Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/rzhang/linux
[deliverable/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)
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 /* failed - set the hijacked flag so that we can use the
94 * pointer as a counter */
95 if (!bitmap->bp[page].map)
96 bitmap->bp[page].hijacked = 1;
97 } else if (bitmap->bp[page].map ||
98 bitmap->bp[page].hijacked) {
99 /* somebody beat us to getting the page */
100 kfree(mappage);
101 return 0;
102 } else {
103
104 /* no page was in place and we have one, so install it */
105
106 bitmap->bp[page].map = mappage;
107 bitmap->missing_pages--;
108 }
109 return 0;
110 }
111
112 /* if page is completely empty, put it back on the free list, or dealloc it */
113 /* if page was hijacked, unmark the flag so it might get alloced next time */
114 /* Note: lock should be held when calling this */
115 static void bitmap_checkfree(struct bitmap_counts *bitmap, unsigned long page)
116 {
117 char *ptr;
118
119 if (bitmap->bp[page].count) /* page is still busy */
120 return;
121
122 /* page is no longer in use, it can be released */
123
124 if (bitmap->bp[page].hijacked) { /* page was hijacked, undo this now */
125 bitmap->bp[page].hijacked = 0;
126 bitmap->bp[page].map = NULL;
127 } else {
128 /* normal case, free the page */
129 ptr = bitmap->bp[page].map;
130 bitmap->bp[page].map = NULL;
131 bitmap->missing_pages++;
132 kfree(ptr);
133 }
134 }
135
136 /*
137 * bitmap file handling - read and write the bitmap file and its superblock
138 */
139
140 /*
141 * basic page I/O operations
142 */
143
144 /* IO operations when bitmap is stored near all superblocks */
145 static int read_sb_page(struct mddev *mddev, loff_t offset,
146 struct page *page,
147 unsigned long index, int size)
148 {
149 /* choose a good rdev and read the page from there */
150
151 struct md_rdev *rdev;
152 sector_t target;
153
154 rdev_for_each(rdev, mddev) {
155 if (! test_bit(In_sync, &rdev->flags)
156 || test_bit(Faulty, &rdev->flags))
157 continue;
158
159 target = offset + index * (PAGE_SIZE/512);
160
161 if (sync_page_io(rdev, target,
162 roundup(size, bdev_logical_block_size(rdev->bdev)),
163 page, READ, true)) {
164 page->index = index;
165 return 0;
166 }
167 }
168 return -EIO;
169 }
170
171 static struct md_rdev *next_active_rdev(struct md_rdev *rdev, struct mddev *mddev)
172 {
173 /* Iterate the disks of an mddev, using rcu to protect access to the
174 * linked list, and raising the refcount of devices we return to ensure
175 * they don't disappear while in use.
176 * As devices are only added or removed when raid_disk is < 0 and
177 * nr_pending is 0 and In_sync is clear, the entries we return will
178 * still be in the same position on the list when we re-enter
179 * list_for_each_entry_continue_rcu.
180 *
181 * Note that if entered with 'rdev == NULL' to start at the
182 * beginning, we temporarily assign 'rdev' to an address which
183 * isn't really an rdev, but which can be used by
184 * list_for_each_entry_continue_rcu() to find the first entry.
185 */
186 rcu_read_lock();
187 if (rdev == NULL)
188 /* start at the beginning */
189 rdev = list_entry(&mddev->disks, struct md_rdev, same_set);
190 else {
191 /* release the previous rdev and start from there. */
192 rdev_dec_pending(rdev, mddev);
193 }
194 list_for_each_entry_continue_rcu(rdev, &mddev->disks, same_set) {
195 if (rdev->raid_disk >= 0 &&
196 !test_bit(Faulty, &rdev->flags)) {
197 /* this is a usable devices */
198 atomic_inc(&rdev->nr_pending);
199 rcu_read_unlock();
200 return rdev;
201 }
202 }
203 rcu_read_unlock();
204 return NULL;
205 }
206
207 static int write_sb_page(struct bitmap *bitmap, struct page *page, int wait)
208 {
209 struct md_rdev *rdev = NULL;
210 struct block_device *bdev;
211 struct mddev *mddev = bitmap->mddev;
212 struct bitmap_storage *store = &bitmap->storage;
213 int node_offset = 0;
214
215 if (mddev_is_clustered(bitmap->mddev))
216 node_offset = bitmap->cluster_slot * store->file_pages;
217
218 while ((rdev = next_active_rdev(rdev, mddev)) != NULL) {
219 int size = PAGE_SIZE;
220 loff_t offset = mddev->bitmap_info.offset;
221
222 bdev = (rdev->meta_bdev) ? rdev->meta_bdev : rdev->bdev;
223
224 if (page->index == store->file_pages-1) {
225 int last_page_size = store->bytes & (PAGE_SIZE-1);
226 if (last_page_size == 0)
227 last_page_size = PAGE_SIZE;
228 size = roundup(last_page_size,
229 bdev_logical_block_size(bdev));
230 }
231 /* Just make sure we aren't corrupting data or
232 * metadata
233 */
234 if (mddev->external) {
235 /* Bitmap could be anywhere. */
236 if (rdev->sb_start + offset + (page->index
237 * (PAGE_SIZE/512))
238 > rdev->data_offset
239 &&
240 rdev->sb_start + offset
241 < (rdev->data_offset + mddev->dev_sectors
242 + (PAGE_SIZE/512)))
243 goto bad_alignment;
244 } else if (offset < 0) {
245 /* DATA BITMAP METADATA */
246 if (offset
247 + (long)(page->index * (PAGE_SIZE/512))
248 + size/512 > 0)
249 /* bitmap runs in to metadata */
250 goto bad_alignment;
251 if (rdev->data_offset + mddev->dev_sectors
252 > rdev->sb_start + offset)
253 /* data runs in to bitmap */
254 goto bad_alignment;
255 } else if (rdev->sb_start < rdev->data_offset) {
256 /* METADATA BITMAP DATA */
257 if (rdev->sb_start
258 + offset
259 + page->index*(PAGE_SIZE/512) + size/512
260 > rdev->data_offset)
261 /* bitmap runs in to data */
262 goto bad_alignment;
263 } else {
264 /* DATA METADATA BITMAP - no problems */
265 }
266 md_super_write(mddev, rdev,
267 rdev->sb_start + offset
268 + page->index * (PAGE_SIZE/512),
269 size,
270 page);
271 }
272
273 if (wait)
274 md_super_wait(mddev);
275 return 0;
276
277 bad_alignment:
278 return -EINVAL;
279 }
280
281 static void bitmap_file_kick(struct bitmap *bitmap);
282 /*
283 * write out a page to a file
284 */
285 static void write_page(struct bitmap *bitmap, struct page *page, int wait)
286 {
287 struct buffer_head *bh;
288
289 if (bitmap->storage.file == NULL) {
290 switch (write_sb_page(bitmap, page, wait)) {
291 case -EINVAL:
292 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
293 }
294 } else {
295
296 bh = page_buffers(page);
297
298 while (bh && bh->b_blocknr) {
299 atomic_inc(&bitmap->pending_writes);
300 set_buffer_locked(bh);
301 set_buffer_mapped(bh);
302 submit_bh(WRITE | REQ_SYNC, bh);
303 bh = bh->b_this_page;
304 }
305
306 if (wait)
307 wait_event(bitmap->write_wait,
308 atomic_read(&bitmap->pending_writes)==0);
309 }
310 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
311 bitmap_file_kick(bitmap);
312 }
313
314 static void end_bitmap_write(struct buffer_head *bh, int uptodate)
315 {
316 struct bitmap *bitmap = bh->b_private;
317
318 if (!uptodate)
319 set_bit(BITMAP_WRITE_ERROR, &bitmap->flags);
320 if (atomic_dec_and_test(&bitmap->pending_writes))
321 wake_up(&bitmap->write_wait);
322 }
323
324 /* copied from buffer.c */
325 static void
326 __clear_page_buffers(struct page *page)
327 {
328 ClearPagePrivate(page);
329 set_page_private(page, 0);
330 page_cache_release(page);
331 }
332 static void free_buffers(struct page *page)
333 {
334 struct buffer_head *bh;
335
336 if (!PagePrivate(page))
337 return;
338
339 bh = page_buffers(page);
340 while (bh) {
341 struct buffer_head *next = bh->b_this_page;
342 free_buffer_head(bh);
343 bh = next;
344 }
345 __clear_page_buffers(page);
346 put_page(page);
347 }
348
349 /* read a page from a file.
350 * We both read the page, and attach buffers to the page to record the
351 * address of each block (using bmap). These addresses will be used
352 * to write the block later, completely bypassing the filesystem.
353 * This usage is similar to how swap files are handled, and allows us
354 * to write to a file with no concerns of memory allocation failing.
355 */
356 static int read_page(struct file *file, unsigned long index,
357 struct bitmap *bitmap,
358 unsigned long count,
359 struct page *page)
360 {
361 int ret = 0;
362 struct inode *inode = file_inode(file);
363 struct buffer_head *bh;
364 sector_t block;
365
366 pr_debug("read bitmap file (%dB @ %llu)\n", (int)PAGE_SIZE,
367 (unsigned long long)index << PAGE_SHIFT);
368
369 bh = alloc_page_buffers(page, 1<<inode->i_blkbits, 0);
370 if (!bh) {
371 ret = -ENOMEM;
372 goto out;
373 }
374 attach_page_buffers(page, bh);
375 block = index << (PAGE_SHIFT - inode->i_blkbits);
376 while (bh) {
377 if (count == 0)
378 bh->b_blocknr = 0;
379 else {
380 bh->b_blocknr = bmap(inode, block);
381 if (bh->b_blocknr == 0) {
382 /* Cannot use this file! */
383 ret = -EINVAL;
384 goto out;
385 }
386 bh->b_bdev = inode->i_sb->s_bdev;
387 if (count < (1<<inode->i_blkbits))
388 count = 0;
389 else
390 count -= (1<<inode->i_blkbits);
391
392 bh->b_end_io = end_bitmap_write;
393 bh->b_private = bitmap;
394 atomic_inc(&bitmap->pending_writes);
395 set_buffer_locked(bh);
396 set_buffer_mapped(bh);
397 submit_bh(READ, bh);
398 }
399 block++;
400 bh = bh->b_this_page;
401 }
402 page->index = index;
403
404 wait_event(bitmap->write_wait,
405 atomic_read(&bitmap->pending_writes)==0);
406 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
407 ret = -EIO;
408 out:
409 if (ret)
410 printk(KERN_ALERT "md: bitmap read error: (%dB @ %llu): %d\n",
411 (int)PAGE_SIZE,
412 (unsigned long long)index << PAGE_SHIFT,
413 ret);
414 return ret;
415 }
416
417 /*
418 * bitmap file superblock operations
419 */
420
421 /* update the event counter and sync the superblock to disk */
422 void bitmap_update_sb(struct bitmap *bitmap)
423 {
424 bitmap_super_t *sb;
425
426 if (!bitmap || !bitmap->mddev) /* no bitmap for this array */
427 return;
428 if (bitmap->mddev->bitmap_info.external)
429 return;
430 if (!bitmap->storage.sb_page) /* no superblock */
431 return;
432 sb = kmap_atomic(bitmap->storage.sb_page);
433 sb->events = cpu_to_le64(bitmap->mddev->events);
434 if (bitmap->mddev->events < bitmap->events_cleared)
435 /* rocking back to read-only */
436 bitmap->events_cleared = bitmap->mddev->events;
437 sb->events_cleared = cpu_to_le64(bitmap->events_cleared);
438 sb->state = cpu_to_le32(bitmap->flags);
439 /* Just in case these have been changed via sysfs: */
440 sb->daemon_sleep = cpu_to_le32(bitmap->mddev->bitmap_info.daemon_sleep/HZ);
441 sb->write_behind = cpu_to_le32(bitmap->mddev->bitmap_info.max_write_behind);
442 /* This might have been changed by a reshape */
443 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
444 sb->chunksize = cpu_to_le32(bitmap->mddev->bitmap_info.chunksize);
445 sb->nodes = cpu_to_le32(bitmap->mddev->bitmap_info.nodes);
446 sb->sectors_reserved = cpu_to_le32(bitmap->mddev->
447 bitmap_info.space);
448 kunmap_atomic(sb);
449 write_page(bitmap, bitmap->storage.sb_page, 1);
450 }
451
452 /* print out the bitmap file superblock */
453 void bitmap_print_sb(struct bitmap *bitmap)
454 {
455 bitmap_super_t *sb;
456
457 if (!bitmap || !bitmap->storage.sb_page)
458 return;
459 sb = kmap_atomic(bitmap->storage.sb_page);
460 printk(KERN_DEBUG "%s: bitmap file superblock:\n", bmname(bitmap));
461 printk(KERN_DEBUG " magic: %08x\n", le32_to_cpu(sb->magic));
462 printk(KERN_DEBUG " version: %d\n", le32_to_cpu(sb->version));
463 printk(KERN_DEBUG " uuid: %08x.%08x.%08x.%08x\n",
464 *(__u32 *)(sb->uuid+0),
465 *(__u32 *)(sb->uuid+4),
466 *(__u32 *)(sb->uuid+8),
467 *(__u32 *)(sb->uuid+12));
468 printk(KERN_DEBUG " events: %llu\n",
469 (unsigned long long) le64_to_cpu(sb->events));
470 printk(KERN_DEBUG "events cleared: %llu\n",
471 (unsigned long long) le64_to_cpu(sb->events_cleared));
472 printk(KERN_DEBUG " state: %08x\n", le32_to_cpu(sb->state));
473 printk(KERN_DEBUG " chunksize: %d B\n", le32_to_cpu(sb->chunksize));
474 printk(KERN_DEBUG " daemon sleep: %ds\n", le32_to_cpu(sb->daemon_sleep));
475 printk(KERN_DEBUG " sync size: %llu KB\n",
476 (unsigned long long)le64_to_cpu(sb->sync_size)/2);
477 printk(KERN_DEBUG "max write behind: %d\n", le32_to_cpu(sb->write_behind));
478 kunmap_atomic(sb);
479 }
480
481 /*
482 * bitmap_new_disk_sb
483 * @bitmap
484 *
485 * This function is somewhat the reverse of bitmap_read_sb. bitmap_read_sb
486 * reads and verifies the on-disk bitmap superblock and populates bitmap_info.
487 * This function verifies 'bitmap_info' and populates the on-disk bitmap
488 * structure, which is to be written to disk.
489 *
490 * Returns: 0 on success, -Exxx on error
491 */
492 static int bitmap_new_disk_sb(struct bitmap *bitmap)
493 {
494 bitmap_super_t *sb;
495 unsigned long chunksize, daemon_sleep, write_behind;
496
497 bitmap->storage.sb_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
498 if (bitmap->storage.sb_page == NULL)
499 return -ENOMEM;
500 bitmap->storage.sb_page->index = 0;
501
502 sb = kmap_atomic(bitmap->storage.sb_page);
503
504 sb->magic = cpu_to_le32(BITMAP_MAGIC);
505 sb->version = cpu_to_le32(BITMAP_MAJOR_HI);
506
507 chunksize = bitmap->mddev->bitmap_info.chunksize;
508 BUG_ON(!chunksize);
509 if (!is_power_of_2(chunksize)) {
510 kunmap_atomic(sb);
511 printk(KERN_ERR "bitmap chunksize not a power of 2\n");
512 return -EINVAL;
513 }
514 sb->chunksize = cpu_to_le32(chunksize);
515
516 daemon_sleep = bitmap->mddev->bitmap_info.daemon_sleep;
517 if (!daemon_sleep ||
518 (daemon_sleep < 1) || (daemon_sleep > MAX_SCHEDULE_TIMEOUT)) {
519 printk(KERN_INFO "Choosing daemon_sleep default (5 sec)\n");
520 daemon_sleep = 5 * HZ;
521 }
522 sb->daemon_sleep = cpu_to_le32(daemon_sleep);
523 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
524
525 /*
526 * FIXME: write_behind for RAID1. If not specified, what
527 * is a good choice? We choose COUNTER_MAX / 2 arbitrarily.
528 */
529 write_behind = bitmap->mddev->bitmap_info.max_write_behind;
530 if (write_behind > COUNTER_MAX)
531 write_behind = COUNTER_MAX / 2;
532 sb->write_behind = cpu_to_le32(write_behind);
533 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
534
535 /* keep the array size field of the bitmap superblock up to date */
536 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
537
538 memcpy(sb->uuid, bitmap->mddev->uuid, 16);
539
540 set_bit(BITMAP_STALE, &bitmap->flags);
541 sb->state = cpu_to_le32(bitmap->flags);
542 bitmap->events_cleared = bitmap->mddev->events;
543 sb->events_cleared = cpu_to_le64(bitmap->mddev->events);
544 bitmap->mddev->bitmap_info.nodes = 0;
545
546 kunmap_atomic(sb);
547
548 return 0;
549 }
550
551 /* read the superblock from the bitmap file and initialize some bitmap fields */
552 static int bitmap_read_sb(struct bitmap *bitmap)
553 {
554 char *reason = NULL;
555 bitmap_super_t *sb;
556 unsigned long chunksize, daemon_sleep, write_behind;
557 unsigned long long events;
558 int nodes = 0;
559 unsigned long sectors_reserved = 0;
560 int err = -EINVAL;
561 struct page *sb_page;
562 loff_t offset = bitmap->mddev->bitmap_info.offset;
563
564 if (!bitmap->storage.file && !bitmap->mddev->bitmap_info.offset) {
565 chunksize = 128 * 1024 * 1024;
566 daemon_sleep = 5 * HZ;
567 write_behind = 0;
568 set_bit(BITMAP_STALE, &bitmap->flags);
569 err = 0;
570 goto out_no_sb;
571 }
572 /* page 0 is the superblock, read it... */
573 sb_page = alloc_page(GFP_KERNEL);
574 if (!sb_page)
575 return -ENOMEM;
576 bitmap->storage.sb_page = sb_page;
577
578 re_read:
579 /* If cluster_slot is set, the cluster is setup */
580 if (bitmap->cluster_slot >= 0) {
581 sector_t bm_blocks = bitmap->mddev->resync_max_sectors;
582
583 sector_div(bm_blocks,
584 bitmap->mddev->bitmap_info.chunksize >> 9);
585 /* bits to bytes */
586 bm_blocks = ((bm_blocks+7) >> 3) + sizeof(bitmap_super_t);
587 /* to 4k blocks */
588 bm_blocks = DIV_ROUND_UP_SECTOR_T(bm_blocks, 4096);
589 offset = bitmap->mddev->bitmap_info.offset + (bitmap->cluster_slot * (bm_blocks << 3));
590 pr_info("%s:%d bm slot: %d offset: %llu\n", __func__, __LINE__,
591 bitmap->cluster_slot, offset);
592 }
593
594 if (bitmap->storage.file) {
595 loff_t isize = i_size_read(bitmap->storage.file->f_mapping->host);
596 int bytes = isize > PAGE_SIZE ? PAGE_SIZE : isize;
597
598 err = read_page(bitmap->storage.file, 0,
599 bitmap, bytes, sb_page);
600 } else {
601 err = read_sb_page(bitmap->mddev,
602 offset,
603 sb_page,
604 0, sizeof(bitmap_super_t));
605 }
606 if (err)
607 return err;
608
609 err = -EINVAL;
610 sb = kmap_atomic(sb_page);
611
612 chunksize = le32_to_cpu(sb->chunksize);
613 daemon_sleep = le32_to_cpu(sb->daemon_sleep) * HZ;
614 write_behind = le32_to_cpu(sb->write_behind);
615 sectors_reserved = le32_to_cpu(sb->sectors_reserved);
616 /* Setup nodes/clustername only if bitmap version is
617 * cluster-compatible
618 */
619 if (sb->version == cpu_to_le32(BITMAP_MAJOR_CLUSTERED)) {
620 nodes = le32_to_cpu(sb->nodes);
621 strlcpy(bitmap->mddev->bitmap_info.cluster_name,
622 sb->cluster_name, 64);
623 }
624
625 /* verify that the bitmap-specific fields are valid */
626 if (sb->magic != cpu_to_le32(BITMAP_MAGIC))
627 reason = "bad magic";
628 else if (le32_to_cpu(sb->version) < BITMAP_MAJOR_LO ||
629 le32_to_cpu(sb->version) > BITMAP_MAJOR_CLUSTERED)
630 reason = "unrecognized superblock version";
631 else if (chunksize < 512)
632 reason = "bitmap chunksize too small";
633 else if (!is_power_of_2(chunksize))
634 reason = "bitmap chunksize not a power of 2";
635 else if (daemon_sleep < 1 || daemon_sleep > MAX_SCHEDULE_TIMEOUT)
636 reason = "daemon sleep period out of range";
637 else if (write_behind > COUNTER_MAX)
638 reason = "write-behind limit out of range (0 - 16383)";
639 if (reason) {
640 printk(KERN_INFO "%s: invalid bitmap file superblock: %s\n",
641 bmname(bitmap), reason);
642 goto out;
643 }
644
645 /* keep the array size field of the bitmap superblock up to date */
646 sb->sync_size = cpu_to_le64(bitmap->mddev->resync_max_sectors);
647
648 if (bitmap->mddev->persistent) {
649 /*
650 * We have a persistent array superblock, so compare the
651 * bitmap's UUID and event counter to the mddev's
652 */
653 if (memcmp(sb->uuid, bitmap->mddev->uuid, 16)) {
654 printk(KERN_INFO
655 "%s: bitmap superblock UUID mismatch\n",
656 bmname(bitmap));
657 goto out;
658 }
659 events = le64_to_cpu(sb->events);
660 if (!nodes && (events < bitmap->mddev->events)) {
661 printk(KERN_INFO
662 "%s: bitmap file is out of date (%llu < %llu) "
663 "-- forcing full recovery\n",
664 bmname(bitmap), events,
665 (unsigned long long) bitmap->mddev->events);
666 set_bit(BITMAP_STALE, &bitmap->flags);
667 }
668 }
669
670 /* assign fields using values from superblock */
671 bitmap->flags |= le32_to_cpu(sb->state);
672 if (le32_to_cpu(sb->version) == BITMAP_MAJOR_HOSTENDIAN)
673 set_bit(BITMAP_HOSTENDIAN, &bitmap->flags);
674 bitmap->events_cleared = le64_to_cpu(sb->events_cleared);
675 strlcpy(bitmap->mddev->bitmap_info.cluster_name, sb->cluster_name, 64);
676 err = 0;
677
678 out:
679 kunmap_atomic(sb);
680 /* Assiging chunksize is required for "re_read" */
681 bitmap->mddev->bitmap_info.chunksize = chunksize;
682 if (err == 0 && nodes && (bitmap->cluster_slot < 0)) {
683 err = md_setup_cluster(bitmap->mddev, nodes);
684 if (err) {
685 pr_err("%s: Could not setup cluster service (%d)\n",
686 bmname(bitmap), err);
687 goto out_no_sb;
688 }
689 bitmap->cluster_slot = md_cluster_ops->slot_number(bitmap->mddev);
690 goto re_read;
691 }
692
693
694 out_no_sb:
695 if (test_bit(BITMAP_STALE, &bitmap->flags))
696 bitmap->events_cleared = bitmap->mddev->events;
697 bitmap->mddev->bitmap_info.chunksize = chunksize;
698 bitmap->mddev->bitmap_info.daemon_sleep = daemon_sleep;
699 bitmap->mddev->bitmap_info.max_write_behind = write_behind;
700 bitmap->mddev->bitmap_info.nodes = nodes;
701 if (bitmap->mddev->bitmap_info.space == 0 ||
702 bitmap->mddev->bitmap_info.space > sectors_reserved)
703 bitmap->mddev->bitmap_info.space = sectors_reserved;
704 if (err) {
705 bitmap_print_sb(bitmap);
706 if (bitmap->cluster_slot < 0)
707 md_cluster_stop(bitmap->mddev);
708 }
709 return err;
710 }
711
712 /*
713 * general bitmap file operations
714 */
715
716 /*
717 * on-disk bitmap:
718 *
719 * Use one bit per "chunk" (block set). We do the disk I/O on the bitmap
720 * file a page at a time. There's a superblock at the start of the file.
721 */
722 /* calculate the index of the page that contains this bit */
723 static inline unsigned long file_page_index(struct bitmap_storage *store,
724 unsigned long chunk)
725 {
726 if (store->sb_page)
727 chunk += sizeof(bitmap_super_t) << 3;
728 return chunk >> PAGE_BIT_SHIFT;
729 }
730
731 /* calculate the (bit) offset of this bit within a page */
732 static inline unsigned long file_page_offset(struct bitmap_storage *store,
733 unsigned long chunk)
734 {
735 if (store->sb_page)
736 chunk += sizeof(bitmap_super_t) << 3;
737 return chunk & (PAGE_BITS - 1);
738 }
739
740 /*
741 * return a pointer to the page in the filemap that contains the given bit
742 *
743 */
744 static inline struct page *filemap_get_page(struct bitmap_storage *store,
745 unsigned long chunk)
746 {
747 if (file_page_index(store, chunk) >= store->file_pages)
748 return NULL;
749 return store->filemap[file_page_index(store, chunk)];
750 }
751
752 static int bitmap_storage_alloc(struct bitmap_storage *store,
753 unsigned long chunks, int with_super,
754 int slot_number)
755 {
756 int pnum, offset = 0;
757 unsigned long num_pages;
758 unsigned long bytes;
759
760 bytes = DIV_ROUND_UP(chunks, 8);
761 if (with_super)
762 bytes += sizeof(bitmap_super_t);
763
764 num_pages = DIV_ROUND_UP(bytes, PAGE_SIZE);
765 offset = slot_number * (num_pages - 1);
766
767 store->filemap = kmalloc(sizeof(struct page *)
768 * num_pages, GFP_KERNEL);
769 if (!store->filemap)
770 return -ENOMEM;
771
772 if (with_super && !store->sb_page) {
773 store->sb_page = alloc_page(GFP_KERNEL|__GFP_ZERO);
774 if (store->sb_page == NULL)
775 return -ENOMEM;
776 }
777
778 pnum = 0;
779 if (store->sb_page) {
780 store->filemap[0] = store->sb_page;
781 pnum = 1;
782 store->sb_page->index = offset;
783 }
784
785 for ( ; pnum < num_pages; pnum++) {
786 store->filemap[pnum] = alloc_page(GFP_KERNEL|__GFP_ZERO);
787 if (!store->filemap[pnum]) {
788 store->file_pages = pnum;
789 return -ENOMEM;
790 }
791 store->filemap[pnum]->index = pnum + offset;
792 }
793 store->file_pages = pnum;
794
795 /* We need 4 bits per page, rounded up to a multiple
796 * of sizeof(unsigned long) */
797 store->filemap_attr = kzalloc(
798 roundup(DIV_ROUND_UP(num_pages*4, 8), sizeof(unsigned long)),
799 GFP_KERNEL);
800 if (!store->filemap_attr)
801 return -ENOMEM;
802
803 store->bytes = bytes;
804
805 return 0;
806 }
807
808 static void bitmap_file_unmap(struct bitmap_storage *store)
809 {
810 struct page **map, *sb_page;
811 int pages;
812 struct file *file;
813
814 file = store->file;
815 map = store->filemap;
816 pages = store->file_pages;
817 sb_page = store->sb_page;
818
819 while (pages--)
820 if (map[pages] != sb_page) /* 0 is sb_page, release it below */
821 free_buffers(map[pages]);
822 kfree(map);
823 kfree(store->filemap_attr);
824
825 if (sb_page)
826 free_buffers(sb_page);
827
828 if (file) {
829 struct inode *inode = file_inode(file);
830 invalidate_mapping_pages(inode->i_mapping, 0, -1);
831 fput(file);
832 }
833 }
834
835 /*
836 * bitmap_file_kick - if an error occurs while manipulating the bitmap file
837 * then it is no longer reliable, so we stop using it and we mark the file
838 * as failed in the superblock
839 */
840 static void bitmap_file_kick(struct bitmap *bitmap)
841 {
842 char *path, *ptr = NULL;
843
844 if (!test_and_set_bit(BITMAP_STALE, &bitmap->flags)) {
845 bitmap_update_sb(bitmap);
846
847 if (bitmap->storage.file) {
848 path = kmalloc(PAGE_SIZE, GFP_KERNEL);
849 if (path)
850 ptr = file_path(bitmap->storage.file,
851 path, PAGE_SIZE);
852
853 printk(KERN_ALERT
854 "%s: kicking failed bitmap file %s from array!\n",
855 bmname(bitmap), IS_ERR(ptr) ? "" : ptr);
856
857 kfree(path);
858 } else
859 printk(KERN_ALERT
860 "%s: disabling internal bitmap due to errors\n",
861 bmname(bitmap));
862 }
863 }
864
865 enum bitmap_page_attr {
866 BITMAP_PAGE_DIRTY = 0, /* there are set bits that need to be synced */
867 BITMAP_PAGE_PENDING = 1, /* there are bits that are being cleaned.
868 * i.e. counter is 1 or 2. */
869 BITMAP_PAGE_NEEDWRITE = 2, /* there are cleared bits that need to be synced */
870 };
871
872 static inline void set_page_attr(struct bitmap *bitmap, int pnum,
873 enum bitmap_page_attr attr)
874 {
875 set_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
876 }
877
878 static inline void clear_page_attr(struct bitmap *bitmap, int pnum,
879 enum bitmap_page_attr attr)
880 {
881 clear_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
882 }
883
884 static inline int test_page_attr(struct bitmap *bitmap, int pnum,
885 enum bitmap_page_attr attr)
886 {
887 return test_bit((pnum<<2) + attr, bitmap->storage.filemap_attr);
888 }
889
890 static inline int test_and_clear_page_attr(struct bitmap *bitmap, int pnum,
891 enum bitmap_page_attr attr)
892 {
893 return test_and_clear_bit((pnum<<2) + attr,
894 bitmap->storage.filemap_attr);
895 }
896 /*
897 * bitmap_file_set_bit -- called before performing a write to the md device
898 * to set (and eventually sync) a particular bit in the bitmap file
899 *
900 * we set the bit immediately, then we record the page number so that
901 * when an unplug occurs, we can flush the dirty pages out to disk
902 */
903 static void bitmap_file_set_bit(struct bitmap *bitmap, sector_t block)
904 {
905 unsigned long bit;
906 struct page *page;
907 void *kaddr;
908 unsigned long chunk = block >> bitmap->counts.chunkshift;
909
910 page = filemap_get_page(&bitmap->storage, chunk);
911 if (!page)
912 return;
913 bit = file_page_offset(&bitmap->storage, chunk);
914
915 /* set the bit */
916 kaddr = kmap_atomic(page);
917 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
918 set_bit(bit, kaddr);
919 else
920 set_bit_le(bit, kaddr);
921 kunmap_atomic(kaddr);
922 pr_debug("set file bit %lu page %lu\n", bit, page->index);
923 /* record page number so it gets flushed to disk when unplug occurs */
924 set_page_attr(bitmap, page->index, BITMAP_PAGE_DIRTY);
925 }
926
927 static void bitmap_file_clear_bit(struct bitmap *bitmap, sector_t block)
928 {
929 unsigned long bit;
930 struct page *page;
931 void *paddr;
932 unsigned long chunk = block >> bitmap->counts.chunkshift;
933
934 page = filemap_get_page(&bitmap->storage, chunk);
935 if (!page)
936 return;
937 bit = file_page_offset(&bitmap->storage, chunk);
938 paddr = kmap_atomic(page);
939 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
940 clear_bit(bit, paddr);
941 else
942 clear_bit_le(bit, paddr);
943 kunmap_atomic(paddr);
944 if (!test_page_attr(bitmap, page->index, BITMAP_PAGE_NEEDWRITE)) {
945 set_page_attr(bitmap, page->index, BITMAP_PAGE_PENDING);
946 bitmap->allclean = 0;
947 }
948 }
949
950 static int bitmap_file_test_bit(struct bitmap *bitmap, sector_t block)
951 {
952 unsigned long bit;
953 struct page *page;
954 void *paddr;
955 unsigned long chunk = block >> bitmap->counts.chunkshift;
956 int set = 0;
957
958 page = filemap_get_page(&bitmap->storage, chunk);
959 if (!page)
960 return -EINVAL;
961 bit = file_page_offset(&bitmap->storage, chunk);
962 paddr = kmap_atomic(page);
963 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
964 set = test_bit(bit, paddr);
965 else
966 set = test_bit_le(bit, paddr);
967 kunmap_atomic(paddr);
968 return set;
969 }
970
971
972 /* this gets called when the md device is ready to unplug its underlying
973 * (slave) device queues -- before we let any writes go down, we need to
974 * sync the dirty pages of the bitmap file to disk */
975 void bitmap_unplug(struct bitmap *bitmap)
976 {
977 unsigned long i;
978 int dirty, need_write;
979
980 if (!bitmap || !bitmap->storage.filemap ||
981 test_bit(BITMAP_STALE, &bitmap->flags))
982 return;
983
984 /* look at each page to see if there are any set bits that need to be
985 * flushed out to disk */
986 for (i = 0; i < bitmap->storage.file_pages; i++) {
987 if (!bitmap->storage.filemap)
988 return;
989 dirty = test_and_clear_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
990 need_write = test_and_clear_page_attr(bitmap, i,
991 BITMAP_PAGE_NEEDWRITE);
992 if (dirty || need_write) {
993 clear_page_attr(bitmap, i, BITMAP_PAGE_PENDING);
994 write_page(bitmap, bitmap->storage.filemap[i], 0);
995 }
996 }
997 if (bitmap->storage.file)
998 wait_event(bitmap->write_wait,
999 atomic_read(&bitmap->pending_writes)==0);
1000 else
1001 md_super_wait(bitmap->mddev);
1002
1003 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1004 bitmap_file_kick(bitmap);
1005 }
1006 EXPORT_SYMBOL(bitmap_unplug);
1007
1008 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed);
1009 /* * bitmap_init_from_disk -- called at bitmap_create time to initialize
1010 * the in-memory bitmap from the on-disk bitmap -- also, sets up the
1011 * memory mapping of the bitmap file
1012 * Special cases:
1013 * if there's no bitmap file, or if the bitmap file had been
1014 * previously kicked from the array, we mark all the bits as
1015 * 1's in order to cause a full resync.
1016 *
1017 * We ignore all bits for sectors that end earlier than 'start'.
1018 * This is used when reading an out-of-date bitmap...
1019 */
1020 static int bitmap_init_from_disk(struct bitmap *bitmap, sector_t start)
1021 {
1022 unsigned long i, chunks, index, oldindex, bit, node_offset = 0;
1023 struct page *page = NULL;
1024 unsigned long bit_cnt = 0;
1025 struct file *file;
1026 unsigned long offset;
1027 int outofdate;
1028 int ret = -ENOSPC;
1029 void *paddr;
1030 struct bitmap_storage *store = &bitmap->storage;
1031
1032 chunks = bitmap->counts.chunks;
1033 file = store->file;
1034
1035 if (!file && !bitmap->mddev->bitmap_info.offset) {
1036 /* No permanent bitmap - fill with '1s'. */
1037 store->filemap = NULL;
1038 store->file_pages = 0;
1039 for (i = 0; i < chunks ; i++) {
1040 /* if the disk bit is set, set the memory bit */
1041 int needed = ((sector_t)(i+1) << (bitmap->counts.chunkshift)
1042 >= start);
1043 bitmap_set_memory_bits(bitmap,
1044 (sector_t)i << bitmap->counts.chunkshift,
1045 needed);
1046 }
1047 return 0;
1048 }
1049
1050 outofdate = test_bit(BITMAP_STALE, &bitmap->flags);
1051 if (outofdate)
1052 printk(KERN_INFO "%s: bitmap file is out of date, doing full "
1053 "recovery\n", bmname(bitmap));
1054
1055 if (file && i_size_read(file->f_mapping->host) < store->bytes) {
1056 printk(KERN_INFO "%s: bitmap file too short %lu < %lu\n",
1057 bmname(bitmap),
1058 (unsigned long) i_size_read(file->f_mapping->host),
1059 store->bytes);
1060 goto err;
1061 }
1062
1063 oldindex = ~0L;
1064 offset = 0;
1065 if (!bitmap->mddev->bitmap_info.external)
1066 offset = sizeof(bitmap_super_t);
1067
1068 if (mddev_is_clustered(bitmap->mddev))
1069 node_offset = bitmap->cluster_slot * (DIV_ROUND_UP(store->bytes, PAGE_SIZE));
1070
1071 for (i = 0; i < chunks; i++) {
1072 int b;
1073 index = file_page_index(&bitmap->storage, i);
1074 bit = file_page_offset(&bitmap->storage, i);
1075 if (index != oldindex) { /* this is a new page, read it in */
1076 int count;
1077 /* unmap the old page, we're done with it */
1078 if (index == store->file_pages-1)
1079 count = store->bytes - index * PAGE_SIZE;
1080 else
1081 count = PAGE_SIZE;
1082 page = store->filemap[index];
1083 if (file)
1084 ret = read_page(file, index, bitmap,
1085 count, page);
1086 else
1087 ret = read_sb_page(
1088 bitmap->mddev,
1089 bitmap->mddev->bitmap_info.offset,
1090 page,
1091 index + node_offset, count);
1092
1093 if (ret)
1094 goto err;
1095
1096 oldindex = index;
1097
1098 if (outofdate) {
1099 /*
1100 * if bitmap is out of date, dirty the
1101 * whole page and write it out
1102 */
1103 paddr = kmap_atomic(page);
1104 memset(paddr + offset, 0xff,
1105 PAGE_SIZE - offset);
1106 kunmap_atomic(paddr);
1107 write_page(bitmap, page, 1);
1108
1109 ret = -EIO;
1110 if (test_bit(BITMAP_WRITE_ERROR,
1111 &bitmap->flags))
1112 goto err;
1113 }
1114 }
1115 paddr = kmap_atomic(page);
1116 if (test_bit(BITMAP_HOSTENDIAN, &bitmap->flags))
1117 b = test_bit(bit, paddr);
1118 else
1119 b = test_bit_le(bit, paddr);
1120 kunmap_atomic(paddr);
1121 if (b) {
1122 /* if the disk bit is set, set the memory bit */
1123 int needed = ((sector_t)(i+1) << bitmap->counts.chunkshift
1124 >= start);
1125 bitmap_set_memory_bits(bitmap,
1126 (sector_t)i << bitmap->counts.chunkshift,
1127 needed);
1128 bit_cnt++;
1129 }
1130 offset = 0;
1131 }
1132
1133 printk(KERN_INFO "%s: bitmap initialized from disk: "
1134 "read %lu pages, set %lu of %lu bits\n",
1135 bmname(bitmap), store->file_pages,
1136 bit_cnt, chunks);
1137
1138 return 0;
1139
1140 err:
1141 printk(KERN_INFO "%s: bitmap initialisation failed: %d\n",
1142 bmname(bitmap), ret);
1143 return ret;
1144 }
1145
1146 void bitmap_write_all(struct bitmap *bitmap)
1147 {
1148 /* We don't actually write all bitmap blocks here,
1149 * just flag them as needing to be written
1150 */
1151 int i;
1152
1153 if (!bitmap || !bitmap->storage.filemap)
1154 return;
1155 if (bitmap->storage.file)
1156 /* Only one copy, so nothing needed */
1157 return;
1158
1159 for (i = 0; i < bitmap->storage.file_pages; i++)
1160 set_page_attr(bitmap, i,
1161 BITMAP_PAGE_NEEDWRITE);
1162 bitmap->allclean = 0;
1163 }
1164
1165 static void bitmap_count_page(struct bitmap_counts *bitmap,
1166 sector_t offset, int inc)
1167 {
1168 sector_t chunk = offset >> bitmap->chunkshift;
1169 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1170 bitmap->bp[page].count += inc;
1171 bitmap_checkfree(bitmap, page);
1172 }
1173
1174 static void bitmap_set_pending(struct bitmap_counts *bitmap, sector_t offset)
1175 {
1176 sector_t chunk = offset >> bitmap->chunkshift;
1177 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1178 struct bitmap_page *bp = &bitmap->bp[page];
1179
1180 if (!bp->pending)
1181 bp->pending = 1;
1182 }
1183
1184 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1185 sector_t offset, sector_t *blocks,
1186 int create);
1187
1188 /*
1189 * bitmap daemon -- periodically wakes up to clean bits and flush pages
1190 * out to disk
1191 */
1192
1193 void bitmap_daemon_work(struct mddev *mddev)
1194 {
1195 struct bitmap *bitmap;
1196 unsigned long j;
1197 unsigned long nextpage;
1198 sector_t blocks;
1199 struct bitmap_counts *counts;
1200
1201 /* Use a mutex to guard daemon_work against
1202 * bitmap_destroy.
1203 */
1204 mutex_lock(&mddev->bitmap_info.mutex);
1205 bitmap = mddev->bitmap;
1206 if (bitmap == NULL) {
1207 mutex_unlock(&mddev->bitmap_info.mutex);
1208 return;
1209 }
1210 if (time_before(jiffies, bitmap->daemon_lastrun
1211 + mddev->bitmap_info.daemon_sleep))
1212 goto done;
1213
1214 bitmap->daemon_lastrun = jiffies;
1215 if (bitmap->allclean) {
1216 mddev->thread->timeout = MAX_SCHEDULE_TIMEOUT;
1217 goto done;
1218 }
1219 bitmap->allclean = 1;
1220
1221 /* Any file-page which is PENDING now needs to be written.
1222 * So set NEEDWRITE now, then after we make any last-minute changes
1223 * we will write it.
1224 */
1225 for (j = 0; j < bitmap->storage.file_pages; j++)
1226 if (test_and_clear_page_attr(bitmap, j,
1227 BITMAP_PAGE_PENDING))
1228 set_page_attr(bitmap, j,
1229 BITMAP_PAGE_NEEDWRITE);
1230
1231 if (bitmap->need_sync &&
1232 mddev->bitmap_info.external == 0) {
1233 /* Arrange for superblock update as well as
1234 * other changes */
1235 bitmap_super_t *sb;
1236 bitmap->need_sync = 0;
1237 if (bitmap->storage.filemap) {
1238 sb = kmap_atomic(bitmap->storage.sb_page);
1239 sb->events_cleared =
1240 cpu_to_le64(bitmap->events_cleared);
1241 kunmap_atomic(sb);
1242 set_page_attr(bitmap, 0,
1243 BITMAP_PAGE_NEEDWRITE);
1244 }
1245 }
1246 /* Now look at the bitmap counters and if any are '2' or '1',
1247 * decrement and handle accordingly.
1248 */
1249 counts = &bitmap->counts;
1250 spin_lock_irq(&counts->lock);
1251 nextpage = 0;
1252 for (j = 0; j < counts->chunks; j++) {
1253 bitmap_counter_t *bmc;
1254 sector_t block = (sector_t)j << counts->chunkshift;
1255
1256 if (j == nextpage) {
1257 nextpage += PAGE_COUNTER_RATIO;
1258 if (!counts->bp[j >> PAGE_COUNTER_SHIFT].pending) {
1259 j |= PAGE_COUNTER_MASK;
1260 continue;
1261 }
1262 counts->bp[j >> PAGE_COUNTER_SHIFT].pending = 0;
1263 }
1264 bmc = bitmap_get_counter(counts,
1265 block,
1266 &blocks, 0);
1267
1268 if (!bmc) {
1269 j |= PAGE_COUNTER_MASK;
1270 continue;
1271 }
1272 if (*bmc == 1 && !bitmap->need_sync) {
1273 /* We can clear the bit */
1274 *bmc = 0;
1275 bitmap_count_page(counts, block, -1);
1276 bitmap_file_clear_bit(bitmap, block);
1277 } else if (*bmc && *bmc <= 2) {
1278 *bmc = 1;
1279 bitmap_set_pending(counts, block);
1280 bitmap->allclean = 0;
1281 }
1282 }
1283 spin_unlock_irq(&counts->lock);
1284
1285 /* Now start writeout on any page in NEEDWRITE that isn't DIRTY.
1286 * DIRTY pages need to be written by bitmap_unplug so it can wait
1287 * for them.
1288 * If we find any DIRTY page we stop there and let bitmap_unplug
1289 * handle all the rest. This is important in the case where
1290 * the first blocking holds the superblock and it has been updated.
1291 * We mustn't write any other blocks before the superblock.
1292 */
1293 for (j = 0;
1294 j < bitmap->storage.file_pages
1295 && !test_bit(BITMAP_STALE, &bitmap->flags);
1296 j++) {
1297 if (test_page_attr(bitmap, j,
1298 BITMAP_PAGE_DIRTY))
1299 /* bitmap_unplug will handle the rest */
1300 break;
1301 if (test_and_clear_page_attr(bitmap, j,
1302 BITMAP_PAGE_NEEDWRITE)) {
1303 write_page(bitmap, bitmap->storage.filemap[j], 0);
1304 }
1305 }
1306
1307 done:
1308 if (bitmap->allclean == 0)
1309 mddev->thread->timeout =
1310 mddev->bitmap_info.daemon_sleep;
1311 mutex_unlock(&mddev->bitmap_info.mutex);
1312 }
1313
1314 static bitmap_counter_t *bitmap_get_counter(struct bitmap_counts *bitmap,
1315 sector_t offset, sector_t *blocks,
1316 int create)
1317 __releases(bitmap->lock)
1318 __acquires(bitmap->lock)
1319 {
1320 /* If 'create', we might release the lock and reclaim it.
1321 * The lock must have been taken with interrupts enabled.
1322 * If !create, we don't release the lock.
1323 */
1324 sector_t chunk = offset >> bitmap->chunkshift;
1325 unsigned long page = chunk >> PAGE_COUNTER_SHIFT;
1326 unsigned long pageoff = (chunk & PAGE_COUNTER_MASK) << COUNTER_BYTE_SHIFT;
1327 sector_t csize;
1328 int err;
1329
1330 err = bitmap_checkpage(bitmap, page, create);
1331
1332 if (bitmap->bp[page].hijacked ||
1333 bitmap->bp[page].map == NULL)
1334 csize = ((sector_t)1) << (bitmap->chunkshift +
1335 PAGE_COUNTER_SHIFT - 1);
1336 else
1337 csize = ((sector_t)1) << bitmap->chunkshift;
1338 *blocks = csize - (offset & (csize - 1));
1339
1340 if (err < 0)
1341 return NULL;
1342
1343 /* now locked ... */
1344
1345 if (bitmap->bp[page].hijacked) { /* hijacked pointer */
1346 /* should we use the first or second counter field
1347 * of the hijacked pointer? */
1348 int hi = (pageoff > PAGE_COUNTER_MASK);
1349 return &((bitmap_counter_t *)
1350 &bitmap->bp[page].map)[hi];
1351 } else /* page is allocated */
1352 return (bitmap_counter_t *)
1353 &(bitmap->bp[page].map[pageoff]);
1354 }
1355
1356 int bitmap_startwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors, int behind)
1357 {
1358 if (!bitmap)
1359 return 0;
1360
1361 if (behind) {
1362 int bw;
1363 atomic_inc(&bitmap->behind_writes);
1364 bw = atomic_read(&bitmap->behind_writes);
1365 if (bw > bitmap->behind_writes_used)
1366 bitmap->behind_writes_used = bw;
1367
1368 pr_debug("inc write-behind count %d/%lu\n",
1369 bw, bitmap->mddev->bitmap_info.max_write_behind);
1370 }
1371
1372 while (sectors) {
1373 sector_t blocks;
1374 bitmap_counter_t *bmc;
1375
1376 spin_lock_irq(&bitmap->counts.lock);
1377 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 1);
1378 if (!bmc) {
1379 spin_unlock_irq(&bitmap->counts.lock);
1380 return 0;
1381 }
1382
1383 if (unlikely(COUNTER(*bmc) == COUNTER_MAX)) {
1384 DEFINE_WAIT(__wait);
1385 /* note that it is safe to do the prepare_to_wait
1386 * after the test as long as we do it before dropping
1387 * the spinlock.
1388 */
1389 prepare_to_wait(&bitmap->overflow_wait, &__wait,
1390 TASK_UNINTERRUPTIBLE);
1391 spin_unlock_irq(&bitmap->counts.lock);
1392 schedule();
1393 finish_wait(&bitmap->overflow_wait, &__wait);
1394 continue;
1395 }
1396
1397 switch (*bmc) {
1398 case 0:
1399 bitmap_file_set_bit(bitmap, offset);
1400 bitmap_count_page(&bitmap->counts, offset, 1);
1401 /* fall through */
1402 case 1:
1403 *bmc = 2;
1404 }
1405
1406 (*bmc)++;
1407
1408 spin_unlock_irq(&bitmap->counts.lock);
1409
1410 offset += blocks;
1411 if (sectors > blocks)
1412 sectors -= blocks;
1413 else
1414 sectors = 0;
1415 }
1416 return 0;
1417 }
1418 EXPORT_SYMBOL(bitmap_startwrite);
1419
1420 void bitmap_endwrite(struct bitmap *bitmap, sector_t offset, unsigned long sectors,
1421 int success, int behind)
1422 {
1423 if (!bitmap)
1424 return;
1425 if (behind) {
1426 if (atomic_dec_and_test(&bitmap->behind_writes))
1427 wake_up(&bitmap->behind_wait);
1428 pr_debug("dec write-behind count %d/%lu\n",
1429 atomic_read(&bitmap->behind_writes),
1430 bitmap->mddev->bitmap_info.max_write_behind);
1431 }
1432
1433 while (sectors) {
1434 sector_t blocks;
1435 unsigned long flags;
1436 bitmap_counter_t *bmc;
1437
1438 spin_lock_irqsave(&bitmap->counts.lock, flags);
1439 bmc = bitmap_get_counter(&bitmap->counts, offset, &blocks, 0);
1440 if (!bmc) {
1441 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1442 return;
1443 }
1444
1445 if (success && !bitmap->mddev->degraded &&
1446 bitmap->events_cleared < bitmap->mddev->events) {
1447 bitmap->events_cleared = bitmap->mddev->events;
1448 bitmap->need_sync = 1;
1449 sysfs_notify_dirent_safe(bitmap->sysfs_can_clear);
1450 }
1451
1452 if (!success && !NEEDED(*bmc))
1453 *bmc |= NEEDED_MASK;
1454
1455 if (COUNTER(*bmc) == COUNTER_MAX)
1456 wake_up(&bitmap->overflow_wait);
1457
1458 (*bmc)--;
1459 if (*bmc <= 2) {
1460 bitmap_set_pending(&bitmap->counts, offset);
1461 bitmap->allclean = 0;
1462 }
1463 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1464 offset += blocks;
1465 if (sectors > blocks)
1466 sectors -= blocks;
1467 else
1468 sectors = 0;
1469 }
1470 }
1471 EXPORT_SYMBOL(bitmap_endwrite);
1472
1473 static int __bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1474 int degraded)
1475 {
1476 bitmap_counter_t *bmc;
1477 int rv;
1478 if (bitmap == NULL) {/* FIXME or bitmap set as 'failed' */
1479 *blocks = 1024;
1480 return 1; /* always resync if no bitmap */
1481 }
1482 spin_lock_irq(&bitmap->counts.lock);
1483 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1484 rv = 0;
1485 if (bmc) {
1486 /* locked */
1487 if (RESYNC(*bmc))
1488 rv = 1;
1489 else if (NEEDED(*bmc)) {
1490 rv = 1;
1491 if (!degraded) { /* don't set/clear bits if degraded */
1492 *bmc |= RESYNC_MASK;
1493 *bmc &= ~NEEDED_MASK;
1494 }
1495 }
1496 }
1497 spin_unlock_irq(&bitmap->counts.lock);
1498 return rv;
1499 }
1500
1501 int bitmap_start_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks,
1502 int degraded)
1503 {
1504 /* bitmap_start_sync must always report on multiples of whole
1505 * pages, otherwise resync (which is very PAGE_SIZE based) will
1506 * get confused.
1507 * So call __bitmap_start_sync repeatedly (if needed) until
1508 * At least PAGE_SIZE>>9 blocks are covered.
1509 * Return the 'or' of the result.
1510 */
1511 int rv = 0;
1512 sector_t blocks1;
1513
1514 *blocks = 0;
1515 while (*blocks < (PAGE_SIZE>>9)) {
1516 rv |= __bitmap_start_sync(bitmap, offset,
1517 &blocks1, degraded);
1518 offset += blocks1;
1519 *blocks += blocks1;
1520 }
1521 return rv;
1522 }
1523 EXPORT_SYMBOL(bitmap_start_sync);
1524
1525 void bitmap_end_sync(struct bitmap *bitmap, sector_t offset, sector_t *blocks, int aborted)
1526 {
1527 bitmap_counter_t *bmc;
1528 unsigned long flags;
1529
1530 if (bitmap == NULL) {
1531 *blocks = 1024;
1532 return;
1533 }
1534 spin_lock_irqsave(&bitmap->counts.lock, flags);
1535 bmc = bitmap_get_counter(&bitmap->counts, offset, blocks, 0);
1536 if (bmc == NULL)
1537 goto unlock;
1538 /* locked */
1539 if (RESYNC(*bmc)) {
1540 *bmc &= ~RESYNC_MASK;
1541
1542 if (!NEEDED(*bmc) && aborted)
1543 *bmc |= NEEDED_MASK;
1544 else {
1545 if (*bmc <= 2) {
1546 bitmap_set_pending(&bitmap->counts, offset);
1547 bitmap->allclean = 0;
1548 }
1549 }
1550 }
1551 unlock:
1552 spin_unlock_irqrestore(&bitmap->counts.lock, flags);
1553 }
1554 EXPORT_SYMBOL(bitmap_end_sync);
1555
1556 void bitmap_close_sync(struct bitmap *bitmap)
1557 {
1558 /* Sync has finished, and any bitmap chunks that weren't synced
1559 * properly have been aborted. It remains to us to clear the
1560 * RESYNC bit wherever it is still on
1561 */
1562 sector_t sector = 0;
1563 sector_t blocks;
1564 if (!bitmap)
1565 return;
1566 while (sector < bitmap->mddev->resync_max_sectors) {
1567 bitmap_end_sync(bitmap, sector, &blocks, 0);
1568 sector += blocks;
1569 }
1570 }
1571 EXPORT_SYMBOL(bitmap_close_sync);
1572
1573 void bitmap_cond_end_sync(struct bitmap *bitmap, sector_t sector, bool force)
1574 {
1575 sector_t s = 0;
1576 sector_t blocks;
1577
1578 if (!bitmap)
1579 return;
1580 if (sector == 0) {
1581 bitmap->last_end_sync = jiffies;
1582 return;
1583 }
1584 if (!force && time_before(jiffies, (bitmap->last_end_sync
1585 + bitmap->mddev->bitmap_info.daemon_sleep)))
1586 return;
1587 wait_event(bitmap->mddev->recovery_wait,
1588 atomic_read(&bitmap->mddev->recovery_active) == 0);
1589
1590 bitmap->mddev->curr_resync_completed = sector;
1591 set_bit(MD_CHANGE_CLEAN, &bitmap->mddev->flags);
1592 sector &= ~((1ULL << bitmap->counts.chunkshift) - 1);
1593 s = 0;
1594 while (s < sector && s < bitmap->mddev->resync_max_sectors) {
1595 bitmap_end_sync(bitmap, s, &blocks, 0);
1596 s += blocks;
1597 }
1598 bitmap->last_end_sync = jiffies;
1599 sysfs_notify(&bitmap->mddev->kobj, NULL, "sync_completed");
1600 }
1601 EXPORT_SYMBOL(bitmap_cond_end_sync);
1602
1603 static void bitmap_set_memory_bits(struct bitmap *bitmap, sector_t offset, int needed)
1604 {
1605 /* For each chunk covered by any of these sectors, set the
1606 * counter to 2 and possibly set resync_needed. They should all
1607 * be 0 at this point
1608 */
1609
1610 sector_t secs;
1611 bitmap_counter_t *bmc;
1612 spin_lock_irq(&bitmap->counts.lock);
1613 bmc = bitmap_get_counter(&bitmap->counts, offset, &secs, 1);
1614 if (!bmc) {
1615 spin_unlock_irq(&bitmap->counts.lock);
1616 return;
1617 }
1618 if (!*bmc) {
1619 *bmc = 2;
1620 bitmap_count_page(&bitmap->counts, offset, 1);
1621 bitmap_set_pending(&bitmap->counts, offset);
1622 bitmap->allclean = 0;
1623 }
1624 if (needed)
1625 *bmc |= NEEDED_MASK;
1626 spin_unlock_irq(&bitmap->counts.lock);
1627 }
1628
1629 /* dirty the memory and file bits for bitmap chunks "s" to "e" */
1630 void bitmap_dirty_bits(struct bitmap *bitmap, unsigned long s, unsigned long e)
1631 {
1632 unsigned long chunk;
1633
1634 for (chunk = s; chunk <= e; chunk++) {
1635 sector_t sec = (sector_t)chunk << bitmap->counts.chunkshift;
1636 bitmap_set_memory_bits(bitmap, sec, 1);
1637 bitmap_file_set_bit(bitmap, sec);
1638 if (sec < bitmap->mddev->recovery_cp)
1639 /* We are asserting that the array is dirty,
1640 * so move the recovery_cp address back so
1641 * that it is obvious that it is dirty
1642 */
1643 bitmap->mddev->recovery_cp = sec;
1644 }
1645 }
1646
1647 /*
1648 * flush out any pending updates
1649 */
1650 void bitmap_flush(struct mddev *mddev)
1651 {
1652 struct bitmap *bitmap = mddev->bitmap;
1653 long sleep;
1654
1655 if (!bitmap) /* there was no bitmap */
1656 return;
1657
1658 /* run the daemon_work three time to ensure everything is flushed
1659 * that can be
1660 */
1661 sleep = mddev->bitmap_info.daemon_sleep * 2;
1662 bitmap->daemon_lastrun -= sleep;
1663 bitmap_daemon_work(mddev);
1664 bitmap->daemon_lastrun -= sleep;
1665 bitmap_daemon_work(mddev);
1666 bitmap->daemon_lastrun -= sleep;
1667 bitmap_daemon_work(mddev);
1668 bitmap_update_sb(bitmap);
1669 }
1670
1671 /*
1672 * free memory that was allocated
1673 */
1674 static void bitmap_free(struct bitmap *bitmap)
1675 {
1676 unsigned long k, pages;
1677 struct bitmap_page *bp;
1678
1679 if (!bitmap) /* there was no bitmap */
1680 return;
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 if (bitmap->sysfs_can_clear)
1722 sysfs_put(bitmap->sysfs_can_clear);
1723
1724 bitmap_free(bitmap);
1725 }
1726
1727 /*
1728 * initialize the bitmap structure
1729 * if this returns an error, bitmap_destroy must be called to do clean up
1730 */
1731 struct bitmap *bitmap_create(struct mddev *mddev, int slot)
1732 {
1733 struct bitmap *bitmap;
1734 sector_t blocks = mddev->resync_max_sectors;
1735 struct file *file = mddev->bitmap_info.file;
1736 int err;
1737 struct kernfs_node *bm = NULL;
1738
1739 BUILD_BUG_ON(sizeof(bitmap_super_t) != 256);
1740
1741 BUG_ON(file && mddev->bitmap_info.offset);
1742
1743 bitmap = kzalloc(sizeof(*bitmap), GFP_KERNEL);
1744 if (!bitmap)
1745 return ERR_PTR(-ENOMEM);
1746
1747 spin_lock_init(&bitmap->counts.lock);
1748 atomic_set(&bitmap->pending_writes, 0);
1749 init_waitqueue_head(&bitmap->write_wait);
1750 init_waitqueue_head(&bitmap->overflow_wait);
1751 init_waitqueue_head(&bitmap->behind_wait);
1752
1753 bitmap->mddev = mddev;
1754 bitmap->cluster_slot = slot;
1755
1756 if (mddev->kobj.sd)
1757 bm = sysfs_get_dirent(mddev->kobj.sd, "bitmap");
1758 if (bm) {
1759 bitmap->sysfs_can_clear = sysfs_get_dirent(bm, "can_clear");
1760 sysfs_put(bm);
1761 } else
1762 bitmap->sysfs_can_clear = NULL;
1763
1764 bitmap->storage.file = file;
1765 if (file) {
1766 get_file(file);
1767 /* As future accesses to this file will use bmap,
1768 * and bypass the page cache, we must sync the file
1769 * first.
1770 */
1771 vfs_fsync(file, 1);
1772 }
1773 /* read superblock from bitmap file (this sets mddev->bitmap_info.chunksize) */
1774 if (!mddev->bitmap_info.external) {
1775 /*
1776 * If 'MD_ARRAY_FIRST_USE' is set, then device-mapper is
1777 * instructing us to create a new on-disk bitmap instance.
1778 */
1779 if (test_and_clear_bit(MD_ARRAY_FIRST_USE, &mddev->flags))
1780 err = bitmap_new_disk_sb(bitmap);
1781 else
1782 err = bitmap_read_sb(bitmap);
1783 } else {
1784 err = 0;
1785 if (mddev->bitmap_info.chunksize == 0 ||
1786 mddev->bitmap_info.daemon_sleep == 0)
1787 /* chunksize and time_base need to be
1788 * set first. */
1789 err = -EINVAL;
1790 }
1791 if (err)
1792 goto error;
1793
1794 bitmap->daemon_lastrun = jiffies;
1795 err = bitmap_resize(bitmap, blocks, mddev->bitmap_info.chunksize, 1);
1796 if (err)
1797 goto error;
1798
1799 printk(KERN_INFO "created bitmap (%lu pages) for device %s\n",
1800 bitmap->counts.pages, bmname(bitmap));
1801
1802 err = test_bit(BITMAP_WRITE_ERROR, &bitmap->flags) ? -EIO : 0;
1803 if (err)
1804 goto error;
1805
1806 return bitmap;
1807 error:
1808 bitmap_free(bitmap);
1809 return ERR_PTR(err);
1810 }
1811
1812 int bitmap_load(struct mddev *mddev)
1813 {
1814 int err = 0;
1815 sector_t start = 0;
1816 sector_t sector = 0;
1817 struct bitmap *bitmap = mddev->bitmap;
1818
1819 if (!bitmap)
1820 goto out;
1821
1822 /* Clear out old bitmap info first: Either there is none, or we
1823 * are resuming after someone else has possibly changed things,
1824 * so we should forget old cached info.
1825 * All chunks should be clean, but some might need_sync.
1826 */
1827 while (sector < mddev->resync_max_sectors) {
1828 sector_t blocks;
1829 bitmap_start_sync(bitmap, sector, &blocks, 0);
1830 sector += blocks;
1831 }
1832 bitmap_close_sync(bitmap);
1833
1834 if (mddev->degraded == 0
1835 || bitmap->events_cleared == mddev->events)
1836 /* no need to keep dirty bits to optimise a
1837 * re-add of a missing device */
1838 start = mddev->recovery_cp;
1839
1840 mutex_lock(&mddev->bitmap_info.mutex);
1841 err = bitmap_init_from_disk(bitmap, start);
1842 mutex_unlock(&mddev->bitmap_info.mutex);
1843
1844 if (err)
1845 goto out;
1846 clear_bit(BITMAP_STALE, &bitmap->flags);
1847
1848 /* Kick recovery in case any bits were set */
1849 set_bit(MD_RECOVERY_NEEDED, &bitmap->mddev->recovery);
1850
1851 mddev->thread->timeout = mddev->bitmap_info.daemon_sleep;
1852 md_wakeup_thread(mddev->thread);
1853
1854 bitmap_update_sb(bitmap);
1855
1856 if (test_bit(BITMAP_WRITE_ERROR, &bitmap->flags))
1857 err = -EIO;
1858 out:
1859 return err;
1860 }
1861 EXPORT_SYMBOL_GPL(bitmap_load);
1862
1863 /* Loads the bitmap associated with slot and copies the resync information
1864 * to our bitmap
1865 */
1866 int bitmap_copy_from_slot(struct mddev *mddev, int slot,
1867 sector_t *low, sector_t *high, bool clear_bits)
1868 {
1869 int rv = 0, i, j;
1870 sector_t block, lo = 0, hi = 0;
1871 struct bitmap_counts *counts;
1872 struct bitmap *bitmap = bitmap_create(mddev, slot);
1873
1874 if (IS_ERR(bitmap))
1875 return PTR_ERR(bitmap);
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 (block = 0; block < blocks; ) {
2039 bitmap_counter_t *bmc_old, *bmc_new;
2040 int set;
2041
2042 bmc_old = bitmap_get_counter(&old_counts, block,
2043 &old_blocks, 0);
2044 set = bmc_old && NEEDED(*bmc_old);
2045
2046 if (set) {
2047 bmc_new = bitmap_get_counter(&bitmap->counts, block,
2048 &new_blocks, 1);
2049 if (*bmc_new == 0) {
2050 /* need to set on-disk bits too. */
2051 sector_t end = block + new_blocks;
2052 sector_t start = block >> chunkshift;
2053 start <<= chunkshift;
2054 while (start < end) {
2055 bitmap_file_set_bit(bitmap, block);
2056 start += 1 << chunkshift;
2057 }
2058 *bmc_new = 2;
2059 bitmap_count_page(&bitmap->counts,
2060 block, 1);
2061 bitmap_set_pending(&bitmap->counts,
2062 block);
2063 }
2064 *bmc_new |= NEEDED_MASK;
2065 if (new_blocks < old_blocks)
2066 old_blocks = new_blocks;
2067 }
2068 block += old_blocks;
2069 }
2070
2071 if (!init) {
2072 int i;
2073 while (block < (chunks << chunkshift)) {
2074 bitmap_counter_t *bmc;
2075 bmc = bitmap_get_counter(&bitmap->counts, block,
2076 &new_blocks, 1);
2077 if (bmc) {
2078 /* new space. It needs to be resynced, so
2079 * we set NEEDED_MASK.
2080 */
2081 if (*bmc == 0) {
2082 *bmc = NEEDED_MASK | 2;
2083 bitmap_count_page(&bitmap->counts,
2084 block, 1);
2085 bitmap_set_pending(&bitmap->counts,
2086 block);
2087 }
2088 }
2089 block += new_blocks;
2090 }
2091 for (i = 0; i < bitmap->storage.file_pages; i++)
2092 set_page_attr(bitmap, i, BITMAP_PAGE_DIRTY);
2093 }
2094 spin_unlock_irq(&bitmap->counts.lock);
2095
2096 if (!init) {
2097 bitmap_unplug(bitmap);
2098 bitmap->mddev->pers->quiesce(bitmap->mddev, 0);
2099 }
2100 ret = 0;
2101 err:
2102 return ret;
2103 }
2104 EXPORT_SYMBOL_GPL(bitmap_resize);
2105
2106 static ssize_t
2107 location_show(struct mddev *mddev, char *page)
2108 {
2109 ssize_t len;
2110 if (mddev->bitmap_info.file)
2111 len = sprintf(page, "file");
2112 else if (mddev->bitmap_info.offset)
2113 len = sprintf(page, "%+lld", (long long)mddev->bitmap_info.offset);
2114 else
2115 len = sprintf(page, "none");
2116 len += sprintf(page+len, "\n");
2117 return len;
2118 }
2119
2120 static ssize_t
2121 location_store(struct mddev *mddev, const char *buf, size_t len)
2122 {
2123
2124 if (mddev->pers) {
2125 if (!mddev->pers->quiesce)
2126 return -EBUSY;
2127 if (mddev->recovery || mddev->sync_thread)
2128 return -EBUSY;
2129 }
2130
2131 if (mddev->bitmap || mddev->bitmap_info.file ||
2132 mddev->bitmap_info.offset) {
2133 /* bitmap already configured. Only option is to clear it */
2134 if (strncmp(buf, "none", 4) != 0)
2135 return -EBUSY;
2136 if (mddev->pers) {
2137 mddev->pers->quiesce(mddev, 1);
2138 bitmap_destroy(mddev);
2139 mddev->pers->quiesce(mddev, 0);
2140 }
2141 mddev->bitmap_info.offset = 0;
2142 if (mddev->bitmap_info.file) {
2143 struct file *f = mddev->bitmap_info.file;
2144 mddev->bitmap_info.file = NULL;
2145 fput(f);
2146 }
2147 } else {
2148 /* No bitmap, OK to set a location */
2149 long long offset;
2150 if (strncmp(buf, "none", 4) == 0)
2151 /* nothing to be done */;
2152 else if (strncmp(buf, "file:", 5) == 0) {
2153 /* Not supported yet */
2154 return -EINVAL;
2155 } else {
2156 int rv;
2157 if (buf[0] == '+')
2158 rv = kstrtoll(buf+1, 10, &offset);
2159 else
2160 rv = kstrtoll(buf, 10, &offset);
2161 if (rv)
2162 return rv;
2163 if (offset == 0)
2164 return -EINVAL;
2165 if (mddev->bitmap_info.external == 0 &&
2166 mddev->major_version == 0 &&
2167 offset != mddev->bitmap_info.default_offset)
2168 return -EINVAL;
2169 mddev->bitmap_info.offset = offset;
2170 if (mddev->pers) {
2171 struct bitmap *bitmap;
2172 mddev->pers->quiesce(mddev, 1);
2173 bitmap = bitmap_create(mddev, -1);
2174 if (IS_ERR(bitmap))
2175 rv = PTR_ERR(bitmap);
2176 else {
2177 mddev->bitmap = bitmap;
2178 rv = bitmap_load(mddev);
2179 if (rv) {
2180 bitmap_destroy(mddev);
2181 mddev->bitmap_info.offset = 0;
2182 }
2183 }
2184 mddev->pers->quiesce(mddev, 0);
2185 if (rv)
2186 return rv;
2187 }
2188 }
2189 }
2190 if (!mddev->external) {
2191 /* Ensure new bitmap info is stored in
2192 * metadata promptly.
2193 */
2194 set_bit(MD_CHANGE_DEVS, &mddev->flags);
2195 md_wakeup_thread(mddev->thread);
2196 }
2197 return len;
2198 }
2199
2200 static struct md_sysfs_entry bitmap_location =
2201 __ATTR(location, S_IRUGO|S_IWUSR, location_show, location_store);
2202
2203 /* 'bitmap/space' is the space available at 'location' for the
2204 * bitmap. This allows the kernel to know when it is safe to
2205 * resize the bitmap to match a resized array.
2206 */
2207 static ssize_t
2208 space_show(struct mddev *mddev, char *page)
2209 {
2210 return sprintf(page, "%lu\n", mddev->bitmap_info.space);
2211 }
2212
2213 static ssize_t
2214 space_store(struct mddev *mddev, const char *buf, size_t len)
2215 {
2216 unsigned long sectors;
2217 int rv;
2218
2219 rv = kstrtoul(buf, 10, &sectors);
2220 if (rv)
2221 return rv;
2222
2223 if (sectors == 0)
2224 return -EINVAL;
2225
2226 if (mddev->bitmap &&
2227 sectors < (mddev->bitmap->storage.bytes + 511) >> 9)
2228 return -EFBIG; /* Bitmap is too big for this small space */
2229
2230 /* could make sure it isn't too big, but that isn't really
2231 * needed - user-space should be careful.
2232 */
2233 mddev->bitmap_info.space = sectors;
2234 return len;
2235 }
2236
2237 static struct md_sysfs_entry bitmap_space =
2238 __ATTR(space, S_IRUGO|S_IWUSR, space_show, space_store);
2239
2240 static ssize_t
2241 timeout_show(struct mddev *mddev, char *page)
2242 {
2243 ssize_t len;
2244 unsigned long secs = mddev->bitmap_info.daemon_sleep / HZ;
2245 unsigned long jifs = mddev->bitmap_info.daemon_sleep % HZ;
2246
2247 len = sprintf(page, "%lu", secs);
2248 if (jifs)
2249 len += sprintf(page+len, ".%03u", jiffies_to_msecs(jifs));
2250 len += sprintf(page+len, "\n");
2251 return len;
2252 }
2253
2254 static ssize_t
2255 timeout_store(struct mddev *mddev, const char *buf, size_t len)
2256 {
2257 /* timeout can be set at any time */
2258 unsigned long timeout;
2259 int rv = strict_strtoul_scaled(buf, &timeout, 4);
2260 if (rv)
2261 return rv;
2262
2263 /* just to make sure we don't overflow... */
2264 if (timeout >= LONG_MAX / HZ)
2265 return -EINVAL;
2266
2267 timeout = timeout * HZ / 10000;
2268
2269 if (timeout >= MAX_SCHEDULE_TIMEOUT)
2270 timeout = MAX_SCHEDULE_TIMEOUT-1;
2271 if (timeout < 1)
2272 timeout = 1;
2273 mddev->bitmap_info.daemon_sleep = timeout;
2274 if (mddev->thread) {
2275 /* if thread->timeout is MAX_SCHEDULE_TIMEOUT, then
2276 * the bitmap is all clean and we don't need to
2277 * adjust the timeout right now
2278 */
2279 if (mddev->thread->timeout < MAX_SCHEDULE_TIMEOUT) {
2280 mddev->thread->timeout = timeout;
2281 md_wakeup_thread(mddev->thread);
2282 }
2283 }
2284 return len;
2285 }
2286
2287 static struct md_sysfs_entry bitmap_timeout =
2288 __ATTR(time_base, S_IRUGO|S_IWUSR, timeout_show, timeout_store);
2289
2290 static ssize_t
2291 backlog_show(struct mddev *mddev, char *page)
2292 {
2293 return sprintf(page, "%lu\n", mddev->bitmap_info.max_write_behind);
2294 }
2295
2296 static ssize_t
2297 backlog_store(struct mddev *mddev, const char *buf, size_t len)
2298 {
2299 unsigned long backlog;
2300 int rv = kstrtoul(buf, 10, &backlog);
2301 if (rv)
2302 return rv;
2303 if (backlog > COUNTER_MAX)
2304 return -EINVAL;
2305 mddev->bitmap_info.max_write_behind = backlog;
2306 return len;
2307 }
2308
2309 static struct md_sysfs_entry bitmap_backlog =
2310 __ATTR(backlog, S_IRUGO|S_IWUSR, backlog_show, backlog_store);
2311
2312 static ssize_t
2313 chunksize_show(struct mddev *mddev, char *page)
2314 {
2315 return sprintf(page, "%lu\n", mddev->bitmap_info.chunksize);
2316 }
2317
2318 static ssize_t
2319 chunksize_store(struct mddev *mddev, const char *buf, size_t len)
2320 {
2321 /* Can only be changed when no bitmap is active */
2322 int rv;
2323 unsigned long csize;
2324 if (mddev->bitmap)
2325 return -EBUSY;
2326 rv = kstrtoul(buf, 10, &csize);
2327 if (rv)
2328 return rv;
2329 if (csize < 512 ||
2330 !is_power_of_2(csize))
2331 return -EINVAL;
2332 mddev->bitmap_info.chunksize = csize;
2333 return len;
2334 }
2335
2336 static struct md_sysfs_entry bitmap_chunksize =
2337 __ATTR(chunksize, S_IRUGO|S_IWUSR, chunksize_show, chunksize_store);
2338
2339 static ssize_t metadata_show(struct mddev *mddev, char *page)
2340 {
2341 if (mddev_is_clustered(mddev))
2342 return sprintf(page, "clustered\n");
2343 return sprintf(page, "%s\n", (mddev->bitmap_info.external
2344 ? "external" : "internal"));
2345 }
2346
2347 static ssize_t metadata_store(struct mddev *mddev, const char *buf, size_t len)
2348 {
2349 if (mddev->bitmap ||
2350 mddev->bitmap_info.file ||
2351 mddev->bitmap_info.offset)
2352 return -EBUSY;
2353 if (strncmp(buf, "external", 8) == 0)
2354 mddev->bitmap_info.external = 1;
2355 else if ((strncmp(buf, "internal", 8) == 0) ||
2356 (strncmp(buf, "clustered", 9) == 0))
2357 mddev->bitmap_info.external = 0;
2358 else
2359 return -EINVAL;
2360 return len;
2361 }
2362
2363 static struct md_sysfs_entry bitmap_metadata =
2364 __ATTR(metadata, S_IRUGO|S_IWUSR, metadata_show, metadata_store);
2365
2366 static ssize_t can_clear_show(struct mddev *mddev, char *page)
2367 {
2368 int len;
2369 spin_lock(&mddev->lock);
2370 if (mddev->bitmap)
2371 len = sprintf(page, "%s\n", (mddev->bitmap->need_sync ?
2372 "false" : "true"));
2373 else
2374 len = sprintf(page, "\n");
2375 spin_unlock(&mddev->lock);
2376 return len;
2377 }
2378
2379 static ssize_t can_clear_store(struct mddev *mddev, const char *buf, size_t len)
2380 {
2381 if (mddev->bitmap == NULL)
2382 return -ENOENT;
2383 if (strncmp(buf, "false", 5) == 0)
2384 mddev->bitmap->need_sync = 1;
2385 else if (strncmp(buf, "true", 4) == 0) {
2386 if (mddev->degraded)
2387 return -EBUSY;
2388 mddev->bitmap->need_sync = 0;
2389 } else
2390 return -EINVAL;
2391 return len;
2392 }
2393
2394 static struct md_sysfs_entry bitmap_can_clear =
2395 __ATTR(can_clear, S_IRUGO|S_IWUSR, can_clear_show, can_clear_store);
2396
2397 static ssize_t
2398 behind_writes_used_show(struct mddev *mddev, char *page)
2399 {
2400 ssize_t ret;
2401 spin_lock(&mddev->lock);
2402 if (mddev->bitmap == NULL)
2403 ret = sprintf(page, "0\n");
2404 else
2405 ret = sprintf(page, "%lu\n",
2406 mddev->bitmap->behind_writes_used);
2407 spin_unlock(&mddev->lock);
2408 return ret;
2409 }
2410
2411 static ssize_t
2412 behind_writes_used_reset(struct mddev *mddev, const char *buf, size_t len)
2413 {
2414 if (mddev->bitmap)
2415 mddev->bitmap->behind_writes_used = 0;
2416 return len;
2417 }
2418
2419 static struct md_sysfs_entry max_backlog_used =
2420 __ATTR(max_backlog_used, S_IRUGO | S_IWUSR,
2421 behind_writes_used_show, behind_writes_used_reset);
2422
2423 static struct attribute *md_bitmap_attrs[] = {
2424 &bitmap_location.attr,
2425 &bitmap_space.attr,
2426 &bitmap_timeout.attr,
2427 &bitmap_backlog.attr,
2428 &bitmap_chunksize.attr,
2429 &bitmap_metadata.attr,
2430 &bitmap_can_clear.attr,
2431 &max_backlog_used.attr,
2432 NULL
2433 };
2434 struct attribute_group md_bitmap_group = {
2435 .name = "bitmap",
2436 .attrs = md_bitmap_attrs,
2437 };
2438
This page took 0.164059 seconds and 5 git commands to generate.