completed = sector;
1588 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1589 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1590 s = 0;
1591 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1592 bitmap_end_sync(bitmap, s, &blocks, 0);
1593 s += blocks;
1594 }
1595 bitmap->last_end_sync = jiffies;
1596 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1597 }
1598 EXPORT_SYMBOL(bitmap_cond_end_sync);
1599
1600 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1601 {
1602 /* For each chunk covered by any of these sectors, set the
1603 * counter to 2 and possibly set resync_needed. They should all
1604 * be 0 at this point
1605 */
1606
1607 sector_t secs;
1608 bitmap_counter_t *bmc;
1609 spin_lock_irq(&bitmap->counts.lock);
1610 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1611 if (!bmc) {
1612 spin_unlock_irq(&bitmap->counts.lock);
1613 return;
1614 }
1615 if (!*bmc) {
1616 *bmc = 2;
1617 bitmap_count_page(&bitmap->counts, offset, 1);
1618 bitmap_set_pending(&bitmap->counts, offset);
1619 bitmap->allclean = 0;
1620 }
1621 if (needed)
1622 *bmc |= NEEDED_MASK;
1623 spin_unlock_irq(&bitmap->counts.lock);
1624 }
1625
1626 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1627 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1628 {
1629 unsigned long chunk;
1630
1631 for (chunk = s; chunk <= e; chunk++) {
1632 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1633 bitmap_set_memory_bits(bitmap, sec, 1);
1634 bitmap_file_set_bit(bitmap, sec);
1635 if (sec < bitmap->mddev->recovery_cp)
1636 /* We are asserting that the array is dirty,
1637 * so move the recovery_cp address back so
1638 * that it is obvious that it is dirty
1639 */
1640 bitmap->mddev->recovery_cp = sec;
1641 }
1642 }
1643
1644 /*
1645 * flush out any pending updates
1646 */
1647 void bitmap_flush(struct mddev *mddev)
1648 {
1649 struct bitmap *bitmap = mddev->bitmap;
1650 long sleep;
1651
1652 if (!bitmap) /* there was no bitmap */
1653 return;
1654
1655 /* run the daemon_work three time to ensure everything is flushed
1656 * that can be
1657 */
1658 sleep = mddev->bitmap_info.daemon_sleep * 2;
1659 bitmap->daemon_lastrun -= sleep;
1660 bitmap_daemon_work(mddev);
1661 bitmap->daemon_lastrun -= sleep;
1662 bitmap_daemon_work(mddev);
1663 bitmap->daemon_lastrun -= sleep;
1664 bitmap_daemon_work(mddev);
1665 bitmap_update_sb(bitmap);
1666 }
1667
1668 /*
1669 * free memory that was allocated
1670 */
1671 static void bitmap_free(struct bitmap *bitmap)
1672 {
1673 unsigned long k, pages;
1674 struct bitmap_page *bp;
1675
1676 if (!bitmap) /* there was no bitmap */
1677 return;
1678
1679 if (bitmap->sysfs_can_clear)
1680 sysfs_put(bitmap->sysfs_can_clear);
1681
1682 if (mddev_is_clustered(bitmap->mddev) && bitmap->mddev->cluster_info &&
1683 bitmap->cluster_slot == md_cluster_ops->slot_number(bitmap->mddev))
1684 md_cluster_stop(bitmap->mddev);
1685
1686 /* Shouldn't be needed - but just in case.... */
1687 wait_event(bitmap->write_wait,
1688 atomic_read(&bitmap->pending_writes) == 0);
1689
1690 /* release the bitmap file */
1691 bitmap_file_unmap(&bitmap->storage);
1692
1693 bp = bitmap->counts.bp;
1694 pages = bitmap->counts.pages;
1695
1696 /* free all allocated memory */
1697
1698 if (bp) /* deallocate the page memory */
1699 for (k = 0; k < pages; k++)
1700 if (bp[k].map && !bp[k].hijacked)
1701 kfree(bp[k].map);
1702 kfree(bp);
1703 kfree(bitmap);
1704 }
1705
1706 void bitmap_destroy(struct mddev *mddev)
1707 {
1708 struct bitmap *bitmap = mddev->bitmap;
1709
1710 if (!bitmap) /* there was no bitmap */
1711 return;
1712
1713 mutex_lock(&mddev->bitmap_info.mutex);
1714 spin_lock(&mddev->lock);
1715 mddev->bitmap = NULL; /* disconnect from the md device */
1716 spin_unlock(&mddev->lock);
1717 mutex_unlock(&mddev->bitmap_info.mutex);
1718 if (mddev->thread)
1719 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1720
1721 bitmap_free(bitmap);
1722 }
1723
1724 /*
1725 * initialize the bitmap structure
1726 * if this returns an error, bitmap_destroy must be called to do clean up
1727 * once mddev->bitmap is set
1728 */
1729 struct bitmap *bitmap_create(struct mddev *mddev, int slot)
1730 {
1731 struct bitmap *bitmap;
1732 sector_t blocks = mddev->resync_max_sectors;
1733 struct file *file = mddev->bitmap_info.file;
1734 int err;
1735 struct kernfs_node *bm = NULL;
1736
1737 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1738
1739 BUG_ON(file && mddev->bitmap_info.offset);
1740
1741 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1742 if (!bitmap)
1743 return ERR_PTR(-ENOMEM);
1744
1745 spin_lock_init(&bitmap->counts.lock);
1746 atomic_set(&bitmap->pending_writes, 0);
1747 init_waitqueue_head(&bitmap->write_wait);
1748 init_waitqueue_head(&bitmap->overflow_wait);
1749 init_waitqueue_head(&bitmap->behind_wait);
1750
1751 bitmap->mddev = mddev;
1752 bitmap->cluster_slot = slot;
1753
1754 if (mddev->kobj.sd)
1755 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1756 if (bm) {
1757 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1758 sysfs_put(bm);
1759 } else
1760 bitmap->sysfs_can_clear = NULL;
1761
1762 bitmap->storage.file = file;
1763 if (file) {
1764 get_file(file);
1765 /* As future accesses to this file will use bmap,
1766 * and bypass the page cache, we must sync the file
1767 * first.
1768 */
1769 vfs_fsync(file, 1);
1770 }
1771 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1772 if (!mddev->bitmap_info.external) {
1773 /*
1774 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1775 * instructing us to create a new on-disk bitmap instance.
1776 */
1777 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1778 err = bitmap_new_disk_sb(bitmap);
1779 else
1780 err = bitmap_read_sb(bitmap);
1781 } else {
1782 err = 0;
1783 if (mddev->bitmap_info.chunksize == 0 ||
1784 mddev->bitmap_info.daemon_sleep == 0)
1785 /* chunksize and time_base need to be
1786 * set first. */
1787 err = -EINVAL;
1788 }
1789 if (err)
1790 goto error;
1791
1792 bitmap->daemon_lastrun = jiffies;
1793 err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1794 if (err)
1795 goto error;
1796
1797 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1798 bitmap->counts.pages, bmname(bitmap));
1799
1800 err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1801 if (err)
1802 goto error;
1803
1804 return bitmap;
1805 error:
1806 bitmap_free(bitmap);
1807 return ERR_PTR(err);
1808 }
1809
1810 int bitmap_load(struct mddev *mddev)
1811 {
1812 int err = 0;
1813 sector_t start = 0;
1814 sector_t sector = 0;
1815 struct bitmap *bitmap = mddev->bitmap;
1816
1817 if (!bitmap)
1818 goto out;
1819
1820 /* Clear out old bitmap info first: Either there is none, or we
1821 * are resuming after someone else has possibly changed things,
1822 * so we should forget old cached info.
1823 * All chunks should be clean, but some might need_sync.
1824 */
1825 while (sector < mddev->resync_max_sectors) {
1826 sector_t blocks;
1827 bitmap_start_sync(bitmap, sector, &blocks, 0);
1828 sector += blocks;
1829 }
1830 bitmap_close_sync(bitmap);
1831
1832 if (mddev->degraded == 0
1833 || bitmap->events_cleared == mddev->events)
1834 /* no need to keep dirty bits to optimise a
1835 * re-add of a missing device */
1836 start = mddev->recovery_cp;
1837
1838 mutex_lock(&mddev->bitmap_info.mutex);
1839 err = bitmap_init_from_disk(bitmap, start);
1840 mutex_unlock(&mddev->bitmap_info.mutex);
1841
1842 if (err)
1843 goto out;
1844 clear_bit(BITMAP_STALE, &bitmap->flags);
1845
1846 /* Kick recovery in case any bits were set */
1847 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1848
1849 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1850 md_wakeup_thread(mddev->thread);
1851
1852 bitmap_update_sb(bitmap);
1853
1854 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1855 err = -EIO;
1856 out:
1857 return err;
1858 }
1859 EXPORT_SYMBOL_GPL(bitmap_load);
1860
1861 /* Loads the bitmap associated with slot and copies the resync information
1862 * to our bitmap
1863 */
1864 int bitmap_copy_from_slot(struct mddev *mddev, int slot,
1865 sector_t *low, sector_t *high, bool clear_bits)
1866 {
1867 int rv = 0, i, j;
1868 sector_t block, lo = 0, hi = 0;
1869 struct bitmap_counts *counts;
1870 struct bitmap *bitmap = bitmap_create(mddev, slot);
1871
1872 if (IS_ERR(bitmap)) {
1873 bitmap_free(bitmap);
1874 return PTR_ERR(bitmap);
1875 }
1876
1877 rv = bitmap_init_from_disk(bitmap, 0);
1878 if (rv)
1879 goto err;
1880
1881 counts = &bitmap->counts;
1882 for (j = 0; j < counts->chunks; j++) {
1883 block = (sector_t)j << counts->chunkshift;
1884 if (bitmap_file_test_bit(bitmap, block)) {
1885 if (!lo)
1886 lo = block;
1887 hi = block;
1888 bitmap_file_clear_bit(bitmap, block);
1889 bitmap_set_memory_bits(mddev->bitmap, block, 1);
1890 bitmap_file_set_bit(mddev->bitmap, block);
1891 }
1892 }
1893
1894 if (clear_bits) {
1895 bitmap_update_sb(bitmap);
1896 /* Setting this for the ev_page should be enough.
1897 * And we do not require both write_all and PAGE_DIRT either
1898 */
1899 for (i = 0; i < bitmap->storage.file_pages; i++)
1900 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
1901 bitmap_write_all(bitmap);
1902 bitmap_unplug(bitmap);
1903 }
1904 *low = lo;
1905 *high = hi;
1906 err:
1907 bitmap_free(bitmap);
1908 return rv;
1909 }
1910 EXPORT_SYMBOL_GPL(bitmap_copy_from_slot);
1911
1912
1913 void bitmap_status(struct seq_file *seq, struct bitmap *bitmap)
1914 {
1915 unsigned long chunk_kb;
1916 struct bitmap_counts *counts;
1917
1918 if (!bitmap)
1919 return;
1920
1921 counts = &bitmap->counts;
1922
1923 chunk_kb = bitmap->mddev->bitmap_info.chunksize >> 10;
1924 seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
1925 "%lu%s chunk",
1926 counts->pages - counts->missing_pages,
1927 counts->pages,
1928 (counts->pages - counts->missing_pages)
1929 << (PAGE_SHIFT - 10),
1930 chunk_kb ? chunk_kb : bitmap->mddev->bitmap_info.chunksize,
1931 chunk_kb ? "KB" : "B");
1932 if (bitmap->storage.file) {
1933 seq_printf(seq, ", file: ");
1934 seq_file_path(seq, bitmap->storage.file, " \t\n");
1935 }
1936
1937 seq_printf(seq, "\n");
1938 }
1939
1940 int bitmap_resize(struct bitmap *bitmap, sector_t blocks,
1941 int chunksize, int init)
1942 {
1943 /* If chunk_size is 0, choose an appropriate chunk size.
1944 * Then possibly allocate new storage space.
1945 * Then quiesce, copy bits, replace bitmap, and re-start
1946 *
1947 * This function is called both to set up the initial bitmap
1948 * and to resize the bitmap while the array is active.
1949 * If this happens as a result of the array being resized,
1950 * chunksize will be zero, and we need to choose a suitable
1951 * chunksize, otherwise we use what we are given.
1952 */
1953 struct bitmap_storage store;
1954 struct bitmap_counts old_counts;
1955 unsigned long chunks;
1956 sector_t block;
1957 sector_t old_blocks, new_blocks;
1958 int chunkshift;
1959 int ret = 0;
1960 long pages;
1961 struct bitmap_page *new_bp;
1962
1963 if (chunksize == 0) {
1964 /* If there is enough space, leave the chunk size unchanged,
1965 * else increase by factor of two until there is enough space.
1966 */
1967 long bytes;
1968 long space = bitmap->mddev->bitmap_info.space;
1969
1970 if (space == 0) {
1971 /* We don't know how much space there is, so limit
1972 * to current size - in sectors.
1973 */
1974 bytes = DIV_ROUND_UP(bitmap->counts.chunks, 8);
1975 if (!bitmap->mddev->bitmap_info.external)
1976 bytes += sizeof(bitmap_super_t);
1977 space = DIV_ROUND_UP(bytes, 512);
1978 bitmap->mddev->bitmap_info.space = space;
1979 }
1980 chunkshift = bitmap->counts.chunkshift;
1981 chunkshift--;
1982 do {
1983 /* 'chunkshift' is shift from block size to chunk size */
1984 chunkshift++;
1985 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1986 bytes = DIV_ROUND_UP(chunks, 8);
1987 if (!bitmap->mddev->bitmap_info.external)
1988 bytes += sizeof(bitmap_super_t);
1989 } while (bytes > (space << 9));
1990 } else
1991 chunkshift = ffz(~chunksize) - BITMAP_BLOCK_SHIFT;
1992
1993 chunks = DIV_ROUND_UP_SECTOR_T(blocks, 1 << chunkshift);
1994 memset(&store, 0, sizeof(store));
1995 if (bitmap->mddev->bitmap_info.offset || bitmap->mddev->bitmap_info.file)
1996 ret = bitmap_storage_alloc(&store, chunks,
1997 !bitmap->mddev->bitmap_info.external,
1998 mddev_is_clustered(bitmap->mddev)
1999 ? bitmap->cluster_slot : 0);
2000 if (ret)
2001 goto err;
2002
2003 pages = DIV_ROUND_UP(chunks, PAGE_COUNTER_RATIO);
2004
2005 new_bp = kzalloc(pages * sizeof(*new_bp), GFP_KERNEL);
2006 ret = -ENOMEM;
2007 if (!new_bp) {
2008 bitmap_file_unmap(&store);
2009 goto err;
2010 }
2011
2012 if (!init)
2013 bitmap->mddev->pers->quiesce(bitmap->mddev, 1);
2014
2015 store.file = bitmap->storage.file;
2016 bitmap->storage.file = NULL;
2017
2018 if (store.sb_page && bitmap->storage.sb_page)
2019 memcpy(page_address(store.sb_page),
2020 page_address(bitmap->storage.sb_page),
2021 sizeof(bitmap_super_t));
2022 bitmap_file_unmap(&bitmap->storage);
2023 bitmap->storage = store;
2024
2025 old_counts = bitmap->counts;
2026 bitmap->counts.bp = new_bp;
2027 bitmap->counts.pages = pages;
2028 bitmap->counts.missing_pages = pages;
2029 bitmap->counts.chunkshift = chunkshift;
2030 bitmap->counts.chunks = chunks;
2031 bitmap->mddev->bitmap_info.chunksize = 1 << (chunkshift +
2032 BITMAP_BLOCK_SHIFT);
2033
2034 blocks = min(old_counts.chunks << old_counts.chunkshift,
2035 chunks << chunkshift);
2036
2037 spin_lock_irq(&bitmap->counts.lock);
2038 /* For cluster raid, need to pre-allocate bitmap */
2039 if (mddev_is_clustered(bitmap->mddev)) {
2040 unsigned long page;
2041 for (page = 0; page < pages; page++) {
2042 ret = bitmap_checkpage(&bitmap->counts, page, 1, 1);
2043 if (ret) {
2044 unsigned long k;
2045
2046 /* deallocate the page memory */
2047 for (k = 0; k < page; k++) {
2048 if (new_bp[k].map)
2049 kfree(new_bp[k].map);
2050 }
2051
2052 /* restore some fields from old_counts */
2053 bitmap->counts.bp = old_counts.bp;
2054 bitmap->counts.pages = old_counts.pages;
2055 bitmap->counts.missing_pages = old_counts.pages;
2056 bitmap->counts.chunkshift = old_counts.chunkshift;
2057 bitmap->counts.chunks = old_counts.chunks;
2058 bitmap->mddev->bitmap_info.chunksize = 1 << (old_counts.chunkshift +
2059 BITMAP_BLOCK_SHIFT);
2060 blocks = old_counts.chunks << old_counts.chunkshift;
2061 pr_err("Could not pre-allocate in-memory bitmap for cluster raid\n");
2062 break;
2063 } else
2064 bitmap->counts.bp[page].count += 1;
2065 }
2066 }
2067
2068 for (block = 0; block < blocks; ) {
2069 bitmap_counter_t *bmc_old, *bmc_new;
2070 int set;
2071
2072 bmc_old = bitmap_get_counter(&old_counts, block,
2073 &old_blocks, 0);
2074 set = bmc_old && NEEDED(*bmc_old);
2075
2076 if (set) {
2077 bmc_new = bitmap_get_counter(&bitmap->counts, block,
2078 &new_blocks, 1);
2079 if (*bmc_new == 0) {
2080 /* need to set on-disk bits too. */
2081 sector_t end = block + new_blocks;
2082 sector_t start = block >> chunkshift;
2083 start <<= chunkshift;
2084 while (start < end) {
2085 bitmap_file_set_bit(bitmap, block);
2086 start += 1 << chunkshift;
2087 }
2088 *bmc_new = 2;
2089 bitmap_count_page(&bitmap->counts,
2090 block, 1);
2091 bitmap_set_pending(&bitmap->counts,
2092 block);
2093 }
2094 *bmc_new |= NEEDED_MASK;
2095 if (new_blocks < old_blocks)
2096 old_blocks = new_blocks;
2097 }
2098 block += old_blocks;
2099 }
2100
2101 if (!init) {
2102 int i;
2103 while (block < (chunks << chunkshift)) {
2104 bitmap_counter_t *bmc;
2105 bmc = bitmap_get_counter(&bitmap->counts, block,
2106 &new_blocks, 1);
2107 if (bmc) {
2108 /* new space. It needs to be resynced, so
2109 * we set NEEDED_MASK.
2110 */
2111 if (*bmc == 0) {
2112 *bmc = NEEDED_MASK | 2;
2113 bitmap_count_page(&bitmap->counts,
2114 block, 1);
2115 bitmap_set_pending(&bitmap->counts,
2116 block);
2117 }
2118 }
2119 block += new_blocks;
2120 }
2121 for (i = 0; i < bitmap->storage.file_pages; i++)
2122 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2123 }
2124 spin_unlock_irq(&bitmap->counts.lock);
2125
2126 if (!init) {
2127 bitmap_unplug(bitmap);
2128 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2129 }
2130 ret = 0;
2131 err:
2132 return ret;
2133 }
2134 EXPORT_SYMBOL_GPL(bitmap_resize);
2135
2136 static ssize_t
2137 location_show(struct mddev *mddev, char *page)
2138 {
2139 ssize_t len;
2140 if (mddev->bitmap_info.file)
2141 len = sprintf(page, "file");
2142 else if (mddev->bitmap_info.offset)
2143 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2144 else
2145 len = sprintf(page, "none");
2146 len += sprintf(page+len, "\n");
2147 return len;
2148 }
2149
2150 static ssize_t
2151 location_store(struct mddev *mddev, const char *buf, size_t len)
2152 {
2153
2154 if (mddev->pers) {
2155 if (!mddev->pers->quiesce)
2156 return -EBUSY;
2157 if (mddev->recovery || mddev->sync_thread)
2158 return -EBUSY;
2159 }
2160
2161 if (mddev->bitmap || mddev->bitmap_info.file ||
2162 mddev->bitmap_info.offset) {
2163 /* bitmap already configured. Only option is to clear it */
2164 if (strncmp(buf, "none", 4) != 0)
2165 return -EBUSY;
2166 if (mddev->pers) {
2167 mddev->pers->quiesce(mddev, 1);
2168 bitmap_destroy(mddev);
2169 mddev->pers->quiesce(mddev, 0);
2170 }
2171 mddev->bitmap_info.offset = 0;
2172 if (mddev->bitmap_info.file) {
2173 struct file *f = mddev->bitmap_info.file;
2174 mddev->bitmap_info.file = NULL;
2175 fput(f);
2176 }
2177 } else {
2178 /* No bitmap, OK to set a location */
2179 long long offset;
2180 if (strncmp(buf, "none", 4) == 0)
2181 /* nothing to be done */;
2182 else if (strncmp(buf, "file:", 5) == 0) {
2183 /* Not supported yet */
2184 return -EINVAL;
2185 } else {
2186 int rv;
2187 if (buf[0] == '+')
2188 rv = kstrtoll(buf+1, 10, &offset);
2189 else
2190 rv = kstrtoll(buf, 10, &offset);
2191 if (rv)
2192 return rv;
2193 if (offset == 0)
2194 return -EINVAL;
2195 if (mddev->bitmap_info.external == 0 &&
2196 mddev->major_version == 0 &&
2197 offset != mddev->bitmap_info.default_offset)
2198 return -EINVAL;
2199 mddev->bitmap_info.offset = offset;
2200 if (mddev->pers) {
2201 struct bitmap *bitmap;
2202 mddev->pers->quiesce(mddev, 1);
2203 bitmap = bitmap_create(mddev, -1);
2204 if (IS_ERR(bitmap))
2205 rv = PTR_ERR(bitmap);
2206 else {
2207 mddev->bitmap = bitmap;
2208 rv = bitmap_load(mddev);
2209 if (rv)
2210 mddev->bitmap_info.offset = 0;
2211 }
2212 mddev->pers->quiesce(mddev, 0);
2213 if (rv) {
2214 bitmap_destroy(mddev);
2215 return rv;
2216 }
2217 }
2218 }
2219 }
2220 if (!mddev->external) {
2221 /* Ensure new bitmap info is stored in
2222 * metadata promptly.
2223 */
2224 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2225 md_wakeup_thread(mddev->thread);
2226 }
2227 return len;
2228 }
2229
2230 static struct md_sysfs_entry bitmap_location =
2231 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2232
2233 /* 'bitmap/space' is the space available at 'location' for the
2234 * bitmap. This allows the kernel to know when it is safe to
2235 * resize the bitmap to match a resized array.
2236 */
2237 static ssize_t
2238 space_show(struct mddev *mddev, char *page)
2239 {
2240 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2241 }
2242
2243 static ssize_t
2244 space_store(struct mddev *mddev, const char *buf, size_t len)
2245 {
2246 unsigned long sectors;
2247 int rv;
2248
2249 rv = kstrtoul(buf, 10, &sectors);
2250 if (rv)
2251 return rv;
2252
2253 if (sectors == 0)
2254 return -EINVAL;
2255
2256 if (mddev->bitmap &&
2257 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2258 return -EFBIG; /* Bitmap is too big for this small space */
2259
2260 /* could make sure it isn't too big, but that isn't really
2261 * needed - user-space should be careful.
2262 */
2263 mddev->bitmap_info.space = sectors;
2264 return len;
2265 }
2266
2267 static struct md_sysfs_entry bitmap_space =
2268 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2269
2270 static ssize_t
2271 timeout_show(struct mddev *mddev, char *page)
2272 {
2273 ssize_t len;
2274 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2275 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2276
2277 len = sprintf(page, "%lu", secs);
2278 if (jifs)
2279 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2280 len += sprintf(page+len, "\n");
2281 return len;
2282 }
2283
2284 static ssize_t
2285 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2286 {
2287 /* timeout can be set at any time */
2288 unsigned long timeout;
2289 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2290 if (rv)
2291 return rv;
2292
2293 /* just to make sure we don't overflow... */
2294 if (timeout >= LONG_MAX / HZ)
2295 return -EINVAL;
2296
2297 timeout = timeout * HZ / 10000;
2298
2299 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2300 timeout = MAX_SCHEDULE_TIMEOUT-1;
2301 if (timeout < 1)
2302 timeout = 1;
2303 mddev->bitmap_info.daemon_sleep = timeout;
2304 if (mddev->thread) {
2305 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2306 * the bitmap is all clean and we don't need to
2307 * adjust the timeout right now
2308 */
2309 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2310 mddev->thread->timeout = timeout;
2311 md_wakeup_thread(mddev->thread);
2312 }
2313 }
2314 return len;
2315 }
2316
2317 static struct md_sysfs_entry bitmap_timeout =
2318 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2319
2320 static ssize_t
2321 backlog_show(struct mddev *mddev, char *page)
2322 {
2323 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2324 }
2325
2326 static ssize_t
2327 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2328 {
2329 unsigned long backlog;
2330 int rv = kstrtoul(buf, 10, &backlog);
2331 if (rv)
2332 return rv;
2333 if (backlog > COUNTER_MAX)
2334 return -EINVAL;
2335 mddev->bitmap_info.max_write_behind = backlog;
2336 return len;
2337 }
2338
2339 static struct md_sysfs_entry bitmap_backlog =
2340 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2341
2342 static ssize_t
2343 chunksize_show(struct mddev *mddev, char *page)
2344 {
2345 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2346 }
2347
2348 static ssize_t
2349 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2350 {
2351 /* Can only be changed when no bitmap is active */
2352 int rv;
2353 unsigned long csize;
2354 if (mddev->bitmap)
2355 return -EBUSY;
2356 rv = kstrtoul(buf, 10, &csize);
2357 if (rv)
2358 return rv;
2359 if (csize < 512 ||
2360 !is_power_of_2(csize))
2361 return -EINVAL;
2362 mddev->bitmap_info.chunksize = csize;
2363 return len;
2364 }
2365
2366 static struct md_sysfs_entry bitmap_chunksize =
2367 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2368
2369 static ssize_t metadata_show(struct mddev *mddev, char *page)
2370 {
2371 if (mddev_is_clustered(mddev))
2372 return sprintf(page, "clustered\n");
2373 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2374 ? "external" : "internal"));
2375 }
2376
2377 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2378 {
2379 if (mddev->bitmap ||
2380 mddev->bitmap_info.file ||
2381 mddev->bitmap_info.offset)
2382 return -EBUSY;
2383 if (strncmp(buf, "external", 8) == 0)
2384 mddev->bitmap_info.external = 1;
2385 else if ((strncmp(buf, "internal", 8) == 0) ||
2386 (strncmp(buf, "clustered", 9) == 0))
2387 mddev->bitmap_info.external = 0;
2388 else
2389 return -EINVAL;
2390 return len;
2391 }
2392
2393 static struct md_sysfs_entry bitmap_metadata =
2394 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2395
2396 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2397 {
2398 int len;
2399 spin_lock(&mddev->lock);
2400 if (mddev->bitmap)
2401 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2402 "false" : "true"));
2403 else
2404 len = sprintf(page, "\n");
2405 spin_unlock(&mddev->lock);
2406 return len;
2407 }
2408
2409 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2410 {
2411 if (mddev->bitmap == NULL)
2412 return -ENOENT;
2413 if (strncmp(buf, "false", 5) == 0)
2414 mddev->bitmap->need_sync = 1;
2415 else if (strncmp(buf, "true", 4) == 0) {
2416 if (mddev->degraded)
2417 return -EBUSY;
2418 mddev->bitmap->need_sync = 0;
2419 } else
2420 return -EINVAL;
2421 return len;
2422 }
2423
2424 static struct md_sysfs_entry bitmap_can_clear =
2425 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2426
2427 static ssize_t
2428 behind_writes_used_show(struct mddev *mddev, char *page)
2429 {
2430 ssize_t ret;
2431 spin_lock(&mddev->lock);
2432 if (mddev->bitmap == NULL)
2433 ret = sprintf(page, "0\n");
2434 else
2435 ret = sprintf(page, "%lu\n",
2436 mddev->bitmap->behind_writes_used);
2437 spin_unlock(&mddev->lock);
2438 return ret;
2439 }
2440
2441 static ssize_t
2442 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2443 {
2444 if (mddev->bitmap)
2445 mddev->bitmap->behind_writes_used = 0;
2446 return len;
2447 }
2448
2449 static struct md_sysfs_entry max_backlog_used =
2450 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2451 behind_writes_used_show, behind_writes_used_reset);
2452
2453 static struct attribute *md_bitmap_attrs[] = {
2454 &bitmap_location.attr,
2455 &bitmap_space.attr,
2456 &bitmap_timeout.attr,
2457 &bitmap_backlog.attr,
2458 &bitmap_chunksize.attr,
2459 &bitmap_metadata.attr,
2460 &bitmap_can_clear.attr,
2461 &max_backlog_used.attr,
2462 NULL
2463 };
2464 struct attribute_group md_bitmap_group = {
2465 .name = "bitmap",
2466 .attrs = md_bitmap_attrs,
2467 };
2468